presentation of diabetic macular edema

Create account

Diabetic Macular Edema

All content on Eyewiki is protected by copyright law and the Terms of Service . This content may not be reproduced, copied, or put into any artificial intelligence program, including large language and generative AI models, without permission from the Academy.


:

1.1 Pathophysiology
1.2 Natural History
1.3 Prevalence
1.4 Risk Factors
1.5 Prevention
2.1 History
2.2 Physical Examination
2.4.1 Vision and DME
2.5.1.1 OCT Biomarkers for DME
2.5.2 2. Fluorescein Angiography
2.6 Laboratory Testing
2.7 Differential Diagnosis
3.1 Medical therapy
3.2.1 DRCR Retina Network anti-VEGF treatment algorithm
3.2.2 Clinical considerations in the anti-VEGF treatment of DME
3.2.3 Steroids
3.3.1 Anti-VEGF
3.3.2 Steroids
3.4 Laser Photocoagulation
3.5 Combined Therapy
3.6 Surgery
3.7.1 Intravitreal injections
3.7.2 Laser Photocoagulation
3.8 Prognosis
3.9 Future Directions
4 Further reading
5 References

Disease Definition

Diabetic macular edema (DME) is the accumulation of excess fluid in the extracellular space within the retina in the macular area, typically in the inner nuclear, outer plexiform, Henle’s fiber layer, and subretinal space. [1] [2]

Pathophysiology

Chronic hyperglycemia-related accumulation of advanced glycated end products (AGEs) disrupts the blood retinal barrier (BRB) characterized by endothelial cell junction breakdown and pericyte loss. The inner BRB is composed of endothelial cells in the retinal capillaries, while the outer BRB is composed of retinal pigment epithelium (RPE) cells. Altered BRB leads to interstitial fluid accumulation within and underneath the retina through leakage of molecules dependent on intact cell to cell junctions ( Figure 1 ). [3] Evidence also shows that DME has an inflammatory component to the disease, with several chemokines and cytokines involved in its development. These factors include vascular endothelial growth factor (VEGF), interleukins (ILs), matrix metalloproteinases (MMPs), and tumor necrosis factor (TNF). Upregulation of multiple pathways leads to increased inflammation, oxidative stress, and vascular dysfunction. [4] There are also significant changes in the neurovascular unit, altering the homeostasis between astrocytes, ganglion cells, Müller cells, retinal vascular endothelial cells, and amacrine cells. [5] Retinal vascular permeability changes also involve the kallikrein-kinin system, which induces vasorelaxation via bradykinin and nitric oxide. [6] [7] [8]

Natural History

DME can develop at any stage of diabetic retinopathy (DR), from mild nonproliferative diabetic retinopathy (NPDR) to proliferative diabetic retinopathy (PDR), but is more frequent as the severity of DR increases. DME threatening or at the fovea is more likely to result in blurred vision and metamorphopsia. When the DME involves or threatens the fovea, the risk of moderate visual loss (MVL, defined as a three-line or more decrease of visual acuity, equivalent to a doubling of the visual angle) over 3 years in the Early Treatment of Diabetic Retinopathy Study (ETDRS) was 24% without treatment. [9] The disease course is variable, with some eyes having chronic persistent DME spanning several years, while other eyes have rapid spontaneous resolution, although the risk of recurrence is always present.

The Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR) found that DME incidence over 25 years among people with type 1 DM (T1DM) was 29%. [10] The Diabetes Control and Complications Trial (DCCT) reported that 27% of people with T1DM had DME within 9 years of onset of diabetes. [11] For people with type 2 DM (T2DM), the WESDR found that 25.4% of those who used insulin and 13.9% of those who did not use insulin had DME. [10] Yau et al. estimated the global prevalence of DME at 6.8% among people with DM. Estimates in the United States are between 2.7% to 3.8%, with non-Hispanic whites less likely to have DME versus non-Hispanic blacks. [12] [13] [14]

Risk Factors

Risk factors for DME and DR are similar. These risks include a longer duration of diabetes mellitus (DM), poor control of DM with elevated hemoglobin A1c (HbA1c), hypertension, and hyperlipidemia. Other secondary risk factors include impaired renal function and the use of thiazolidinediones. [15] [16] [17]

Primary prevention of DME involves intensive control of DM, blood glucose, hypertension, blood lipids, and other systemic risk factors. The Diabetic Retinopathy Clinical Research (DRCR) Protocol W investigated whether aflibercept injections in eyes with baseline moderate to severe NPDR could prevent the eventual development of center-involved DME (ci-DME) with vision loss. [18] Vision loss was defined as a 10 letter or more decrease in visual acuity (VA) at 1 visit, or a 5 to 9 letter decrease at 2 consecutive visits, with the decrease in vision attributed to the ci-DME. The 2-year cumulative probability of developing ci-DME with vision loss was 14.8% in the sham group versus 4.1% in the aflibercept group. However, there was no significant difference in mean change of visual acuity (VA) between both groups. From baseline to the 2-year checkup, eyes treated with sham had a mean (SD) change of -2.0 (6.1) letters, while aflibercept treated eyes had a -0.9 (5.8) letter change (adjusted mean difference 0.5 letters, [97.5% CI: -1.0 to 1.9 letters, p=0.47]. At present, the use of anti-VEGF injections for the prevention of ci-DME is still not the standard of care.

DME is suspected in patients with any level of DR who present with blurred vision or metamorphopsias. A detailed history including the approximate date of onset of diabetes, the use of insulin versus oral antihyperglycemic agents, and the quality of metabolic control (e.g., HbA1c level) should be elicited. Any associated medical problems such as hypertension, hypercholesterolemia, renal disease, and thyroid disease should be identified, along with a thorough review of medications. It should be noted that mild to extensive DME may be present without symptoms evident to the patient.

Physical Examination

Patients undergo a detailed biomicroscopic examination using the slit lamp biomicroscope and indirect ophthalmoscope. Historically, DME classifications were based on the ETDRS definitions of clinically significant macular edema (CSME). The specific criteria for diagnosing CSME were:

Retinal thickening at or within 500 μm of the center of the fovea
Hard exudates at or within 500 μm of the center of the fovea if adjacent to an area of retinal thickening
Retinal thickening of at least 1 disc area any portion of which is within 1500 μm (approximately 1 disc diameter) from the center of the fovea

Thus, CSME as defined by the ETDRS in the 1980s is a clinical diagnosis made by slit-lamp examination using a contact lens. While clinical examinations remain essential for the full evaluation of DME, Optical Coherence Tomography (OCT) is now routinely used to complement physical examination in the diagnosis of DME. The management of DME is currently based on the central subfoveal thickness (CST) on the macular OCT.

Macular thickening with or without hard exudates may be seen with stereo biomicroscopy. However, some eyes may present without apparent signs of retinal thickening on clinical examination despite significant DME as observed using OCT. Thickening can occur in various patterns: focal, multifocal, and diffuse areas of retina thickening. Despite these terms being frequently used, there are no well-established standard definitions, and different authors use them inconsistently. [19] In the ETDRS, focal leakage results from microaneurysms that may be treated with fluorescein angiography (FA) guided focal laser, while diffuse capillary leakage is from a more widespread breakdown of the BRB, which may be treated with grid laser. [20] Hard exudates in various patterns may also be seen, including circinate rings and focal aggregations of exudates ( Figures 2-3 ). Hard exudates consist of lipoprotein residues of serous leakage from damaged vessels, serving as biomarkers for DME.

DME may present with decreased visual acuity (VA), metamorphopsia, changes in color perception, and difficulty reading, although it may also present asymptomatically.

Vision and DME

Studies show a poor to modest correlation between visual acuity and central subfoveal thickness (CST) on OCT, within the range of 0.3-0.5. [21] [22] Over two years, only around 12-14% of the change in VA can be attributed to the change in OCT thickness for eyes with DME. In a study by the DRCR Network, the slope of the best fit line shows an approximately 4.4 letter improvement (95% CI: 3.5, 5.3) for every 100-micron decrease in CST at baseline. [22] Interestingly, on follow-up after treatment with macular laser photocoagulation, there were some eyes with paradoxical improvement in VA with increased CST (7-17% at different time points) and paradoxical worsening of VA with decreased CST (18-26% at different time points). These findings highlight how OCT measurements cannot be used as a perfect surrogate for visual acuity; however, in clinical and research settings, the technology remains an important tool. The DRCR has recommended a 10% change in CST to indicate a real change that can be considered in clinical decision-making.

Diagnostic Procedures

1. optical coherence tomography (oct).

OCT within recent years has quickly become an important ancillary procedure in the diagnosis and treatment of DME. Three basic structural changes can be seen: retinal swelling, cystoid macular edema, and subretinal fluid ( Figure 4 ). Macular scans can quickly and accurately identify even subtle areas of thickening, along with quantitative metrics for different areas. Changes in the anatomic distribution of DME can be monitored over time, especially the fluid’s relationship to the fovea. This information has proven crucial regarding clinical and research implications for the evaluation and management of DME. More recently, the International Council of Ophthalmology (ICO) guidelines for diabetic eye care in 2018 have adopted the clinical entity of center-involved DME (ci-DME) versus non-center involved DME (non-ciDME) for the evaluation of macular fluid ( Figure 5 ). The classification is:

Center-involved DME: Retinal thickening in the macula that involves the central subfield zone (1 mm in diameter)
Non-center involved DME: Retinal thickening in the macula that does not involve the central subfield zone (1 mm in diameter)

Aside from the location of the swelling, the DRCR retina network has given recommendations on CST treatment thresholds based on sex-matched standards. [23] The thresholds were different per OCT machine since thickness measurements cannot be compared between different devices, with each device having its own normative database and algorithms. In DRCR Protocol T, in conjunction with visual acuity, treatment eligibility thresholds per machine were:

Heidelberg Spectralis - 320 μm for men or 305 μm for women
Zeiss Cirrus OCT - 305 μm for men or 290 μm for women
Zeiss Stratus OCT - 250 μm for both men and women

Moreover, OCT is a more sensitive method for objective evaluation of vitreomacular interface abnormalities (VMIA), which include vitreomacular adhesion (VMA), vitreomacular traction (VMT), and epiretinal membrane (ERM). Identifying VMIA is crucial when diagnosing the etiology of macular edema, whether it is primarily from DME, from secondary causes of VMIA, or combined mechanism macular edema.

OCT Biomarkers for DME

DME's different OCT features have been associated with disparate prognostic outcomes and treatment responses.

Disorganization of the retinal inner layers (DRIL) is thought to represent damaged cells within the inner retinal layers, indicating a disruption in the normal visual pathway from the photoreceptors to the ganglion cells. [24] DRIL is identified on OCT when there is disruption of the demarcating interface lines between the ganglion cell-inner plexiform complex (GCL-IPL), inner nuclear layer (INL), outer plexiform layer (OPL), and outer nuclear layer (ONL). Eyes with DRIL have a worse baseline and final VA despite anti-Vascular Endothelial Growth Factor (VEGF) injections and have almost eight times greater risk for poor visual recovery. In contrast, the resolution of DRIL has resulted in better visual outcomes ( Figure 6 ). [25] [26] [27] DRIL was better correlated to VA when compared to glycemic status and CST. [24]
External limiting membrane (ELM) and ellipsoid zone (EZ) disruption strongly correlate with baseline and final BCVA. At the same time, restoration of the EZ was a requirement for good visual recovery in some studies ( Figure 7 ). [28] [29] [30] [31] [32] The ELM connects a row of zonular adherents to the photoreceptor cell bodies, while the EZ represents the isthmus between the outer and inner segments of photoreceptors. [33]

2. Fluorescein Angiography

Fluorescein angiography (FA) is performed to identify leaking microaneurysms or capillaries to guide laser treatment, and areas of retinal ischemia. Leakage on the angiogram is not synonymous with retinal edema. Focal DME is characterized by focal leakage from microaneurysms or capillaries. In contrast, diffuse DME is diagnosed when poorly demarcated areas of capillary leakage are present ( Figure 9 ). Recently, there has been a decreasing trend in the use of FA in the management of DME, likely due to the procedure being more invasive and time-consuming compared to OCT. [45] Contraindications to the use of FA include pregnancy and allergy associated with the contrast dye.

Laboratory Testing

Primary: HbA1c (Glycated Hemoglobin), Blood pressure, Lipid Profile. Secondary: Hemoglobin (anemia exacerbates diabetic retinopathy and may be associated with diabetic nephropathy), Fasting Blood Sugar (FBS), Post Prandial Blood Sugar (PPBS), Urea, Creatinine, Urine microalbumin levels, Thyroid panel.

Differential Diagnosis

Other causes of macular edema include retinal vein occlusion, ruptured macroaneurysm, Irvine-Gass syndrome, radiation retinopathy, hypertensive retinopathy, subfoveal choroidal neovascularization, retinal vein occlusion, and VMIA. OCT and FA ancillary diagnostics can help differentiate between these differential diagnoses ( Figures 10-11 ).

Medical therapy

Strict control of diabetes, blood glucose, hypertension, and hypercholesterolemia [46]
Diet Modification
Weight Loss

Pharmacotherapy

At present, anti-VEGF agents are the first-line treatment for DME requiring treatment. Since 2005, intravitreal bevacizumab has been used off-label for ocular conditions. FDA approved ranibizumab for DME in 2012, Aflibercept in 2014 and Brolucizumab and Faricimab in 2022.

DRCR Retina Network anti-VEGF treatment algorithm

Six monthly injections are given unless VA is 20/20 or better, and CST is <320μm for men or <305μm for women on the Heidelberg Spectralis, in which case treatment may be withheld starting the 4 th month. At the 6 th monthly visit, if stability is achieved in vision or CST from the previous 2 or more visits, or DME has resolved, further treatment may be withheld. Stability is defined as:

No BCVA increase of ≥5 letters (approximately 1 line on a Snellen chart)
No decrease in OCT CST ≥10%
BCVA decreases ≥5 letters (approximately 1 line on a Snellen chart) in the setting of persistent DME
OCT CST increases ≥10%

If there is worsening on follow-up, anti-VEGF injections are resumed. In general, eyes showing stability may be followed-up in 3-4 months, and sooner as needed. Withholding injections does not require 20/20 vision or a dry macula, only evidence of stability over the previous 2 or more visits. DRCR Protocol I published the first anti-VEGF PRN treatment algorithm for DME. Long-term data show that even without using a monthly treatment protocol, eyes with persistent DME often maintain good BCVA over the long term in clinical trial settings. [47] Outside of trial settings, the five-year extension study of Protocol I showed that CST remained stable among eyes given anti-VEGF; however, mean BCVA worsened between the 5 th and 2nd-year time points. [48] On average, patients are given 6-8 injections in the first year, 2-4 injections in the second year, and 0-1 injections beginning the third year. [49]

Clinical considerations in the anti-VEGF treatment of DME

Observation is recommended for eyes with ci-DME and visual acuity of 20/25 or better. This recommendation is based on the findings of DRCR Protocol V, where patients were randomized to receive either aflibercept injections, macular laser, or observation. At the 2-year endpoint, the mean BCVA was 20/20, and the rates of ≥ 5 letter vision loss (16-19%) were similar among all three groups. Given the risks of ocular infection and the costs associated with intravitreal injections, observation is a viable treatment option in these patients if they can reliably follow up and receive appropriate therapy when there is clinical deterioration. In contrast, among eyes with ci-DME presenting with 20/50 vision or worse, Protocol T results showed that aflibercept was superior to both ranibizumab and bevacizumab, with area under the curve analysis showing better visual outcomes over 1 year. [50] However, the superiority of aflibercept over ranibizumab was not maintained at the 2-year checkup. [23] Bevacizumab thinned the retina the least; however, all 3 medications had similar visual outcomes among eyes with baseline vision of 20/40 or better.

Not all treatment results in complete resolution of DME, with 40% of eyes still showing persistent DME (defined as never having a CST < 250μm through 6 months on time-domain OCT) in Protocol I over 2 years. [47] Eyes given bevacizumab were more likely to have persistent DME than eyes given aflibercept. The percentage of eyes with complete resolution increases with each ranibizumab or aflibercept injection. [51] At the 3-year follow-up, eyes with chronic persistent DME still had a significant 7 letter mean improvement compared to baseline, lower than the 13 letter mean improvement in eyes with complete DME resolution. [52] Importantly, only 3.4% of eyes with chronic persistent DME lost ≥ 2 lines of vision over 2 years regardless of the type of anti-VEGF they received. [53] To summarize, in clinical trial settings, a substantial proportion of eyes with persistent DME still have ≥ 2 lines of improvement in vision over the long term, with very few eyes developing considerable vision loss.

For eyes that do not respond optimally, switching to another anti-VEGF drug is a treatment option. The switch is typically from bevacizumab to either ranibizumab or aflibercept, or ranibizumab to aflibercept. The rationale for switching is that aflibercept has 100 times more affinity to VEGF-A than bevacizumab or ranibizumab, while concurrently binding placental growth factor (PIGF) and VEGF-B. [54] Aflibercept also has a longer half-life and can negate more cytokines that promote DME development. Most studies looking at switching are retrospective and show a significant reduction in CST; however, visual outcomes vary. [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] Some eyes may be delayed responders to anti-VEGF. These eyes need more than the initial 3-6 monthly injections to show significant improvement but eventually catch up with immediate responders. [69] [70] It may be viable to continue treatment with the same medication if there is at least a trend of continued improvement, as bevacizumab and ranibizumab are more cost-effective options. Caution is warranted for switching, as any clinical improvement may be due to continued treatment among delayed responders instead of the shift in medications.

DRCR protocol AC was a randomized controlled trial evaluating the relative efficacy of administering aflibercept monotherapy compared with bevacizumab first with a switch to aflibercept (step therapy) in eyes with suboptimal response despite treatment. A total of 312 eyes with moderate vision loss from ci-DME were enrolled. Over a 2-year period, the mean improvement in VA was 14.0 letters in the bevacizumab-first group and 15.0 letters in the aflibercept-monotherapy group (adjusted difference, 0.8 letters; 95% confidence interval, −0.9 to 2.5; P=0.37). At 2 years, the mean changes in VA and retinal CST were similar in the two groups. [71]

Intravitreal steroids improve vision and decrease retinal thickness, as there is an inflammatory component to DME. Steroids have powerful anti-edematous and anti-inflammatory effects as they decrease several pro-inflammatory mediators (IL-6, IL-8, TNF-α, MCP-1, ICAM-1, VEGF, etc). [72] However, long-term results have not shown maintenance of initial clinical improvement. In DRCR Protocol I, although the group given intravitreal triamcinolone had a similar response with ranibizumab in the first 6 months, vision declined due to cataracts. Even with cataract surgery, the final vision did not recover to the levels comparable to the group given ranibizumab. [47] In Protocol U eyes with persistent DME despite at least 6 ranibizumab injections were randomized to continue ranibizumab monotherapy, or shift to ranibizumab plus dexamethasone (Ozurdex). [73] At 6 months average visual gain was similar in both groups, although the improvement in CST was greater in the combination group. Phakic eyes given intravitreal steroids often develop cataracts needing surgery, and are at risk for intraocular pressure (IOP) elevations leading to glaucoma.

Pharmacotherapy Landmark Studies

Bevacizumab: Bevacizumab is a recombinant humanized monoclonal IgG1 antibody that binds VEGF. Bevacizumab is given off-label for the treatment of DME, and remains the most cost-effective treatment option among anti-VEGF medications. [74] In the BOLT study, intravitreal Bevacizumab (1.25 mg) at 6-week intervals was reported to be more effective than modified ETDRS focal/grid laser in terms of improvement in visual acuity at 12 months. [75] In DRCR Protocol T, a comparison between bevacizumab, ranibizumab, and aflibercept, 1-year results showed that bevacizumab thinned the retina the least. [23] However, all 3 medications had similar visual outcomes among eyes with baseline vision of 20/40 or better. [53] Intravitreal bevacizumab doses of 1.25 to 2.5mg have shown improvement in best-corrected visual acuity and reduced macular thickness on OCT at 24 months in The Pan-American Collaborative Retina Study Group. [76]
DRCR Protocol I was the first definitive phase III study demonstrating the effectiveness of ranibizumab for DME treatment. [47] Four treatment arms were compared: ranibizumab with immediate grid/focal laser, ranibizumab with macular laser given only for at least 6 months of persistent DME, intravitreal triamcinolone with immediate macular laser, and macular laser with sham injections. Results showed that ranibizumab in an as-needed treatment protocol was superior to laser therapy. Eyes treated with ranibizumab gained an average of 8-9 letters, versus an average of only 3 letters gained with laser therapy at the 1-year endpoint. Protocol I revolutionized the treatment protocol for DME when it was published in 2011. Before this study, macular laser was considered the 1 st line therapy for DME, a treatment protocol originating from the original ETDRS papers first published in 1985.
RISE and RIDE - The two-year results for ranibizumab in DME showed that 98% of patients maintained vision (lost less than 15 letters) with 0.3mg monthly injections, 34-45% of patients gained at least 15 letters, and mean visual acuity gain was 10.9 to 12.5 letters. [78] Only 45-49% of patients needed macular laser compared with 91-94% in the control group. No additional benefit was seen with 0.5mg monthly versus 0.3mg ranibizumab monthly. [79]
Aflibercept: Aflibercept is a soluble decoy receptor that binds VEGF-A, VEGF-B, and placental growth factor with high affinity. [79] It is termed a decoy receptor or a “VEGF-trap” as VEGF mistakenly binds with aflibercept instead of the body’s native receptors, reducing VEGF’s activity. It is FDA approved for the treatment of DME. In the Da Vinci study, DME patients were assigned randomly to 1 of 5 treatment regimens: Aflibercept 0.5mg every 4 weeks (0.5q4); 2mg every 4 weeks (2q4); 2mg every 8 weeks after 3 initial monthly doses (2q8); or 2mg dosing as needed after 3 initial monthly doses (2prn), or macular laser photocoagulation. [80] Mean improvements in BCVA in the aflibercept groups at week 52 were 11.0, 13.1, 9.7, and 12.0 letters for 0.5q4, 2q4, 2q8, and 2prn regimens, respectively, versus -1.3 letters for the laser group. The proportion of eyes with gains in BCVA of 15 or more ETDRS letters at week 52 in the aflibercept groups were 40.9%, 45.5%, 23.8%, and 42.2% versus 11.4% for laser. Mean reductions in CST in the aflibercept groups at week 52 were -165.4μm, -227.4μm, -187.8μm, and -180.3μm versus -58.4μm for laser. [79] The PHOTON trial evaluated the use of high-dose 8mg aflibercept every 12 (8q12) or 16 (8q16) weeks, compared with conventional dosing of 2mg every 8 weeks. [81] Results showed that the 8mg extended interval dosage was non-inferior to conventional treatment when comparing gains in BCVA and retinal thinning at 48 weeks. There was an increase of 9.2, 8.8, and 7.9 letters, and a mean CST reduction of -165μm, -172μm, and -148μm, for the 2q8, 8q12, and 8q16 groups at 48 weeks, respectively. Patients in the high-dose group with a 50μm increase in CST or a 10-letter loss from week 12 onwards had their dosing schedule shortened to either 12 or 8 weeks, but 93% of patients remained on dosing intervals of 12 weeks or more. There was no increase in hypertension in the high-dose group.
Faricimab: Faricimab is a novel combined mechanism inhibitor that binds both Angiopoietin-2 (Ang-2) and VEGF-A with high specificity and affinity. The 1-year results from the phase II BOULEVARD study show that in patients given the 6.0mg dose, there was a statistically significant gain of 3.6 letters over ranibizumab. [82] Moreover, faricimab showed improvement in DR severity, reduced CST, and had a longer time to retreatment than ranibizumab. There were no unexpected safety concerns noted. The phase III YOSEMITE and RHINE non-inferiority trials compared faricimab with aflibercept in patients with DME. [83] The non-inferiority primary endpoint of BCVA was achieved with faricimab given every 8 weeks. More than 50% of patients achieved dosing of every 16 weeks, and more than 70% achieved dosing of every 12 weeks or longer. These results demonstrate the potential of faricimab to decrease the treatment burden on patients by extending the intraocular durability of anti-VEGF. FDA approved the use of faricimab for DME as 4 monthly loading doses followed by injections every 1-4 months, or a loading dose of 6 monthly injections followed up with injections every 2 months.
Brolucizumab: Brolucizumab is a single-chain antibody fragment with a high affinity for VEGF. Brolucizumab’s low molecular weight of 26kDa allows more of the drug to be delivered intraocularly per injection, with the potential for increased durability and effective tissue penetration in the eye. The phase III KESTREL and KITE 52 week results showed that brolucizumab was non-inferior to aflibercept in the mean improvement of BCVA, more patients had CST <280μm without persistent macular fluid, and >50% of brolucizumab 6mg patients were maintained on q12 weekly dosage after loading. [84] However, reports of retinal vascular occlusion (RVO) and retinal vasculitis (RV) with intraocular inflammation (IOI) have been reported. [85] [86] Risk factors for developing these adverse reactions include a prior history of IOI or RVO in the 12 months before brolucizumab initiation, female sex, and same-day bilateral injections. [87] [88] FDA approved use of Brolucimab 6 mg for the treatment of DME in June 2022 for an injection every 8-12 weeks after a loading dose of 5 injections 6 weeks apart.
Triamcinolone (1 mg or 4 mg) preservative-free intravitreal injection was less effective and had more side-effects for most patients with DME than focal/grid photocoagulation at 2-years follow-up (Protocol B). [89]
Dexamethasone (Ozurdex) 0.7 mg biodegradable implant improved vision by at least 15 letters in 22% of patients at 3 years in the phase III MEAD study. [90] The FDA approved the 0.7 mg implant for DME, and at this dose, 41.5% of patients needed anti-glaucoma medications, with 0.6% needing glaucoma surgery. Around 60% of eyes in the 0.7 mg implant group had cataract surgery. The implant is effective for approximately 3-6 months intraocularly.
Fluocinolone acetonide (Iluvien) 0.19 mg non-biodegradable implant is a sustained release device effective for up to 3 years intraocularly. It is FDA approved to treat DME in patients who have been previously treated with a course of corticosteroids and did not have a clinically significant rise in IOP. [91] [92] At the 0.2 μg/day dosage BCVA improvement of at least 15 letters was found in 28.7% of patients, with 38.4% needing anti-glaucoma medications and 4.8% needing glaucoma surgery. The latter figure was reduced to 0% among eyes that did not have a history of IOP elevation with a prior steroid challenge. The implant can have complications like the dexamethasone implant, including cataracts and implant migration.

Laser Photocoagulation

Before developing anti-VEGF for DME, the standard treatment for CSME was macular laser photocoagulation since the ETDRS was published in 1985. In “focal” CSME, a focal laser pattern is used to treat leaking microaneurysms identified on the FA that contribute to the retinal edema ( Figure 12 ). In “diffuse” CSME, intraretinal leakage is noted on the FA from dilated retinal capillary beds or intraretinal microvascular abnormalities (IRMA) without isolated, discrete foci of leakage. Macular grid is done for diffuse macular edema ( Figure 13 ). Laser photocoagulation has been shown to decrease the risk of moderate visual loss (loss of 15 or more ETDRS letters) from 24% to 12% by 3 years. [9] After laser treatment, the follow-up examination is at three months. If residual CSME is noted, OCT and FA may be performed to evaluate the benefit and location of repeat laser treatment. With the FDA approval of anti-VEGF for DME, focal/grid laser is only indicated in patients with non-ciDME. Especially in resource-limited countries with decreased access to anti-VEGF agents, macular laser remains a viable treatment option for patients with DME.

Combined Therapy

Intravitreal ranibizumab with laser: Intravitreal ranibizumab with prompt (within 1 week) or deferred (after 24 weeks) laser is more effective compared to focal/grid laser alone for the treatment of ci-DME (Protocol I). [47] Ranibizumab is injected intravitreally at baseline with prompt laser, followed by monthly ranibizumab injections for 4 months followed by the continuation of injections at 16 weeks if the OCT central subfield thickness is ≥250 um with visual acuity worse than 20/20.
Steroid with laser: Intravitreal triamcinolone (IVT, 4 mg) with focal/grid laser within a week is more effective than laser alone at 4 months (Protocol B). However, long-term benefit with this combined therapy was not seen, with mean BCVA better in the laser monotherapy group than the combined therapy group from the 16-month timepoint until the 2-year endpoint. Complications of steroid therapy included cataracts, ocular hypertension, and glaucoma. The difference in BCVA could not be attributed fully to the development of cataracts. In DRCR Protocol I, visual acuity improvement in eyes given IVT with prompt laser were comparable to eyes given ranibizumab at 6 months, but vision declined afterward until the 2-year timepoint. [47] [93] Subgroup analysis in pseudophakic eyes given IVT showed visual improvement was significantly better than in phakic eyes, with results comparable to pseudophakic eyes given ranibizumab at the 1 and 2-year time points.
Intravitreal ranibizumab with peripheral targeted retinal photocoagulation (TRP): The DAVE study was a phase I/II clinical trial that evaluated if ranibizumab with TRP could reduce the number of required anti-VEGF injections compared to ranibizumab monotherapy. [94] Peripheral TRP was defined as retinal photocoagulation administered outside the macula to areas of retinal capillary nonperfusion identified on widefield FA. The nonperfused hypoxic retina is thought to upregulate hypoxia-inducible factors and cytokines, including VEGF and erythropoietin, with small studies suggesting that TRP can decrease the anti-VEGF treatment burden. [95] [96] [97] [98] [99] At the 3-year endpoint, there was no evidence that combined ranibizumab plus TRP reduced treatment burden or improved vision outcomes compared to ranibizumab alone.

No well-constructed studies show a definitive benefit of pars plana vitrectomy (PPV) for managing DME. The theoretical basis for PPV as a treatment option comes from reports that it increases vitreous oxygenation in ischemia, leading to decreased VEGF production, and from the observation that DME is less common among eyes with PVD. [100] [101] [102] [103] [104] [105] Vitreous viscosity also significantly decreases, which may bring about a greater diffusion of pro-inflammatory cytokines away from the macula. [106] Other authors suggest PPV plus internal limiting membrane (ILM) peeling should be attempted, as its removal brings better resolution of the tractional forces at the vitreoretinal interface known to worsen DME. This procedure also prevents proliferating astrocytes from using the ILM as a scaffold which may lead to ERM. [107] In a systematic review looking at PPV for DME, CST was significantly decreased by 102 μm, and a non-significant VA increase of 2 letters was observed. [108] However, the anatomic benefit was not maintained by the 12-month timepoint. A similar meta-analysis looking at PPV plus ILM peeling versus PPV alone showed no significant difference in postoperative vision and macular thickness. [109] DRCR Protocol D, a prospective study of eyes with DME and VMT, found that at 6 months postop, 43% of eyes had a reduction in central subfield thickness to <250 μm. [110] However, the median VA did not change at 6 months. In 38% of eyes the median visual acuity improved by ≥10 letters, and in 22% the median VA decreased by ≥10 letters. Posthoc analysis of DRCR Protocol I showed that previously vitrectomized eyes given anti-VEGF for ci-DME had no improved clinical outcomes compared to non-vitrectomized eyes. [111]

Treatment Complications

Complications, listed below, may arise from the various treatment modalities. The per-injection risk of developing complications is also listed below, when available.

Intravitreal injections

Endophthalmitis (0.019-0.05%) [112] [113] [114]
Intraocular inflammation (0.09-0.4%) [115]
Retinal tears/detachment (0.01-0.08%) [116] [117] [118]
Increase in intraocular pressure (For intravitreal steroids, IOP ≥21 mmHg in 45% at 1 month, 20% at 3 months, and 13% of eyes 6 months post-injection.) [119]
Cataract (Common after intravitreal steroid injections. By year three, 83% of patients given IVT had cataract surgery.) [120]
Subconjunctival hemorrhage (10%) [121]
Vitreous hemorrhage
Subretinal fibrosis
Extension of the laser scar into the fovea
Choroidal neovascular membrane
Paracentral scotoma
Decreased visual acuity
Retinal tears and retinal detachment
Elevated intraocular pressure
Endophthalmitis

Clinical factors associated with better visual outcomes with anti-VEGF treatment include lower hemoglobin A1c, younger age, less severe DR, absence of ERM, quick and consistent CST decreases with anti-VEGF therapy, and absence of prior panretinal photocoagulation (PRP). [122] [123] In terms of anatomic outcomes, eyes with hard exudates within the 6mm foveal center had a larger CST decrease at the end of 1 year. The presence of exudates may be a marker of BRB abnormalities typical of DME responsive to anti-VEGF. Eyes that lack exudates may have other underlying mechanisms of retinal thickening, including cystoid degeneration, traction, or ischemia. In contrast, eyes with high baseline CST ≥ 570μm had a significantly higher chance of developing persistent DME despite monthly ranibizumab therapy. [124] In terms of response to macular laser therapy, the ETDRS reported that worse clinical outcomes were associated with higher blood lipid levels, presence of hard exudates, and diffuse edema. [125]

Matsunaga et al. looked at eyes with DME treated with at least 1 dose of anti-VEGF, and then afterward were lost to follow-up (LTFU) for at least 6 months before returning to the clinic. [126] Their vision worsened significantly at the initial return visit to ~20/69 from ~20/52, however at the 3, 6, and 12-month follow-up, and final checkup, their vision recovered and had no significant difference with their baseline vision. OCT CST also showed similar trends. Gao et al. found that 25% of patients with NPDR and DME had no follow-up for at least 1 year after receiving 1 anti-VEGF injection. [127] Factors associated with LTFU included being Hispanic (OR 1.66), American Indian, Pacific Islander, multiple races (OR 2.6), and unknown race (OR 1.59) compared to whites. Lower adjusted gross income and decreasing baseline vision were also factors significantly associated with LTFU.

Some studies have shown increased risks of end stage renal disease in patients with concurrent DME. Assessment of renal function should be strongly encouraged in patients with chronic DME.

Future Directions

Numerous pharmacotherapy trials for DME treatment are underway. The development of a long-acting anti-VEGF that could remain effective in the vitreous for multiple months or years would significantly decrease the treatment burden for patients needing monthly injections. Novel pharmacotherapies based on different mechanisms of action include anti-VEGF designed ankyrin repeat proteins (DARPins), growth factor inhibitors, anti-inflammatory medications, hormone modulators, acetylcholine receptor blockers, IGF-1 receptor blockers, and neuroprotective/antiapoptotic agents. [128] Port delivery systems FDA approved for wet AMD are in development to treat DR and DME. [129] These experimental therapies have the potential to significantly decrease the burden of treatment while restoring vision to DME patients in the coming years ahead.

Teleophthalmology and Artificial intelligence (AI) are being used to develop screening tools for diabetic retinopathy and DME. These modalities can detect the retinal complications of diabetes remotely, and if this technology is placed in the offices of internists and endocrinologists, it may allow for early detection and timely intervention. [130] [131]

Diabetic Macular Edema: What to Know

Frequently asked questions.

Diabetic macular edema (DME) is a complication of diabetic retinopathy . It occurs when fluid builds up in the macula , the center portion of the retina. The retina is the layer of cells at the back of your eye that helps convert light into the images you see.

Diabetes is the leading cause of new blindness in the United States. DME is the most common cause of vision loss in people with diabetes.

Thankfully, there are steps you can take to help prevent diabetic macular edema. If you’re already experiencing diabetic macular edema, there are treatment options that can help alleviate your symptoms.

This article will discuss the symptoms, causes, treatments, and risk factors of diabetic macular edema.

Verywell / Daniel Fishel

Types of Diabetic Macular Edema

There are two key types of diabetic macular edema:

Non-central-involved diabetic macular edema (mild)
Central-involved diabetic macular edema (severe)

Both types are characterized by thickening of the retina that is 1 millimeter in diameter or larger. In non-central-involved diabetic macular edema, the thickening doesn't involve the central subfield zone of the macula. The central subfield is the 1 mm area around the center point of the fovea, a small depression in your macula where your vision is sharpest.

In central-involved diabetic macular edema, the retinal thickening does involve the central subfield of the macula. This often causes progressive vision loss.

Tests for Diabetic Macular Edema

To determine if you have diabetic macular edema, your healthcare provider may use one or more of the following tests:

Visual acuity test : This is the standard test you typically take at your yearly eye exam. It consists of a chart with letters and numbers that decrease in size as you read from the top down. Your healthcare professional usually asks you to cover one eye and read from the lowest line you can see clearly. The test is done on both eyes.
The Amsler Grid : With this test, you can wear reading glasses if you typically use them. The grid is placed at the same distance you’d normally place a newspaper or a book when reading. You’ll cover one eye and mark any parts of the grid that aren’t clear. You’ll switch eyes and again mark the chart.
Dilated eye exam : Your healthcare provider will administer eye drops that will cause your pupils to widen. Then, your healthcare provider will examine the retina to see if there’s fluid buildup around the macula.
Optical coherence tomography : Using a special light and camera, your healthcare provider will look to see if the retina is thick (and how thick) and if the macula is swollen.
Fluorescein angiography : To see if there’s any damage to the macula, your healthcare provider will inject dye into your arm. They dye travels through the blood vessels until it reaches your eye. A camera photographs your retina to see if there’s damage or disease related to diabetic macular edema.

Symptoms of Diabetic Macular Edema

Vision changes are the primary symptoms of diabetic macular edema. These include blurred or distorted vision near or in the center of your field of vision. Colors may also appear dull or washed out.

Causes of Diabetic Macular Edema

Diabetic macular edema is a complication of diabetes. It evolves over time and is caused by poor blood sugar control. It doesn’t occur on its own. Consistently high blood sugar can damage the small blood vessels in your eye.

Initially, this damage starts as diabetic retinopathy, which can impair your vision. If left untreated, fluid from these damaged blood vessels can leak into the macula. This is what causes diabetic macular edema.

Risk Factors for Diabetic Macular Edema

Anyone with type 1 or type 2 diabetes can get diabetic macular edema.

Prevalence of Diabetic Macular Edema

According to the National Eye Institute, approximately 7.7 million Americans have diabetic retinopathy. Of those, approximately 10% have diabetic macular edema.

Non-Hispanic African Americans are three times more likely to have diabetic macular edema than non-Hispanic Whites.

Several risk factors could lead to diabetic macular edema. These include:

Kidney disease
Very high blood pressure
High fat levels in the blood
Fluid retention
Pregnancy

Treating Diabetic Macular Edema

Treatments for diabetic macular edema start with treating diabetes itself. However, depending on the type or severity of your diabetic macular edema, you may need additional treatments.

Injectable Anti-VEGF Medications

Anti-VEGF medications work to block new blood vessels from developing. They also stop leakage from abnormal blood vessels that could flood the macula.

These drugs are injected directly into the eye, so you may experience some mild pressure. Your healthcare provider will determine how many injections you’ll need.

The American Diabetes Association recommends anti–VEGF injections as a first-line treatment for most people with central-involved diabetic macular edema that impairs vision.

Laser Therapy

Using laser light, your healthcare provider will try to close and destroy any blood vessels that are leaking into the macula. Typically pain-free, this procedure can help slow or stop the growth of new blood vessels that may further damage your vision.

Laser therapy can help protect your vision and possibly even improve it some. There’s also the possibility you could have permanent blind spots, however.

Anti-Inflammatory Medications

Corticosteroids can be used to reduce swelling of the blood vessels that leads to increased fluid in the macula. You and your healthcare provider will decide on the best way to administer them.

Corticosteroid eye drops are a non-invasive way to help reduce retinal thickness and improve vision. They can be used long-term.

Corticosteroids may also be administered by injection. Because diabetic macular edema is a chronic or long-term problem, multiple injections are typically required.

To reduce the number of injections, an implant that contains a sustained-release corticosteroid is another option.

If your eye doesn’t respond to steroids or you experience side effects due to steroids, your healthcare provider may try nonsteroidal anti-inflammatory drugs (NSAIDs).

According to the American Diabetes Association, persons who have persistent diabetic macular edema despite receiving anti–VEGF therapy (or those who are not candidates for this treatment) may benefit from laser therapy or corticosteroid injections into the eye.

Preventing Diabetic Macular Edema

The best prevention for diabetic macular edema is managing your diabetes and following a healthy lifestyle. Maintaining proper blood sugar levels, coupled with keeping your cholesterol and blood pressure in check, is important in preventing diabetic macular edema.

It’s also important to get regular eye exams, including the dilated eye exam. This way your healthcare provider can monitor any changes to your vision and check for possible eye damage.

How Often Should I Get an Eye Exam?

The American Diabetes Association recommends:

Adults with type 1 diabetes receive an initial dilated and comprehensive eye examination within five years of diagnosis
Persons newly diagnosed with type 2 diabetes undergo an eye exam shortly after diagnosis

Those with normal exams and well-controlled blood sugar levels may be screened every one to two years. If diabetic retinopathy is present, dilated retinal examinations should be performed at least annually, and possibly more often.

If you’re pregnant, particularly if you have diabetes, it’s crucial to have an eye exam with dilation during your pregnancy and undergo close monitoring if needed.

Regardless of type, people with diabetes are at higher risk of developing diabetic macular edema, particularly if they already have diabetic retinopathy. Monitoring your eye health by receiving annual eye exams with dilation is crucial to detecting vision changes or damage to blood vessels that could lead to diabetic macular edema.

If you’re diagnosed with diabetic macular edema, there are treatments to effectively treat the condition and preserve your vision.

A Word From Verywell

The onset of diabetic macular edema doesn’t mean you’ll lose your vision. There are effective therapies to treat and repair damage to the macula and blood vessels in your eye to prevent diabetic macular edema from becoming severe.

The key to preventing diabetic macular edema is to properly manage your diabetes to avoid consistently high blood sugar levels. Keeping your blood pressure and cholesterol levels in check will also aid in preventing diabetic macular edema.

If you’ve noticed changes in your vision or have concerns regarding diabetic macular edema, speak to your primary healthcare provider. They can assess your condition and determine next steps to prevent or alleviate diabetic macular edema.

While diabetic macular edema can’t be cured, the damage to the macula can be treated and reduced to preserve your vision. It’s important to maintain good blood sugar levels and follow a healthy lifestyle to keep the condition from becoming more severe.

While macular edema is often caused by diabetes, there are other possible causes. For instance, if you have eye surgery for conditions such as cataracts, glaucoma , or retinal disease, you could develop macular disease following the surgery. Age-related macular degeneration and inflammatory diseases also could lead to macular edema.

Yes, bananas deliver nutrients that could boost eye health and help prevent macular degeneration. These include folic acid and vitamin B6, as well as carotenoids, which are antioxidants that protect your eyes. A healthy diet of fruits and vegetables will boost eye health and help ward off macular degeneration.

Prevent Blindness. Eye diseases & conditions: diabetes-related macular edema .

International Diabetes Federation. Care and prevention .

Elsevier.health. Diabetic retinopathy .

National Eye Institute. Macular edema .

American Diabetes Association Professional Practice Committee. 12. Retinopathy, neuropathy, and foot care: Standards of medical care in diabetes—2022 . Diabetes Care . 2022;45(Supplement_1):S185-S194. doi:10.2337/dc22-S012

Abe S, Nishitsuka K, Yamashita H. Long-term outcome of steroid eye drop therapy for diabetic macular edema . Invest Ophthalmol Vis Sci . 2018;59(9):4827-.

Al Dhibi HA, Arevalo JF. Clinical trials on corticosteroids for diabetic macular edema . World J Diabetes . 2013;4(6):295-302. doi:10.4239/wjd.v4.i6.295

Agrón E, Mares J, Clemons TE, Swaroop A, Chew EY, Keenan TDL. Dietary nutrient intake and progression to late age-related macular degeneration in the age-related eye disease studies 1 and 2 . Ophthalmology . 2021;128(3):425-442. doi:10.1016/j.ophtha.2020.08.018

By Karon Warren Warren is a freelance health writer based in Georgia with a bachelor's degree in journalism from the University of Southern Mississippi.

Blurry Vision
Conjunctivitis - Pink Eye
Corneal Abrasions
Eye Discharge
Dilated Pupils
Eye Infections
Eye Twitching
Ocular Migraines
Swollen Eyelids
How to Get Rid of a Stye
Blepharitis
Eye Floaters
Myopia in Children
Myopia (nearsightedness)
Astigmatism
Contrast sensitivity testing
Refractive errors and refraction
Visual Acuity: 20/20 Vision
Hyperopia (farsightedness)
Digital Eye Strain
Computer Glasses
Kids & Screen Time
Blue Light and Vision
Blue Light Glasses
How to Choose Eyeglass Lenses
Anti-reflective Lenses
Progressive Lenses
Eyeglass Lens Coatings
Photochromic Lenses
Multi-focal Lenses
Eyeglass Frame Materials
How to Clean Your Glasses
What do the numbers on your eyeglass frames mean?
Eyeglass temples: How do you know if they're the right length?
Can you be allergic to eyeglasses?
The Best Glasses for Your Face Shape
Pupillary Distance
Glasses for Round Faces
Glasses for Small Faces
Glasses for Long Faces
Sunglasses for Oval Faces
How Much Do Glasses Cost?
Cheap Glasses
Choosing Eyeglasses
Best Places to Buy Glasses
How to Solve Problems with New Glasses
Styles: Trends in Men's Glasses
Polarized Lenses
Prescription Sunglasses Guide
Ray-Ban Wayfarer Guide
Performance Sunglasses
Are Designer Shades Worth It?
How to Spot Fake Wayfarer Sunglasses
Guide to High-Quality Sunglasses
Tint Guide for Sports Sunglasses
UV Radiation and Your Eyes
Contact Lens Basics
Reading a Contact Lens Prescription
Soft Contact Lens Care
Daily Disposable Contacts
Buying Contacts Without a Prescription?
Bifocal Contacts
Toric Contact Lenses
Colored Contacts
Multifocal Contacts
Gas Permeable Contact Lenses
Scleral GP Contact Lenses
What is vision insurance?
How to Use Vision Insurance Benefits
Check Your Vision Insurance Benefits
EyeMed Vision Insurance
VSP Vision Insurance
Vision Insurance or Vision Benefits Plan
Vision Insurance Provider List
Provider Networks and Vision Insurance
What is Covered by Vision Insurance?
Medicare & Medicaid Vision Benefits
Medicare Advantage Plans (Part C Plans)
Medicaid: Eligibility and Vision Benefits
How to Get Free Eye Exams and Glasses
LASIK Eye Surgery Guide
Cost of LASIK
LASIK Risks
How Long Does LASIK Last?
Does LASIK Hurt?
Do I Have To Be Awake During LASIK?
PRK Surgery
SMILE Laser Surgery
Epi-LASIK Surgery
Lens Replacement Surgery
LASEK Eye Surgery
PresbyLASIK
Cataract Surgery
Laser Cataract Surgery
Cataract Surgery Video
Cataract Surgery Recovery
Cataract Surgery Complications
Cataract Surgery Cost
What to Expect During Eye Exams
Free Eye Exams
What is an Eye Test?
How to Read Your Eyeglass Prescription
How to Choose an Eye Doctor
Infant vision development
Are contact lenses a good choice for kids?
Reading glasses: Tips before you buy
How older drivers can improve their driving at night
Lutein & Zeaxanthin
Eye Vitamins and Supplements
Omega-3 Fatty Acids
Safety Glasses
Sports Glasses & Goggles
Eye Safety Basics
Contact Lenses for Sports
Shooting Glasses and Hunting Eyewear
Ski Goggles
Eye Diagram
Baby's Eye Colors
Dominant Eye Test
Color Blindness
Color Blind Tests

Diabetic retinopathy

By Sharon Peralta , medically reviewed by Thomas J. Stokkermans, OD, PhD, FAAO

What is diabetic retinopathy?

Diabetic retinopathy (DR) is a serious eye disease that can affect people with diabetes mellitus. It occurs when high levels of glucose (sugar) in the blood cause changes to the network of blood vessels located in the retina . These changes can damage the retina, leading to vision loss and blindness.

Diabetes is a condition characterized by the chronic elevation of glucose levels in the blood. In the body, blood sugar levels are controlled by insulin. When there is too little insulin produced or the body is resistant to insulin, diabetes may result. Diabetes can impact many organs and systems of the body, including the eyes.

The retina is the light-sensing membrane located in the back of the eye. It converts light and visual images into special signals. These signals are then sent to the brain to process what you see.

Changes to the retina, such as those that occur from diabetes, can impact its ability to function properly. These changes can result in issues that lead to visual impairment.

Diabetic retinopathy is the most common eye condition among individuals with diabetes. It is also one of the top causes of blindness in the U.S. adult population. Diabetes-related retinopathy affects 93 million people worldwide.

SEE RELATED: Other types of retinopathy

Causes of diabetic retinopathy

Click image to enlarge.

Several factors may lead to the development of diabetic retinopathy. However, prolonged hyperglycemia (high blood sugar levels) is thought to be the leading reason behind its onset.

High blood sugar can cause the blood vessels in the retina to become enlarged and leak fluid. In other cases, the growth of new, abnormal vessels in the retina may occur. At times, the vessels in the retina may become blocked. This prevents blood flow through this part of the eye, depriving it of oxygen and nutrients. Each of these factors can damage the blood vessels in the retina, causing diabetic retinopathy.

Who is at risk?

Diabetic retinopathy can affect people with Type 1 or Type 2 diabetes. Women who develop pregnancy-related diabetes (gestational diabetes) may also be at risk. DR may occur in those with both diagnosed and undiagnosed diabetic conditions.

The longer you have diabetes, the higher your risk of developing DR.

The biggest risk factors include:

Poorly controlled blood sugar levels

Long-term diabetes

High cholesterol

High blood pressure

Age (50 and over)

Family history of diabetes

Ethnicity may also increase a person’s risk. People within the following populations may be more likely to develop DR:

Hispanic/Latino

African American

American Indian

Alaska Native

SEE RELATED: Diabetic eye disease: What’s your risk?

Signs and symptoms

The early stage of DR may not cause any symptoms. When diabetic retinopathy symptoms do occur, however, they may include:

Distorted vision

Blurred vision

Double vision

Eye floaters

Flashes of light

Blind spots

Eye pain or pressure

Eye redness

Fluctuating vision

Poor night vision

Changes in color vision

Partial or total vision loss

People with this condition typically experience symptoms in both eyes.

What is the first sign of diabetic retinopathy?

One of the first signs of diabetic retinopathy is the presence of microaneurysms in the blood vessels of the retina. These are bulges within the vessels that leak fluid or blood. They often look like small red dots sprinkled across the retina when viewed through diagnostic equipment.

Microaneurysms are a characteristic feature of diabetic retinopathy. Other signs include:

Blood in the retina ( retinal hemorrhage )

Thickening of the macula ( macular edema )

Fatty deposits (hard exudates) in or under the retina

Cotton wool spots (soft exudates) in the retina

As the condition progresses, further changes to the retina may develop. These include:

Abnormal vessels and vascular changes within the retina

New vessels growing within the retina or iris

Blood in the vitreous ( vitreous hemorrhage )

Scarring of the retina

The blockage or loss of macular blood vessels, depriving the macula of nutrients and oxygen (macular ischemia)

READ MORE: Types of diabetic eye problems

Main types and stages of diabetic retinopathy

Diabetes-related retinopathy is classified into two primary types. These include non-proliferative diabetic retinopathy and proliferative diabetic retinopathy. The main difference between the two is whether new, unhealthy blood vessels have developed in the retina.

The condition is further categorized into four stages of diabetic retinopathy progression.

Type: Non-proliferative diabetic retinopathy (NPDR)

NPDR is the early stage of the general condition. It occurs in three stages of progression — mild, moderate and severe. The key factors in this stage are leaky retinal vessels and blockages in small blood vessels called capillaries. This leads to swelling, inflammation and tissue damage in the retina.

Stage 1: Mild non-proliferative diabetic retinopathy

Mild NPDR is the earliest stage of the condition. A few microaneurysms are generally present at this stage of the disease.

Stage 2: Moderate non-proliferative diabetic retinopathy

In this stage, the number of microaneurysms and areas of bleeding in the retina increases. Doctors refer to these areas of bleeding as dot-and-blot hemorrhaging. Other retinal changes may also arise. These include the presence of fatty deposits and cotton wool spots.

Increased swelling of the blood vessels may occur. The vessels may also become blocked.

Stage 3: Severe non-proliferative diabetic retinopathy

The condition is in the severe NPDR stage when the following factors are present:

Microaneurysms and scattered hemorrhaging occur within all four quadrants (quarters) of the retina.

Venous beading (where the vessels take on a sausage-link appearance) is present in at least two quadrants of the retina.

Intraretinal microvascular abnormalities (IRMA) develop in at least one quadrant of the retina.

Doctors refer to this as the “4-2-1 rule.”

The noticeable effects of non-proliferative diabetic retinopathy may include:

Gradual loss of vision

NPDR can eventually trigger the growth of new blood vessels in the retina (neovascularization). It can also lead to diabetic macular edema (DME).

Diabetic macular edema

DME is swelling and thickening of the macula and is the leading cause of vision loss in people with DR. It can develop during any DR stage and may require treatment, even if it appears in mild NPDR.

Neovascularization occurs during the advanced stage of DR. This stage is called proliferative diabetic retinopathy.

SEE RELATED: What are Roth spots?

Type: Proliferative diabetic retinopathy (PDR)

PDR is characterized by the development of new but abnormal blood vessels in the retina.

Stage 4: Proliferative diabetic retinopathy

When normal blood vessels in the retina become blocked or damaged, the body tries to find another way to help blood reach these tissues. These changes trigger the release of a protein called vascular endothelial growth factor (VEGF), which aids in the creation of new (but abnormal) vessels. This is called neovascularization and indicates the fourth stage of the disease.

The development of abnormal vessels can cause scar tissue to form. The scar tissue can pull or tug on the retina, leading to a condition called retinal detachment .

The vessels may also leak, allowing blood to enter the vitreous (the gel-like fluid in the back of the eye). Known as vitreous hemorrhage, this can prevent light from reaching certain points on the retina. Vitreous hemorrhage can impact your vision, from causing a mild increase in floaters to completely blocking out your vision. It can clear up by itself over time, but its cause will always require some type of treatment.

Proliferative diabetic retinopathy may lead to further eye conditions, including:

Neovascular glaucoma

Other effects of proliferative diabetic retinopathy may include:

Eye flashes or floaters

Sudden loss of vision

Significant vision impairment or blindness

Diabetic macular ischemia

Diabetic macular ischemia (DMI) is a common complication of diabetic retinopathy. It most often presents in advanced cases but may develop at any DR stage.

DMI occurs when blood vessels in the macula become narrowed or blocked, deteriorate or are destroyed. This impedes blood flow into the macula, depriving it of necessary oxygen and nutrients. DMI can cause mild to severe vision loss, depending on its location and severity.

SEE RELATED: The link between diabetes and glaucoma

Diagnosis, management and treatment

The earlier diabetic retinopathy is detected and treated, the better the chances of keeping it from progressing. Recognizing early stage diabetes eye symptoms can lead to timely treatment and management of the condition. Your doctor may be able to detect signs of DR even before you have symptoms.

How is the condition diagnosed?

Diagnosing diabetic retinopathy typically begins with a comprehensive eye exam . Your eye doctor may use the following tests to diagnose the disease:

Visual acuity exam – This exam uses eye charts to test your ability to see objects clearly at different distances.

Amsler grid test – This test uses a square grid made of straight lines, rather than a chart with letters. It allows the doctor to assess your central vision for distortions or blind spots.

Dilated retinal exam – Your doctor may place drops in your eyes to temporarily dilate (widen) your pupils. This allows them to see into the back part of your eye, where the retina sits.

Tonometry – Tonometry measures eye pressure (intraocular pressure) and is used to screen for glaucoma. It may be part of the exam since people with diabetes have a higher risk of developing glaucoma.

Ophthalmoscopy – With this technique, your doctor uses a special device called an ophthalmoscope. This device has magnifying lenses used to examine the retina.

Fundus photography – Your doctor may take pictures of the fundus (back of the eye) to check the retina and other structures.

Fluorescein angiography – Fluorescein (a yellow dye) is administered into a vein. The dye moves through your bloodstream and into the vessels in your eyes. A special camera is then used to take pictures of the retina as the dye moves through the vessels. With this exam, your doctor can determine if new blood vessels have developed. It also allows them to see any areas of blood vessel blockage or leakage.

Optical coherence tomography (OCT) – Using light wave technology, OCT produces cross-sectional images of the eye’s structures. These images allow your doctor to view the layers of the retina to check for changes or damage. It also measures the thickness of the retinal layers.

In some instances, artificial intelligence (AI) and teleophthalmology are used to screen for DR. Teleophthalmology is a form of telemedicine. It allows doctors to deliver eye care through digital methods. Medical providers can use these tools to detect signs of the condition remotely, which could lead to an earlier diagnosis.

SEE RELATED: Retinal imaging and scans

Regular eye exams are essential to detect diabetic retinopathy vision problems early and prevent further vision loss. Your doctor may use the results of your eye exam to determine if or which diabetic retinopathy treatments are needed.

Diabetic retinopathy treatment is commonly provided by a retinal specialist , which is an ophthalmologist specializing in the retina.

What treatment options are available?

Patients with mild diabetic retinopathy generally do not need treatment. At this stage, controlling blood sugar and blood pressure, and blood cholesterol and lipid levels is key to managing the condition.

Your eye doctor may recommend one or more of the following treatments for advanced cases of DR:

Laser surgery closes off blood vessel leaks and reduces retinal swelling (macular edema). These procedures may also help shrink abnormal blood vessels and reduce the risk of retinal detachment.

There are two main types of laser treatment for diabetic retinopathy. These include:

Focal or grid laser photocoagulation – This treats certain cases of macular edema. It involves focusing laser energy directly on the affected area or in a grid-like pattern on the retina. Focal laser treatments target specific microaneurysms. Grid laser treatments address larger areas of blood vessel leakage.

Scatter (panretinal) laser photocoagulation – Advanced cases of DR, or proliferative diabetic retinopathy, may be treated with panretinal photocoagulation (PRP). This is a more rigorous type of laser diabetic retinopathy surgery. Laser surgery with PRP is designed to reduce the amount of VEGF produced by the retina. In this way, it lowers the risk of new, abnormal vessel development (neovascularization).

Laser diabetic retinopathy treatments are generally performed in an ophthalmologist’s office or an outpatient treatment center.

SEE RELATED: Laser coagulation eye surgery

A vitrectomy involves surgically removing some or all of the vitreous fluid from the eye.

This may be recommended in advanced cases of diabetic retinopathy, such as when blood has leaked into the vitreous gel and is impairing vision. The vitreous cavity is then filled with a clear fluid to help maintain the shape of the eye.

A vitrectomy may also be conducted if scar tissue has formed in the retina.

Vitrectomies are typically performed at a hospital or an outpatient surgery center.

Anti-VEGF eye injections

Eye injections of medications known as anti-VEFGs can help stop the overproduction of VEGF in the eye. They help inhibit the growth of abnormal blood vessels and shrink those that have already developed.

Anti-VEGFs are injected into the vitreous of the eye. Along with preventing blood vessel growth, these eye injections help control macular swelling. They also reduce the risk of vessel leakage and retinal scarring.

Anti-VEGF medications used for DR are often the same ones utilized for macular degeneration injections. Some of the more common medications used for diabetic retinopathy include:

Eylea (aflibercept)

Lucentis (ranibizumab)

Avastin (bevacizumab)

Eye injections are provided in your ophthalmologist’s office. They may be recommended on a monthly basis, depending on the type of medication used and other factors. Your eye doctor will tell you how often you should have diabetic retinopathy medication injections.

Corticosteroids

Corticosteroid medications may also be used as a diabetic retinopathy treatment. Steroids reduce fluid leakage and swelling in the retina to help control macular edema.

These diabetic retinopathy medication treatments may be given in the form of an eye injection on a monthly basis. As an alternative, an intraocular implant may be surgically placed in the eye. The implant is filled with a steroid such as dexamethasone or fluocinolone acetonide. Steroid implants may offer a longer-acting option over monthly eye injections.

Corticosteroid implants are generally used in cases of DR that are resistant to anti-VEGF injection therapy. They provide a controlled release of medication over a certain period of time.

Intravitreal steroid eye injections are performed in your ophthalmologist’s office. Implants may also be placed in your eye doctor’s office. In more complex cases, implant surgery may take place in a hospital or surgical center.

Can diabetic retinopathy be reversed?

The very early stage of diabetic retinopathy may be reversed with tight blood sugar control. Eye injections and surgery may reverse certain effects of the condition.

In some cases, treatment can help improve vision. But in many instances, vision loss is permanent, and the goal of treatment is to prevent additional vision loss

Yearly diabetic eye exams allow your doctor to detect the condition as early as possible. Early diagnosis and treatment can reduce the risk of disease progression.

How is the disease managed?

While diabetic retinopathy cannot be cured, you can take steps to manage the disease and help keep it from advancing.

Diabetes is a chronic, lifelong condition that can cause significant health issues if it is not controlled. Managing this disease through diet, exercise and medical care (including vision care) can help preserve your health and protect your eyesight. Controlling diabetes can slow the development and progression of diabetic eye diseases, such as retinopathy.

Other factors key to managing diabetic retinopathy include:

Controlling blood sugar levels – This is central to managing both diabetes and diabetic eye conditions. Be sure to eat a healthy diet and take diabetes medication or insulin as prescribed by your doctor to control glucose levels.

Controlling high blood pressure and blood lipids – Ensuring your blood pressure and blood lipids (such as cholesterol and triglycerides) remain under control can lower your chance of developing retinopathy.

Managing diabetes-related kidney problems – Treating kidney or other health issues caused by diabetes is essential to maintaining the condition.

Getting yearly diabetic eye exams – Having regular eye exams allows your doctor to detect signs of retinal diseases or other problems caused by diabetes. The earlier DR is detected, the better the chances of preserving your vision.

See your eye doctor for more on managing diabetic retinopathy and its effects.

Health risks and complications

Diabetic retinopathy can threaten your vision and cause visual impairment. It can also increase the risk of other eye conditions. However, the risks and complications of the disease affect health factors that go beyond your eyesight.

Diabetes-related retinopathy can also increase the risk of other health problems. Some of these include:

Cardiovascular (heart) disease

Diabetic peripheral neuropathy (nerve dysfunction in the extremities)

Diabetic nephropathy (kidney dysfunction)

Diabetic foot syndrome

You could experience a range of other complications due to diabetes. Eye problems related to the disease can involve one or more of the following conditions:

Tractional retinal detachment

Macular edema

Vision loss or blindness

There may be other medical problems associated with DR. Consult your doctor to learn more about your personal risks or health concerns.

Some of the steps aimed at managing DR are the same used to help prevent it. It might not always be possible to prevent the onset of diabetic retinopathy. However, you may be able to reduce the risk of disease progression and vision loss. Factors that may lower these risks include:

Getting yearly eye exams – Having your eyes examined at least once every 12 months allows your doctor to detect and treat problems as early as possible. In fact, many people first learn they have diabetes through a routine eye exam .

Manage diabetes – Controlling diabetes is key to lowering the risks associated with DR. It is important to monitor and control your blood sugar and take diabetes medication or insulin as prescribed by your doctor. A healthy diet and proper nutrition are also important for controlling blood sugar.

Manage high blood pressure and blood lipids – Controlling your blood pressure and blood lipid levels can help lower the risk of diabetic retinopathy.

Get eye exams during pregnancy – If you are pregnant and have diabetes, get your eyes examined during each trimester. Inform your eye doctor if you develop gestational diabetes, as this is also a risk factor. Your eye doctor may recommend more frequent eye exams within the first year after giving birth.

Follow a healthy diet – Eating a healthy, balanced diet (such as the Mediterranean diet ) may lower your risk for diabetes-related eye conditions. Certain vitamins and minerals may also help support your eye health. Talk to your doctor before taking any new dietary supplements.

Get regular exercise – Exercising regularly may lower the risk of eye diseases such as diabetic retinopathy. Physical activity also helps lower blood glucose levels in the body.

Reduce ultraviolet (UV) exposure – Prolonged exposure to UV rays can damage the eye’s tissues, including the retina. Reducing the amount of time spent in the sun may help lower your chance of developing DR.

Avoid tobacco use – Some studies have found that smoking may increase the risk of diabetic retinopathy. Quitting or avoiding smoking is important for your eye health and general wellness.

Speak with your eye doctor for further details on lowering your risk of diabetic eye disease.

When to call an eye doctor

Diabetic retinopathy is a serious eye condition that needs prompt care from an eye care professional. If you have symptoms of diabetic retinopathy or notice vision changes, visit an eye doctor right away. Contact your doctor to schedule an exam or find an eye doctor near you if you do not yet have one. Prompt care can help prevent vision loss and maintain your eyesight.

READ NEXT: 8 ways to protect your eyesight

Notes and References

Diabetic retinopathy . StatPearls [Internet]. August 2023.

Diabetic retinopathy: Causes, symptoms, treatment . EyeSmart . American Academy of Ophthalmology. November 2023.

Diabetes . A.D.A.M. Medical Encyclopedia [Internet]. February 2023.

Anatomy, head and neck: Eye retina . StatPearls [Internet]. August 2023.

Diabetic retinopathy . Johns Hopkins Medicine. Accessed February 2024.

Retinal physiology and circulation: Effect of diabetes . Comprehensive Physiology . July 2020.

Vision loss and diabetes . National Center for Chronic Disease Prevention and Health Promotion; Diabetes. CDC. May 2024.

Diabetic eye disease . EyeSmart . American Academy of Ophthalmology. October 2021.

Diabetic eye problems . MedlinePlus [Internet]. December 2023.

Diabetic retinopathy in the aging population: A perspective of pathogenesis and treatment . Clinical Interventions in Aging . July 2021.

Risk of diabetic retinopathy and retinal neurodegeneration in individuals with type 2 diabetes: Beichen Eye Study . Frontiers in Endocrinology . May 2023.

Association of obesity with diabetic retinopathy in US adults with diabetes in a national survey . Endocrine Connections . June 2021.

Diabetic retinopathy symptoms . Stanford Medicine. Accessed February 2024.

Signs and symptoms of eye microaneurysms . EyeSmart . American Academy of Ophthalmology. May 2023.

Role of microaneurysms in the pathogenesis and therapy of diabetic macular edema: A descriptive review . Medicina . February 2023.

What is the difference between drusen and exudates? EyeSmart . American Academy of Ophthalmology. September 2023.

Diabetic retinopathy . EyeWiki . American Academy of Ophthalmology. July 2023.

Diabetic retinopathy . Merck Manual Consumer Version. April 2022.

Diabetic macular ischemia . EyeWiki . American Association of Ophthalmology. March 2024.

Diabetic retinopathy pathophysiology . EyeWiki . American Academy of Ophthalmology. February 2022.

The role of inflammation and therapeutic concepts in diabetic retinopathy — A short review . International Journal of Molecular Sciences . January 2023.

Ferroptosis in the ageing retina: A malevolent fire of diabetic retinopathy . Ageing Research Reviews . January 2024.

Update on management of non-proliferative diabetic retinopathy without diabetic macular edema; Is there a paradigm shift? Journal of Ophthalmic & Vision Research . January 2022.

PDR: Public health considerations . Optometry Times . August 2022.

Diabetic retinopathy screening . EyeWiki . American Academy of Ophthalmology. November 2023.

Visual field test. EyeSmart . American Academy of Ophthalmology. March 2022.

Diabetic retinopathy diagnosis . Stanford Medicine. Accessed February 2024.

Fundus camera . StatPearls [Internet]. August 2023.

What is optical coherence tomography? EyeSmart . American Academy of Ophthalmology. April 2023.

Diabetic macular edema . EyeWiki . American Academy of Ophthalmology. November 2023.

Tele-ophthalmology: Digital care in the digital world . Stanford Medicine. Accessed March 2024.

Referring patients with DR/DME to a retina specialist . Ophthalmology Times . February 2023.

Panretinal photocoagulation . EyeWiki . American Academy of Ophthalmology. June 2024.

Diabetic retinopathy . Kellogg Eye Center. University of Michigan Health. Accessed February 2024.

Lasers (surgery) . EyeWiki . American Academy of Ophthalmology. December 2023.

Diabetic retinopathy treatment . Stanford Medicine. Accessed February 2024.

Vitrectomy . National Eye Institute. October 2022.

What is vitrectomy? EyeSmart . American Academy of Ophthalmology. July 2023.

Anti-VEGF treatments . EyeSmart . American Academy of Ophthalmology. July 2023.

Injections for wet macular degeneration: What you need to know . BrightFocus Foundation. January 2023.

Intravitreal injection . A.D.A.M. Medical Encyclopedia [Internet]. November 2022.

Anti-VEGF injections: Start to finish . Review of Optometry . February 2020.

Aflibercept . EyeWiki . American Academy of Ophthalmology. January 2024.

Ranibizumab . EyeWiki . American Academy of Ophthalmology. October 2023.

Bevacizumab . EyeWiki . American Academy of Ophthalmology. February 2024.

Intravitreal implants . StatPearls [Internet]. July 2023.

Diabetic retinopathy . National Eye Institute. February 2024.

Type 2 diabetes . A.D.A.M. Medical Encyclopedia [Internet]. February 2023.

Diabetic nephropathy (kidney disease) . Johns Hopkins Medicine. Accessed February 2024.

Diabetic retinopathy as the leading cause of blindness and early predictor of cascading complications — risks and mitigation . EPMA Journal . February 2023.

Peripheral diabetic neuropathy . StatPearls [Internet]. September 2023.

Optometry’s role in the diabetes epidemic . Review of Optometry . July 2020.

Diabetes and eye disease . A.D.A.M. Medical Encyclopedia . February 2023.

What else can we do to prevent diabetic retinopathy? Diabetologia . June 2023.

Page published on Wednesday, August 14, 2024

Page updated on Tuesday, August 20, 2024

Medically reviewed on Wednesday, July 3, 2024

Schedule an exam

Women’s vision: Understanding risks and ways to protect it

Alice in Wonderland syndrome depiction of a large fish in a small bowl, and a small fish in a large bowl, which is how a person with Alice in Wonderland syndrome might perceive themselves in their surroundings

Alice in Wonderland syndrome (AIWS)

Police officer conducting a field sobriety test

Horizontal gaze nystagmus (HGN)

How to limit your child’s screen time with digital apps, ray-ban stories: features, pros and cons, how dirty are your facial tools.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

View all journals
Explore content
About the journal
Publish with us
Sign up for alerts
Data Descriptor
Open access
Published: 23 August 2024

MAPLES-DR: MESSIDOR Anatomical and Pathological Labels for Explainable Screening of Diabetic Retinopathy

Gabriel Lepetit-Aimon ORCID: orcid.org/0009-0000-4110-8227 1 ,
Clément Playout 2 , 3 ,
Marie Carole Boucher 2 , 3 ,
Renaud Duval 2 , 3 ,
Michael H. Brent 4 &
Farida Cheriet 1

Scientific Data volume 11 , Article number: 914 ( 2024 ) Cite this article

145 Accesses

1 Altmetric

Metrics details

Diagnostic markers

Reliable automatic diagnosis of Diabetic Retinopathy (DR) and Macular Edema (ME) is an invaluable asset in improving the rate of monitored patients among at-risk populations and in enabling earlier treatments before the pathology progresses and threatens vision. However, the explainability of screening models is still an open question, and specifically designed datasets are required to support the research. We present MAPLES-DR (MESSIDOR Anatomical and Pathological Labels for Explainable Screening of Diabetic Retinopathy), which contains, for 198 images of the MESSIDOR public fundus dataset, new diagnoses for DR and ME as well as new pixel-wise segmentation maps for 10 anatomical and pathological biomarkers related to DR. This paper documents the design choices and the annotation procedure that produced MAPLES-DR, discusses the interobserver variability and the overall quality of the annotations, and provides guidelines on using the dataset in a machine learning context.

Self-supervised retinal thickness prediction enables deep learning from unlabelled data to boost classification of diabetic retinopathy

An interpretable multiple-instance approach for the detection of referable diabetic retinopathy in fundus images

Automated image curation in diabetic retinopathy screening using deep learning

Background & summary.

Diabetic retinopathy (DR) is a complication of diabetes mellitus (DM) that damages the retinal microvasculature and can lead to vision impairment. DR develops gradually and is clinically characterized by stages according to the presence of lesions in the retina. Non-invasive fundus imaging can detect these lesions, and regular screening of at-risk populations is necessary to ensure early treatment and preserve their vision 1 . Despite widespread screening programs in North America 2 , 3 , 4 , 40% of patients with DM are still not monitored for DR. Studies based on these programs indicate that teleophthalmology screening increases the rate of monitored patients 5 , and that algorithms for automatic diagnosis of DR can enable further improvements by reducing costs and increasing the frequency of examinations, as well as ensuring timelier management of referred patients 1 .

In the last decade, machine learning models have been successfully applied to the automatic diagnosis of DR in fundus images. At the core of the development of these supervised algorithms are public datasets of annotated images, used for training and validation 6 . Among them, the Eyepacs 7 and MESSIDOR 8 datasets have enabled the development of automatic screening algorithms that exceed the performance required by the FDA 6 , and even outperform human experts 9 . Yet, medical personnel still lack confidence in these technologies, which do not meet the standards of explainable AI. We believe that to overcome this limitation, improving training labels should not be overlooked. Indeed, uncertainty around validation methods and generalization issues call for more diverse and bias-aware validation datasets and better documentation of their labeling process. The weak explainability of screening models calls, in addition, for more exhaustive and clinically relevant training labels beyond simple diagnostic classification. In its report “Four principles of Explainable AI”, the National Institute of Standards and Technology underlines that a proper explanation of algorithm outcomes should be meaningful to the target audience 10 , namely, in our case, it must be formulated using a vocabulary familiar to ophthalmologists. Restricting training labels to only DR grades is a major shrinkage of medical vocabulary, especially when clinical justifications rely heavily on identifying the anatomical and pathological structures of the retina (vessels, macula, red or bright lesions, etc.) 11 , 12 , 13 . Therefore, whether it is to guide models a priori so that they learn representations compatible with clinical knowledge or to interpret these representations a posteriori by comparing them with known biomarkers, datasets with pixel-wise annotations of such structures play a key role in the development of explainable screening models for DR. We designed MAPLES-DR (MESSIDOR Anatomical and Pathological Labels for Explainable Screening of Diabetic Retinopathy) with this objective in mind.

Annotating pixel-wise labels requires considerably more time and effort than image-level diagnostic labels; consequently, fundus datasets with such labels are few. Most of them publish annotations for only one biomarker (e.g. vessels 14 , 15 , 16 , 17 , optic disc 18 , 19 , 20 , exudates 21 , microaneurysms 22 ). To our knowledge, only four datasets provide labels for four lesions that are symptomatic of DR: microaneurysms, hemorrhages, exudates, and cotton-wool spots (CWS). FGADR 23 and Retinal Lesions 24 comprise numerous images (1842 and 1593 respectively), but pathological structures are mostly annotated as large clusters of lesions. In contrast, IDRiD 25 and DDR 26 provide precise segmentations of each lesion, but for fewer images (81 and 757 images, respectively). The articles accompanying these datasets focus on the performance gain they bring to existing screening methods, but provide only shallow descriptions of the protocols used to produce their annotations. None of them contain annotations of anatomical structures.

The MAPLES-DR dataset extends the MESSIDOR 8 dataset by complementing its image-wise diagnostic labels with novel pixel-level annotations of 10 classes of anatomical and pathological structures along with new DR and macular edema (ME) grades for 198 of its images (see Fig. 1 ). As far as we know, no other public dataset offers such a comprehensive range of anatomical and pathological labels for fundus images. This paper extensively documents the design choices and labeling protocol of this dataset, as well as its potential biases and inaccuracies. By thoroughly documenting our approach and releasing all the tools and software that produced the annotations along with this paper, we aim to help other research teams build more such datasets for explainable DR screening.

Schematic overview of the MAPLES-DR annotation protocol and of the content of the dataset.

MAPLES-DR does not provide new fundus images, but extends MESSIDOR labels by releasing annotations of new biomarkers for 198 fundus images of the original MESSIDOR dataset 8 published in 2014. The publication of this new annotations dataset was approved by the MESSIDOR consortium. This section describes the annotation process that produced MAPLES-DR.

Fundus images selection

Of the 1,200 images available in the original MESSIDOR dataset, 200 were selected for annotation in MAPLES-DR. We randomly chose 30 healthy images, 59 R1 images, 55 R2 images, and 56 R3 images, according to the DR grades provided by MESSIDOR. This selection is not representative of the prevalence of the pathology in either the MESSIDOR dataset or an actual screened population (see Fig. 2 ). However, it provides a set of images for each pathology stage and focuses on those critical for screening (R1 and R2). In the last stage of the annotation campaign, we found two duplicate entries in the dataset (the original MESSIDOR dataset contains several duplicate images stored with different names, which was later fixed in MESSIDOR-2 27 ). We excluded them from MAPLES-DR, reducing the number of images to 198.

Disparities among DR and ME grade distributions (converted to MESSIDOR grades) between MAPLES-DR subset (purple), MESSIDOR complete dataset (teal blue) and prevalence from first screening of the Toronto tele-ophthalmology program 40 (gray).

Procedure to grade DR and ME

The diagnoses provided with the MESSIDOR database for DR and ME are based on an uncommon grading system. To facilitate comparison with other public fundus image databases, we regraded MAPLES-DR images following the guidelines for Canadian teleophthalmology screening, whose grades are closer to international standards (i.e. ICDR 11 or the Scottish system 12 ).

These guidelines distinguish six grades for DR (R0: absent, R1: mild, R2: moderate, R3: severe, R4A: proliferative, R4S: stable treated proliferative) and three for ME (M0: absent, M1: mild, M2: moderate). Each grade is associated with a recommended course of action (from rescreening in 12-24 months to immediate referral to an ophthalmologist) and is determined by the number and position of red lesions (hemorrhages, microaneurysms, neovessels) and bright lesions (exudates, cotton wool spots) visible in the retina. Images with insufficient quality for diagnosis could also be flagged as R6 or M6. A comprehensive definition of this grading system was published by Boucher et al . 13 .

The variability between observers in the diagnosis of DR and ME is not negligible and is well documented in the literature 28 , 29 , 30 , 31 , 32 . Our DR and ME grading procedures accounted for this by splitting diagnosis labeling into two phases. First, each image was individually graded by three senior retinologists and a common diagnosis was determined by majority voting. The few cases of complete disagreement between the three retinologists (5 cases of DR and 12 cases of ME out of 198 images) were resolved in a second phase by deliberation between the experts to establish a consensual diagnosis.

Procedure to annotate Retinal Structures

The main contribution of MAPLES-DR is the annotation of retinal structures symptomatic of DR at a pixel-level, which are absent from the original MESSIDOR dataset.

Selection of anatomical and pathological biomarkers

The decision to include or exclude retinal structures from the list of annotated biomarkers was based on clinical understanding of their role in DR histopathology and screening procedures. See Figs. 3 and 4 for examples of pathological structures as annotated in MAPLES-DR.

Pathological structures from image 20060411_59190_0200_PP . (Left: the fundus image without annotations).

Pathological structures from image 20051214_57940_0100_PP .

Diabetes mellitus affects the walls of the vessels, eventually causing microvascular dysfunctions that manifest in the retina as microaneurysms, hemorrhages, intraretinal microvascular abnormalities (IRMA), or neovessels. We refer to these pathological structures as “ red lesions ”. Microaneurysms appear as small circular dilations of the capillaries. They are early signs of microvascular dysfunction and are commonly used to detect mild DR. Intraretinal Hemorrhages develop in the more advanced stages of the pathology. They are usally divided into dot hemorrhages (circular and well-defined spots, typically caused by a microaneurysm rupture) and blot hemorrhages (larger with less defined borders). Both were simply annotated as hemorrhages in MAPLES-DR. IRMA can appear starting from the moderate non-proliferative stage (R2), but their proliferation coincides with the next stage of the disease (R3), indicating a 50% risk of developing neovascularization within one year. The emergence of Neovessels (NV) signals a transition to the proliferative stage (R4A), the most severe stage of the DR screening, which requires immediate referral to an ophthalmologist. Indeed, extensive NVs can leak, causing preretinal or vitreous hemorrhages and leading to major visual loss if left untreated. The distinction between NVs and IRMAs is difficult to establish using fundus images alone and normally requires fluorescein angiography. In the absence of this imaging modality, IRMAs are not differentiated from NVs in MAPLES-DR.

In the severe stages of DR, the retina thickens and “bright lesions” can appear. Hard exudates usually arise from leakage from damaged capillaries, potentially causing loss of visual acuity. Furthermore, ischemia may cause a blockage in axonal transport in the optic nerve fiber layer, which can in turn lead to axoplasmic accumulation and the appearance of lesions known as Cotton Wool Spots (CWS). These lesions are characterized by their white appearance and blurry borders. While the principal etiology is diabetic retinopathy, CWS can also be observed in other vascular diseases (systemic artial hypertension, vein obstruction, coagulopathies, etc.). Finally, we provide annotations of Drusens . While these bright lesions are not symptomatic of DR and are more commonly associated with Age-related Macular Degeneration (AMD), their aspect is similar to exudates and can be mistaken for them. They usually appear around the macula and are histologically situated at the interface with the Retinal Pigment Epithelium (RPE). It is supposed that they originate from degenerative products of the RPE’s cells and are composed of lipids and glycoproteins.

Besides pathological biomarkers, MAPLES-DR also includes segmentation of anatomical structures: the optic disc and cup, the macula, and retinal vessels. Although these anatomical structures are present in all images, including healthy ones, their appearance and proximity to lesions provide valuable diagnostic information. Retinal vessels morphology is indicative of stages of DR: an increase in arteriolar tortuosity is associated with mild and moderate stages 33 , while venous beading and dilation are symptoms of severe proliferative stages. In addition, the severity of a lesion often depends on its position relative to the Optic disc or the Macula . ME is graded by counting the number of lesions within one or two optic disk diameters from the macula. Some clinical guidelines distinguish disc neovascularization from other neovascularization, etc. Fig. 5 shows an example of MAPLES-DR annotations of the retinal anatomical structures.

Anatomical structures from image 20051205_35354_0400_PP . ( Dark blue : optic disc, light blue : optic cup, dark purple : macula, light purple : vessels).

The choice of jointly annotating anatomical biomarkers and pathological biomarkers is, to our knowledge, unprecedented for fundus datasets, and distinguishes MAPLES-DR from existing datasets of DR annotations such as FGADR 23 or IDRiD 25 .

Recruiting annotators and designing the annotation tools

The segmentation maps of MAPLES-DR were annotated by seven Canadian retinologists affiliated with five different hospitals: Hôpital Maisonneuve Rosemont, Montréal; University Health Network, Toronto; Centre Hospitalier Universitaire de Montréal; Université de Sherbrooke; and Centre Hospitalier Universitaire Saint-Justine, Montréal). All the annotators were seniors retinologists and were recruited thanks to their involvement in teleophthalmology programs for the detection of DR in Quebec and Ontario.

The annotations were collected on a custom web-based annotation platform that we developed specifically for this project. It was designed to support a “scalable” annotation protocol, capable of being extended to much more ambitious annotation campaigns, such as labeling large Canadian telemedicine databases containing tens of thousands of images. With this in mind, its web interface was designed to allow users to work collaboratively on the same images regardless of their geographical location. Its specialized drawing and visualization tools provided an intuitive yet effective workflow to segment biomarkers in fundus images.

Preprocessing MESSIDOR fundus images

The dimensions of MESSIDOR fundus images varies from 1440×960 to 2304×1536 pixels. To standardize the resolution of the MAPLES-DR segmentation maps, all fundus images were cropped and resized to 1500×1500 pixels prior to the annotation process.

Then, two image enhancement algorithms were applied in order to assist the annotators. The first corrected illumination variations while preserving image hues, by subtracting the median-filtered image and using contrast-limited adaptive histogram equalization (CLAHE) on the brightness channel only. The second maximized the contrast by independently normalizing the intensities of each RGB channel so that they spread over their full range of values (see Fig. 6 ). These two preprocessing algorithms could be activated individually or in combination on the annotation interface. The decision of how and when to use these image enhancements was left to the clinician’s discretion.

Fundus image preprocessed with the visual enhancement algorithms available in the annotation platform.

Reviewing and correcting preannotated segmentation maps

Labeling anatomical and pathological retinal structures at a pixel level is an intrinsically tedious task. So instead of labeling from scratch, retinologists were assigned the less laborious task of reviewing AI-generated segmentations maps (we will refer to them as preannotated maps ) and correcting any segmentation errors. Preannotated maps were only provided for exudates, microaneurysms, hemorrhages, and blood vessels. Labeling the other anatomical biomarkers (macula, optic disc, and optic cup) was more straightforward and didn’t require preannotation. As for the other pathological biomarkers (neovessels, drusen and CWS), the databases published at the time were not sufficient to train segmentation models and therefore those biomarkers were labeled from scratch.

In practice, generating the preannotated maps was entrusted to two models: one responsible for segmenting lesions 34 , the other for segmenting vessels 35 . Both models were state-of-the-art in 2018. The final segmentation thresholds of both models were slightly reduced to favor false positives over false negatives. Indeed, we anticipated that reviewing all preannotated structures and removing the incorrect ones would be an easier task than spotting and annotating all those missing. Conversely, no lesions were pre-annotated for healthy images (according to MESSIDOR’s DR labels), to avoid the need of removing them.

Upon reaching 100 reviewed and corrected vessel segmentations, the weights of the pre-annotation model were fine-tuned on those new labels. The resulting improvements to the automated segmentation maps further reduced by 10% the average time required to annotate vessels.

Instructions provided to retinologists

In preparation of the annotation process, we separated the 10 biomarkers into 4 categories: 1 . red lesions (micro-aneurysms, hemorrhages, and neovessels), 2 . bright lesions (exudates, drusen, cotton wool spots), 3 . vessels, and 4 . other anatomical biomarkers outside vessels (optic disc, optic cup, macula). An additional “uncertain” channel was added to the two pathological biomarker categories (i.e. uncertain bright and uncertain red ) in anticipation of ambiguous cases. Here, annotators could annotate any structures that appeared pathological, but whose exact nature was unclear.

Each retinologist was assigned one of four categories and focused on annotating only the corresponding biomarkers. Consequently, each of the 198 images was seen by several retinologists but each type of biomarker was only reviewed by a single expert: one segmenting vessels, another the red lesions, etc. By splitting the annotation tasks and specializing each clinician in the labeling of a few biomarkers, we intended to simplify the learning curve for the annotation tools and speed up the entire annotation process. This measure, combined with preannotated labeling, reduced the cumulative time needed to annotate an image to an average of 22 minutes.

Besides the list of biomarkers to label and a list of recommendations on using the annotation tools, no explicit labeling instructions were provided to clinicians. We trusted their diagnostic expertise and did not provide instructions on the clinical definition of biomarkers or the level of detail expected. Yet, it is worth noting that the platform’s design induced implicit guidelines. Lesions in the preannotated maps were small (the median lesion diameter was 13 pixels) and numerous; this influenced the annotators to label each lesion individually (instead of circling the general area) and to pay attention to all lesions, even the smallest ones. As Radsch et al . 36 recently reported, providing expert annotators with exemplary images instead of extended text descriptions significantly improves the quality of biomedical annotations.

Data Records

The MAPLES-DR dataset is available as a figshare dataset 37 and contains two archives: MAPLES-DR.zip and AdditionalData.zip . The fundus images themselves are not included in these archives, as requested by the MESSIDOR consortium, but they are freely available to researchers who request them on the Consortium’s website .

MAPLES-DR.zip

The first archive provides the aggregated and final data of MAPLES-DR. It is intended to be used in direct conjunction with the MESSIDOR-1 archive to provide DR and ME labels and segmentation maps to use with their fundus images.

The archive MAPLES-DR.zip contains two folders: train/ and test/ , which are organized with a common structure (see Fig. 7 for an overview). Each includes a csv table named diagnosis.csv which associates the image names (following the MESSIDOR naming convention) with their corresponding DR and ME consensus grades. Each folder contains 12 subfolders: one for each segmented biomarker: the four anatomical biomarkers, the six pathological biomarkers, as well as RedUncertains and BrightUncertains for the ambiguous pathological structures. In those subfolders, the segmentation maps are saved as binary images in PNG format with the same naming convention and at the same resolution as their MESSIDOR counterparts. Note that because the MESSIDOR images vary in dimensions (ranging from 960 × 1440 up to 1536 × 2304 pixels), the images in MAPLES-DR.zip do as well.

Overview of the MAPLES-DR principal archive: MAPLES-DR.zip .

To divide the dataset into train and test sets, a multilabel vector was built for each image indicating the presence or absence of each type of lesion. The sets were then obtained by applying iterative multilabel-shuffle stratification as introduced by Sechidis et al . 38 . The resulting training and testing sets comprise, respectively, 138 and 60 samples. However, one segmentation of the macula is missing from the training set because the fundus image is centered on the disk, and four training and two testing maps of the optic cup are also unavailable as their boundaries were too fuzzy to be segmented. The corresponding files were simply removed from the folders: train/Macula , train/OpticCup and test/OpticCup , reducing their file counts to 137, 134, and 58 respectively.

AdditionalData.zip

The second archive contains all additional information and data collected throughout the annotation process, including the two image duplicates (see Fig. 8 for an overview). It is a complementary resource to the main archive, intended for multiple applications. For instance, these data can serve to conduct variability studies on non-aggregated DR and ME diagnoses or on the segmentation of retinal structures in the duplicate images; to better appreciate the annotation work by consulting the preannotated maps or the time spent correcting them; or to complement the annotations themselves with the textual comments left by retinologists.

Overview of MAPLES-DR complementary archive: AdditionalData.zip .

It contains the following files and directories:

biomarkers_annotations_infos.xls : identify which Retinologist performed the annotation of a given biomarker category, the Time spent on each annotation (in seconds), any Comment they left, and the Annotation# rank (1 for the first image annotated, 200 for the last).

diagnosis_infos.xls : contains the DR and ME grades annotated by each Retinologist and the consensus they reached. It also includes any Comment left by the retinologists while grading.

pre_annotations/ : contains the automatic segmentation of Vessels , Exudates , Hemorrhages , and Microaneurysms provided as preannotations to the retinologists.

annotations/ : stores 12 subfolders (one for each biomarker), which in turn contains all 200 segmentation maps annotated by the retinologists, without distinction between the train, test and duplicate sets. The images are stored as PNG binary masks at the resolution at which they were annotated (1500 × 1500 pixels) using the ROIs provided in the MESSIDOR-ROIs.csv .

MESSIDOR-ROIs.csv : provides the bounding boxes extracted from the MESSIDOR images to obtain square regions of interest and remove as much of the blank borders as possible around the circular fundus area. For each image identified by a name , its bounding box is stored as a top-left ( x0 , y0 ) and bottom-right ( x1 , y1 ) pair of coordinates in pixels. The original resolution ( W , H ) of the MESSIDOR fundus images is also included (resp. width and height in pixel). Those coordinates are necessary in order to either pad and resize the annotations/ and pre_annotations/ maps to match MESSIDOR resolution, or conversely, to crop and resize the MESSIDOR images to match the 1500 × 1500 resolution of MAPLES-DR segmentation maps.

dataset_record.yaml : lists the biomarkers names, the image filenames in the train , test and duplicates sets, as well as the filename of the samples that are missing an optic cup or macula segmentation map (resp. no_cup and no_macula ).

Technical Validation

Dr and me grades: overview.

Of the 198 images selected from MESSIDOR, all were judged to be of sufficient quality to receive a DR grade. Only one was deemed insufficient for ME grading. Our team of retinologists diagnosed a large majority of images as stage R1: Mild DR and none as stage R4S: Stable Treated Proliferative (cf. Fig. 9 ). From this perspective, the discrepancy between the grade distribution published in MESSIDOR that we used to select the 200 images (cf. Fig. 2 ) and the MAPLES-DR grades may seem significant, but two factors must be considered to understand this disparity. First, the visualization tools available in our annotation platform promote high sensitivity of lesions detection: MAPLES-DR retinologists reported that CLAHE preprocessing was particularly helpful for spotting small red lesions (i.e. microaneurysms and small hemorrhages) at first glance. Thanks to these tools, they identified enough microaneurysms in half of the MESSIDOR R0 (healthy) images to reclassify them as R1. Second, to be diagnosed as R2 in MAPLES-DR, an image must contain at least 4 hemorrhages (according to international grading guidelines), but it only needs to contain 1 hemorrhage or 5 microaneurysms to receive the same grade in MESSIDOR. Due to this difference between the two grading systems, many images classified as R2 or R3 in MESSIDOR were reclassified as R1 in MAPLES-DR. The opposite trend is seen for the macular edema scores: Although most of the selected images were classified as M0 in MESSIDOR, many were reclassified as M2 in MAPLES-DR because the former grading focuses on exudates, while the latter grading also considers microaneurysms and hemorrhages within the macula. These results underline the impact of the grading systems on global pathological labels and validate a posteriori our choice to reannotate the images in a system comparable to the international standard.

Image counts for each DR and ME grade in MAPLES-DR and number of retinologists agreeing with the final grade. (Deliberation was required when all three retinologists were in disagreement).

Our DR and ME grading procedure required a deliberation phase when all three retinologists were in disagreement. This step was necessary in five cases for DR and 12 cases for ME, and the deliberation always resulted in a low-severity grade: R0, R1 or M0 (see Fig. 9 ). Although not negligible, the disagreement rate between our three retinologists was consistent with the scores found in the literature: the quadratic Cohen kappa scores between individual and consensus grades for our three retinologists were 0.873, 0.862, and 0.619, whereas Krause et al . 31 measured an average of 0.87 for the same metric. More recently Teoh et al . 32 measured an intragroup Gwet AC2 coefficient of 0.628, while our retinologists achieved 0.964 (CI 95%: 0.947-0.981) for DR and 0.844 (CI 95%: 0.799-0.888) for ME.

Retinal structures: overview of annotation work

The pixel-wise annotation of DR biomarkers was performed primarily by the three retinologists who also graded DR and ME: one specialist annotated the bright lesions on 87% of the images, the second labeled the red lesions on 62% of them, while the third performed the segmentation of the vessels for all 198 images. The other four retinologists contributed mainly to the annotation of the optic disc and macula, as well as a few red and bright lesions (cf. Fig. 10 ). In total, the team dedicated 69 hours to reviewing preannotated structures, erasing false positives, manually adding annotations that were missing from the preannotated maps, and correcting lesions that were correctly segmented but misclassified (e.g., microaneurysms instead of hemorrhages).

Proportion of images annotated by each retinologist per category of biomarker.

The macula and optic disc (including the cup) were the fastest to be annotated, with an average of 2 minutes per image, followed by vessel segmentation, with an average of 6 minutes per image. For the latter, preannotation appears to have been a valuable aid, since 77.5% of the pixels found in the vascular maps published in MAPLES-DR were preannotated and less than 8.3% of pixels from the initial preannotation maps were manually erased as false positive. However, the vessel labeling process still required substantial manual corrections: On average, an area of 56k pixels was manually added to each segmentation map (cf. Fig. 11 ), most of which consisted of small vessels (see Fig. 12a for an example of manual corrections of the vessels preannotations). In terms of image area, anatomical structures account for most of the MAPLES-DR annotations because, in addition to being typically larger than lesions, they are present in all images. Note that there are six images in which the optic cup segmentation is missing because its boundaries were too uncertain, and one image in which the macula is not visible as the image is centered on the optic disc.

Overview of the annotation work performed by the retinologists for each category of biomarkers: red lesions, bright lesions, vessels, and other anatomical structures. Manually added annotations appear darker while manually erased pre-annotations appear lighter. Both are labeled with the average number of corrected pixels per image. The final area covered by each type of biomarker after correction is displayed as a fraction of the total image area of the dataset.

Examples of corrections performed on image 20060412_61593_0200_PP when reviewing its preannotated segmentation maps. ( White : unchanged preannotation, teal : manually added segmentation, purple : manually removed preannotation).

Despite covering only a small fraction of the MAPLES-DR images (0.43% of pixels for red lesions and 0.41% for bright lesions), pathological structures required the greatest annotation effort. The time spent on each image varied greatly: from a few dozen seconds for healthy images to an hour for the most severe cases. Similarly to vessels, manual annotation of pre-segmented lesions (microaneurysms, hemorrhages, and exudates) relied heavily on the automatic segmentations. Only a quarter of the pixels were manually added, while the rest came directly from the preannotated maps. An analysis of the size distribution of lesions added manually by retinologists reveals that many small lesions (i.e. microaneurysms, exudates, or drusens) were annotated as 10px diameter circles. This shape corresponds to the default configuration of the annotation tool and is not representative of the exact shape of these small lesions. Unlike vessels, preannotation lesion maps contained more false positives and required a more careful review by retinologists to eliminate them. More than 58% of the preannotated hemorrhages were erased, as were 47% of the microaneurysms and 40% of the exudates (cf. Fig. 11 ). Most of the erased hemorrhages and microaneurysms were vessels erroneously detected as pathological, as seen in Fig. 12b . Overall, the annotation of pathological structures took an average of 6 minutes for bright lesions and 10 minutes for red lesions.

Retinal Lesions: correlation with DR grade

To assess the relevance of the annotations of pathological structures, we calculated the number of individual lesions per image, as well as their total area, and compared the distribution of both with respect to the severity of diabetic retinopathy (cf. Fig. 13 ). The general trend of these distributions matches the clinical intuition: the number of lesions per image and the area they cover increase with the severity of the pathology. Furthermore, the distribution of individual lesions is in line with the clinical definition of each grade. The number of microaneurysms, characteristic of low-severity grades, increases substantially between R0 and R1, and again between R1 and R2. For hemorrhages and exudates, whose count distinguishes the more severe stages, the most significant transitions are between stages R1 and R2, and between stages R3 and R4A. Similarly, the average surface area covered by the neovessels increases between stages R3 and R4A. Such transitions are absent in the cotton wool spots and drusen distributions, as there are too few of them in MAPLES-DR to observe any trends (<0.01% of total pixels).

Distribution of the number of individual lesions and their total area per image for each type of pathological structure, grouped by DR severity.

Retinal structures: semantic segmentation baseline

The learnability of MAPLES-DR labels of retinal structures was tested on a semantic segmentation task by training a simple UNet model to jointly segment them all as a multiclass map. We used a straightforward training protocol: the model was trained with a learning rate of 0.003, using the Dice coefficient as the loss function and stochastic gradient descent (SGD) as the optimizer. To accelerate the training, the encoder’s initial weights were pretrained on ImageNet.

The semantic segmentation performances of the model trained on the MAPLES-DR training set and tested on its test set are summarized in Fig. 14 . The model was able to successfully segment the four anatomical structures, as well as exudates, microaneurysms, and hemorrhages. Segmentation performances for these last two classes may appear low but are, in fact, comparable to the scores obtained when learning to segment these lesions with other public databases. Actually, the average segmentation precision of microaneurysms of a model trained and tested on MAPLES-DR – where microaneurysms are particularly well represented – is significantly higher (56%, see Fig. 14(a) ) than that obtained with the model that won the IDRiD challenge (50.2%, see Table 2 ). Unfortunately, not all lesions provided in MAPLES-DR are as easily learnable. Our simple model was unable to provide satisfactory segmentation for CWS, drusen, and neovessels. Indeed, because MAPLES-DR contains few images of severe or proliferative DR, examples of these lesions are rare in its training set and even rarer in its test set.

Summary of the semantic segmentation performances of a simple UNet model trained to segment all retinal structures available in MAPLES-DR.

Retinal lesions segmentation: training on MAPLES-DR, testing on public datasets

To put MAPLES-DR’s relevance to model learning in perspective with other popular public datasets, we trained five models to segment retinal lesions (i.e. Micro-Aneurysms, Hemorrhages, Exudates and CWS) on MAPLES-DR, IDRiD 25 , DDR 26 , FGADR 23 , and Retinal Lesions 24 . All models were then evaluated on the individual test sets associated with each dataset. The measured mIoUs for every combination of train and test sets are compiled in Table 1 .

For each test set, the model that achieves the best performance is the one trained on the corresponding train set. This result was expected given the discrepancies among the characteristics of the databases with respect to fundus images (ethnicity, image quality and resolution, colorimetry, etc) and labeling style (coarse segmentations for Retinal Lesions, precise ones for MAPLES-DR, IDRiD, and DDR and a mixture of them for FGADR) 39 . In spite of those differences, models trained on one dataset can, to varying degrees, generalize to the others. We measured this ability by ranking the models by mIoU on each test set and then averaging those ranks. The model trained on MAPLES-DR achieves the second-best average rank of 2.8, behind IDRiD, whose average rank is 2.2. Interestingly, those were the two models trained with the smallest numbers of training samples, but on datasets containing only good-quality images and labeled with precise segmentations of lesions.

Retinal lesions segmentation: pretraining with MAPLES-DR

As a final validation step, we illustrate how MAPLES-DR can be used in conjunction with another existing public dataset to improve segmentation performance. Because of the aforementioned distribution misalignments between datasets, it is often not relevant to simply combine training sets, and transfer learning is generally a better choice to take advantage of features learned on one dataset to improve training on another. We tested the benefit of MAPLES-DR in this context by training two simple UNet models on the IDRiD retinal lesions dataset. While both models were initialized with weights pre-trained on ImageNet, one was also pretrained on the MAPLES-DR training set before being fine-tuned on IDRiD. In both cases, we used the same basic training protocol as in the previous section. The resulting average precision scores for the segmentation of microaneurysms, hemorrhages, exudates, and cotton wool spots are presented in Table 2 , along with the best model from the IDRiD competition leaderboard.

While pretraining with MAPLES-DR did not have much effect on microaneurysms and exudates, it significantly improved the segmentation of hemorrhages and cotton wool spots by a margin of 11%. For these two lesions, the model pre-trained on MAPLES-DR even outperformed the winning submission of the IDRiD challenge. Such performances are remarkable considering the rudimentary training method we used, compared to the arsenal of performance optimization techniques usually involved in challenge submissions: k-fold cross-validation, model ensembles, sophisticated regularizations, extensive hyperparameter tuning, etc. All these techniques would very likely significantly improve the performances we obtained, but are outside the scope of this study. Still, even without them, a simple UNet pre-trained on MAPLES-DR and finetuned on IDRiD equals the IDRiD challenge winner according to the competition ranking metric (i.e. the mean over all four lesions of the average precision score).

Usage Notes

The fundus images corresponding to the diagnostic labels and biomarker segmentation maps of MAPLES-DR are the property of the MESSIDOR consortium and are freely available from the Consortium’s website . Users must be careful to download the MESSIDOR original dataset, not MESSIDOR-2 (MAPLES-DR uses several images that were removed in MESSIDOR-2). For more information on the download procedure, refer to the dedicated page of the MAPLES-DR documentation .

The pathological segmentation structures present in MAPLES-DR are unevenly represented: while microaneurysms, hemorrhages and exudates are numerous, this is not the case for CWS, drusens, and neovessels (see Fig. 13 ). For the latter biomarkers, MAPLES-DR might be more appropriate as a pre-training set (cf. Section pre-training with MAPLES-DR) rather than as a training or test set. In addition, to identify neovascularization examples in MAPLES-DR images, one can rely on a DR grade of R4A or on the annotating clinician’s comments pertaining to vessels.

We also wish to draw the reader’s attention to the non-negligible number of bright lesions annotated as BrightUncertains when they were clearly pathological but difficult to classify as drusen, exudates or CWS. We recommend including them in the training and testing sets of models that segment or detect bright lesions without classifying them. For image-specific information on bright lesions annotated as uncertain, please consult the comments left by the annotators.

Finally, the MAPLES-DR segmentation maps were padded and resized to heterogeneous sizes and resolutions to match the original MESSIDOR fundus images. However, libraries specialized in training segmentation models commonly expect a standardized image size. We advise researchers who wish to use MAPLES-DR in such a context to crop and resize MESSIDOR fundus images with the same ROI we used to annotate them. We realize that such operations require a certain amount of boiler plate code; therefore, we have published a Python library, maples_dr , which automates downloading the MAPLES-DR labels, and extracting and preprocessing the MESSIDOR fundus images to a uniform image resolution. See the Code Availability section for more information.

Code availability

MAPLES-DR annotations must be paired with MESSIDOR fundus images to be exploited to their full potential. To simplify the joint usage of both datasets, we developed a python package named maples_dr , which is available on pipy and GitHub . The key features of this library are: loading and exporting the fundus images with their annotations in formats suited for machine learning applications, preprocessing the fundus images, and providing a pythonic access to all MAPLES-DR additional data (retinologist comments, annotation times, preannotations, etc). Quick-start and code examples are available in the maples_dr documentation , along with an API reference. The repository also contains the code that produced the figures in this paper.

The algorithms used to generate the preannotation maps cannot be made public as they were implemented with the Theano library, which is no longer maintained. For transparency purposes, the original preannotations are available in the AdditionalData.zip archive of MAPLES-DR. However, we have released up-to-date Python segmentation models (trained with MAPLES-DR and other public datasets) in two Github repositories: gabriel-lepetitaimon/fundus-vessels-toolkit provides automatic segmentation and graph extraction of the retinal vasculature; ClementPla/fundus-lesions-toolkit provides automatic segmentation and visualization of retinal lesions.

Finally the annotation platform code and documentation are freely available to any research team that wishes to use it in the Github repository: LIV4D/AnnotationPlatform .

Lanzetta, P. et al . Fundamental principles of an effective diabetic retinopathy screening program. Acta Diabetologica 57 , 785–798, https://doi.org/10.1007/s00592-020-01506-8 (2020).

Article PubMed PubMed Central Google Scholar

Cheng, A. Y. Y. Canadian Diabetes Association 2013 clinical practice guidelines for the prevention and management of diabetes in Canada. Canadian Journal of Diabetes 37 , S1–S3 (2013).

Article ADS PubMed Google Scholar

Hooper, P. et al . Excerpt from the Canadian Ophthalmological Society evidence-based clinical practice guidelines for the management of diabetic retinopathy. Canadian Journal of Ophthalmology 52 , S45–S74, https://doi.org/10.1016/j.jcjo.2017.09.027 (2017).

Article PubMed Google Scholar

Egunsola, O. et al . Diabetic retinopathy screening: A systematic review of qualitative literature. Canadian Journal of Diabetes 45 , 725–733.e12, https://doi.org/10.1016/j.jcjd.2021.01.014 (2021).

Avidor, D., Loewenstein, A., Waisbourd, M. & Nutman, A. Cost-effectiveness of diabetic retinopathy screening programs using telemedicine: a systematic review. Cost Effectiveness and Resource Allocation 18 , https://doi.org/10.1186/s12962-020-00211-1 (2020).

Raman, R. et al . Fundus photograph-based deep learning algorithms in detecting diabetic retinopathy. Eye 33 , 97–109, https://doi.org/10.1038/s41433-018-0269-y (2019).

Dugas, E., Jorge, J. & Cukierski, W. Diabetic retinopathy detection. https://kaggle.com/competitions/diabetic-retinopathy-detection (2015).

Decencière, E. et al . Feedback on a publicly distributed image database: the MESSIDOR database. Image Analysis & Stereology 33 , 231–234, https://doi.org/10.5566/ias.1155 (2014).

Article Google Scholar

Gulshan, V. et al . Performance of a deep-learning algorithm vs manual grading for detecting diabetic retinopathy in India. JAMA Ophthalmology 137 , 987, https://doi.org/10.1001/jamaophthalmol.2019.2004 (2019).

Phillips, P. J., Hahn, C. A., Fontana, P. C., Broniatowski, D. A. & Przybocki, M. A. Four principles of explainable artificial intelligence. Tech. Rep. NISTIR 8312, National Institute of Standards and Technology, Gaithersburg, Maryland https://doi.org/10.6028/NIST.IR.8312 (2020).

Wilkinson, C. P. et al . Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales. Ophthalmology 110 , 1677–1682, https://doi.org/10.1016/S0161-6420(03)00475-5 (2003).

Article CAS PubMed Google Scholar

Zachariah, S., Wykes, W. & Yorston, D. Grading diabetic retinopathy (DR) using the Scottish grading protocol. Community Eye Health 28 , 72 (2015).

PubMed PubMed Central Google Scholar

Boucher, M. et al . Evidence-based Canadian guidelines for tele-retina screening for diabetic retinopathy: recommendations from the Canadian Retina Research Network (CR2N) Tele-Retina Steering Committee. Canadian Journal of Ophthalmology 55 , 14–24, https://doi.org/10.1016/j.jcjo.2020.01.001 (2020).

Staal, J., Abràmoff, M. D., Niemeijer, M., Viergever, M. A. & Van Ginneken, B. Ridge-based vessel segmentation in color images of the retina. IEEE Transactions on Medical Imaging 23 , 501–509, https://doi.org/10.1109/TMI.2004.825627 (2004).

Budai, A. et al . Robust vessel segmentation in fundus images. International Journal of Biomedical Imaging 2013 https://doi.org/10.1155/2013/154860 (2013).

Jin, K. et al . Fives: A fundus image dataset for artificial intelligence based vessel segmentation. Scientific Data 9 , 475, https://doi.org/10.1038/s41597-022-01564-3 (2022).

Lyu, X., Cheng, L. & Zhang, S. The reta benchmark for retinal vascular tree analysis. Scientific Data 9 , 397, https://doi.org/10.1038/s41597-022-01507-y (2022).

Orlando, J. I. et al . Refuge challenge: A unified framework for evaluating automated methods for glaucoma assessment from fundus photographs. Medical Image Analysis 59 , 101570 (2020).

Kovalyk, O. et al . Papila: Dataset with fundus images and clinical data of both eyes of the same patient for glaucoma assessment. Scientific Data 9 , 291, https://doi.org/10.1038/s41597-022-01388-1 (2022).

Kumar, J. H. et al . Chákṣu: A glaucoma specific fundus image database. Scientific Data 10 , 70, https://doi.org/10.1038/s41597-023-01943-4 (2023).

Lin, L. et al . The SUSTech-SYSU dataset for automated exudate detection and diabetic retinopathy grading. Scientific Data 7 , https://doi.org/10.1038/s41597-020-00755-0 (2020).

Decencière, E. et al . Teleophta: Machine learning and image processing methods for teleophthalmology. IRBM 34 , 196 – 203, https://doi.org/10.1016/j.irbm.2013.01.010 (2013). Special issue : ANR TECSAN : Technologies for Health and Autonomy.

Zhou, Y., Wang, B., Huang, L., Cui, S. & Shao, L. A benchmark for studying diabetic retinopathy: segmentation, grading, and transferability. IEEE Transactions on Medical Imaging 40 , 818–828, https://doi.org/10.1109/TMI.2020.3037771 (2020).

Wei, Q. et al . Learn to segment retinal lesions and beyond. In 2020 25th International Conference on Pattern Recognition (ICPR) , 7403–7410 https://doi.org/10.1109/ICPR48806.2021.9412088 (IEEE, 2021).

Porwal, P. et al . Indian diabetic retinopathy image dataset (IDRiD): a database for diabetic retinopathy screening research. Data 3 , 25, https://doi.org/10.3390/data3030025 (2018).

Li, T. et al . Diagnostic assessment of deep learning algorithms for diabetic retinopathy screening. Information Sciences 501 , 511–522, https://doi.org/10.1016/j.ins.2019.06.011 (2019).

Abràmoff, M. D. et al . Automated Analysis of Retinal Images for Detection of Referable Diabetic Retinopathy. JAMA Ophthalmology 131 , 351–357, https://doi.org/10.1001/jamaophthalmol.2013.1743 (2013). https://jamanetwork.com/journals/jamaophthalmology/articlepdf/1668203/ecs120076_351_357.pdf .

Ruamviboonsuk, P., Teerasuwanajak, K., Tiensuwan, M. & Yuttitham, K. Interobserver agreement in the interpretation of single-field digital fundus images for diabetic retinopathy screening. Ophthalmology 113 , 826–832, https://doi.org/10.1016/j.ophtha.2005.11.021 (2006).

Bragge, P., Gruen, R. L., Chau, M., Forbes, A. & Taylor, H. R. Screening for presence or absence of diabetic retinopathy: a meta-analysis. Archives of Ophthalmology 129 , 435–444, https://doi.org/10.1001/archophthalmol.2010.319 (2011).

Shi, L. et al . Telemedicine for detecting diabetic retinopathy: a systematic review and meta-analysis. British Journal of Ophthalmology 99 , 823–831, https://doi.org/10.1136/bjophthalmol-2014-305631 (2015).

Krause, J. et al . Grader variability and the importance of reference standards for evaluating machine learning models for diabetic retinopathy. Ophthalmology 125 , 1264–1272, https://doi.org/10.1016/j.ophtha.2018.01.034 (2018).

Teoh, C. S. et al . Variability in grading diabetic retinopathy using retinal photography and its comparison with an automated deep learning diabetic retinopathy screening software. Healthcare 11 , 1697, https://doi.org/10.3390/healthcare11121697 (2023).

Sasongko, M. B. et al . Retinal vascular tortuosity in persons with diabetes and diabetic retinopathy. Diabetologia 54 , 2409–2416, https://doi.org/10.1007/s00125-011-2200-y (2011).

Playout, C., Duval, R. & Cheriet, F. A novel weakly supervised multitask architecture for retinal lesions segmentation on fundus images. IEEE Transactions on Medical Imaging 38 , 2434–2444, https://doi.org/10.1109/TMI.2019.2906319 (2019).

Lepetit-Aimon, G., Duval, R. & Cheriet, F. Large receptive field fully convolutional network for semantic segmentation of retinal vasculature in fundus images. In Computational Pathology and Ophthalmic Medical Image Analysis , 201–209 (Springer, https://doi.org/10.1007/978-3-030-00949-6_24 2018).

Rädsch, T. et al . Labelling instructions matter in biomedical image analysis. Nature Machine Intelligence 5 , 273–283, https://doi.org/10.1038/s42256-023-00625-5 (2023).

Lepetit-Aimon, G., Playout, C. & Cheriet, F. MAPLES-DR: MESSIDOR Anatomical and Pathological Labels for Explainable Screening of Diabetic Retinopathy. figshare https://doi.org/10.6084/m9.figshare.24328660 (2023).

Sechidis, K., Tsoumakas, G. & Vlahavas, I. On the stratification of multi-label data. In Machine Learning and Knowledge Discovery in Databases , 145–158, https://doi.org/10.1007/978-3-642-23808-6_10 (Springer, Berlin, Heidelberg, 2011).

Playout, C. & Cheriet, F. Cross-dataset generalization for retinal lesions segmentation. arXiv preprint arXiv:2405.08329 https://doi.org/10.48550/arXiv.2405.08329 (2024).

Cao, J., Felfeli, T., Merritt, R. & Brent, M. H. Sociodemographics associated with risk of diabetic retinopathy detected by tele-ophthalmology: 5-year results of the Toronto tele-retinal screening program. Canadian Journal of Diabetes 46 , 26–31, https://doi.org/10.1016/j.jcjd.2021.05.001 (2022).

Download references

Acknowledgements

This study was funded by the Natural Science and Engineering Research Council of Canada as well as Diabetes Action Canada and FROUM (Fonds de recherche en ophtalmologie de l’Université de Montréal). The original MESSIDOR dataset was kindly provided by the Messidor program partners (see https://www.adcis.net/en/third-party/messidor/ . The authors would like to thank Dr. Marie Carole Boucher, Dr. Michael H Brent, Dr. Renaud Duval as well as Dr. Karim Hammamji, Dr. Ananda Kalevar, Dr. Cynthia Qian, and Dr. David Wong for their time and effort labeling the MAPLES-DR dataset. We also thank Dr. Fares Antaky and Dr. Daniel Milad for participating in a inter-observer variability study that helped us assess the quality of lesions segmentations of MAPLES-DR. We thank Philippe Debanné for editing this manuscript and Emmanuelle Richer, Zacharie Legault, and Fantin Girard for their valuable input on the technical validation and figures.

Author information

Authors and affiliations.

Department of Computer and Software Engineering, Polytechnique Montréal, Montréal, QC, Canada

Gabriel Lepetit-Aimon & Farida Cheriet

Department of Ophthalmology, Université de Montréal, Montréal, Canada

Clément Playout, Marie Carole Boucher & Renaud Duval

Centre Universitaire d’Ophtalmologie, Hôpital Maisonneuve-Rosemont, Montréal, Canada

Department of Ophthalmology and Vision Science, University of Toronto, Toronto, Canada

Michael H. Brent

You can also search for this author in PubMed Google Scholar

Contributions

M.C.B. and F.C. conceptualized and designed the study. G.L.A. and C.P. designed and supervised the development of the annotation plateform. M.C.B. and R.D. and M.B. annotated the dataset. G.L.A. supervised the annotation campaigns, documented its procedure, and finally exported and analyzed the dataset. C.P. performed and analyzed the trainability experiments. G.L.A. drafted the initial manuscript. All authors reviewed and approved the final manuscript as submitted.

Corresponding author

Correspondence to Gabriel Lepetit-Aimon .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ .

Reprints and permissions

About this article

Cite this article.

Lepetit-Aimon, G., Playout, C., Boucher, M.C. et al. MAPLES-DR: MESSIDOR Anatomical and Pathological Labels for Explainable Screening of Diabetic Retinopathy. Sci Data 11 , 914 (2024). https://doi.org/10.1038/s41597-024-03739-6

Download citation

Received : 20 December 2023

Accepted : 05 August 2024

Published : 23 August 2024

DOI : https://doi.org/10.1038/s41597-024-03739-6

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Quick links

Explore articles by subject
Guide to authors
Editorial policies

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Publications
Account settings
My Bibliography
Collections
Citation manager

Save citation to file

Email citation, add to collections.

Create a new collection
Add to an existing collection

Add to My Bibliography

Your saved search, create a file for external citation management software, your rss feed.

Search in PubMed
Search in NLM Catalog
Add to Search

Evaluating Ocular Blood Flow in Diabetic Macular Edema using Three-dimensional Pseudocontinuous Arterial Spin Labeling

Affiliations.

1 Department of Ophthalmology, Beijing Friendship Hospital, Capital Medical University, Beijing 100050, China.
2 Department of Radiology, Beijing Chaoyang Hospital, Capital Medical University, Beijing 100020, China.
PMID: 39185661
DOI: 10.2174/0115734056307305240403060707

Background: Alterations in ocular blood flow play an important role in the pathogenesis of diabetic macular edema; however, this remains unclear.

Objectives: This study aimed to investigate ocular blood flow in eyes with or without diabetic macular edema using arterial spin labeling.

Methods: This cross-sectional study included 118 eyes of 65 patients with diabetic retinopathy analyzed between November 2018 and December 2019. We included a total of 53 eyes without diabetic macular edema (mean [SD] age, 57.83 [7.23] years; 29 men [54.7%]) and 65 eyes with diabetic macular edema (mean [SD] age, 60.11 [7.63] years; 38 men [58.5%]). Using a 3.0-T magnetic resonance imaging, participants were imaged with arterial spin labeling with multiple post-labeling delays.

Results: The mean ocular blood flow at post-labeling delays of 1.5 and 2.5 s was significantly lower in eyes with diabetic macular edema among patients with diabetic retinopathy compared with the remaining subgroups (P=0.022 and P <0.001, respectively). The mean ocular blood flow exhibited a significant decrease in eyes with diabetic macular edema when the post-labeling delay was set at 2.5 s in the nonproliferative and proliferative diabetic retinopathy groups, compared with the remaining subgroups (P=0.005 and P=0.002, respectively). The cutoff points of ocular blood flow at post-labeling delays of 1.5 s and 2.5 s were 9.40 and 11.10 mL/100 g/min, respectively.

Conclusion: Three-dimensional pseudocontinuous arterial spin labeling can identify differences in the ocular blood flow of patients with diabetic eyes with and without diabetic macular edema.

Keywords: Arterial spin labeling; Diabetes.; Diabetic macular edema; Magnetic resonance imaging; Ocular blood flow; Retinopathy.

PubMed Disclaimer

Search in MeSH

Related information

Linkout - more resources.

Genetic Alliance
MedlinePlus Health Information
Citation Manager

NCBI Literature Resources

MeSH PMC Bookshelf Disclaimer

The PubMed wordmark and PubMed logo are registered trademarks of the U.S. Department of Health and Human Services (HHS). Unauthorized use of these marks is strictly prohibited.

Get new issue alerts Get alerts
Submit a Manuscript

Secondary Logo

Journal logo.

Colleague's E-mail is Invalid

Your message has been successfully sent to your colleague.

Save my selection

Diabetic macular edema

Evidence-based management.

Browning, David J; Stewart, Michael W 1, ; Lee, Chong

Charlotte Eye, Ear, Nose, and Throat Associates, Charlotte, North Carolina, USA

1 Department of Ophthalmology, Mayo Clinic, Jacksonville, Florida, USA

Correspondence to: Dr. Michael W Stewart, Department of Ophthalmology, Mayo Clinic, Jacksonville, Florida 32224, USA. E-mail: [email protected]

Received July 23, 2018

Accepted August 03, 2018

Diabetic macular edema (DME) is the most common cause of vision loss in patients with diabetic retinopathy with an increasing prevalence tied to the global epidemic in type 2 diabetes mellitus. Its pathophysiology starts with decreased retinal oxygen tension that manifests as retinal capillary hyperpermeability and increased intravascular pressure mediated by vascular endothelial growth factor (VEGF) upregulation and retinal vascular autoregulation, respectively. Spectral domain optical coherence tomography (SD-OCT) is the cornerstone of clinical assessment of DME. The foundation of treatment is metabolic control of hyperglycemia and blood pressure. Specific ophthalmic treatments include intravitreal anti-VEGF drug injections, intravitreal corticosteroid injections, focal laser photocoagulation, and vitrectomy, but a substantial fraction of eyes respond incompletely to all of these modalities resulting in visual loss and disordered retinal structure and vasculature visible on SD-OCT and OCT angiography. Efforts to close the gap between the results of interventions within randomized clinical trials and in real-world contexts, and to reduce the cost of care increasingly occupy innovation in the social organization of ophthalmic care of DME. Pharmacologic research is exploring other biochemical pathways involved in retinal vascular homeostasis that may provide new points of intervention effective in those cases unresponsive to current treatments.

Epidemiology and Risk Factors

Diabetic macular edema (DME) is the most common cause of visual loss in those with diabetic retinopathy and is increasing in prevalence globally.[ 1 2 3 ] The prevalence of DME in patients with diabetic retinopathy is 2.7%–11%[ 4 5 6 7 8 ] and it depends on the type of diabetes and the duration of the disease, but for both types 1 and 2 after 25-years duration, it approximates 30%. Systemic factors associated with DME include longer duration of diabetes, higher systolic blood pressure, and higher hemoglobin A1C. The sole ocular factor associated with DME is diabetic retinopathy severity as increasing severity is associated with increasing prevalence of DME.[ 9 10 11 ]

Genetics, Pathoanatomy, and Pathophysiology

The hypothesis that genetic risk and protective alleles exist for development of DME has not been tested with genome wide association studies of adequate size, but studies are underway.[ 12 ]

The capillaries in the macula are distributed in four strata within the inner retina with the exception of the single-level arrangement bordering the foveal avascular zone within the ganglion cell layer.[ 13 ] Farther from the fovea, the three additional levels of capillaries are found within the deep ganglion cell layer, inner plexiform layer/superficial inner nuclear layer, and deep inner nuclear layer, respectively.[ 14 ] These strata can be imaged by optical coherence tomography (OCT).[ 15 ]

Eighty percent of diabetes-related microaneurysms originate in the inner nuclear layer and its border zones[ 16 ] and are commonly found on the edges of nonperfused retina. Microaneurysms in DME do not preferentially cluster in any particular quadrant.[ 17 ] In DME, spectral domain (SD)-OCT angiography has shown microaneurysms and abnormal deep capillary networking in the superficial outer nuclear layer, a normally avascular zone.[ 18 ]

In center-involved diabetic macular edema (CIDME), the central macula is often thickest, an inversion of the normal morphology. In the foveal avascular zone, the only mechanism for extracellular fluid resorption is the retinal pigment epithelial (RPE) pump, which may explain the greater accumulation of edema fluid at this location.[ 19 ] An associated fundus sign is the appearance of the macular lipid star [ Fig. 1 ]. In 15%–30% of cases of DME, a subfoveal serous retinal detachment is present. Although the explanation for the subfoveal location of fluid is conjectural, one theory posits an impaired RPE pump due to decreased subfoveal choroidal circulation.[ 20 21 ]

Breakdown of the inner blood–retina barrier rather than outer blood–retina barrier breakdown is more important to the formation of DME.[ 22 ] Diabetes causes a redistribution of occludin in retinal vascular endothelium.[ 23 ] The Muller cells proliferate in epiretinal membranes that exert traction on microvessels and increase their permeability. Astrocytes, which wrap their end feet around microvessels, decrease their production of glial fibrillary acidic protein in diabetes, which may alter the blood–retina barrier.[ 23 ]

Diabetes-related abnormalities of the vitreoretinal interface may promote the development DME. During the process of vitreous separation, the macula and the disk may adhere to the posterior hyaloid more firmly, and traction may contribute to blood retinal barrier breakdown and provide a scaffold for cellular proliferation, which further increases traction on the macula.[ 24 25 26 ] In eyes with DME, the internal limiting membrane has more adherent cellular elements on its vitreous side, is thicker, and has more heparin sulfate proteoglycan compared with the internal limiting membrane from nondiabetic eyes. Fibromuscular cells found in epiretinal blocks of tissue biopsied at the time of vitrectomy for DME may be the basis for tangential traction on the retina with concomitant increases in capillary permeability.[ 25 26 ]

In DME, the macula is thickened due to increased extracellular fluid derived from hyperpermeable retinal capillaries.[ 27 ] Prolonged hyperglycemia leads to reduced inner retinal oxygen tension, venous dilation, increased VEGF concentration within the retina, leukocyte stasis, and dysregulated growth factor levels, which together are associated with increased exudation of serum out of the retinal vasculature and into the extracellular space.[ 28 29 ] The RPE pump is overwhelmed by the exudation of serum and macular swelling results.[ 30 31 ]

A framework for understanding the pathophysiology of diabetic macular edema (DME) is the oxygen theory.[ 32 ] Prolonged periods of hyperglycemia lead to reduced perfusion of the inner retina and decreased inner retinal oxygen tension. The autoregulatory response of the retinal arterioles causes dilation, which leads to increased hydrostatic pressure in the intraretinal capillaries and venules as specified by Poiseuille's law.[ 31 ] The elevated intravascular pressure experienced by the capillaries may damage them.[ 31 32 ] Concomitantly, the decrease in retinal oxygen tension upregulates the synthesis of VEGF and other permeability factors, which increases microvasculature leakage. According to Starling's law, increased intravascular pressure and vascular permeability result in a net flow of water, ions, and macromolecules from the intravascular space into the extravascular space. Extracellular fluid is resorbed by re-entering the retinal vessels further downstream or through the choroid via the pumping action of the RPE.[ 32 33 ]

Many variables are suspected to modulate this process. The duration of diabetes and the integrated elevation of blood glucose reflected in the glycated hemoglobin (HbA1C) have proven pathophysiological importance. Retinal neurons and glial cells increase their production of VEGF, even before ophthalmoscopic evidence of capillary loss, associated with reduced occludin in capillary endothelial tight junctions.[ 23 34 ] Increased inflammation, characterized by leukostasis, accumulation of macrophages, intercellular adhesion molecule-1 activation (ICAM-1), and prostacyclin upregulation, is associated with capillary nonperfusion and breakdown of the blood–retina barrier.[ 29 35 ] Patients with DME have elevated vitreous levels of VEGF, ICAM-1, interleukin-6 (IL-6), and monocyte chemoattractant protein-1 compared with nondiabetic patients.[ 36 ] Inflammatory cytokines such as tumor necrosis factors alpha and beta, alpha 4 integrin, nitric oxide, and interleukin-1β mediate vascular permeability.[ 23 37 38 39 ] Many other small molecules and growth factors may contribute to the development of DME, although the details of the pertinent pathways are incompletely understood.[ 37 38 40 41 ] High lipid levels may cause endothelial dysfunction and increased vascular permeability through a local inflammatory response and higher levels of advanced glycation end products.[ 42 ] In addition to extracellular edema, intracellular edema may be relevant for DME. Dysregulated metabolism is associated with accumulation of intracellular osmotically active solutes that draw in water and cause cellular swelling.[ 31 ]

Decrease in subfoveal choroidal blood flow in type 2 diabetic patients with retinopathy may be relevant in the pathophysiology of DME. Eyes with DME have been reported to have a greater decrease in choroidal blood flow than eyes without DME, suggesting relative hypoxia of the RPE and outer retina, and consequent increased permeability of the outer blood retinal barrier.[ 20 ]

The vitreous may play a role in the pathogenesis of DME. Cross-linking and protein glycation are increased in the diabetic vitreous, which may explain instances of tangential macular traction that may induce DME.[ 43 44 ] Besides the direct effect of traction causing leakage from blood vessels or macular elevation with subretinal fluid, vitreous adherent to the macula may loculate mediators of vessel permeability in proximity to macular capillaries and may impede oxygenation of the retina, thereby causing venous dilation and increased edema via Starling's law or by upregulation of VEGF.[ 33 45 46 47 48 49 ]

This account of the pathophysiology of DME informs an understanding of how treatments for DME work [ Fig. 2 ]. Grid laser increases the oxygenation of the inner retina both by reducing the number of oxygen-consuming photoreceptors and by shortening the diffusion pathway to the inner retina for oxygen originating in the choroid.[ 32 50 ] Focal photocoagulation presumably works by destroying leakage sources such as microaneurysms but may also improve RPE pumping of sodium ions and water outward toward the choroid.[ 32 51 52 ] Anti-VEGF drugs work by blocking the permeability inducing effects of VEGF.[ 53 ] Corticosteroids reduce expression of the VEGF gene, differentially regulate expression of the various VEGF receptors, and have other non-VEGF-mediated effects such as decreasing leukocyte recruitment and production of ICAM-1, and inhibiting collagenase induction that reduce the permeability of retinal microvessels.[ 54 55 56 57 58 59 ] Vitrectomy may work by increasing intravitreal and secondarily inner retinal oxygen levels, leading to downregulation of VEGF synthesis, which decreases the permeability of microvessels.[ 32 60 ] In addition, vitrectomy may open compartments of loculated cytokines and relieve traction exerted on the macula by an altered vitreous.[ 60 61 ]

Clinical Definitions

Retinal thickening within one disk diameter of the center of the macula or definite hard exudates in this region.[ 62 ]

Center-involved diabetic macular edema

DME in which the fovea is involved.

Clinically significant macular edema

The situation in which at least one of the following criteria is fulfilled:

Retinal thickening within 500 μm of the center of the macula
Hard exudates within 500 μm of the center of the macula with adjacent retinal thickening
One disk area of retinal thickening any part of which is within one disk diameter of the center of the macula.[ 63 ]

Focal and diffuse diabetic macular edema

These two terms have not been defined consistently in the literature.[ 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 ] Focal edema is said to arise from microaneurysms, whereas diffuse edema is said to arise from generally dilated and hyperpermeable capillaries throughout the macula.[ 83 ] Focal DME has been reported to be more common than diffuse DME, but many cases of DME have mixed features making a clear distinction difficult.[ 51 80 84 85 86 87 ] Additional confusion may ensue because the term focal is used to describe a technique of applying laser directly to microaneurysms when treating DME with focal/grid photocoagulation.[ 63 88 ]

Other classifications of DME have been proposed. One scheme differentiates diffuse edema, cystoid macular edema, and serous retinal detachment based on OCT.[ 89 ] Attempts to correlate these subgroups to treatment outcomes have yielded inconsistent results, no consensus guidance exists on interventions for the proposed subtypes, and no DRCR network study has shown these DME classifications to help in clinical decision-making.[ 90 ]

Subclinical diabetic macular edema (SCDME)

The severity of DME may not reach the definition of CSME or CIDME. Clinical assessment of macular edema and OCT assessment of macular edema frequently disagree in this group of patients.[ 91 92 ] In addition, some eyes do not have clinically recognized DME, but macular thickening is detectable by OCT.[ 4 ] The term subclinical DME has been used to define both of these classes of DME that are less severe than clinically significant DME.[ 92 93 ]

Persistent diabetic macular edema

DME that has been treated without complete resolution is defined as persistent.[ 94 95 96 97 ] Persistent DME has been noted in a proportion of eyes treated by any modality, including focal laser photocoagulation, intravitreal injection of anti-VEGF drugs or corticosteroids, and vitrectomy. Different criteria have been applied for the number of treatments or duration of treatment required before applying the term. Some eyes have persistent edema despite all known treatments for DME.

Recurrent diabetic macular edema

Many cases exist in which DME resolves after treatment, but subsequently recurs.[ 98 99 ] Although DME can resolve spontaneously without treatment, and then recur, the term recurrent DME is used with reference to treated eyes with recurrences.

Methods of detection of DME

DME can be detected by stereoscopic slit-lamp examination using a fundus lens.[ 51 63 100 ] Direct ophthalmoscopy allows detection of lipid exudates but lacks stereopsis. Although lipid suggests associated macular thickening, the two findings are not synonymous; presence of lipid alone is an unreliable surrogate for DME.[ 101 102 ] Stereoscopic fundus photographs and fluorescein angiography can be used to assess the presence of DME but have been largely supplanted by OCT.[ 24 103 104 105 ]

The importance of OCT in diagnosing and managing DME cannot be overemphasized. The clinical diagnosis of DME as practiced in the Early Treatment Diabetic Retinopathy Study (ETDRS) era before OCT was beset by variability among clinicians.[ 51 106 ] In contrast, measurements made with OCT are highly reproducible.[ 107 108 109 110 111 ] In general, any change of macular thickness greater than 11% of a previous measurement exceeds OCT measurement variability and can be assumed to be a real change in macular thickness.[ 109 ] In addition to measurement variability, there is short-term fluctuation in macular thickness in DME. This refers to the variability noted over the course of days to even weeks when there is no trend in the changes.[ 112 ] Short-term fluctuation in DME is dependent on actual macular thickness and is larger than measurement variability.[ 113 ]

Of the many OCT indices that can be followed in the course of DME, the central subfield mean thickness (CST) is the best single measure.[ 114 115 ] It is more reproducible than center point thickness, yet is highly correlated ( r = 0.99) with the latter.[ 114 115 ] Total macular volume (TMV) correlates somewhat less well with CST ( r = 0.76), and there have been no conclusions drawn from analyzing TMV that would not have been drawn by studying CST instead.[ 94 104 ]

OCT was originally developed using time domain acquisition of images.[ 116 ] Subsequently instruments using spectral domain acquisition of images (SD-OCT) and swept-source OCT (SS-OCT) have been developed. SD-OCT and SS-OCT allow faster acquisition of images, denser sampling of the macula, and better imaging of the choroid and outer retina.[ 117 118 119 120 ] The normal values for SD-OCT and SS-OCT differ because the segmentation algorithms define the retina layers differently, and measurements are not interconvertible across instruments made by different companies.[ 118 119 121 ] The axial resolution of SD-OCT is 2–5 μm.[ 118 122 ] For the central subfield, the mean coefficient of variation of SD-OCT has been reported to be 0.66%.[ 118 ] The coefficient of repeatability for the central subfield thickness of SD-OCT is 5 μm.[ 123 ]

OCT is good for objectively measuring macular thickness, but macular thickening is only modestly correlated with visual acuity ( r = −0.52) perhaps due to variable duration of edema and ischemia.[ 23 124 ] Photoreceptor outer segment length, defined as the length between the ellipsoid zone and the RPE, and outer retinal layer thickness, defined as the length between the external limiting membrane (ELM) and the RPE, correlate better with visual acuity ( r = −0.81 and −0.65 to −0.8787, respectively).[ 125 126 127 ] Disorganization of the inner retinal layers (DRIL), defined as lack of definition of boundaries between ganglion cell-inner plexiform layer or inner-nuclear-outer-plexiform layers in ≥ 50% of the 1 mm central subfield, has been associated with worse visual acuity and less response to injections with bevacizumab or ranibizumab.[ 128 129 130 ] On average, each additional 100 μm of DRIL is associated with 6 ETDRS letters lost.[ 130 ]

Besides its usefulness in the detection of macular edema, OCT has value in following DME over time. SD-OCT provides enough detail regarding the outer retina that correlations of intactness of structures with visual outcomes are possible. Increased disruption of the ELM and ellipsoid zone (EZ) are associated with worse visual acuity outcomes.[ 131 132 ]

Natural History

The ETDRS provided natural history data regarding DME. Over 3 years of follow-up, the rate of moderate visual loss (15 letters on the ETDRS chart) was 8% per year.[ 63 ] Rates of visual loss increased according to the baseline visual acuity, with worse seeing eyes losing vision at a higher rate.[ 63 ] Rates of visual loss also increased according to baseline retinopathy severity, with eyes having more severe retinopathy losing vision at higher rates than eyes with less severe retinopathy.[ 63 ] Rates of visual acuity gain of at least 6 ETDRS letters in untreated eyes with DME and visual acuity of ≤ 20/40 over three years of follow-up were 20%–25%.[ 63 ] Of eyes with DME less severe than CSME (one subset of what has been termed subclinical DME) and observed without treatment, 22% and 25% progressed to CIDME at 1 and 3 years of follow-up, respectively.[ 63 ] In the OCT era, 31% of eyes with SCDME progressed to CSME over a median follow-up of 14 months.[ 93 ]

Chronic, untreated DME and refractory DME can lead to subretinal fibrosis, particularly if hard exudates are present, and by more subtle RPE pigmentary changes.[ 133 134 135 136 137 ]

Metabolic control and effects of drugs

Recognition of the risk factors for DME led to randomized clinical trials of better blood pressure control in attempts to reduce the prevalence of the condition. The Diabetes Control and Complications Trial showed that tight blood glucose control in patients with type 1 diabetes reduced the cumulative incidence of macular edema at 9-year follow-up by 29% and reduced the application of focal laser treatment for DME by half.[ 138 139 ] The United Kingdom Prospective Diabetes Study was an analogous randomized clinical trial of patients with type 2 diabetes. It showed that tighter blood glucose control reduced the requirement for laser treatment at 10 years by 29%, compared with looser control; 78% of the laser treatments were for DME.[ 140 ] It also showed that a mean systolic blood pressure reduction of 10 mm Hg and a diastolic blood pressure reduction of 5 mm Hg over a median follow-up of 8.4 years led to a 35% reduction in retinal laser treatments, of which 78% were for DME.[ 141 ]

Increased serum cholesterol levels are associated with increased severity and risk of retinal hard exudates.[ 142 143 ] Patients with abnormally elevated triglycerides and HDL cholesterol had worse visual acuity outcomes after focal/grid photocoagulation than did patients with normal levels in one small prospective study.[ 144 ]

Thiazolidinediones are oral agents used to treat type 2 diabetes. They are peroxisome proliferator-activated receptor γ agonists that work by enhancing insulin sensitivity. Pioglitazone and rosiglitazone are members of this class of drugs in common use. They have been associated with peripheral edema, pulmonary edema, and/or congestive heart failure, especially when used in combination with insulin. Plasma VEGF levels are higher in patients on thiazolidinediones than in patients not on these drugs.[ 145 ] Case reports and retrospective database cohort studies suggest that they can be associated with new or worsened DME as well, but in the ACCORD study, use of thiazolidinediones was not associated with prevalence of DME at baseline or incidence of DME over 4 years of follow-up.[ 146 147 ]

Improved control of diabetes, hypertension, and serum lipids is frequently underemphasized by the ophthalmologist because changes in systemic disease management are usually made by the internist, yet there is an intimate connection between these changes and retinal effects. A multifactorial intervention aimed at reducing glycated hemoglobin, elevated blood pressure, and elevated serum lipids can produce measurable effects in macular thickness in as little as 6 weeks and forms a rational foundation on which to apply specific ophthalmic interventions.[ 148 ]

Specific Ophthalmic Treatments

Focal/grid laser photocoagulation.

The ETDRS demonstrated superior visual outcomes with focal/grid laser for CSME compared with the natural history. Laser thus became the standard of care over the next 30 years.[ 63 ] Treatments were repeated at 4-month intervals if CSME persisted and treatable lesions or untreated, thickened, and nonperfused retina remained. The average patient received between three and four focal/grid laser treatments. ETDRS style focal/grid photocoagulation for DME has potential side effects including paracentral scotomas, subretinal fibrosis, and secondary choroidal neovascularization.[ 134 149 150 151 152 ]

The technique of focal/grid argon laser treatment has been modified over time. The most significant changes are embodied in the DRCR.net protocols that employ focal/grid photocoagulation. Rather than burns that can vary from 50 to 200 μm, all contemporary burns are 50 μm and they are less intense.[ 153 ] Yellow wavelength laser is acceptable in addition to green, but blue-green is not used because of concern over absorption by macular luteal pigment. Use of a guiding fluorescein angiogram is optional.[ 51 74 149 154 ] On average, for mild CIDME with CST in the range of 300–350 μm, one can expect that focal/grid laser will produce ~25 μm of macular thinning at the usual first follow-up interval of 3–4 months. For every 100 μm of additional baseline macular thickening above this threshold, one can expect that focal/grid laser will yield approximately 10 μm of additional macular thinning at the 3-to 4-month follow-up visit.[ 19 ] Visual acuity at this follow-up visit is, on average, unchanged from baseline.[ 154 155 156 157 ]

Subthreshold Laser Photocoagulation

Besides focal/grid suprathreshold laser treatment, diode laser micropulse laser has been used in case series and small randomized clinical trials.[ 158 159 160 ] Its advantages are absence of RPE scarring, no subsequent choroidal neovascularization, and elimination of paracentral visual field scotomas.[ 160 161 ] The disadvantages are that there is no visible endpoint for treatment, making it difficult to determine where treatment has and has not been given, and that there is no standardized, consensus set of treatment parameters or guidelines with respect to treatment within the foveal avascular zone. In addition, the reduction in macular edema after subthreshold laser photocoagulation occurs with a slower time course and more treatments are necessary to achieve elimination of edema.[ 160 ]

Intravitreal Injections of Corticosteroids

Corticosteroids were first used to treat DME in 2001.[ 162 ] Triamcinolone, dexamethasone, and fluocinolone have been used in many forms, including particulate suspensions, viscoelastic mixtures, and solid slow-release devices.[ 113 163 164 165 166 ] Many dosages and intervals between triamcinolone injections have been tried.[ 167 ] Although enthusiasm for serial intravitreal triamcinolone injections was initially high, protocol B of the DRCR network showed that focal laser led to superior visual acuity outcomes at 3 years relative to either triamcinolone 1 or 4 mg.[ 157 163 ] Since then, therapy with corticosteroids has taken a secondary role to anti-VEGF therapy. Side effects of cataract in phakic eyes and intraocular pressure elevation have accompanied all steroids studied, although to varying degrees.[ 157 168 ]

Slowly dissolving intravitreal dexamethasone implants (Ozurdex ® , Allergan, Irvine, CA, USA) are effective in treating DME although the visual acuity gains are generally less than with anti-VEGF injections.[ 90 169 ] In a 3-year randomized controlled trial, the 0.7 mg dexamethasone implant was associated with ≥ 15 letter improvement in best corrected visual acuity (BCVA) in 22.2% of patients compared to 12.0% in the sham group.[ 169 ] Over three years in phakic patients, 59.2% of eyes required cataract surgery; 41.5% of eyes required the use of ocular hypotensive therapy.[ 169 ] The long-term visual outcome of intravitreal dexamethasone implant therapy correlates with the 3-month treatment response.[ 170 ]

Intravitreal fluocinolone acetonide implants (Iluvien ® , Alimera, Alpharetta, GA, USA) last 3 years and, unlike the dexamethasone implant, do not dissolve. In the FAME trial, patients with persistent DME despite macular laser were randomized to low-dose (0.2 μg/day), high-dose (0.5 μg/day), or sham treatment. The percentage of eyes gaining at least 15 ETDRS letters at 24 months was 28.7% compared with 16.2% in the sham group. Cataract surgery was required in 74.9% of the low-dose fluocinolone group compared with 23.1% in the sham group. Glaucoma developed in 1.6% of eyes compared with 0.5% of sham eyes.[ 171 ]

Intravitreal Injections of Anti-VEGF Drugs

Anti-VEGF drugs include aptamers (pegaptanib), antibodies to VEGF (bevacizumab), antibody fragments to VEGF (ranibizumab), and fusion proteins, which combine a receptor for VEGF with the Fc fragment of an immunoglobulin (aflibercept and conbercept). The antibodies and fusion proteins bind all isoforms of VEGF-A; fusion proteins additionally bind VEGF-B and placental growth factor. Fusion proteins have higher affinity for VEGF and the potential for less frequent injection frequency in the treatment of DME.[ 172 173 ]

Bevacizumab (Avastin ® , Genentech, S. San Francisco, CA, USA/Roche, Basel, SW) is Food and Drug Administration (FDA)-approved for treatment of advanced solid cancers, but is widely used off-label in the treatment of DME. It is much less expensive than the FDA-approved ocular anti-VEGF drugs.[ 174 ] Ziv-aflibercept (Zaltrap ® , Regeneron, Tarrytown, NY, USA) is systemically formulated aflibercept in a buffered solution with a higher osmolarity (1,000 mOsm/L) than aflibercept.[ 175 ] In a rabbit model, intravitreal injection of ziv-aflibercept did not affect serum or intraocular osmolarity, and human studies are beginning to be published.[ 172 173 ]

The first anti-VEGF drug used to treat DME was pegaptanib (Macugen ® , Bausch and Lomb, Rochester, NY, USA), which selectively blocks the 165-isoform of VEGF.[ 176 ] Its promise was superseded by superior results obtained with anti-VEGF drugs that blocked all isoforms of VEGF. The efficacy of bevacizumab and ranibizumab were proven in randomized controlled clinical trials in 2010 and that of aflibercept in 2014.[ 177 178 179 ] A prospective, randomized, comparative effectiveness trial of these three drugs showed no difference in efficacy of the three drugs in eyes with center-involved DME and visual acuity of 20/40 or better at 1 or 2 years of follow-up.[ 174 ] However, in eyes with visual acuity of 20/50 or worse, aflibercept was superior to ranibizumab and bevacizumab at 1 year, whereas at 2 years, aflibercept was no longer superior to ranibizumab but remained superior to bevacizumab.[ 174 180 ] An example illustrating effectiveness of aflibercept, persistence of DME, and SD-OCT correlates of suboptimal visual acuity outcomes is shown in Fig. 3 .

Approaches aimed at increasing the intravitreal concentration of anti-VEGF agents have not proved beneficial. The READ-3 clinical trial examining two doses of ranibizumab (0.5 and 2.0 mg) in DME showed that at 2 years, the 0.5 mg dose was associated with a better visual outcome.[ 181 182 ] Focal laser added from the outset to anti-VEGF does not improve visual acuity outcomes relative to its use in a deferred manner if incomplete drying of the macula occurs with anti-VEGF therapy.[ 183 ] Randomized clinical trials demonstrate that these general results apply across various racial and ethnic groups.[ 174 184 ] As a result, in 2018, serial anti-VEGF intravitreal injection monotherapy is the standard of care for treating DME in developed countries.

Although serial injections of anti-VEGF drugs are first-line therapy for DME, some patients do not respond or respond incompletely. In the RISE and RIDE trials, persistent macular thickening was found in 20%–25% of patients.[ 178 ] A secondary analysis of protocol T comparing intravitreal aflibercept, bevacizumab, and ranibizumab for CI-DME found that persistent DME through 24 weeks was found in 31.6%, 65.6%, and 41.5% of eyes receiving aflibercept, bevacizumab, and ranibizumab, respectively.[ 97 ] Despite their incomplete responses, the visual acuity outcomes of eyes with chronic persistent DME were similar to those of eyes with complete resolution of edema.[ 97 ] Similar results were found in a secondary analysis of protocol I comparing intravitreal ranibizumab with prompt or deferred focal laser to intravitreal triamcinolone with prompt focal laser for CI-DME.[ 185 ]

A concomitant effect of anti-VEGF treatment for DME is improvement in retinopathy severity or slowing of the rate of progression of retinopathy. This effect has been noted with ranibizumab and aflibercept.[ 179 ] For aflibercept, there is an association between baseline retinopathy severity and proportion of patients achieving ≥ 2-step severity score improvement.[ 186 ] Another concomitant effect is thinning of the choroid.[ 187 188 189 ] In treatment naïve CIDME, 3–6 months of bevacizumab or ranibizumab was associated with choroidal thinning.[ 190 191 ]

No better results have been reported than those of RISE and RIDE using a monthly injections regimen. In the READ-2 trial, when less than monthly injection frequency after 2 years was succeeded by 1 year of monthly injections, additional statistically significant improvement in visual acuity was attainable (mean of 3.1 additional ETDRS letters).[ 192 ] However, RESOLVE, RESTORE, and DRCR network protocols I and T have demonstrated that similar outcomes can be achieved with monthly injections for 3–4 months followed by OCT and visual acuity guided prn follow-up treatment that decreases the number of injections required to produce the visual outcome.[ 193 194 ] Despite safety concerns that intravitreal anti-VEGF drugs could raise the risk of cardiovascular complications in patients with diabetes, there is no consistent evidence that this is the case.[ 174 194 195 196 ]

Bevacizumab is more cost-effective in treating DME than ranibizumab or aflibercept.[ 197 198 ] Medicare reimbursement for anti-VEGF drugs varies widely. In 2012, Medicare reimbursement was $50 for bevacizumab and $1,903 for ranibizumab.[ 174 ] The unit dose cost of aflibercept approximates that for ranibizumab for the treatment of macular degeneration, but the smaller approved dose of ranibizumab (0.3 mg) in the US means that the cost of ranibizumab is approximately 60% that of aflibercept. The cost differences arise because bevacizumab is not approved for intraocular use by the FDA, whereas ranibizumab and aflibercept are FDA-approved for intraocular use. Factors that influence which drugs are used include patient-factors and physician-factors. Patient-factors include Medigap insurance coverage and out-of-pocket costs. Physician-factors include Medicare drug repayment policies, industry economic incentives, and risks associated with compounding of bevacizumab.[ 199 ]

Combined Intravitreal Anti-VEGF and Corticosteroid Injections

Combination intravitreal bevacizumab and triamcinolone has not been found to improve outcomes compared with intravitreal bevacizumab monotherapy.[ 200 ] The addition of an intravitreal dexamethasone sustained release device to a regimen of ranibizumab injections did not improve visual acuity outcomes at 24 weeks, although macular thinning was greater than with intravitreal ranibizumab (IVR) alone.[ 96 ]

The idea that vitreomacular adhesion might promote DME arose from the observation that eyes with DME have a lower prevalence of posterior vitreous detachment than eyes without DME.[ 24 ] The subsequent observation that resolution of DME could occur after posterior vitreous detachment strengthened the plausibility that surgical induction of a vitreomacular separation might improve DME.[ 201 202 ] With the advent of OCT, vitreomacular adhesion was shown to be a risk factor for DME.[ 87 ] Vitrectomy for DME was first reported in 1992.[ 203 ] Since then, many small retrospective and prospective case series, several small clinical trials, but no large, multi-centered, randomized, controlled trials of the approach have been published.[ 46 49 61 94 95 98 99 136 204 205 206 207 208 209 210 211 212 213 214 215 ] Vitrectomy was introduced for eyes with a taut posterior hyaloid adherent to the macula, often associated with shallow traction macular detachment, which had failed previous focal/grid laser.[ 45 48 99 136 203 216 ] Later, it was explored as a therapy for eyes with an attached but non-thickened, non-taut posterior hyaloid or for eyes with persistent DME despite previous focal laser or intravitreal triamcinolone injection regardless of the status of the posterior hyaloid [Figs. 4 and 5 ].[ 95 99 204 205 207 215 217 ] Most recently, the treatment has been studied as a potential primary therapy in eyes with more severe edema and greater visual acuity loss at presentation.[ 46 136 208 210 213 215 218 219 ] The relative frequencies of the various candidate groups have been reported as follows: refractory DME in eyes with attached but non-taut posterior hyaloid 68%, refractory DME in eyes with posterior vitreous detachment 22%, refractory DME in eyes with a taut posterior hyaloid 5%, and refractory DME in eyes with an epiretinal membrane 5%.[ 220 ]

A controversy exists regarding the effects of vitrectomy for DME. Several groups of investigators have reported data to suggest that vitrectomy reduces macular thickening but does not improve visual acuity.[ 135 211 215 218 220 ] Others report improved visual acuities simultaneous with decreases in macular thickening or lagging behind macular thinning by a few months.[ 94 136 208 221 ] Others report improved visual acuity in cases with macular traction, but no visual improvement in cases without traction.[ 48 222 ]

The largest prospective observational study with standardized data collection was performed by the DRCR network. In this study, which was not a randomized trial, 241 eyes were followed to a primary outcome visit at 6 months. Baseline median CSMT was 491 μm, interquartile range (IQR) (356, 586). Baseline visual acuity was 57 letters, IQR (45, 66). At 6 months follow-up, the median change in CSMT was -97 μm, IQR (−8, +10). The median change in ETDRS letter score was + 1 letter, IQR (−8, +10).

As has been reported for all other treatments for DME, recurrence of edema after initial improvement, incomplete resolution of macular thickening, and failure to respond at all to treatment also occur with vitrectomy, but the rates of these undesirable outcomes may be reduced compared with focal laser and intravitreal triamcinolone injections alone.[ 4 9 10 28 ]

Although there is general acceptance that vitrectomy has a role in the management of at least some cases of DME, there is also consensus that it has no role in many cases, including cases of mild edema with minimal visual compromise and in cases with large submacular hard exudates, in which chronic RPE atrophic changes limit the potential for improvement even after specifically removing these exudates through small retinotomies.[ 137 223 ] A prospective, multicenter, randomized clinical trial is needed to define the role of vitrectomy surgery in the management of DME.

Discrepancy between Outcomes in Randomized Controlled Trials and Real-World Conditions

The outcomes obtained in the treatment of DME in Randomized Controlled Trials (RCTs) and under real-world conditions are different. In real-world conditions, inferior visual acuity gains associated with less frequent intravitreal injections have been reported, a relationship that has been consistently noted internationally.[ 224 225 226 227 228 229 230 ] There are many factors that possibly explain the discrepancy. In clinical trials, patients are preselected for their commitment to complete the schedule of visits, costs are borne in most cases by the entity performing the study, and subsidies for travel are often provided. In real life, lack of time and means could contribute to lower treatment intensity, the nonmedical costs for patients are onerous especially for patients of lower economic means and who are motivated not to miss work for doctor visits, and the need to manage other comorbidities.[ 227 228 231 ] Both non-elderly and elderly patients with DME have higher rates of comorbidity and loss of work time and personal time compared with diabetic patients without DME.[ 232 ] For example, non-elderly patients with DME had an average of 24.7 annual days with healthcare visits compared with 14.4 for age-matched controls with diabetes but no DME.[ 232 ] The average direct medical cost ratio, adjusted for age, sex, race, geographic region, and comorbidity, for Medicare patients with DME over 3 years was 1.31 times that for diabetic controls without DME.[ 233 ] In a retrospective claims analysis of 2,733 newly diagnosed patients with DME conducted over the interval 2008 through 2010, the mean annual numbers of bevacizumab injections were 2.2, 2.5, and 3.6 for the years 2008, 2009, and 2010, respectively, fewer than in major clinical trials of anti-VEGF agents.[ 227 ] Similarly, in a retrospective study of 121 eyes of 110 patients with a new diagnosis of DME receiving anti-VEGF injection therapy for the first time between 2007 and 2012 from the Geisinger Health System database, a mean of 3.1 ± 2.4 injections per study eye were given in the first year of treatment. The mean change in corrected visual acuity was 4.7 ± 12.3 approximate ETDRS letters, where approximate ETDRS letters are calculated from Snellen visual acuity. Higher numbers of anti-VEGF injections in the first 12 months after diagnosis correlate with improved visual outcomes, implying that real-world outcomes usually lag those in RCTs.[ 227 ] Other factors that may contribute to the discrepancies include lack of protocol refractions in many real-world visits, and the variability in treatment regimens and follow-up used by real-world clinicians compared with standardized regimens in clinical trials.[ 228 ]

In a German study looking at pooled anti-VEGF injections for DME, the mean change in VA at 12 months was -1.3 letters with a median of 6 injections.[ 225 ] Both the number of injections and the visual acuity outcomes are inferior to those reported in RCTs. In a study of Medicare claims data from 2008 through 2010, the mean number of claims per year for anti-VEGF injections for DME was 3.1–4.6.[ 226 ] A US commercial database claims study over 2008 through 2010 reported mean numbers of bevacizumab injections for DME varying from 2.2 to 3.6.[ 227 ] By comparison, the number of injections of ranibizumab in RISE and RIDE was 12 in the first year, and of ranibizumab, bevacizumab, or aflibercept was 9–10 in DRCR protocol T.[ 174 178 ] A Danish study of IVR for DME at 12 months reported a median number of injections of 5 and a median change in BCVA of +5 ETDRS letters.[ 229 ] An Italian study of IVR for eyes with unilateral DME reported a mean ± SD number of injections of 4.15 ± 1.99 over 18 months of follow-up with a worsening of visual acuity on average.[ 230 ]

New Directions

Genetic mutations that render patients more or less susceptible to DME as a complication of diabetes mellitus are likely to be defined. The physiological pathways contributing to DME and not mediated by VEGF are a likely focus of future research. Using OCT and OCT angiography, it should be possible to define at almost histological levels the retinal changes occurring in DME and determine, which if any changes associate with visual outcomes. Clinical trials of new drugs initiated by drug companies and comparative effectiveness research by organizations like the DRCR network will provide an evidential basis for rational therapy. Efforts to close the gap between randomized clinical trials and real-world outcomes and to reduce the cost of care will draw increasing attention.

Summary of Key Points

The prevalence of DME is increasing worldwide, mainly because of increasing type 2 diabetes.
Understanding retinal anatomy helps in analyzing clinical presentations of DME based on the effects of the avascularity of the central macula, the locations of the microvessels in the inner retinal layers, the importance of the pigment epithelial layer, and the role of the vitreoretinal interface.
The oxygen theory of DME is the most comprehensive pathophysiologic schema and VEGF is the single most important mediator in that pathway, although not the sole mediator.
OCT is critical in managing DME.
Macular thickening has an imperfect correlation with visual acuity probably due to factors currently difficult to assess such as duration of edema and degree of macular ischemia.
Metabolic control of blood glucose, blood pressure, and serum lipids is the foundation of therapy for DME, and specific ocular treatments are most effective when this foundation is optimized first.
Serial injections of anti-VEGF drugs are first-line therapy for DME. Focal/grid laser, intravitreal injections of corticosteroids, and vitrectomy have secondary roles in particular cases.

Financial support and sponsorship

Conflicts of interest.

There are no conflicts of interest.

Cited Here |
PubMed | CrossRef |
Google Scholar
View Full Text | PubMed | CrossRef |
View Full Text | PubMed |

Diabetes; diabetic retinopathy; macular edema

+ Favorites
View in Gallery

Readers Of this Article Also Read

Diabetic macular edema treatment guidelines in india: all india..., toxic optic neuropathy, current approach in diagnosis and management of anterior uveitis, in situ</em> keratomileusis', 'parmar pragya; salman, amjad; rajmohan, m; jesudasan, nelson ca', 'indian journal of ophthalmology', 'jul-aug 2009', '57', '4' , 'p 320-322');" onmouseout="javascript:tooltip_mouseout()" class="ejp-uc__article-title-link">fibrinous anterior uveitis following laser in situ keratomileusis, reduction in astigmatism in manual small incision cataract surgery through....

Open access
Published: 28 August 2024

Optimized deep CNN for detection and classification of diabetic retinopathy and diabetic macular edema

V Thanikachalam 1 ,
K Kabilan 1 &
Sudheer Kumar Erramchetty 1

BMC Medical Imaging volume 24 , Article number: 227 ( 2024 ) Cite this article

Metrics details

Diabetic Retinopathy (DR) and Diabetic Macular Edema (DME) are vision related complications prominently found in diabetic patients. The early identification of DR/DME grades facilitates the devising of an appropriate treatment plan, which ultimately prevents the probability of visual impairment in more than 90% of diabetic patients. Thereby, an automatic DR/DME grade detection approach is proposed in this work by utilizing image processing. In this work, the retinal fundus image provided as input is pre-processed using Discrete Wavelet Transform (DWT) with the aim of enhancing its visual quality. The precise detection of DR/DME is supported further with the application of suitable Artificial Neural Network (ANN) based segmentation technique. The segmented images are subsequently subjected to feature extraction using Adaptive Gabor Filter (AGF) and the feature selection using Random Forest (RF) technique. The former has excellent retinal vein recognition capability, while the latter has exceptional generalization capability. The RF approach also assists with the improvement of classification accuracy of Deep Convolutional Neural Network (CNN) classifier. Moreover, Chicken Swarm Algorithm (CSA) is used for further enhancing the classifier performance by optimizing the weights of both convolution and fully connected layer. The entire approach is validated for its accuracy in determination of grades of DR/DME using MATLAB software. The proposed DR/DME grade detection approach displays an excellent accuracy of 97.91%.

Peer Review reports

Introduction

Diabetes Mellitus (DM) has reached epidemic proportions in terms of global incidence and predominance in recent years, and the study show the expected range will be in 2030 more than 360 million people who are expected to be affected by DM around the world [ 1 ]. DM is a condition in which the blood glucose level increases excessively in response to insulin insufficiency, leading to impairment of the functioning of the retina, nerves, heart and kidneys. With changes in lifestyle and dietary habits coupled with factors such as physical inactivity and obesity, DM has become more prevalent and has surpassed the status of being a disease just confined to the rich [ 2 , 3 ]. DM patients are highly susceptible to developing DR, which results in abnormal retinal blood vessel growth and has a debilitating effect on vision. This progressive microvascular disorder leads to physical complications such as Diabetic Macular Edema (DME), retinal neovascularization, retinal permeability and retinal ischemia. In DR, abnormal blood vessel growth is caused by the need to supply oxygenated blood to the hypoxic retina. In addition, retinal thickening in the macular regions causes DME. It is an undisputable fact that medical treatments are more successful when diseases are discovered in their early stages.

Thereby, it is crucial to cure DR and DME in their earlier stages to prevent the serious consequence of vision loss in patients. Moreover, prior to complete blindness, there are rarely any visual or ophthalmic symptoms related to DR [ 4 , 5 , 6 ]. The high blood sugar levels seen in a DM patient, damages the retinal blood vessels, resulting in the leakage and accumulation of fluids such as soft exudates, hard exudates, haemorrhages and microaneurysms in the eye. The volume of these accumulated fluid defines the grade of DR, while the distance between macula and hard exudate defines the degree of DME [ 7 ]. Through early detection of DR, almost 90% of visual impairment cases are possible to be prevented. Additionally, through proper classification of DME/DR intensity, devising a suitable treatment for the DM patients is accomplished [ 8 ].

Consequently, patients with diabetes are recommended to undertake regular retinal fundal photography, in which retinal images are gathered and analysed by an ophthalmologist. Following the Airlie House DR classification, the Early Treatment Diabetic Retinopathy Study (ETDRS) group and the literature by Diabetic Retinopathy Study (DRS) group presents the classification of grades of DR using retinal fundus imaging. A conventional film camera was used in earlier days for capturing fundus images, which was later substituted by a digital camera. The fundus photography captured using Scanning Laser Ophthalmoscope (SLO) is popular nowadays [ 9 , 10 ]. The manual analysis of fundus images by ophthalmologist are ineffectual in terms of high throughput screening, therefore several automatic machine learning and deep learning fundal photography-based DR/DME screening techniques are introduced [ 11 , 12 , 13 ].

The image processing approach is the most effective technique for identifying the grades of DME/DR owing to its promising attributes of excellent adaptability, quicker processing time and maximum reliability. In case of image processing approach, the input retinal fundus image undergoes five different stages namely pre-processing methods, segmentation, feature extraction techniques, feature selection process and efficient classification. The pre-processing technique is carried out with the intention of enhancing the quality of the input image by minimizing the noises. The mean filter is one of the prominently used filter for pre-processing owing to its effectiveness in lessening pixel intensity variations and removing redundant pixels.

However, its application is limited due to the drawback of initiating pseudo noise edges [ 14 ]. The linear filters are inept for pre-processing, since it blurs the edges and contrast of the image, while the non-linear filters such as median filter [ 15 ] and adaptive mean filter [ 16 ] are effective in minimizing the noises in the image, however on the downside, the blurring of vital and edge regions leads to information loss. Therefore, to overcome the drawbacks, DWT is used as the pre-processing technique. The accuracy of identification of grades of DR/DME is further improved with the aid of an appropriate segmentation technique, effective in accurate segmentation of the retinal vessels and lesions. The segmentation of the retinal fundus image is hindered by several obstacles such as non-uniform illumination, undefined artefacts, improper image acquisition, complex components and lesion shape variability [ 17 ].

The Fuzzy C-Means clustering methods presented [ 18 ] is a predominantly used segmentation technique in recent research work, which forms diverse clusters through image pixel division. The complex nature of this technique however prevents its wide scale implementation. Here, in this work, ANN is used for segmentation in response to its simple structure and high accuracy in segmentation. Some of the commonly used feature extraction techniques are sparse representation [ 19 ], global histogram normalization [ 20 ] and Fourier Transform [ 21 ]. However, these techniques are inept in terms of retinal vein recognition. Gabor filter is suitable for retinal vein extraction, but its application is hindered due to the difficulty experienced in parameter configuration. Hence, Adaptive Gabor Filter (AGF), that resolves the complications in parameter configuration of conventional Gabor filter is used in this work for feature extraction.

The choice of an appropriate feature selection technique significantly improves the classification accuracy of the classifier. The feature selection approaches like Maximize Relevancy and Minimize Redundancy (mRMR) and Relief operates with excellent computational efficiency but less accuracy in terms of feature selection. The Genetic Algorithm [ 22 ] is an also a commonly used approach for feature selection, but it is in efficient in handling huge input samples due to computational complexity. The neural network techniques like Recurrent Neural network (RNN) and Probabilistic Neural Network (PNN) require large training data sets and display weak interpretability. Thereby, in this work, RF is selected for feature selection in view of its implementational ease and robust generalization capability. After feature selection comes the process of classification. The machine learning based Logistic Regression [ 23 ] Classifier is an efficient technique with excellent discriminative potential, but it is incapable of solving linear problems. The CNN [ 24 , 25 ] is a highly accurate technique, capable of quickly identifying and classifying any medical disorder. However, it requires large number of training images. Hence, a Deep CNN based classification is proposed in this work for the accurate classification of grades of DR/DME. Moreover, the working of the Deep CNN classifier is optimized using Chicken Swarm Algorithm (CSA).

A novel automatic DR/DME detection approach using optimized Deep CNN is proposed in this work. The different phases of the proposed image processing approach involve DWT for pre-processing, ANN for segmentation, AGF for feature extraction, RF for feature selection and finally CS optimized Deep CNN for classification. The retinal fundus images are provided as input for the proposed diagnosis model, and it is evaluated for its performance using MATLAB software.

As shown below, we provide numerous major breakthroughs and additions in this work that greatly improve our model’s efficacy and applicability for the identification of DME and DR:

While some literature utilizes various optimization techniques, such as Genetic Algorithms or Harris Hawks Optimization, this paper uses the Chicken Swarm Algorithm (CSA) to optimize the deep CNN model, which is unique.

The paper combines several techniques, including DWT for preprocessing, AGF for feature extraction, and RF for feature selection. While these methods have been individually used in other studies, the combination and the specific workflow are distinct.

The novelty lies in the integrated approach combining DWT, ANN for segmentation, AGF, RF, and CSA-optimized Deep CNN for classifying the grades of DR/DME. This combination of methods aims to enhance the detection accuracy.

The proposed method achieves a high accuracy rate of 97.91% in detecting and classifying DR/DME grades, which is presented as an improvement over existing methods.

The paper highlights the effectiveness of using CSA to optimize the Deep CNN classifier, which is a novel application of this algorithm in this context.

Literature study

DR and DME are two common complications of diabetes that can lead to vision loss and blindness if not detected and treated early. In recent research studies, the application of CNNs has shown promising results in the early detection and classification of DR and DME, ultimately contributing to the development of more effective and automated screening processes in diabetic eye care. Sundaram et al., [ 26 ] discusses an artificial intelligence-based approach for the detection of DR and DME. This model utilizes preprocessing, blood vessel segmentation, feature extraction, and classification techniques. It also introduces a contrast enhancement methodology using the Harris hawks optimization technique. The model was tested on two datasets, IDRiR and Messidor, and evaluated based on its accuracy, precision, recall, F-score, computational time, and error rate. This technology aims to assist in the early detection of these severe eye conditions, which are common causes of vision impairment in the working population, and it suggests a significant positive impact on the healthcare sector by enabling timely and cost-effective diagnosis.

He et al., [ 27 ] discusses a deep learning approach to classify DR severity and DME risk from fundus images. Three independent CNN’s were developed for classifying DR grade, DME risk, and a combination of both. They introduced a fusion method to combine features extracted by the CNNs, aiming to assist clinicians with real-time, accurate assessments of DR. The paper highlights the potential for automated systems to enhance early detection and treatment, and reports classification accuracy rates of 0.65 for DR grade and 0.72 for DME risk. Reyes et al., [ 28 ] discusses a system designed to classify DR and DME, which are common causes of blindness in diabetic patients. The system employs the Inception v3 transfer learning model and MATLAB digital image processing to analyze retinal images without the need for dilating drops, which can have side effects. Tested by medical professionals in the Philippines, the system showed reliable and accurate results, indicating its potential as an assistive diagnostic device for endocrinologists and ophthalmologists.

Kiruthikadevi et., [ 29 ] discusses the development and implementation of a system designed to detect and assess DR and DME from color fundus images using CNN’s. The system aims to automate the detection process to support early diagnosis and effective treatment, as substantially manual diagnosis by clinicians is not feasible at scale, particularly in resource-limited settings. The proposed two-stage approach first verifies the presence of Hemorrhages and Exudates in fundus images, and then evaluates the macular region to determine the risk of DME. The methodology includes image preprocessing to reduce noise, extraction of regions of interest focusing on the macular area, and generation of motion patterns to imitate the human visual system, all with the broader goal of contributing to the prevention of vision loss due to diabetes-related complications.

Sudha Abirami R and Suresh Kumar G [ 30 ] provides a comprehensive overview of the application of deep learning and machine learning models for the detection and classification of diabetic eye diseases, with a primary focus on DR. Various public datasets, like EyePACS and Messidor, and image preprocessing techniques are used to enhance the images before they are input into machine learning models like CNN’s. Transfer learning is emphasized as a critical technique to improve model performance, with most of the past work highlighting the need for classification of all types of diabetic eye diseases, not just DR. Despite powerful commercial AI solutions available, the review identifies a gap in affordable methods and suggests further development of computer-aided diagnostic models that are efficient and reliable for categorizing various diabetic eye conditions.

Lihteh Wu et al., [ 31 ] discusses the importance of categorizing and staging the severity of DR to provide adequate treatment and prevent visual loss. The paper emphasizes the global epidemic of diabetes mellitus and the associated risk of DR, a leading cause of blindness in the working-age population. DR is characterized by progressive microvascular changes leading to retinal ischemia, neovascularization, and macular edema. The International Clinical Disease Severity Scale for DR is highlighted as a simple and evidence-based classification system that facilitates communication among various healthcare providers involved in diabetes care without the need for specialized examinations. The scale is based on the Early Treatment of DR Study’s 4:2:1 rule relying on clinical examination.

This work [ 32 ] introduces a new framework for classifying DR and DME from retinal images. Using deep learning methods, particularly CNN’s, coupled with a modified Grey Wolf Optimizer (GWO) algorithm with variable weights, the research seeks to improve the precision and performance of the classification. This approach addresses the urgent problem of early detection and treatment of diabetic eye diseases, which are the major causes of blindness worldwide. The experimental results show that the suggested approach is an effective method for the accurate diagnosis of DR and DME, highlighting its potential in improving the diagnostic capabilities and care of patients in ophthalmology.

The paper [ 33 ] proposes a robust framework for classifying retinopathy grade and assessing the risk of macular edema in DR images. The study introduces a comprehensive approach that integrates image preprocessing, feature extraction, and machine learning algorithms to accurately classify retinal images and predict the likelihood of macular edema. By leveraging a combination of handcrafted features and deep learning techniques, such as CNN’s, the framework achieves high classification accuracy and robustness. The proposed methodology addresses the urgent need for automated and accurate diagnosis of DR, providing a valuable tool for clinicians in assessing disease severity and guiding treatment decisions. Experimental results demonstrate the effectiveness of the proposed framework in accurately classifying retinopathy grade and predicting macular edema risk, highlighting its potential for enhancing clinical workflows and improving patient outcomes in diabetic eye care.

In summary, CNN’s are a highly effective method for the classification and grading of DR and DME, with various approaches including feature reduction, attention mechanisms, and network fusion methods contributing to their success. The integration of deep learning techniques with traditional image processing methods and novel architectures has led to significant improvements in the accuracy and efficiency of diagnosing these conditions.

Proposed system framework

The disease of DM has become a prominent disorder found in many middle aged and older generations due to the drastic unhealthy changes witnessed in food habits and lifestyle of humans. Thus, the DM is no longer considered to be the disease only confined to the rich. The person who develops DM are affected many complications among which DR and DME are the one that has direct impact over the vision. The effects of DR and DME are highly critical, since it eventually leads to a complete blindness. Through a timely accurate identification of degree of DR/DME in a diabetic patient, the condition of blindness is greatly prevented [ 34 ]. Thereby, an accurate DR/DME grade detection approach as illustrated in Fig. 1 is proposed in this work.

Automatic DR/DME grade detection using optimized Deep CNN architecture

The proposed approach using DWT for pre-processing of the retinal fundus image. Through pre-processing, the unwanted noises that affects the retinal photography is removed and an enhanced image with uniform resolution is obtained as output. Next the pre-processed image is subjected to ANN segmentation, which is highly effective in isolation of the required region of interest. Subsequently, AGF with high reginal vein recognition capability is used for feature extraction. Moreover, the vital features that assists classification are selected among all the extracted features using the approach of RF. Finally, the degree of DR/DME is accurately detected using CS optimized Deep CNN classifier. The CSA is used for optimizing the weights of both convolution and fully connected layer, resulting in the improvement of the classification performance of Deep CNN. Moreover, the entire technique is validated in MATLAB software for ascertaining its significance in identification of DR/DME grades.

Preprocessing using DWT

Pre-processing is one of the crucial steps undertaken in image processing to improve the image quality and thereby enhance the accuracy of DR and DME identification. Here, the pre-processing of fundus images is done using DWT [ 35 ], which is characterized with an excellent image decomposition property. Initially the images are resized to obtain uniform resolution and increased processing speed. Then the green channel image that has vital information are extracted before undergoing histogram equalization. The resultant image with improved dynamic range and contrast are made noise free through filtering.

The fundus image is decomposed into several sub band images. At the end of every computed value in decomposition stage, the frequency resolution is twice, and the computed time resolution is halved. The products of decomposition are detail coefficients and approximation coefficients, where the latter is further decomposed into detail coefficients and values of approximation coefficients in every later level. The approximation coefficient is the first sub-band image, while the remaining coefficient are detailed coefficients, so resulting in the formation of several sub-band images. The translation parameters and discrete set of scale used in DWT are $\:\left(\tau\:=n{2}^{-m}\right)$ and $\:\left(s={2}^{-m}\right)$ respectively. The wavelet family is given as,

The $\:\:x\left[n\right]$ decomposition is given as,

Where the scaling and wavelet coefficients are specified as $\:{\:d}_{j,k}j=1\dots\:J$ and $\:{\:c}_{j,k}j=1\dots\:J$ respectively.

Where, the scaling sequence, wavelet and complex conjugate are expressed as $\:\:{h}_{J}\left[n-{2}^{J}k\right]$ , $\:{g}_{j}\left[n-{2}^{j}k\right]$ and (*) respectively. The DWT is implemented separately for every column and row of the image. The image $\:X$ is decomposed into high frequency detail coefficients $\:\:{X}_{H}^{1},\:{X}_{V}^{1}\:and\:{X}_{D}^{1}$ and low frequency approximation coefficient $\:\:{X}_{A}^{1}$ .

The image after $\:{N}^{th}$ level decomposition is expressed as,

The preprocessed image is then segmented using ANN.

Segmentation using ANN

The process of segmentation is also a crucial procedure like pre-processing and is vital for the precise detection of DR and DME owing to its significant role in understanding the complex areas of interest of retinal fundus images. This image subdivision process ceases with the complete isolation of the required object of interest. In this work, ANN is used for segmentation, and it segments the pre-processed fundus images into areas and pixel groups that stands for micro aneurysms, lesions like haemorrhages, retinal blood vessels, optic disc and fovea in addition to hard and soft exudates. The ANN can impersonate the working of human brain in resolving complicated real-world problems and its structure encompasses three connected sequential layers normally called as input layer, hidden layer and output layer as presented in Fig. 2 [ 36 ].

Structure of ANN

The number of multipliers in ANN characterised with N output nodes, W hidden layer nodes and M inputs is given as,

The computational complexity of operation and calculation in each layer is reduced with the implementation of multipliers using add and shift operations rather than floating point numbers. Weights are quantized on the assumption that only a small number of shift and add operations are permitted due to the complexity of design hardware implementation. As a result, the quantization value of an original number is chosen to be the closest to it. Consider the following scenario: the maximum number of shift and add operations is 3, and the weights in the ANN are integers 0.8735 and 0.3811. The following new addition and shift operation representation may be used to represent these numbers:

With this form, every weight is converted into a sum of power-2 integers that can be executed using shift and add operations. The ANN’s multiplier modules are therefore broken down into a few adder and shifter modules, one for each multiplier that is necessary. Even if the computational complexity is reduced by a straightforward quantization with regard to the number of power-2 operations, an error is still produced, which might be problematic in some circumstances. To solve this issue, a potential error compensation approach is shown below.

Average quantization error reduction

Weights are quantized using only their values in the typical kind of quantization. As a result, there can be a considerable loss of accuracy due to accumulating quantization errors. Consequently, a compensating error approach is suggested [ 37 ]. There might be some accuracy decrease with each quantization. However, each image region is similar, and subsequent weight quantization can make up for the accuracy loss caused by weight quantization. By doing this, both average error and accuracy loss may be decreased. This is accomplished by distributing the generated mistake in the subsequent weight quantization, which comes after each weight has been quantized. Take the following instance into consideration. Three different weight coefficients of 0.8000, 0.4250, and 0.4050 are considered, and only three shift and add operations are permitted. It is displayed how close the closest quantized value is as shift and add number.

Consequently, the average quantization error is

Diffusion of each quantization mistake during the subsequent phases of weight quantization might lower the average quantization error. In the instance of example that has.

The current quantization step considers all quantization faults from earlier levels. Consequently, + 0.0500 is added to current value of 0.4250. The present quantization considers the values (+ 0.0500 and 0.0750). This implies that 0.4050 is added to previous values of + 0.0500 and 0.0750. Because the prior quantization mistakes are considered in the current weight quantization in this case, the average error is lowered. The overall quantization error can be decreased using this method.

Activation function linearization

The most popular ANN activation function is hyperbolic tangent, which has the following form.

Thus, a floating-point division and an exponential operation both need to be computed. It may be effective to lower the overall computation volume by linearizing and simplifying activation function. The four intervals that make up domain of tanh(x) function in this chapter are utilised to create a linear approximation function in each interval.

With the aid of pricewise linear function, computation is accomplished leaving division and multiplication and all operations are in shift or addition form.

Feature extraction using adaptive gabor filter (AGF)

The AGF is used for feature extraction of the ANN segmented retinal fundus images [ 38 ]. Because it resembles the receptive field profiles in human cortical simple cells, Gabor filtering is an effective computer vision feature analysis function. Gabor filters have been effectively used by earlier academics to exploit a variety of biometric traits. A complex sinusoidal grating that is directed and modulated by a 2D Gaussian function is known as a circular AGF.

Where, the term j = $\:\sqrt{-1}$ and $\:{g}_{\sigma\:}\left(x,y\right)$ refers to Gaussian envelope,

The span-limited sinusoidal grating frequency $\:\:\mu\:$ , the direction in the range of $\:\:{0}^{^\circ\:}-{180}^{^\circ\:}$ , and the standard deviation of a Gaussian envelope which is indicated by $\:\sigma\:$ . The $\:{G}_{\sigma\:,\mu\:,\theta\:}\left(x,y\right)$ term may be divided into a real part, $\:{R}_{\sigma\:,\mu\:,\theta\:}(x,y)$ and an imaginary part, $\:{I}_{\sigma\:,\mu\:,\theta\:}$ (x, y), using Euler’s formula, as illustrated in (6)–(8). In a picture, the genuine portion may be used for ridge detection while the fictitious portion is useful for edge detection.

Regions of uniform brightness, however, cause a negligible response from AGF. Direct current (DC) is what being used here. DC component is eliminated by using Eq. (9) so that Gabor filter would be insensitive to illumination:

Where $\:(2k+1{)}^{2}$ is 2Dd Gabor filter size. As a result, the definition of a Gabor transform with robust illumination is given in (26), where $\:I(x,\:y)$ is an image.

According to earlier studies, AGF-based edge identification performs best when filter parameters match the direction $\:\:\theta\:$ , variance $\:\:\sigma\:$ , and center frequency $\:\:\mu\:$ of input picture texture. After AGF based feature extraction, the process of feature selection RF is carried out.

Feature selection using random forest

The feature selection process aids in the identification of the smallest feature subset, which is pivotal to predict DR and DME with higher degree of accuracy by eliminating other irrelevant or redundant features. Thus, the choice of an effective feature selection process complements the classifier performance in identifying the DR/DME grades. The RF technique is adopted in this work for feature selection on account of its robust anti-interference and generalization capability [ 39 ]. This model aggregation-based machine learning algorithm is well suited for ill-posed and high-dimensional regression tasks. The RF when employed for feature selection, evaluates the importance score of every feature and determines their impact on the classification prediction. The RF builds decision trees using gini index and determines the final class in every tree. The impurity of node $\:\:v$ is estimated using the gini index,

Where, the fraction of $\:class-i$ records are specified as $\:\:{f}_{i}$ . For splitting the tree node $\:v$ , the Gini gain information of feature $\:{\:X}_{i}$ is given as,

Where, the right and left child node of node $\:v$ is specified as $\:{\:v}^{R}$ and $\:{\:v}^{L}$ respectively, while the node $\:v$ impurity is specified as $\:Gini\left({X}_{i},v\right)$ . The child nodes are assigned with fraction of examples referred as $\:{\:W}_{R}$ and $\:{\:W}_{L}$ . The splitting feature is the one that maximizes impurity reduction. The $\:gain\left({X}_{i},v\right)$ is used for calculating the importance score of $\:{\:X}_{i}$ ,

Where, the split nodes and ensemble size is specified as $\:k \epsilon S{x}_{i}$ and $\:{\:n}_{tree}$ respectively. The normalization of the importance score is,

Here, the maximum importance is specified as $\:{\:Imp}_{max}$ [ $\:{0\le\:Imp}_{max}\le\:1$ ]. The weight $\:gain\left({X}_{i},v\right)$ utilizes the importance score of preliminary RFs, thereby the penalized gini information gain is estimated as,

The regularization level is regulated by the base coefficient of $\:{\:X}_{i}$ , which is represented as $\:{\:\lambda\:}_{i} \epsilon \left[\text{0,1}\right]$ .

The weight of $\:{\:Imp}_{norm}$ is controlled by the importance coefficient represented as $\:\:\gamma\: \epsilon \left[\text{0,1}\right]$ . For an $\:{\:X}_{i}$ without maximum $\:{\:Imp}_{norm}$ , smaller $\:{\:\lambda\:}_{i}$ is effectuated by larger $\:\:\gamma\:$ , ultimately leading to a larger penalty on $\:{\:gain}_{G}\left({X}_{i},v\right)$ . In case of maximum penalty,

The $\:{\:gain}_{G}\left({X}_{i},v\right)$ is,

By injecting the normalized importance score, the Gini information gain weighting is achieved. Thus, the smallest and appropriate features are selected using RF and these features are used for enhancing the classification using CS optimized Deep CNN.

Classification using chicken swarm optimized deep CNN

The CS optimized Deep CNN model that are widely used for the detection are employed for classifying the grades of DME and DR. The CS algorithm is employed for optimizing the kernel values of convolution layer and optimizing the weights of the fully connected layer [ 40 ]. The features extracted using RF is provided as input to the CS optimized Deep CNN. The architecture of CNN comprises of distinct layers like convolution and pooling layers, which are grouped as modules. These modules are then subsequently followed by the fully connected layer that ultimately provides the class labels as outcomes. Modules are usually stacked on top of each other to build a deep model, which is becoming more and more popular. The structure of CS optimized Deep CNN used for the detection of DR/DME grades is given in Fig. 3 .

Architecture of CNN

Convolution layers

The convolution layer observes and analyses the features of the given input and performs the operation of a feature extractor. This layer comprises of several neurons that are grouped as feature maps. Each neuron belonging to a particular feature map is connected to the other neurons in the vicinity (previous layer) using their receptive field and the filter bank, which is a trainable weight set. In this layer, the weights and inputs are combined, and the output is moved to the successive layer using a non-linear activation function. The weights of the neurons grouped in a feature map are required to be uniform, but this is not the case due to the presence of distinct feature maps with different weights, enabling the extraction of multiple features from a specific region. The $\:\:{e}^{th}$ output feature map is expressed as,

Where, the terms $\:F{M}_{e},*and\:{I}_{M}^{seg}$ represents the $\:{\:e}^{th}$ feature map associated convolution filter, convolution operator and the input image respectively. The non-linear activation function is represented using the term $\:f(\bullet\:)$ .

Pooling layers

The pooling layers aids with attaining the spatial invariance to translation and distortion in the input. Moreover, the feature map’s spatial resolution is decreased in this layer. Initially, it is a common norm to employ average pooling layer for broadcasting the input average of small region of the image to the successive layer. The pooling layer output is given as,

Where, down sampling layer and the convolution layer are specified as $\:PL-1$ and $\:PL$ respectively. The input features of down sampling layer are represented as $\:\:{x}^{PL-1}$ , while the additive bias and kernel maps of the convolution layer is specified as $\:\:{Bi}^{PL}$ and $\:{\:K}_{ij}$ respectively. The input map selection is referred as $\:{\:M}_{j}$ , the output and input are indicated as $\:\:i$ and $\:\:j$ respectively. The crucial element of a field is chosen using max pooling.

Fully connected layers

Several convolution and pooling layers are stacked with one another to obtain optimal feature representation. These feature representations are fully analysed by the fully connected layer to accomplish operation of high-level reasoning. The accuracy of the Deep CNN is further improved with the aid of CS optimization. The flowchart of CS optimized Deep CNN for identification DR/DME grades is shown in Fig. 4 .

Flowchart of CS optimized Deep CNN

Chicken swarm (CS) optimization

The CS optimization algorithm enhances the classification accuracy of the Deep CNN through optimization of the fully connected layer and convolution layer. The characteristic traits of a chicken swarm that encompasses roosters, chicks and hens forms the basis of this algorithm. The rules associated with this algorithm is given as:

The rooster is the head of a chicken swarm, which comprises of numerous chicks and hens.

The fitness value of the chicken determines its individuality and aids in distinguishing itself from the others. The chief rooster is the one with the best fitness value, while chicks are the ones with worst fitness value. The rest are termed as hens and a casual mother-child relationship is created between the chicks and hens.

After several steps, each of their status gets updated.

The rooster guides the others in search of their food, while the chick forages for its food by staying in the vicinity of their mothers. In a dimensional space (DS), at a time step $\:ts$ , the positions of the N virtual hens are represented as,

Where, the mother hens, the chicks, hens and roosters are represented using the terms $\:NM,\:NC,\:NHl$ and $\:NR$ respectively. The chance of obtaining the food is more for the rooster with best fitness value.

Where, the fitness value associated with A is specified as $\:fv$ , the rooster index is specified as $\:\:l$ , the smallest constant used for evading the zero-division error is specified as $\:\:\epsilon\:$ and the gaussian distribution with SD $\:{\:\sigma\:}^{2}$ and mean 0 is represented as $\:Randn(0,{\sigma\:}^{2})$ .

Where, a random number between [0,1] is specified as $\:\:Rand$ . The randomly selected index from the swarm and the rooster index is represented as $\:\:ro2 \epsilon [1,\dots\:,N]$ and $\:ro1 \epsilon [1,\dots\:,N]$ respectively. Furthermore, $\:f{v}_{m}>f{v}_{ro1}$ and $\:f{v}_{m}>f{v}_{ro2}$ , hence $\:\:S2<1<S1$ . The probability of the chick staying nearby its mother is specified using the term FL, which lies between [0, 2].

Results and discussion

The proposed automatic DR/DME grade detection model was confirmed for its effectiveness by executing in MATLAB. The dataset having 2072 high resolution retinal fundus images is collected from MESSIDOR [ 41 ] to assess the performance of research work proposed under CS optimized Deep CNN based diagnostic technique. Among the gathered 2072 image samples, 1402 samples belong to healthy people without diabetic condition, while 520 samples belong to diabetic patients having DR/DME. A total of 150 retinal fundus images is considered as testing data. The overall details of the selected dataset are tabulated in Table 1 .

Input Image

The provided input retinal fundus image seen in Fig. 5 , undergoes the process of pre-processing initially. The several stages involved in pre-processing is displayed in Fig. 6 . The images are resized in view of supporting a uniform resolution. Then the resized input image undergoes gray scale conversion, noise reduction and filtering to obtain a pre-processed retinal fundus image of enhanced quality. In addition to obtaining a pristine noise-free image, the DWT based pre-processing also aids with reducing the processing time required for the execution of the entire technique.

Stages of Pre-processing

The DWT pre-processing is compared against prominent techniques including the filer methods such as Mean filter, Median filter, Wiener filter and Hilbert Transform in terms of Root Mean Square Error (RMSE), Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM) and Mean Square Error (MSE). The results obtained are taken for comparison in Table 2 .

On analyzing the observations given in Table 2 , it is concluded that the DWT performs better than all the other commonly used pre-processing techniques. Thus, the DWT technique is successful in its role of enhancing the accuracy of the proposed automatic DR/DME diagnostic system.

Segmentation using ANN outputs

The output obtained using ANN based segmentation is provided in Fig. 7 . From the obtained segmented retinal image, it is noted that the ANN is capable of accurately segmenting lesions affecting the eyes. Moreover, it is also seen that the ANN is effective in accurate segmentation of the DR/DME affected regions without compromising the image clarity. The different grades of DR are Proliferative DR, Severe Non-Proliferative DR (NPDR), Moderate NPDR and mild NPDR. Moreover, the DME is categorized in to three different grades namely mild DME, moderate DME and severe DME. So, the final classified output of the CS optimized Deep CNN classifier is shown in Fig. 8 .

As seen in Fig. 8 , the Deep CNN accurately classifies the retinal fundus image as Severe NPDR condition. The influence of CS optimized CNN in classification is verified by comparing with the existing classifier techniques and the concerned results are tabulated in Table 3 and is also graphically represented in Fig. 9 . The developed CS optimized Deep CNN has an enhanced accuracy of 97.91, sensitivity of 97.82%, specificity of 98.64%, Precision value of 0.97 and F1 score of value 0.98. Moreover, it is also noted that the CSA is effective in improving the overall performance of Deep CNN.

CS optimized Deep CNN classifier output

Classifier comparison in terms of ( a ) Accuracy ( b ) Sensitivity ( c ) Specificity ( d ) Precision and ( e ) F1 Score

To assess the effect of the Random Forest feature selection procedure on the functionality of our model, we conducted an ablation study. The findings projected in Table 4 showed that adding feature selection increased the accuracy of the model from 93.85 to 97.91%, along with gains in precision, recall, and F1-score. This proves how well the feature selection process works to improve the model’s ability to correctly categorize the various grades of diabetic macular oedema (DME) and diabetic retinopathy (DR), underscoring the crucial role that feature selection plays in the overall performance of the classification process.

Recent discoveries in deep learning and medical imaging, such as Zhang et al. [ 42 ] and Zhang et al. [ 43 ], have shown the usefulness of region-based integration-and-recalibration networks for nuclear cataract categorization for AS-OCT images. These investigations emphasize the increasing significance of advanced image processing methods in raising diagnostic precision, as does the work of Xiao et al. [ 44 ], who presented a multi-style spatial attention module for cortical cataract classification.

In contrast with existing research, which mainly concentrates on AS-OCT pictures, our study improves feature extraction from retinal images by using CNNs in conjunction with Discrete Wavelet Transform (DWT). To further set our method apart, we also used the Chicken Swarm Algorithm (CSA) for model weight optimization. Our strategy provides a unique combination of DWT and CSA, exceeding the performance metrics stated in the referenced publications, which focus on attention mechanisms and recalibration.

Furthermore, our results highlight the potential of deep learning methods in real-time clinical settings, especially in automated DR and DME detection, which hasn’t been thoroughly studied with the attention mechanisms employed in existing studies, as far as we came to know. This demonstrates how innovative our methodology is in bringing these approaches to a new setting in medical imaging and advances the area of automated medical diagnosis.

An automatic DR/DME grade detection approach using optimized Deep CNN is introduced in this article. The rise seen in patients affected by DM in recent times has in turn resulted in an increased risk of early age blindness because of DR and DME. Thereby, the proposed work has an impact in aiding with the earlier detection of this serious medical condition. Through prompt detection and proper treatment, a substantial number of DM patients are saved from a potential sightless dark future. In this approach, the input retinal fundus images are initially pre-processed using DWT, resulting in the deliverance of noise-free sharp contrast retinal images. Then with the application of ANN, the exact region of interest is found and segmented. The vital features that support effective classification is obtained using AGF, while RF is used as the feature selection technique in this work. Ultimately, the grades of DR/DME are identified using CS optimized Deep CNN classifier. The entire approach is evaluated for its accuracy using MATLAB software and from the derived results, it is concluded that the CSA is successful in improving the classification accuracy of the Deep-CNN classifier. The proposed automatic DR/DME grade detection technique works with an outstanding accuracy of 97.91%.

Availability of data and materials

IDRiR Dataset: https://ieee-dataport.org/open-access/indian-diabetic-retinopathy-image-dataset-idrid . Messidor Dataset: https://www.adcis.net/en/third-party/messidor/ .

Abbreviations

Diabetic Retinopathy
Diabetic Macular Edema

Discreate Wavelet Transform

Artificial Neural Network

Adaptive Gabor Filter
Random Forest

Convolutional Neural Network

Chicken Swarm Algorithm

Diabetes Mellitus

Scanning Laser Ophthalmoscope

Maximize Relevancy and Minimize Redundancy

Recurrent Neural Network

Probabilistic Neural Network

Grey Wolf Optimizer

Root Mean Square Error

Peak Signal to Noise Ratio

Mean Square Error

Structural Similarity Index

Wu L, Fernandez-Loaiza P, Sauma J, Hernandez-Bogantes E, Masis M. Classification of diabetic retinopathy and diabetic macular edema. World J Diabetes. 2013;4(6):290.

Article PubMed PubMed Central Google Scholar

Zou Q, Qu K, Luo Y, Yin D, Ju Y, Tang H. Predicting diabetes mellitus with machine learning techniques. Front Genet. 2018;9:515.

Cole JB, Jose C. Florez. Genetics of diabetes mellitus and diabetes complications. Nat Rev Nephrol. 2020;16:377–90.

Li X, Zhu XHLYL, Fu C-W, Pheng-Ann H. CANet: cross-disease attention network for joint diabetic retinopathy and diabetic macular edema grading. IEEE Trans Med Imaging. 2019;39(5):1483–93.

Article PubMed Google Scholar

Markan A, Agarwal A, Arora A, Bazgain K. Vipin Rana, and Vishali Gupta. Novel imaging biomarkers in diabetic retinopathy and diabetic macular edema. Therapeutic Adv Ophthalmol. 2020;12:2515841420950513.

Article Google Scholar

Everett LA, Yannis M. Paulus. Laser therapy in the treatment of diabetic retinopathy and diabetic macular edema. Curr Diab Rep. 2021;21(9):1–12.

Chaudhary PK, Pachori RB. Automatic diagnosis of different grades of diabetic retinopathy and diabetic macular edema using 2-D-FBSE-FAWT. IEEE Transact Instrument Measure. 2022;71:1–9.

Tu Z, Gao S, Zhou K, Chen X, Fu H, Gu Z, Cheng J, Zehao Yu, Liu J. SUNet: A lesion regularized model for simultaneous diabetic retinopathy and diabetic macular edema grading. In: 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI). 2020. p. 1378–82.

Chapter Google Scholar

Kobat SG, Baygin N, Yusufoglu E, Baygin M, Barua PD, Dogan S, Yaman O, et al. Automated diabetic retinopathy detection using horizontal and vertical patch division-based pre-trained DenseNET with digital fundus images. Diagnostics. 2022;12(8):1975.

Horie S, Ohno-Matsui K. Progress of imaging in diabetic retinopathy—from the past to the present. Diagnostics. 2022;12:1684.

Mustafa H, Ali SF, Bilal M, Hanif MS. Multi-Stream Deep Neural Network for Diabetic Retinopathy Severity Classification Under a Boosting Framework. IEEE Access. 2022;10:113172–83.

Wang J, Bai Y, Xia B. Simultaneous diagnosis of severity and features of diabetic retinopathy in fundus photography using deep learning. IEEE J Biomed Health Inform. 2020;24(12):3397–407.

Abdelsalam MM, Zahran MA. A novel approach of diabetic retinopathy early detection based on multifractal geometry analysis for OCTA macular images using support vector machine. In IEEE Access. 2021;9:22844–58.

Thanh DN, Hoang, Serdar Engínoğlu. An iterative mean filter for image denoising. IEEE Access. 2019;7:167847–59.

Tang H, Ni R, Zhao Y, Li X. Median filtering detection of small-size image based on CNN. J Vis Commun Image Represent. 2018;51:162–8.

Rakshit M. An efficient ECG denoising methodology using empirical mode decomposition and adaptive switching mean filter. Biomed Signal Process Control. 2018;40:140–8.

He Y, Jiao W, Shi Y, Lian J, Zhao B, Zou W, Zhu Y, Zheng Y. Segmenting diabetic retinopathy lesions in multispectral images using low-dimensional spatial-spectral matrix representation. IEEE J Biomed Health Inform. 2019;24(2):493–502.

Cai W, Zhai B, Liu Y, Liu R, Xin Ning. Quadratic polynomial guided fuzzy C-means and dual attention mechanism for medical image segmentation. Displays. 2021;70: 102106.

Zhai S, Jiang T. Sparse representation-based feature extraction combined with support vector machine for sense‐through‐foliage target detection and recognition. IET Signal Proc. 2014;8(5):458–66.

Menotti D, Najman L, Facon J, Arnaldo de A, Araújo. Multi-histogram equalization methods for contrast enhancement and brightness preserving. IEEE Trans Consum Electron. 2007;53(3):1186–94.

Islam, Nahidul MD, Sulaiman N, Rashid M, Bari BS, Jahid Hasan MD, Mustafa M, Jadin MS. "Empirical mode decomposition coupled with fast fourier transform based feature extraction method for motor imagery tasks classification. In: 2020 IEEE 10th International Conference on System Engineering and Technology (ICSET). 2020. p. 256–61.

Ullah N, Mohmand MI, Ullah K, Gismalla MSM, Ali L, Khan SU, Ullah N. Diabetic Retinopathy Detection Using Genetic Algorithm-Based CNN Features and Error Correction Output Code SVM Framework Classification Model. In: Wireless Communications and Mobile Computing 2022. 2022.

Google Scholar

Leontidis G, Hunter A. A new unified framework for the early detection of the progression to diabetic retinopathy from fundus images. Comput Biol Med. 2017;90:98–115.

Khalil H, El-Hag N, Sedik A, El-Shafie W, Mohamed AE, Khalaf AAM, El-Banby GM, Abd El-Samie FI, El-Fishawy AS. Classification of Diabetic Retinopathy types based on Convolution Neural Network (CNN). Menoufia Journal of Electronic Engineering Research, 28(ICEEM2019-Special Issue). 2019:126–53. https://doi.org/10.21608/mjeer.2019.76962 .

Khan S, Haris Z, Abbas, Danish Rizvi SM. Classification of diabetic retinopathy images based on customised CNN architecture. In: 2019 Amity International conference on artificial intelligence (AICAI). 2019. p. 244–8.

S Sundaram, et al. Diabetic Retinopathy and Diabetic Macular Edema Detection Using Ensemble Based Convolutional Neural Networks. Multidisciplinary Digital Publishing Institute. 2023;13(5):1001–1001. https://doi.org/10.3390/diagnostics13051001 .

J. He, L. Shen, X. Ai and X. Li. "Diabetic Retinopathy Grade and Macular Edema Risk Classification Using Convolutional Neural Networks". Jul. 2019. https://doi.org/10.1109/icpics47731.2019.8942426 .

Reyes ACS et al. Sep. SBC Based Diabetic Retinopathy and Diabetic Macular Edema Classification System using Deep Convolutional Neural Network. vol. 9. no. 3. pp. 9–16. 2020. https://doi.org/10.35940/ijrte.c4195.099320 .

Kiruthikadevi K. Convolutional neural networks for diabetic retinopathy macular edema from color fundus image. Int J Res Appl Sci Eng Technol (IJRASET). 2021;9(3):1436–40. https://doi.org/10.22214/ijraset.2021.33514 .

Kumar GS, SSAR 1. “A comprehensive review on detecting diabetic eye diseases using deep learning and machine learning models.” Int J Res Appl Sci Eng Technol (IJRASET). 2023;11(9):49–58. https://doi.org/10.22214/ijraset.2023.55596 .

L Wu. Classification of diabetic retinopathy and diabetic macular edema. 2013;4(6):290–290. https://doi.org/10.4239/wjd.v4.i6.290 .

Reddy VPC, Gurrala KK. Joint DR-DME classification using deep learning-CNN based modified grey-wolf optimizer with variable weights. Biomed Signal Process Control. 2022;73:103439–103439.

Balasuganya B, Chinnasamy A, Sheela D. An effective framework for the classification of retinopathy grade and risk of macular edema for diabetic retinopathy images. J Med Imaging Health Inf. 2022;12:138–48. https://doi.org/10.1166/jmihi.2022.3933 .

Gangaputra S, Lovato JF, Larry Hubbard, Davis MD, Esser BA, Ambrosius WT, Chew EY, Greven C, Perdue LH, Wong WT, Audree Condren, Wilkinson CP, Agrón E, Adler S, Danis RP, ACCORD Eye Research Group. Comparison of standardized clinical classification with fundus photograph grading for the assessment of diabetic retinopathy and diabetic macular edema severity. Retina. 2023;33(7):1393–9. https://doi.org/10.1097/IAE.0b013e318286c952 .

Xu J, Yang W, Wan C, Shen J. Weakly supervised detection of central serous chorioretinopathy based on local binary patterns and discrete wavelet transform. Comput Biol Med. 2020;127: 104056. https://doi.org/10.1016/j.compbiomed.2020.104056 . Epub 2020 Oct 14. PMID: 33096297.

Abiodun OI, Jantan A, Omolara AE, Dada KV, Mohamed NA, Arshad H. State-of-the-art in artificial neural network applications: a survey. Heliyon. 2018;4(11): e00938. https://doi.org/10.1016/j.heliyon.2018.e00938 . PMID: 30519653; PMCID: PMC6260436.

Shen H, Mellempudi N, He X, Gao Q, Wang C, Wang M. Efficient post-training quantization with fp8 formats. ArXiv. /abs/2309.14592. 2023.

Belgacem R, et al. Applying a set of gabor filter to 2D- retinal Fundus image to detect the Optic nerve Head (ONH). Ann Med Health Sci Res. 2018;8:48–58.

Chen RC, Dewi C, Huang SW, et al. Selecting critical features for data classification based on machine learning methods. J Big Data. 2020;7:52. https://doi.org/10.1186/s40537-020-00327-4 .

Article CAS Google Scholar

Wang H, Chen Z, Liu G. An Improved Chicken Swarm Optimization Algorithm for Feature Selection. In: Qian, Z., Jabbar, M., Li, X, editors Proceeding of 2021 International Conference on Wireless Communications, Networking and Applications. WCNA 2021. Lecture Notes in Electrical Engineering. Springer, Singapore; 2022. https://doi.org/10.1007/978-981-19-2456-9_19 .

Decencière, et al. Feedback on a publicly distributed database: the Messidor database. Image Analysis & Stereology. 2014;v. 33(n. 3):231–4 ISSN 1854–5165.

Zhang X, Xiao Z, Fu H, et al. Attention to region: region-based integration-and-recalibration networks for nuclear cataract classification using AS-OCT images. Med Image Anal. 2022;80: 102499.

Zhang X, Xiao Z, Yang B, et al. Regional context-based recalibration network for cataract recognition in AS-OCT. Pattern Recogn. 2024;147: 110069.

Xiao Z, Zhang X, Zheng B, et al. Multi-style spatial attention module for cortical cataract classification in AS-OCT image with supervised contrastive learning. Comput Methods Programs Biomed. 2024;244: 107958.

Download references

Acknowledgements

We would like to thank VIT Chennai for providing funding for open access publication.

Open access funding provided by Vellore Institute of Technology. This research received funding from VIT Chennai for open access publication.

Author information

Authors and affiliations.

School of Computer Science & Engineering, Vellore Institute of Technology, Chennai, India

V Thanikachalam, K Kabilan & Sudheer Kumar Erramchetty

You can also search for this author in PubMed Google Scholar

Contributions

All authors contributed significantly to the development and completion of this manuscript. Their specific contributions are detailed below:- Thanikachalam V: Conceptualization, methodology, formal analysis, investigation, and writing—original draft preparation.- Kabilan K: Data curation, software implementation, visualization, and writing—review and editing.- Sudheer Kumar Erramchetty: Supervision, project administration, funding acquisition, and writing—review and editing.Each author has approved the submitted version and has agreed to be personally accountable for their contributions to the work, ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Corresponding authors

Correspondence to V Thanikachalam or Sudheer Kumar Erramchetty .

Ethics declarations

Ethics approval and consent to participate.

Not applicable. This study did not involve any human or animal subjects that require ethics approval.

Consent for publication

Competing interests.

The authors declare no competing interests.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Thanikachalam, V., Kabilan, K. & Erramchetty, S.K. Optimized deep CNN for detection and classification of diabetic retinopathy and diabetic macular edema. BMC Med Imaging 24 , 227 (2024). https://doi.org/10.1186/s12880-024-01406-1

Download citation

Received : 03 July 2024

Accepted : 21 August 2024

Published : 28 August 2024

DOI : https://doi.org/10.1186/s12880-024-01406-1

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Retinal Fundus Image
Discreate Wavelet transform
Artificial neural network
Deep convolutional neural network

BMC Medical Imaging

ISSN: 1471-2342

General enquiries: [email protected]

Macular Edema in Diabetes

Author: Emmanouil Mavrikakis, MD, PhD; Chief Editor: Andrew A Dahl, MD, FACS more...
Sections Macular Edema in Diabetes
Practice Essentials
Pathophysiology
Epidemiology
Physical Examination
Approach Considerations
Intravitreal Treatment
Laser Treatments
Future Therapies
Pars Plana Vitrectomy
Complications
Medication Summary
Corticosteroids, Ophthalmic
Ophthalmics, VEGF Inhibitors

Macular edema in diabetes, defined as retinal thickening within two disc diameters of the center of the macula, results from retinal microvascular changes that compromise the blood-retinal barrier, causing leakage of plasma constituents into the surrounding retina and, consequently, retinal edema. [ 1 ]

Focal edema is associated with hard exudate rings caused by leakage from microaneurysms. Diffuse edema is caused by leakage from microaneurysms, retinal capillaries, and arterioles.

Diabetes is the leading cause of new blindness in the United States, with clinically significant macular edema (CSME) contributing greatly to this vision loss.

Signs and symptoms

The following findings indicate the presence of clinically significant macular edema (CSME), as defined by the Early Treatment Diabetic Retinopathy Study (ETDRS):

Retinal thickening within 500 µm of the center of the fovea

Hard, yellow exudates within 500 µm of the center of the fovea with adjacent retinal thickening

At least one disc area of retinal thickening, any part of which is within one disc diameter of the center of the fovea

See Clinical Presentation for more detail.

Diabetic macular edema (DME) is diagnosed by funduscopic examination. The following studies can also be performed, to provide information for treatment and follow-up:

Optical coherence tomography (OCT): Captures reflected light from retinal structures to create a cross-sectional image of the retina, which is comparable to histologic sections as seen with a light microscope; it can demonstrate three basic structural changes of the retina from diabetic macular edema: retinal swelling, cystoid edema, and serous retinal detachment

Fluorescein angiography: Distinguishes and localizes areas of focal versus diffuse leakage, thereby guiding the placement of laser photocoagulation

Color stereo fundus photographs: Can be used to evaluate long-term changes in the retina

Visual acuity should also be measured. Although it does not aid in the diagnosis of CSME—initially, at least, patients may have a visual acuity of 20/20—it is an important parameter in following the progression of macular edema.

Laboratory studies

Protein levels: Proteinuria is a good marker for the development of diabetic retinopathy; thus, patients with diabetic nephropathy should be observed more closely

Lipid and triglyceride levels: Elevated triglyceride and lipid levels increase the risk for retinopathy, whereas normalization of lipid levels reduces retinal leakage and deposition of exudates

See Workup for more detail.

Pharmacologic treatment

Intravitreal treatments for macular edema include the following:

Intravitreal corticosteroids (eg, triamcinolone acetonide, dexamethasone, fluocinolone acetonide)

Intravitreal anti-VEGF agents (eg, aflibercept, brolucizumab, faricimab, or ranibizumab)

Laser treatment

Laser photocoagulation is a well-proven therapy to reduce the risk for vision loss from diabetic macular edema. Treatments include the following:

Focal treatment: Addresses leaking microaneurysms

Grid pattern photocoagulation: Used for diffuse leakage

See Treatment and Medication for more detail.

The Early Treatment Diabetic Retinopathy Study (ETDRS) set the guidelines for the treatment of diabetic macular edema (DME). Since that time, the standard of treatment for diabetic macular edema has been glycemic control as demonstrated by the Diabetes Control and Complications Trial (DCCT), optimal blood pressure control as demonstrated by the United Kingdom Prospective Diabetes Study (UKPDS), and macular focal/grid laser photocoagulation.

In ETDRS, laser photocoagulation reduced the risk for moderate visual loss from diabetic macular edema by 50% (from 24% to 12% 3 years after initiation of treatment). [ 1 ] Nevertheless, some patients suffer permanent visual loss even after intensive treatment.

Research has started to focus on the use of anti-vascular endothelial growth factor (VEGF) therapy to treat DME. As new and promising treatment options emerge, these treatments will need to be reevaluated.

It is imperative for patients with diabetes to understand that a healthy lifestyle and compliance with medical care can greatly reduce the development and progression of complications of their disease, in the eyes as well as other organs.

For patient education information, see the Diabetes Center , as well as Diabetic Eye Disease .

For further clinical information, see the Medscape Reference articles Diabetes Mellitus, Type 1 , Diabetes Mellitus, Type 2 , and Diabetic Retinopathy .

Diabetic macular edema results from retinal microvascular changes. Thickening of the basement membrane and reduction in the number of pericytes are believed to lead to increased permeability and incompetence of the retinal vasculature. This compromise of the blood-retinal barrier leads to the leakage of plasma constituents into the surrounding retina, with subsequent retinal edema. [ 1 ] Hypoxia produced by this mechanism can also stimulate the production of vascular endothelial growth factor (VEGF). There is evidence that VEGF is up-regulated in diabetic macular edema and proliferative diabetic retinopathy . [ 2 ]

A study suggests that the pathogenesis of diabetic macular edema is not only related to VEGF dependency but also to other inflammatory and angiogenic cytokine levels that can be suppressed by corticosteroids. [ 3 ]

Diabetes is the leading cause of new blindness in the United States, and clinically significant macular edema (CSME) contributes greatly to this vision loss. In the absence of ophthalmologic treatment, persons with diabetes have a 25%-30% risk for moderate vision loss. With treatment, the risk drops by 50%. According to 2007 data, 23.6 million people in the United States have diabetes, but only 17.9 million have been diagnosed. [ 4 ] About 50% of those with diagnosed diabetes do not receive appropriate eye care. The World Health Organization estimates that worldwide, more than 150 million people have diabetes.

Although diabetes is more common in Hispanics, African Americans, and Native Americans than in Whites, no data describe a greater risk of developing macular edema among diabetic patients of any one racial group. Likewise, no data describe a difference in risk for diabetic macular edema between the sexes.

Diabetic retinopathy , not specifically diabetic macular edema, generally occurs in persons older than 40 years. It rarely occurs before puberty.

Albert DM, Jakobiec FA. Principles and Practice of Ophthalmology . 2nd ed. Philadelphia: WB Saunders Co; 2000.

Aiello LP, Avery RL, Arrigg PG, Keyt BA, Jampel HD, Shah ST, et al. Vascular endothelial growth factor in ocular fluid of patients with diabetic retinopathy and other retinal disorders. N Engl J Med . 1994 Dec 1. 331(22):1480-7. [QxMD MEDLINE Link] .

Sohn HJ, Han DH, Kim IT, et al. Changes in aqueous concentrations of various cytokines after intravitreal triamcinolone versus bevacizumab for diabetic macular edema. Am J Ophthalmol . 2011 Oct. 152(4):686-94. [QxMD MEDLINE Link] .

National Diabetes Information Clearinghouse. National Diabetes Statistics, 2007. [Full Text] .

Otani T, Kishi S, Maruyama Y. Patterns of diabetic macular edema with optical coherence tomography. Am J Ophthalmol . 1999 Jun. 127(6):688-93. [QxMD MEDLINE Link] .

Early Treatment Diabetic Retinopathy Study Research Group. Treatment techniques and clinical guidelines for photocoagulation of diabetic macular edema. Early Treatment Diabetic Retinopathy Study Report Number 2. Early Treatment Diabetic Retinopathy Study Research Group. Ophthalmology . 1987 Jul. 94(7):761-74. [QxMD MEDLINE Link] .

Chew EY, Klein ML, Ferris FL 3rd, Remaley NA, Murphy RP, Chantry K, et al. Association of elevated serum lipid levels with retinal hard exudate in diabetic retinopathy. Early Treatment Diabetic Retinopathy Study (ETDRS) Report 22. Arch Ophthalmol . 1996 Sep. 114(9):1079-84. [QxMD MEDLINE Link] .

Bonini-Filho MA, Jorge R, Barbosa JC, Calucci D, Cardillo JA, Costa RA. Intravitreal injection versus sub-Tenon's infusion of triamcinolone acetonide for refractory diabetic macular edema: a randomized clinical trial. Invest Ophthalmol Vis Sci . 2005 Oct. 46(10):3845-9. [QxMD MEDLINE Link] .

Jonas JB, Martus P, Degenring RF, Kreissig I, Akkoyun I. Predictive factors for visual acuity after intravitreal triamcinolone treatment for diabetic macular edema. Arch Ophthalmol . 2005 Oct. 123(10):1338-43. [QxMD MEDLINE Link] .

Patelli F, Fasolino G, Radice P, Russo S, Zumbo G, DI Tizio FM, et al. Time course of changes in retinal thickness and visual acuity after intravitreal triamcinolone acetonide for diffuse diabetic macular edema with and without previous macular laser treatment. Retina . 2005 Oct-Nov. 25(7):840-5. [QxMD MEDLINE Link] .

Avitabile T, Longo A, Reibaldi A. Intravitreal triamcinolone compared with macular laser grid photocoagulation for the treatment of cystoid macular edema. Am J Ophthalmol . 2005 Oct. 140(4):695-702. [QxMD MEDLINE Link] .

Tucker M. FDA OKs Implant (Ozurdex) for Some With Diabetic Macular Edema. Medscape Medical News. Available at https://www.medscape.com/viewarticle/827594 . Accessed: July 5, 2014.

Ozurdex (dexamethasone intravitreal implant) [package insert]. Irvine, CA: Allergan, Inc. September, 2014. Available at [Full Text] .

Boyer DS, Yoon YH, Belfort R Jr, Bandello F, Maturi RK, Augustin AJ, et al. Three-year, randomized, sham-controlled trial of dexamethasone intravitreal implant in patients with diabetic macular edema. Ophthalmology . 2014 Oct. 121(10):1904-14. [QxMD MEDLINE Link] .

Reuters. Eye Implant Iluvien for Diabetic Macular Edema Clears FDA. Medscape Medical News. Available at https://www.medscape.com/viewarticle/832439 . Accessed: October 6, 2014.

Campochiaro PA, Brown DM, Pearson A, Chen S, Boyer D, Ruiz-Moreno J, et al. Sustained delivery fluocinolone acetonide vitreous inserts provide benefit for at least 3 years in patients with diabetic macular edema. Ophthalmology . 2012 Oct. 119(10):2125-32. [QxMD MEDLINE Link] .

Wykoff CC, Abreu F, Adamis AP, and the, YOSEMITE and RHINE Investigators. Efficacy, durability, and safety of intravitreal faricimab with extended dosing up to every 16 weeks in patients with diabetic macular oedema (YOSEMITE and RHINE): two randomised, double-masked, phase 3 trials. Lancet . 2022 Jan 21. [QxMD MEDLINE Link] .

Massin P, Bandello F, Garweg JG, et al. Safety and efficacy of ranibizumab in diabetic macular edema (RESOLVE Study): a 12-month, randomized, controlled, double-masked, multicenter phase II study. Diabetes Care . 2010 Nov. 33(11):2399-405. [QxMD MEDLINE Link] . [Full Text] .

Nguyen QD, Shah SM, Heier JS, et al. Primary End Point (Six Months) Results of the Ranibizumab for Edema of the mAcula in diabetes (READ-2) study. Ophthalmology . 2009 Nov. 116(11):2175-81.e1. [QxMD MEDLINE Link] .

Lang G, RESTORE study group. Safety and efficacy of ranibizumab as monotherapy or adjunctive to laser photocoagulation in diabetic macular edema: 12-month results of the RESTORE study. Late-breaker presentation at EASDec Meeting. May 22, 2010.

Elman MJ, Aiello LP, Beck RW, et al. Randomized trial evaluating ranibizumab plus prompt or deferred laser or triamcinolone plus prompt laser for diabetic macular edema. Ophthalmology . 2010 Jun. 117(6):1064-1077.e35. [QxMD MEDLINE Link] . [Full Text] .

Elman MJ, Bressler NM, Qin H, et al. Expanded 2-Year Follow-up of Ranibizumab Plus Prompt or Deferred Laser or Triamcinolone Plus Prompt Laser for Diabetic Macular Edema. Ophthalmology . 2011 Apr. 118(4):609-14. [QxMD MEDLINE Link] .

Elman MJ, Qin H, Aiello LP, Beck RW, Bressler NM, Ferris FL 3rd, et al. Intravitreal Ranibizumab for Diabetic Macular Edema with Prompt versus Deferred Laser Treatment: Three-Year Randomized Trial Results. Ophthalmology . 2012 Sep 19. [QxMD MEDLINE Link] .

Nguyen QD, Brown DM, Marcus DM, Boyer DS, Patel S, Feiner L, et al. Ranibizumab for diabetic macular edema: results from 2 phase III randomized trials: RISE and RIDE. Ophthalmology . 2012 Apr. 119(4):789-801. [QxMD MEDLINE Link] .

Maturi RK, Glassman AR, Liu D, and the, Diabetic Retinopathy Clinical Research Network. Effect of Adding Dexamethasone to Continued Ranibizumab Treatment in Patients With Persistent Diabetic Macular Edema: A DRCR Network Phase 2 Randomized Clinical Trial. JAMA Ophthalmol . 2018 Jan 1. 136 (1):29-38. [QxMD MEDLINE Link] . [Full Text] .

Korobelnik JF, Do DV, Schmidt-Erfurth U, Boyer DS, Holz FG, Heier JS, et al. Intravitreal Aflibercept for Diabetic Macular Edema. Ophthalmology . 2014 Jul 8. [QxMD MEDLINE Link] .

Ghorayeb G. Intravitreal aflibercept 8 mg for diabetic macular edema: Week 48 efficacy outcomes by baseline demographics in the phase 2/3 PHOTON trial (abstract). Presented at the 2023 ARVO Annual Meeting. New Orleans, LA. Investigative Ophthalmology & Visual Science . June 2023. 64(2727): [Full Text] .

Cunningham ET Jr, Adamis AP, Altaweel M, Aiello LP, Bressler NM, D'Amico DJ, et al. A phase II randomized double-masked trial of pegaptanib, an anti-vascular endothelial growth factor aptamer, for diabetic macular edema. Ophthalmology . 2005 Oct. 112(10):1747-57. [QxMD MEDLINE Link] .

Michaelides M, Kaines A, Hamilton RD, et al. A prospective randomized trial of intravitreal bevacizumab or laser therapy in the management of diabetic macular edema (BOLT study) 12-month data: report 2. Ophthalmology . 2010 Jun. 117(6):1078-1086.e2. [QxMD MEDLINE Link] .

Rajendram R, Fraser-Bell S, Kaines A, Michaelides M, Hamilton RD, Esposti SD, et al. A 2-Year Prospective Randomized Controlled Trial of Intravitreal Bevacizumab or Laser Therapy (BOLT) in the Management of Diabetic Macular Edema: 24-Month Data: Report 3. Arch Ophthalmol . 2012 Apr 9. [QxMD MEDLINE Link] .

Diabetic Retinopathy Clinical Research Network, Wells JA, Glassman AR, Ayala AR, Jampol LM, Aiello LP, et al. Aflibercept, bevacizumab, or ranibizumab for diabetic macular edema. N Engl J Med . 2015 Mar 26. 372 (13):1193-203. [QxMD MEDLINE Link] .

Wells JA, Glassman AR, Ayala AR, et al. Aflibercept, bevacizumab or ranibizumab for diabetic macular edema: two-year results from comparative effectiveness randomized clinical trial. Ophthalmology . June 2016. 123(6):1351-1359. [QxMD MEDLINE Link] . [Full Text] .

Baker CW, Glassman AR, Beaulieu WT, Antoszyk AN, Browning DJ, Chalam KV, et al. Effect of Initial Management With Aflibercept vs Laser Photocoagulation vs Observation on Vision Loss Among Patients With Diabetic Macular Edema Involving the Center of the Macula and Good Visual Acuity: A Randomized Clinical Trial. JAMA . 2019 May 21. 321 (19):1880-1894. [QxMD MEDLINE Link] .

Diabetic Retinopathy Clinical Research Network. A randomized trial comparing intravitreal triamcinolone acetonide and focal/grid photocoagulation for diabetic macular edema. Ophthalmology . 2008 Sep. 115(9):1447-9, 1449.e1-10. [QxMD MEDLINE Link] . [Full Text] .

O'Day R, Barthelmes D, Zhu M, Wong TY, McAllister IL, Arnold JJ, et al. Baseline central macular thickness predicts the need for retreatment with intravitreal triamcinolone plus laser photocoagulation for diabetic macular edema. Clin Ophthalmol . 2013. 7:1565-70. [QxMD MEDLINE Link] . [Full Text] .

Hsu J. Drug delivery methods for posterior segment disease. Curr Opin Ophthalmol . 2007 May. 18(3):235-9. [QxMD MEDLINE Link] .

Lewis H, Abrams GW, Blumenkranz MS, Campo RV. Vitrectomy for diabetic macular traction and edema associated with posterior hyaloidal traction. Ophthalmology . 1992 May. 99(5):753-9. [QxMD MEDLINE Link] .

Tachi N, Ogino N. Vitrectomy for diffuse macular edema in cases of diabetic retinopathy. Am J Ophthalmol . 1996 Aug. 122(2):258-60. [QxMD MEDLINE Link] .

Fong DS, Segal PP, Myers F, Ferris FL, Hubbard LD, Davis MD. Subretinal fibrosis in diabetic macular edema. ETDRS report 23. Early Treatment Diabetic Retinopathy Study Research Group. Arch Ophthalmol . 1997 Jul. 115(7):873-7. [QxMD MEDLINE Link] .

Takagi H, Otani A, Kiryu J, Ogura Y. New surgical approach for removing massive foveal hard exudates in diabetic macular edema. Ophthalmology . 1999 Feb. 106(2):249-56; discussion 256-7. [QxMD MEDLINE Link] .

Contributor Information and Disclosures

Emmanouil Mavrikakis, MD, PhD Ophthalmologist-in-Chief, Consultant Vitreoretinal Surgeon, Ophthalmology Department, G Gennimatas Athens General Hospital, Greece Emmanouil Mavrikakis, MD, PhD is a member of the following medical societies: American Academy of Ophthalmology , American Society of Retina Specialists , European Society of Cataract and Refractive Surgery Disclosure: Nothing to disclose.

Baseer U Khan, MD Associate Professor of Ophthalmology, University of Toronto Faculty of Medicine; Ophthalmologist, Clarity Eye Institute, Canada Baseer U Khan, MD is a member of the following medical societies: Canadian Ophthalmological Society Disclosure: Nothing to disclose.

Wai-Ching Lam, MD, FRCSC Professor, Department of Ophthalmology and Vision Sciences, University of Toronto Faculty of Medicine, Canada Wai-Ching Lam, MD, FRCSC is a member of the following medical societies: American Academy of Ophthalmology , Canadian Ophthalmological Society , Royal College of Physicians and Surgeons of Canada Disclosure: Received honoraria from Novartis for speaking and teaching; Received honoraria from Allergan for review panel membership; Received honoraria from Alcon for review panel membership; Received honoraria from Bayer for consulting; Received honoraria from Alcon for review panel membership; Received honoraria from Novartis for review panel membership.

Francisco Talavera, PharmD, PhD Adjunct Assistant Professor, University of Nebraska Medical Center College of Pharmacy; Editor-in-Chief, Medscape Drug Reference Disclosure: Received salary from Medscape for employment. for: Medscape.

Steve Charles, MD Founder and CEO of Charles Retina Institute; Clinical Professor, Department of Ophthalmology, University of Tennessee College of Medicine Disclosure: Received royalty and consulting fees for: Alcon Laboratories.

Andrew A Dahl, MD, FACS Assistant Professor of Surgery (Ophthalmology), New York College of Medicine (NYCOM); Director of Residency Ophthalmology Training, The Institute for Family Health and Mid-Hudson Family Practice Residency Program; Staff Ophthalmologist, Telluride Medical Center Andrew A Dahl, MD, FACS is a member of the following medical societies: American Academy of Ophthalmology , American College of Surgeons , American Intraocular Lens Society, American Medical Association , American Society of Cataract and Refractive Surgery , Contact Lens Association of Ophthalmologists , Medical Society of the State of New York , New York State Ophthalmological Society , Outpatient Ophthalmic Surgery Society Disclosure: Nothing to disclose.

What would you like to print?

Print this section
Print the entire contents of
Print the entire contents of article

Fast Five Quiz: Diabetic Macular Edema
Fast Five Quiz: Type 2 Diabetes and Glucose Monitoring
Fast Five Quiz: Hyperglycemia Associated With Type 2 Diabetes
Fast Five Quiz: Glycemic Control in Type 2 Diabetes
Fast Five Quiz: Do You Know What to Watch for and How Best to Treat Diabetic Foot Ulcers?
Fast Five Quiz: Type 2 Diabetes and Hypertension
Fast Five Quiz: Type 2 Diabetes and Coronary Heart Disease
Processed Foods Linked to Diabetic Complications
Mental Health and Diabetic Complications: A Two-Way Link
New Perspective on Diabetic Neuropathy Emerges

Bariatric Surgery Still Best Option for Some With Obesity

Drug Interaction Checker
Pill Identifier
Calculators

2001/s/viewarticle/987718 Healthy Habits Lower T2D Microvascular Risks: Cohort Study
2003/viewarticle/998209Clinical Review Medscape Now! Hot Topics in Family Medicine November 2023 Part 1 0.5 CME / CE / ABIM MOC Credits Clinical Review You are being redirected to Medscape Education Yes, take me there 0.5 CME / CE / ABIM MOC Medscape Now! Hot Topics in Family Medicine November 2023 Part 1
2001/viewarticle/processed-foods-linked-diabetic-complications-2024a1000exxnews news Processed Foods Linked to Diabetic Complications
2001https://www.medscape.com/viewcollection/37483Conference Coverage Conference Coverage European Association for the Study of Diabetes (EASD) 2024 Annual Meeting

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Publications
Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

Advanced Search
Journal List
Rom J Ophthalmol
v.59(3); Jul-Sep 2015

DIABETIC MACULAR EDEMA

Ovidiu musat.

* Central Military Emergency University Hospital „Dr. Carol Davila”, Ophthalmology Department, Bucharest, Romania

Corina Cernat

Mahdi labib, andreea gheorghe, madalina zamfir, ana maria boureanu.

Diabetic macular edema (DME) remains the most common cause of vision loss among diabetic patients. New understanding of the underlying pathophysiology has interest in the potential benefits of the specific pharmacologic therapy, such as treatment with intraocular steroids, anti-vascular endothelial growth factor (VEGF), and protein kinase C-beta (PKCβ) inhibition. At the last time, laser photocoagulation, according to the guidelines of the Early Treatment of Diabetic Retinopathy Study (ETDRS), continues to be primary standard care treatment in most communities.

Optical coherence tomography (OCT) is very useful in monitoring macular edema progression and response to treatment.

Diabetic macular edema (DME) is manifested as retinal thickening caused by the accumulation of intraretinal fluid, primarily in the inner and outer plexiform layers. It is believed to be a result of hyperpermeability of the retinal vasculature. DME can be present with any level of diabetic retinopathy.

ETDRS Criteria for Clinically Significant Macular Edema (CSME)(1):

• Retinal thickening at the center of the macula

• Retinal thickening and/or adjiacent hard exudates at or within 500 µ of the center of the macula ( Fig. 1 )

An external file that holds a picture, illustration, etc.
Object name is RomJOphthalmol-59-133-g001.jpg

• An area of retinal thickening greater than or equal to one disc area, any part of which is within 1 disc diameter of the center of the macula

Prevalence and incidence [ 1 ] [ 2 ]

In USA: The WHO (World Health Organization) estimates 15 million DME half undiagnosed and 50% of 8 million without eye care, 25-30% risk of vision loss from CSME. International, WHO estimates more than 150 million patients with diabetes worldwide. However, the absolute prevalence of DME might be increasing due to the overall increased prevalence of diabetes in industrialized nations. It is expected that the incidence of DME will decrease as excellent metabolic control is increasingly embraced as a therapeutic goal by patients and health care workers.

Pathology and pathophysiology of dme

Normal retinal circulation is unique: retinal capillaries are non-fenestrated and capillary endothelial cells have tight jonctions; normal capillaries do not leak fluid, blood. There is no lymphatc system in the retina, so in the presence of retinal pathology, leaking fluid can accumulate and cause edema or swelling. Retina responds to ischemia by stimulating growth factors to produce new vessels (called neovascularization).

DME is the result of microvascular changes in diabetes leading to incompetence of vessels, edema. Hypoxic state stimulate VEGF causing more edema.

Thus, 2 key changes occur:

• Vessel permeability

- Damaged endothelial wall becomes more porous

- Vessel leaks fluid, lipids, erythrocytes

- Accumulation of the fluid results in edema (macular edema if located within the central region of the retina)

• Vessel closure

- Supply of oxigen and nutrients are decreased

New fragile growth occurs (secondary to ischemia)

An external file that holds a picture, illustration, etc.
Object name is RomJOphthalmol-59-133-g002.jpg

Photomicrograph of cystoids spaces and subretinal fluid in the retina of a diabetic patient with severe DME

Clinical associations and risk factors

Macular edema is strongly positively associated with diabetic retinopathy severity. Glycemic control is a conclusively identifies risk factor for retinopathy progression as well as for DME. Duration of diabetes is strongly correlated with prevalence and incidence of macular edema, retinopathy progression, and other diabetic complications. The diagnosis of diabetes in type 2 subjects occasionally occurs sometime after subclinical diabetes has been manifest, which yields a small proportion of patients who may present with macular edema at the time of diagnosis, or even have decreased vision from macular edema at the presenting sign. In contrast, persons with type 1 diabetes are very unlikely to experience advanced retinopathy and macular edema before 5 years of duration.

Clinical Associations with Diabetic Macular Edema Severity: [ 2 ] [ 3 ]

• Duration of Diabetes – increased risk of diabetic retinopathy

• Glycemic control – The Diabetes Control and Complication Trial (DCCT) clearly demonstrated that tighter control of blood sugar is associated with reduced incidence of diabetic retinopathy (Glycosylated hemoglobin (HbA1c) should be less than 7%)

• Nephropathy – proteinuria is a good marker for development of diabetic retinopathy; thus, patients with diabetic with nephropathy should be observed more closely

• Hypertension – increased risk of retinopathy (diabetic retinopathy with superimposed hypertensive retinopathy)

• Dislipidemia – normalization of lipid levels reduces retinal leakage and exudates deposition

• Pregnancy – diabetic retinopathy can progress rapidly in pregnant women, especially those with preexisting diabetic retinopathy

• Intraocular surgery

• Uveitis

• Panretinal Photocoagulation

Clinical presentation of diabetic macular edema

Patients with DME present with a range of visual symptoms depending on the degree to which the fovea is involved and the chronicity of the edema. If the macula center is not involved patients are rarely symptomatic; only a few very observant individuals may notice relative paracentral scotomas corresponding to focal edema and hard exudates. Some patients with central macular involvement have excellent acuity and no visual complains, presumably because of only recent involvement of the center. Over time, patients experience a gradual progressive vision loss over weeks to month. Patients may complain of loss of color vision, poor night vision and washing-out of vision in bright sunlight with poor dark-light adaptation.

Metamorphopsia is not uncommon. Frequently, patients with center involved DME note fluctuation of vision from day-to-day or even over the course of a day. In some cases, the patient may relate such changes to fluid retention, hyper or hypoglycemia, or ambient lighting.[ 4 ][ 5 ]

On fundus examination with slit lamp biomicroscopy or contact lens, retinal thickening may present in some commonly identified patterns. Focal edema often occurs associated with a cluster of microaneurysms, sometimes surrounded by an incomplete ring of hard exudates. Diffuse DME may be very difficult to identify clinically if the retina is of uniform thickness, due to the lack of reference landmarks. Clues include the height of the retinal blood vessels over the pigment epithelium, loss of the foveal depression or even cystoids spaces. Other features sometimes seen with macular edema include variable loss of retinal transparency, a large burden of microaneurysms and intraretinal hemorrhages, and dispersed flecks of hard exudates.

An external file that holds a picture, illustration, etc.
Object name is RomJOphthalmol-59-133-g003.jpg

Color photograph of a diabetic patient with focal macular edema, with circinate hard exudates roughly circumscribing the area of retinal thickening

An external file that holds a picture, illustration, etc.
Object name is RomJOphthalmol-59-133-g004.jpg

Color photograph of a diabetic patient with diffuse macular edema

Stereoscopic fundus photographs provide an opportunity to evaluate long-term changes in the retina.

Fluoresceine angiography is useful in demonstrating the breakdown of the blood-retinal barrier by delineating retinal capillary leakage and capillary nonperfusion. Fluorescein angiography is not relevant in aiding in the diagnosis of CSME but should be performed if treatment of CMSE is being considered.

Optical coherence tomography (OCT) is able to demonstrate a moderate correlation between retinal thickness and best corrected visual acuity, and it is able to demonstrate 3 basic structural changes of the retina from diabetic macular edema, that is, retinal swelling, cystoid edema, and serous retinal detachment. OCT is not currently required to establish a diagnosis and is not prescribed by current practice guideline; however, OCT has gained widespread acceptance as an additional modality to help identify and evaluate macular pathology. Quantitative measurement of macular thickness and subjective analysis of the foveal architecture allow a precise and reproducible way to monitor macular edema.

An external file that holds a picture, illustration, etc.
Object name is RomJOphthalmol-59-133-g005.jpg

OCT scan demonstrates cystoid edema

Optimizing diabetic, hypertensive, and lipid control has been shown to positively impact diabetic retinopathy.

The ETDRS conclusively demonstrated that focal/grid laser photocoagulation was safe and effective in reducing vision loss due to DME. Significant visual improvement is uncommon; the goal of macular laser treatment is to reduce progression. [ 6 ] Photocoagulation reduced the risk of moderate visual loss from diabetic macular edema by 50%, from 24% to 12%, 3 years after initiation of treatment. Laser treatment is most effective when initiated before visual acuity is lost from diabetic macular edema; this emphasizes the need for diligent monitoring and follow-up care. Fluorescein angiography and fundus photos are obtained prior to initiation of laser theraphy. Ophthalmologist views the FA to guide treatment of CSME: for focal leakage, direct laser theraphy using green-only Argon laser is applied to all leaking microaneurysm between 500 and 3000 µm from the center of the macula; for diffuse leakage of capillary nonperfusion adjiacent to the macula, a light-intensity grid pattern using green-only Argon laser is applied to all areas of diffuse leakage more than 500µm from the center of the macula and 500µm from the temporal margin of the optic disc. Multiple sessions spread out over many months are frequently necessary for resolution of DME.

Given the importance of VEGF in vascular permeability and its up regulation in diabetic retinopathy, the rationale for use of anti-VEGF drugs is clear. Current specific anti-VEGF therapy is given intravitreal at frequent intervals, which may temporarily blunt the effects of VEGF and lessen macular edema.

Intravitreal triamcinolone acetonide (IVTA) has been shown to significantly reduce macular edema and to improve visual acuity, particularly when macular edema is pronounced. Some studies advocate IVTA as primary therapy, whereas others label it as adjunctive therapy to macular photocoagulation.[ 7 ][ 8 ]

A subset of patients with DME has coexistent epiretinal membranes and/or partial posterior vitreous detachment with retinal traction. These patients may benefit from pars plana vitrectomy to address the mechanical issues contributing to the retinal edema. Even without obvious retinal traction, some clinicians believe that many cases of DME respond to removal of the vitreous, with or without removal of the internal limiting membrane. No large randomized trials have evaluated the treatment to date.[ 9 ]

The landscape of DME management is rapidly changing with the advent of research advances leading to better understanding of pathophysiologic mechanisms and discovery of potential therapeutic compounds. There are several challenges that remain in bringing forth new treatments with adequate evidence base to guide clinicians in timely patient care.

The News Feed

Researchers use algorithm to create new severity scale for diabetic macular ischemia, it’s based on the distribution of capillary nonperfusion on oct-a..

The progression characteristics of diabetic macular ischemia aren’t well understood, but it is known that the condition is accompanied by capillary nonperfusion leading to visual impairment. Recently, researchers aiming to develop a severity scale for diabetic macular ischemia published their findings in the journal Ophthalmology Science . They used a technique called uniform manifold approximation and projection, which can reinterpret complex multidimensional data as a lower-dimensional representation for ease of visualization and data clustering.

The single-center prospective case series included 301 eyes from 301 patients with diabetic retinopathy. The researchers obtained 3x3mm swept source OCT-A images and used them to create en face images in a 2.5mm circle. This area was divided into 15x15-pixel squares. Nonperfusion squares had no retinal vessels. Using the algorithm, the researchers classified eyes into severity groups: initial, mild, superficial, moderate and severe.

The researchers reported that nonperfusion square counts in the deep layer increased in a stepwise fashion in eyes with initial, mild, moderate and severe algorithmic classification. No significant changes in nonperfusion square counts in the superficial layer were seen between the initial and mild groups.

The researchers observed, however, that in the intermediate stage, the superficial group had higher nonperfusion square counts in the central sector of the superficial layer versus those in the moderate group. The moderate group also exhibited a foveal avascular zone that extended into the temporal subfield of the deep layer. The eyes in the severe group had worse visual acuity and had more proliferative disease than the other groups.

“We proposed a novel severity scale for diabetic capillary nonperfusion in the macula on OCTA images,” the researchers wrote in their Ophthalmology Science article. “This preliminary study would contribute to the further investigation to establish the diagnostic and staging criteria for diabetic macular ischemia.”

Yoshida M, Murakami T, Nishikawa K, et al. Severity scale of diabetic macular ischemia based on the distribution of capillary nonperfusion in OCT angiography. Ophthalmol Sci 2024. [Epub ahead of print].

Hormone Therapy in Postmenopausal Women Linked with Later Glaucoma Diagnosis

Extent of crora with central foveal involvement may inform visual prognosis, incorporating oct imaging can enhance cataract assessment accuracy, new study finds, macular ganglion cell analysis can monitor glaucoma.

Information

Author Services

Initiatives

You are accessing a machine-readable page. In order to be human-readable, please install an RSS reader.

All articles published by MDPI are made immediately available worldwide under an open access license. No special permission is required to reuse all or part of the article published by MDPI, including figures and tables. For articles published under an open access Creative Common CC BY license, any part of the article may be reused without permission provided that the original article is clearly cited. For more information, please refer to https://www.mdpi.com/openaccess .

Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature Paper should be a substantial original Article that involves several techniques or approaches, provides an outlook for future research directions and describes possible research applications.

Feature papers are submitted upon individual invitation or recommendation by the scientific editors and must receive positive feedback from the reviewers.

Editor’s Choice articles are based on recommendations by the scientific editors of MDPI journals from around the world. Editors select a small number of articles recently published in the journal that they believe will be particularly interesting to readers, or important in the respective research area. The aim is to provide a snapshot of some of the most exciting work published in the various research areas of the journal.

Original Submission Date Received: .

Active Journals
Find a Journal
Proceedings Series
For Authors
For Reviewers
For Editors
For Librarians
For Publishers
For Societies
For Conference Organizers
Open Access Policy
Institutional Open Access Program
Special Issues Guidelines
Editorial Process
Research and Publication Ethics
Article Processing Charges
Testimonials
Preprints.org
SciProfiles
Encyclopedia

Article Menu

Subscribe SciFeed
Google Scholar
on Google Scholar

Find support for a specific problem in the support section of our website.

Please let us know what you think of our products and services.

Visit our dedicated information section to learn more about MDPI.

Article Versions Notes

Action	Date	Notes	Link
article xml file uploaded	28 August 2024 11:16 CEST	Original file	-
article xml uploaded.	28 August 2024 11:16 CEST	Update
article pdf uploaded.	28 August 2024 11:16 CEST	Version of Record
article supplementary file uploaded.	28 August 2024 11:16 CEST	-
article html file updated	28 August 2024 11:18 CEST	Original file

Nasimi, S.; Nasimi, N.; Grauslund, J.; Vergmann, A.S.; Subhi, Y. Real-World Efficacy of Intravitreal Faricimab for Diabetic Macular Edema: A Systematic Review. J. Pers. Med. 2024 , 14 , 913. https://doi.org/10.3390/jpm14090913

Nasimi S, Nasimi N, Grauslund J, Vergmann AS, Subhi Y. Real-World Efficacy of Intravitreal Faricimab for Diabetic Macular Edema: A Systematic Review. Journal of Personalized Medicine . 2024; 14(9):913. https://doi.org/10.3390/jpm14090913

Nasimi, Safiullah, Nasratullah Nasimi, Jakob Grauslund, Anna Stage Vergmann, and Yousif Subhi. 2024. "Real-World Efficacy of Intravitreal Faricimab for Diabetic Macular Edema: A Systematic Review" Journal of Personalized Medicine 14, no. 9: 913. https://doi.org/10.3390/jpm14090913

Article Metrics

Article access statistics, further information, mdpi initiatives, follow mdpi.

Subscribe to receive issue release notifications and newsletters from MDPI journals

COMMENTS

Diabetic Macular Edema: Diagnosis and Management
Download PDF. Diabetic retinopathy (DR) is the leading cause of new cases of blindness among adults aged 18 to 64 years in the United States. 1 Diabetic macular edema (DME), a severe complication of DR that occurs specifically as a result of inadequately treated diabetes mellitus (DM), has overtaken proliferative diabetic retinopathy as the ...
Diabetic Macular Edema
Figure 1: Diabetic macular edema. A: Schematic diagram of DME. Microaneurysms or damaged capillaries resulting from the breakdown of the blood-retina barrier leak fluid to the extracellular space, resulting in a swollen retina. Resorption of DME is dependent on the adjacent capillaries and retinal pigment epithelium. Resorption of fluid may ...
Diabetic Macular Edema
Diabetic macular edema (DME) is one of the leading causes of vision loss worldwide. It affects over 75,000 Americans yearly, and nearly 100 million people worldwide show some signs of macular edema secondary to diabetes. ... The most important is the initial presenting severity of the edema: the more severe the initial presentation, the more ...
Macular Edema in Diabetes Clinical Presentation
After 20 years of disease, nearly all patients with type 1 and 60% of patients with type 2 diabetes have some degree of retinopathy. The risk increases with the duration of disease. Hence, diabetic retinopathy is more likely to be present in patients older than 40 years. The Diabetes Control and Complications Trial (DCCT) clearly demonstrated ...
Diabetic macular edema: Evidence-based management
Diabetic macular edema (DME) is the most common cause of vision loss in patients with diabetic retinopathy with an increasing prevalence tied to the global epidemic in type 2 diabetes mellitus. ... has been studied as a potential primary therapy in eyes with more severe edema and greater visual acuity loss at presentation ...
Macular Edema
Macular edema is a retinal condition characterized by fluid buildup in the central part of the retina responsible for sharp, central vision, and accompanies various retinal diseases such as diabetic retinopathy, retinal vascular occlusions, and uveitis.[1][2] This condition causes decreased visual acuity and, when persistent, can lead to severe vision loss.[3] Fluid accumulates due to a ...
Diabetic Macular Edema: Diagnosis and Management
Macular Edema: Diagnosis and ManagementDiabeticretinopathy (DR) is the leading cause of new cases of blindness among adults aged 18 to 64 years in the United States.1 Dia betic macular edema (DME), a severe complication of DR that occurs specifi cally as a result of inadequately treated diabetes mellitus (DM), has overtaken proliferative ...
PDF Your guide to diabetic macular oedema
The Macular Society is the only charity dedicated to beating macular disease. This guide explains how diabetes can affect the eyes. Being diagnosed with diabetic macular oedema can be distressing and worrying but, with the right information and support, people can cope very well. Macular oedema is painless and treatments are available.
What To Know About Diabetic Macular Edema
Diabetes can damage the blood vessels in your retina (the very back layer of your eyeball). That's called diabetes-related retinopathy. One way that the blood vessels are damaged is when the ...
Diabetic macular oedema
Diabetic macular oedema, characterised by exudative fluid accumulation in the macula, is the most common form of sight-threatening retinopathy in people with diabetes. It affects one in 15 people with diabetes resulting in more than 20 million cases worldwide. Few epidemiological studies have been done to specifically investigate risk factors for diabetic macular oedema, although poor ...
Diabetic Macular Edema: Types, Symptoms, Treatment
Diabetic macular edema (DME) is a complication of diabetic retinopathy. It occurs when fluid builds up in the macula, the center portion of the retina. The retina is the layer of cells at the back of your eye that helps convert light into the images you see. Diabetes is the leading cause of new blindness in the United States.
Diabetic Macular Edema: Pathophysiology and Novel ...
Diabetic macular edema (DME) is the major cause of vision loss in diabetic persons. Alteration of the blood-retinal barrier is the hallmark of this disease, characterized by pericyte loss and endothelial cell-cell junction breakdown. Recent animal and clinical studies strongly indicate that DME is an inflammatory disease. Multiple cytokines and chemokines are involved in the pathogenesis ...
Diabetic Macular Edema
One of the most common optical coherence tomography (OCT) patterns of diabetic macular edema is diabetic cystoid macular edema (DCME) where intraretinal fluid accumulates in cystoid-like spaces.1 Diabetic cystoid macular edema has been associated with a worse visual outcome,1 presumably owing to retinal cell damage and subsequent degenerative changes. Although OCT offers a simple and reliable ...
Diabetic macular edema: pathogenesis and treatment
Diabetic macular edema is a major cause of visual impairment. The pathogenesis of macular edema appears to be multifactorial. Laser photocoagulation is the standard of care for macular edema. However, there are cases that are not responsive to laser therapy. Several therapeutic options have been proposed for the treatment of this condition.
Diabetic Retinopathy Clinical Presentation
Macular edema is the leading cause of visual impairment in patients with diabetes. It may result from functional damage and necrosis of retinal capillaries. In cases of PDR, edema also may be caused by retinal traction if the retina is sufficiently elevated away from the retinal pigment epithelium.
Guidelines for the Management of Center-Involving Diabetic Macular
Introduction. Diabetic macular edema (DME) commonly leads to impairment of visual acuity in patients with non-proliferative diabetic retinopathy (NPDR) and is more prevalent in patients with type 2 diabetes mellitus (DM2) than type 1 (DM1), accounting to nearly 12.9% and 7.86% cases, respectively. 1 The causes of DME are multifactorial; however, the predominant reason is attributable to blood ...
Diabetes-Related Retinopathy: Causes, Treatment, Signs & Symptoms
Diabetic macular edema. DME is swelling and thickening of the macula and is the leading cause of vision loss in people with DR. It can develop during any DR stage and may require treatment, even if it appears in mild NPDR. Neovascularization occurs during the advanced stage of DR. This stage is called proliferative diabetic retinopathy.
PDF Faricimab in Diabetic Macular Edema: Two-Year Results From the ...
YOSEMITE: NCT03622580; RHINE: NCT03622593. a CST was measured as the distance from the internal limiting membrane to Bruch's membrane.b BCVA was measured using the ETDRS visual acuity chart at a starting distance of 4 m. c Previously anti-VEGF-treated eyes (treated ≥ 3 months before day 1) were limited to 25% of the total enrolment.d Primary efficacy endpoint: adjusted mean BCVA change ...
MAPLES-DR: MESSIDOR Anatomical and Pathological Labels for ...
Reliable automatic diagnosis of Diabetic Retinopathy (DR) and Macular Edema (ME) is an invaluable asset in improving the rate of monitored patients among at-risk populations and in enabling ...
Evaluating Ocular Blood Flow in Diabetic Macular Edema using ...
The mean ocular blood flow exhibited a significant decrease in eyes with diabetic macular edema when the post-labeling delay was set at 2.5 s in the nonproliferative and proliferative diabetic retinopathy groups, compared with the remaining subgroups (P=0.005 and P=0.002, respectively). The cutoff points of ocular blood flow at post-labeling ...
Diabetic macular edema: Evidence-based management
Diabetic macular edema (DME) is the most common cause of vision loss in patients with diabetic retinopathy with an increasing prevalence tied to the global epidemic in type 2 diabetes mellitus. Its pathophysiology starts with decreased retinal oxygen tension that manifests as retinal capillary hyperpermeability and increased intravascular ...
Nutrients
Diabetic macular edema (DME), the main cause of vision loss in diabetic individuals, is characterized by a gradual increase in retinal thickening, ... Additionally, the heterogeneity of disease presentation and response to treatment underscores the complexity of DR management . The current literature emphasizes the efficacy of intravitreal ...
Optimized deep CNN for detection and classification of diabetic
Diabetic Retinopathy (DR) and Diabetic Macular Edema (DME) are vision related complications prominently found in diabetic patients. The early identification of DR/DME grades facilitates the devising of an appropriate treatment plan, which ultimately prevents the probability of visual impairment in more than 90% of diabetic patients. Thereby, an automatic DR/DME grade detection approach is ...
Macular Edema in Diabetes
Practice Essentials. Macular edema in diabetes, defined as retinal thickening within two disc diameters of the center of the macula, results from retinal microvascular changes that compromise the blood-retinal barrier, causing leakage of plasma constituents into the surrounding retina and, consequently, retinal edema. [ 1]
Changes in Ocular Blood Flow After Intravitreal Injection for Diabetic
Diabetic retinopathy (DR) is the primary cause of visual impairment among working-age populations in industrialized nations. Vision loss may result from various mechanisms, but the most prevalent cause is diabetic macular edema (DME). Citation 1. The treatment landscape for DME has advanced significantly over the past decade.
DIABETIC MACULAR EDEMA
Clinical presentation of diabetic macular edema. Patients with DME present with a range of visual symptoms depending on the degree to which the fovea is involved and the chronicity of the edema. If the macula center is not involved patients are rarely symptomatic; only a few very observant individuals may notice relative paracentral scotomas ...
Caregiver Experience Survey of Anti-Vascular Endothelial Growth Factor
Abstract. Introduction: Diabetic macular edema (DME) and neovascular age-related macular degeneration (nAMD) require frequent treatment/monitoring visits. We aimed to understand the burden of treatment on caregivers. Methods: This multinational, noninterventional study used a cross-sectional survey of adult patients with DME or nAMD treated with anti-vascular endothelial growth factor ...
Researchers Use Algorithm to Create New Severity Scale for Diabetic
The progression characteristics of diabetic macular ischemia aren't well understood, but it is known that the condition is accompanied by capillary nonperfusion leading to visual impairment. Recently, researchers aiming to develop a severity scale for diabetic macular ischemia published their findings in the journal Ophthalmology Science ...
Enrollment, dosing complete in phase 1 trial of AIV007 for AMD ...
AiViva Biopharma announced that it has completed enrollment and dosing of the last patient in a phase 1 clinical trial of AIV007 for wet age-related macular degeneration and diabetic macular edema ...
JPM
Efficacity and Safety of the Fluocinolone Acetonide Implant in Uveitic Macular Edema: A Real-Life Study from the French Uveitis Network. Journals. Active Journals Find a Journal Proceedings Series. ... A.S.; Subhi, Y. Real-World Efficacy of Intravitreal Faricimab for Diabetic Macular Edema: A Systematic Review. J. Pers. Med. 2024, 14, 913 ...

Diabetic Macular Edema

Disease Definition

Pathophysiology

Natural History

Risk Factors

Physical Examination

Vision and DME

Diagnostic Procedures

OCT Biomarkers for DME

2. Fluorescein Angiography

Laboratory Testing

Differential Diagnosis

Medical therapy

Pharmacotherapy

DRCR Retina Network anti-VEGF treatment algorithm

Clinical considerations in the anti-VEGF treatment of DME

Pharmacotherapy Landmark Studies

Laser Photocoagulation

Combined Therapy

Treatment Complications

Intravitreal injections

Future Directions

Further reading

Diabetic Macular Edema: What to Know

Types of Diabetic Macular Edema

Tests for Diabetic Macular Edema

Symptoms of Diabetic Macular Edema

Causes of Diabetic Macular Edema

Risk Factors for Diabetic Macular Edema

Prevalence of Diabetic Macular Edema

Treating Diabetic Macular Edema

Injectable Anti-VEGF Medications

Laser Therapy

Anti-Inflammatory Medications

Preventing Diabetic Macular Edema

How Often Should I Get an Eye Exam?

A Word From Verywell

Diabetic retinopathy

What is diabetic retinopathy?

Causes of diabetic retinopathy

Who is at risk?

Signs and symptoms

What is the first sign of diabetic retinopathy?

Main types and stages of diabetic retinopathy

Type: Non-proliferative diabetic retinopathy (NPDR)

Type: Proliferative diabetic retinopathy (PDR)

Diagnosis, management and treatment

How is the condition diagnosed?

What treatment options are available?

Can diabetic retinopathy be reversed?

How is the disease managed?

Health risks and complications

When to call an eye doctor

Notes and References

Women’s vision: Understanding risks and ways to protect it

Alice in Wonderland syndrome (AIWS)

Horizontal gaze nystagmus (HGN)

MAPLES-DR: MESSIDOR Anatomical and Pathological Labels for Explainable Screening of Diabetic Retinopathy

Similar content being viewed by others

Self-supervised retinal thickness prediction enables deep learning from unlabelled data to boost classification of diabetic retinopathy

An interpretable multiple-instance approach for the detection of referable diabetic retinopathy in fundus images

Automated image curation in diabetic retinopathy screening using deep learning

Fundus images selection

Procedure to grade DR and ME

Procedure to annotate Retinal Structures

Selection of anatomical and pathological biomarkers

Recruiting annotators and designing the annotation tools

Preprocessing MESSIDOR fundus images

Reviewing and correcting preannotated segmentation maps

Instructions provided to retinologists

Data Records

MAPLES-DR.zip

AdditionalData.zip

Technical Validation

Retinal structures: overview of annotation work

Retinal Lesions: correlation with DR grade

Retinal structures: semantic segmentation baseline

Retinal lesions segmentation: training on MAPLES-DR, testing on public datasets

Retinal lesions segmentation: pretraining with MAPLES-DR

Usage Notes