CauDR: A causality-inspired domain generalization framework for fundus-based diabetic retinopathy grading.

Diabetic retinopathy (DR) is the most common diabetic complication, which usually leads to retinal damage, vision loss, and even blindness. A computer-aided DR grading system has a significant impact on helping ophthalmologists with rapid screening and diagnosis. Recent advances in fundus photography have precipitated the development of novel retinal imaging cameras and their subsequent implementation in clinical practice. However, most deep learning-based algorithms for DR grading demonstrate limited generalization across domains. This inferior performance stems from variance in imaging protocols and devices inducing domain shifts. We posit that declining model performance between domains arises from learning spurious correlations in the data. Incorporating do-operations from causality analysis into model architectures may mitigate this issue and improve generalizability. Specifically, a novel universal structural causal model (SCM) was proposed to analyze spurious correlations in fundus imaging. Building on this, a causality-inspired diabetic retinopathy grading framework named CauDR was developed to eliminate spurious correlations and achieve more generalizable DR diagnostics. Furthermore, existing datasets were reorganized into 4DR benchmark for DG scenario. Results demonstrate the effectiveness and the state-of-the-art (SOTA) performance of CauDR. Diabetic retinopathy (DR) is the most common diabetic complication, which usually leads to retinal damage, vision loss, and even blindness. A computer-aided DR grading system has a significant impact on helping ophthalmologists with rapid screening and diagnosis. Recent advances in fundus photography have precipitated the development of novel retinal imaging cameras and their subsequent implementation in clinical practice. However, most deep learning-based algorithms for DR grading demonstrate limited generalization across domains. This inferior performance stems from variance in imaging protocols and devices inducing domain shifts. We posit that declining model performance between domains arises from learning spurious correlations in the data. Incorporating do-operations from causality analysis into model architectures may mitigate this issue and improve generalizability. Specifically, a novel universal structural causal model (SCM) was proposed to analyze spurious correlations in fundus imaging. Building on this, a causality-inspired diabetic retinopathy grading framework named CauDR was developed to eliminate spurious correlations and achieve more generalizable DR diagnostics. Furthermore, existing datasets were reorganized into 4DR benchmark for DG scenario. Results demonstrate the effectiveness and the state-of-the-art (SOTA) performance of CauDR.

ASSOCIATION OF MACULAR STRUCTURE WITH MICROPERIMETRY SENSITIVITY FOLLOWING VITRECTOMY FOR PROLIFERATE DIABETIC RETINOPATHY.

PURPOSE: To evaluate macular sensitivity using microperimetry in patients with proliferate diabetic retinopathy following vitrectomy and to investigate the relationship between the sensitivity and foveal microstructures with optical coherence tomography/angiography. METHODS: Eighty-four eyes of 84 patients with proliferative diabetic retinopathy, who were indicated for vitrectomy, had no intraocular surgery history 3 months preoperatively, and were able to ensure fundus examination after the vitrectomy, were included. A logMAR best-corrected visual acuity, macular sensitivity of microperimetry, macular retinal thickness, and macular vessel perfusion using optical coherence tomography/angiography were examined at 1 week, 1 month, and 3 months postoperatively. RESULTS: The logMAR best-corrected visual acuity and mean macular sensitivity of patients with proliferative diabetic retinopathy improved postoperatively (P < 0.05). There was a significant correlation between best-corrected visual acuity and mean sensitivity (P < 0.05). Postoperative mean macular sensitivity was significantly correlated with outer retinal thickness in the 0 to 6 mm macular area (P < 0.05) and also significantly correlated with deep capillary plexus perfusion (P < 0.05). Fixation stability and mean macular sensitivity did not show any correlation with glycated hemoglobin, triglyceride, serum total cholesterol, carbamide, and creatinine and duration of diabetes mellitus (P > 0.05). CONCLUSION: Postoperative mean macular sensitivity was significantly correlated with outer retinal thickness and deep capillary plexus perfusion for patients with proliferative diabetic retinopathy. The authors found that the visual performance of patients can be evaluated by the outer retinal thickness and deep capillary plexus perfusion, so optical coherence tomography/angiography examination can be an important prognostic factor for visual performance in patients.Clinical Trial Registration: This trial is registered with the Chinese Clinical Trial Registry (http://www.chictr.org.cn; Registration No.: ChiCTR2100043399).

CLINICAL CHARACTERIZATION OF AQUEOUS AND VITREOUS RETINOL-BINDING PROTEIN 3 CONCENTRATIONS IN RELATION TO DIABETIC RETINOPATHY SEVERITY, RETINAL STRUCTURES, AND SYSTEMIC COMPLICATIONS.

PURPOSE: To evaluate Retinol-Binding Protein 3 (RBP3) from photoreceptors in aqueous and its association with vitreous concentrations, diabetic retinopathy (DR) severity, retinal layer thickness, and clinical characteristics in people with diabetes. METHODS: RBP3 concentration was measured by custom-developed enzyme-linked immunosorbent assay in aqueous and correlated with vitreous concentrations in patients from the 50-Year Medalist study and Beetham Eye Institute at Joslin Diabetes Center. RESULTS: Aqueous RBP3 concentration (N = 131) was elevated in eyes with no to mild DR (mean ± SD 0.7 nM ± 0.2) and decreased in eyes with moderate to severe DR (0.65 nM ± 0.3) and proliferative DR (0.5 nM ± 0.2, P < 0.001) compared to eyes without diabetes. Aqueous and vitreous RBP3 concentrations correlated with each other (r = 0.34, P = 0.001) and between fellow eyes (P < 0.0001). History of retinal surgery did not affect aqueous RBP3 concentrations, but cataract surgery affected both vitreous and aqueous levels. Elevated aqueous RBP3 concentration associated with increased thickness of the outer nuclear layer (P = 0.004) and correlated with hemoglobin A1c, whereas vitreous RBP3 concentrations correlated with diabetic systemic complications. CONCLUSION: These findings suggest that aqueous RBP3 concentration may be an important endogenous clinical retinal protective factor, a biomarker for DR severity, and a promising VEGF-independent clinical intervention target in DR.

A novel fusion of genetic grey wolf optimization and kernel extreme learning machines for precise diabetic eye disease classification.

In response to the growing number of diabetes cases worldwide, Our study addresses the escalating issue of diabetic eye disease (DED), a significant contributor to vision loss globally, through a pioneering approach. We propose a novel integration of a Genetic Grey Wolf Optimization (G-GWO) algorithm with a Fully Convolutional Encoder-Decoder Network (FCEDN), further enhanced by a Kernel Extreme Learning Machine (KELM) for refined image segmentation and disease classification. This innovative combination leverages the genetic algorithm and grey wolf optimization to boost the FCEDN's efficiency, enabling precise detection of DED stages and differentiation among disease types. Tested across diverse datasets, including IDRiD, DR-HAGIS, and ODIR, our model showcased superior performance, achieving classification accuracies between 98.5% to 98.8%, surpassing existing methods. This advancement sets a new standard in DED detection and offers significant potential for automating fundus image analysis, reducing reliance on manual examination, and improving patient care efficiency. Our findings are crucial to enhancing diagnostic accuracy and patient outcomes in DED management.

Comparison of 21 artificial intelligence algorithms in automated diabetic retinopathy screening using handheld fundus camera.

BACKGROUND: Diabetic retinopathy (DR) is a common complication of diabetes and may lead to irreversible visual loss. Efficient screening and improved treatment of both diabetes and DR have amended visual prognosis for DR. The number of patients with diabetes is increasing and telemedicine, mobile handheld devices and automated solutions may alleviate the burden for healthcare. We compared the performance of 21 artificial intelligence (AI) algorithms for referable DR screening in datasets taken by handheld Optomed Aurora fundus camera in a real-world setting. PATIENTS AND METHODS: Prospective study of 156 patients (312 eyes) attending DR screening and follow-up. Both papilla- and macula-centred 50° fundus images were taken from each eye. DR was graded by experienced ophthalmologists and 21 AI algorithms. RESULTS: Most eyes, 183 out of 312 (58.7%), had no DR and mild NPDR was noted in 21 (6.7%) of the eyes. Moderate NPDR was detected in 66 (21.2%) of the eyes, severe NPDR in 1 (0.3%), and PDR in 41 (13.1%) composing a group of 34.6% of eyes with referable DR. The AI algorithms achieved a mean agreement of 79.4% for referable DR, but the results varied from 49.4% to 92.3%. The mean sensitivity for referable DR was 77.5% (95% CI 69.1-85.8) and specificity 80.6% (95% CI 72.1-89.2). The rate for images ungradable by AI varied from 0% to 28.2% (mean 1.9%). Nineteen out of 21 (90.5%) AI algorithms resulted in grading for DR at least in 98% of the images. CONCLUSIONS: Fundus images captured with Optomed Aurora were suitable for DR screening. The performance of the AI algorithms varied considerably emphasizing the need for external validation of screening algorithms in real-world settings before their clinical application.

What is already known on this topic? Diabetic retinopathy (DR) is a common complication of diabetes. Efficient screening and timely treatment are important to avoid the development of sight-threatening DR. The increasing number of patients with diabetes and DR poses a challenge for healthcare.What this study adds? Telemedicine, mobile handheld devices and artificial intelligence (AI)-based automated algorithms are likely to alleviate the burden by improving efficacy of DR screening programs. Reliable algorithms of high quality exist despite the variability between the solutions.How this study might affect research, practice or policy? AI algorithms improve the efficacy of screening and might be implemented to clinical use after thorough validation in a real-life setting.

[Biomechanics of diabetic vitreopapillary traction syndrome]. / Biomekhanika diabeticheskogo vitreopapillyarnogo traktsionnogo sindroma.

Diabetic vitreopapillary traction syndrome (VPT) is a variant of diabetic retinopathy (DR) that can lead to vision loss in advanced stages. This review reports on the biomechanics of the vitreous in the pathogenesis of proliferative DR, in particular diabetic VPT. The article analyzes and summarizes literature data, presents the views of different authors on this problem, and provides the results of Russian and foreign scientific research on this pathology. It is concluded that further research in this area can lead to a significant improvement in the results of therapy, timely diagnosis, and preservation of vision in patients with DR.

Clinical features, management, and outcomes of patients with sterile endophthalmitis associated with intravitreal bevacizumab injection: retrospective case series.

PURPOSE: To evaluate clinical features, treatment protocol, outcomes, and complications that developed in this case series of 24 patients who had consecutive sterile endophthalmitis after intravitreal bevacizumab (IVB) injection. METHODS: In this retrospective case series, IVB was repackaged in individual aliquots from the three batches that were used on the same day. IVB was injected into 26 eyes of 26 patients due to diabetic macular edema, age-related macular degeneration, and branch retinal vein occlusion. All patients had intraocular inflammation. Patients were divided into two groups severe and moderate inflammation according to the intraocular inflammation. The medical records of all patients were reviewed. At each follow-up visit, the complete ophthalmologic examination was performed, including best corrected visual acuity (BCVA), intraocular pressure, biomicroscopy, and posterior fundus examination. RESULTS: Twenty-four of 26 patients were included in the study. Two patients were excluded from this study since they didn't come to follow-up visits. The mean BCVA was 1.00 ± 0.52 Log MAR units before IVB. At the final visit, the BCVA was 1.04 ± 0.47 Log MAR units. These differences were not significant (p = 0.58). Of the 24 eyes, 16 eyes had severe, and 8 eyes had moderate intraocular inflammation. Eleven eyes in the severe inflammation group underwent pars plana vitrectomy due to intense vitreous opacity. Smear, culture results, and polymerase chain reaction results were negative. CONCLUSION: Sterile endophthalmitis may occur after IVB injection. Differential diagnosis of sterile endophthalmitis from infective endophthalmitis is crucial to adjust the appropriate treatment and prevent long-term complications due to unnecessary treatment.

Comparison of the cytokines levels in aqueous humor in vitrectomized eyes versus non-vitrectomized eyes with diabetic macular edema.

BACKGROUND: This study was conducted to compare concentrations of VEGF family growth factors, inflammation-related factors, and adhesion molecules in the aqueous humor of eyes with diabetic macular edema (DME), with and without prior vitrectomy. METHODS: A total of 31 eyes were included, 11 with DME that had undergone vitrectomy, 9 with DME but without vitrectomy, and 11 from age-related cataract patients as controls. The concentrations of cytokines including TNF-α, IL-6, IL-8, IP-10, MCP-1, IFN-γ, MIP-1 α, MIP-1 ß, PECAM-1, MIF, VCAM-1, ICAM-1, PIGF were quantified using Luminex Human Discovery Assay. Central macular thickness (CMT) values of all eyes were measured using optical coherence tomography (OCT). RESULTS: (1) Vitrectomized DME eyes exhibited significantly higher levels of IL-6 and IL-8 compared to non-vitrectomized eyes (P < 0.05). (2) In vitrectomized group, after Benjamini-Hochberg correction, there was a significant positive correlation between the levels of VEGF and PlGF (rs = 0.855, P < 0.05), as well as the levels of TNF-α and IFN-γ (rs = 0.858, P < 0.05). In non-vitrectomized group, significant positive correlations were found between VEGF and PlGF levels after correcting for multiple comparisons (rs = 0.9, P < 0.05). (3) In non-vitrectomized group, the concentrations of VEGF and PlGF in aqueous humor were significantly positively correlated with CMT values (rs = 0.95, P < 0.05; rs = 0.9, P < 0.05, respectively). CONCLUSIONS: The concentrations of IL-6 and IL-8 in the aqueous humor were significantly higher in vitrectomized DME eyes compared to nonvitrectomized DME eyes and the levels of VEGF were similar in the two groups, suggesting that inflammation after vitrectomy may be a key factor in the occurrence and development of DME.

[Outcomes of bimanual vitreoretinal surgery in the treatment of diabetic retinopathy complications]. / Rezul'taty bimanual'noi vitreoretinal'noi khirurgii v lechenii oslozhnenii diabeticheskoi retinopatii.

The incidence of diabetic retinopathy (DR) requiring vitreorentinal surgery is increasing. The search for new effective and safe methods of treatment, the choice of the optimal time for surgery, and the assessment of long-term treatment outcomes are relevant problems. PURPOSE: This study evaluates the long-term results of vitreorentinal surgery using the bimanual technique in DR with different stages of fibrovascular proliferation. MATERIAL AND METHODS: The study included 135 patients (135 eyes) who were divided into groups depending on the predominant type of proliferation - vascular or fibrous. Patients underwent vitrectomy with membranectomy using the bimanual technique, with peripheral panretinal endolaser coagulation of the retina and tamponade of the vitreous cavity with balanced salt solution. The postoperative observation period lasted up to 12 months. RESULTS: Both groups showed statistically significant improvement in visual function and anatomical changes in central retinal thickness. A statistically significant improvement in best corrected visual acuity (BCVA) was found in patients with initially predominantly vascular proliferation. Correlation analysis showed that initially higher BCVA tends to persist in the postoperative period. A negative correlation was found between the final BCVA and the presence of type 2 diabetes mellitus, fibrous stage of proliferation, high central retinal thickness, and the presence of diabetic macular edema (DME) - both initially and after treatment. The frequency of complications in the groups was comparable, except for postoperative DME, which was more often detected in patients with fibrous proliferation. CONCLUSION: The bimanual technique of vitreorentinal surgery for complications of DR allows achieving high anatomical and functional results. Higher BCVA is noted in patients with the vascular stage of proliferation and initially high BCVA. The obtained data allow us to form a hypothesis about the possibility of earlier surgery in patients with high BCVA, but require further investigation.

The effect of adjuvant Mitomycin C during vitrectomy on functional and anatomical outcomes in patients with severe diabetic tractional retinal detachment.

PURPOSE: To evaluate the effect of adjuvant Mitomycin C (MMC) use on the anatomical and functional success of vitreoretinal surgery (VRS) in severe diabetic tractional retinal detachment (dTRD) patients. METHODS: A retrospective analysis of consecutive patients undergoing VRS due to severe dTRD was conducted. Patients were categorized into those who received 20 µg/0.1 mL MMC via MMC sandwich method (Group 1) and those who did not (Group 2). Demographics, surgical characteristics, visual outcomes, and complications that may related to MMC were analyzed. RESULTS: A total of 25 eyes were included, 13 in Group 1 and 12 in Group 2. No statistical difference was observed in baseline characteristics between the groups. The mean best-corrected visual acuity was 1.90 ± 0.43 logMAR and 1.93 ± 0.41 logMAR preoperatively and 1.60 ± 0.78 logMAR and 1.56 ± 0.78 logMAR postoperatively in Groups 1 and 2, respectively (p = 0.154). The postoperative mean intraocular pressure was 16.23 ± 2.55 mmHg and 13.08 ± 4.94 mmHg in Groups 1 and 2, respectively (p = 0.225). The rate of re-surgery was significantly lower in Group 1 (0% vs. 41.7% in Group 2, p = 0.015). Retina was attached in all patients at the last visit. No MMC-related complication was recorded. CONCLUSION: Intraoperative adjuvant MMC application for severe dTRD significantly reduces re-surgery rates with good anatomical and functional outcomes safely.

Intraretinal Hyper-Reflective Foci Are Almost Universally Present and Co-Localize With Intraretinal Fluid in Diabetic Macular Edema.

Purpose: In diabetic macular edema (DME), hyper-reflective foci (HRF) has been linked to disease severity and progression. Using an automated approach, we aimed to investigate the baseline distribution of HRF in DME and their co-localization with cystoid intraretinal fluid (IRF). Methods: Baseline spectral-domain optical coherence tomography (SD-OCT) volume scans (N = 1527) from phase III clinical trials YOSEMITE (NCT03622580) and RHINE (NCT03622593) were segmented using a deep-learning-based algorithm (developed using B-scans from BOULEVARD NCT02699450) to detect HRF. The HRF count and volume were assessed. HRF distributions were analyzed in relation to best-corrected visual acuity (BCVA), central subfield thickness (CST), and IRF volume in quartiles, and Diabetic Retinopathy Severity Scores (DRSS) in groups. Co-localization of HRF with IRF was calculated in the central 3-mm diameter using the en face projection. Results: HRF were present in most patients (up to 99.7%). Median (interquartile range [IQR]) HRF volume within the 3-mm diameter Early Treatment Diabetic Retinopathy Study ring was 1964.3 (3325.2) pL, and median count was 64.0 (IQR = 96.0). Median HRF volumes were greater with decreasing BCVA (nominal P = 0.0109), and increasing CST (nominal P < 0.0001), IRF (nominal P < 0.0001), and DRSS up to very severe nonproliferative diabetic retinopathy (nominal P < 0.0001). HRF co-localized with IRF in the en face projection. Conclusions: Using automated HRF segmentation of full SD-OCT volumes, we observed that HRF are a ubiquitous feature in DME and exhibit relationships with BCVA, CST, IRF, and DRSS, supporting a potential link to disease severity. The spatial distribution of HRF closely followed that of IRF.

Lipidomic studies revealing serological markers associated with the occurrence of retinopathy in type 2 diabetes.

PURPOSE: The duration of type 2 diabetes mellitus (T2DM) and blood glucose levels have a significant impact on the development of T2DM complications. However, currently known risk factors are not good predictors of the onset or progression of diabetic retinopathy (DR). Therefore, we aimed to investigate the differences in the serum lipid composition in patients with T2DM, without and with DR, and search for potential serological indicators associated with the development of DR. METHODS: A total of 622 patients with T2DM hospitalized in the Department of Endocrinology of the First Affiliated Hospital of Xi'an JiaoTong University were selected as the discovery set. One-to-one case-control matching was performed according to the traditional risk factors for DR (i.e., age, duration of diabetes, HbA1c level, and hypertension). All cases with comorbid chronic kidney disease were excluded to eliminate confounding factors. A total of 42 pairs were successfully matched. T2DM patients with DR (DR group) were the case group, and T2DM patients without DR (NDR group) served as control subjects. Ultra-performance liquid chromatography-mass spectrometry (LC-MS/MS) was used for untargeted lipidomics analysis on serum, and a partial least squares discriminant analysis (PLS-DA) model was established to screen differential lipid molecules based on variable importance in the projection (VIP) > 1. An additional 531 T2DM patients were selected as the validation set. Next, 1:1 propensity score matching (PSM) was performed for the traditional risk factors for DR, and a combined 95 pairings in the NDR and DR groups were successfully matched. The screened differential lipid molecules were validated by multiple reaction monitoring (MRM) quantification based on mass spectrometry. RESULTS: The discovery set showed no differences in traditional risk factors associated with the development of DR (i.e., age, disease duration, HbA1c, blood pressure, and glomerular filtration rate). In the DR group compared with the NDR group, the levels of three ceramides (Cer) and seven sphingomyelins (SM) were significantly lower, and one phosphatidylcholine (PC), two lysophosphatidylcholines (LPC), and two SMs were significantly higher. Furthermore, evaluation of these 15 differential lipid molecules in the validation sample set showed that three Cer and SM(d18:1/24:1) molecules were substantially lower in the DR group. After excluding other confounding factors (e.g., sex, BMI, lipid-lowering drug therapy, and lipid levels), multifactorial logistic regression analysis revealed that a lower abundance of two ceramides, i.e., Cer(d18:0/22:0) and Cer(d18:0/24:0), was an independent risk factor for the occurrence of DR in T2DM patients. CONCLUSION: Disturbances in lipid metabolism are closely associated with the occurrence of DR in patients with T2DM, especially in ceramides. Our study revealed for the first time that Cer(d18:0/22:0) and Cer(d18:0/24:0) might be potential serological markers for the diagnosis of DR occurrence in T2DM patients, providing new ideas for the early diagnosis of DR.

The association in diabetic retinopathy and stroke finding from NHANES evidence.

PURPOSE: Diabetic retinopathy and stroke are both vascular pathologies, and this study intends to investigate the relationship between diabetic retinopathy and stroke. METHODS: The NHANES database was used to find the relationship between diabetic retinopathy and stroke with 1948 individuals aged 40 years or older. The sensitivity of the data was verified by multiple interpolation, further analysis was done by subgroup analyses, and possible links were investigated with mediation studies. RESULTS: Diabetes retinopathy was found to be closely associated with stroke, with the PDR group having a higher stroke incidence than the NPDR group. After controlling for covariates, there were still substantial differences in the risk of stroke among patients with NPDR and PDR. Overall, subgroup analysis revealed DR group showed an important distinction, compared to the non-DR (OR = 1.76, 95% CI 1.15-2.64). The results of the mediation research indicated that the connection between DR and stroke was mediated by the frailty index and hypertension. CONCLUSION: This study demonstrated a statistically significant correlation between DR and stroke, which persisted even after DR staging and was more prevalent in PDR patients than in NPDR patients. Stroke prevention may benefit from DR health management.

Risk factors of internal carotid artery stenosis in patients with proliferative diabetic retinopathy: an analysis using optical coherence tomography and optical coherence tomography angiography.

BACKGROUND: This research investigates the correlation between the severity of internal carotid artery (ICA) stenosis and retinal parameters in patients with proliferative diabetic retinopathy (PDR), aiming to uncover potential risk factors. METHODS: A retrospective analysis of 68 patients (136 eyes) diagnosed with bilateral PDR from January 1, 2017, to December 31, 2021, was conducted. Carotid artery stenosis (CAS) was assessed using neck computed tomography angiography (CTA) and carotid duplex ultrasound (CDUS), with stenosis classified into two groups: normal (group 1) and mild or above (group 2), based on the North American Symptomatic Carotid Endarterectomy Trial (NASCET) criteria. Optical coherence tomography (OCT) and OCT angiography (OCTA) measured several retinal parameters, including sub foveal choroidal thickness (SFCT), retinal nerve fiber layer (RNFL) thickness, ganglion cell-inner plexiform layer (GCIPL) thickness, vessel density (VD), and foveal avascular zone (FAZ) area. Statistical analyses determined correlations between ICA degrees and retinal parameters. RESULTS: This study showed significant differences between groups in total VD, FAZ area, total RNFL thickness, and temporal RNFL thickness, indicating that patients with more severe ICA stenosis had noticeable retinal changes. Other parameters such as hyperlipidemia, total cholesterol levels, and intraocular pressure (IOP) also differed significantly, while no notable differences were observed in SFCT, central VD, average GCIPL, and superior, nasal, and inferior RNFL thickness. CONCLUSION: The study findings highlight retinal changes, such as an increased FAZ area, decreased total VD, and a total and thinner temporal RNFL, which suggest the need for carotid artery evaluation in patients. These findings have important clinical implications for the need for carotid work up in patients with PDR.

Attention-based deep learning framework for automatic fundus image processing to aid in diabetic retinopathy grading.

BACKGROUND AND OBJECTIVE: Early detection and grading of Diabetic Retinopathy (DR) is essential to determine an adequate treatment and prevent severe vision loss. However, the manual analysis of fundus images is time consuming and DR screening programs are challenged by the availability of human graders. Current automatic approaches for DR grading attempt the joint detection of all signs at the same time. However, the classification can be optimized if red lesions and bright lesions are independently processed since the task gets divided and simplified. Furthermore, clinicians would greatly benefit from explainable artificial intelligence (XAI) to support the automatic model predictions, especially when the type of lesion is specified. As a novelty, we propose an end-to-end deep learning framework for automatic DR grading (5 severity degrees) based on separating the attention of the dark structures from the bright structures of the retina. As the main contribution, this approach allowed us to generate independent interpretable attention maps for red lesions, such as microaneurysms and hemorrhages, and bright lesions, such as hard exudates, while using image-level labels only. METHODS: Our approach is based on a novel attention mechanism which focuses separately on the dark and the bright structures of the retina by performing a previous image decomposition. This mechanism can be seen as a XAI approach which generates independent attention maps for red lesions and bright lesions. The framework includes an image quality assessment stage and deep learning-related techniques, such as data augmentation, transfer learning and fine-tuning. We used the architecture Xception as a feature extractor and the focal loss function to deal with data imbalance. RESULTS: The Kaggle DR detection dataset was used for method development and validation. The proposed approach achieved 83.7 % accuracy and a Quadratic Weighted Kappa of 0.78 to classify DR among 5 severity degrees, which outperforms several state-of-the-art approaches. Nevertheless, the main result of this work is the generated attention maps, which reveal the pathological regions on the image distinguishing the red lesions and the bright lesions. These maps provide explainability to the model predictions. CONCLUSIONS: Our results suggest that our framework is effective to automatically grade DR. The separate attention approach has proven useful for optimizing the classification. On top of that, the obtained attention maps facilitate visual interpretation for clinicians. Therefore, the proposed method could be a diagnostic aid for the early detection and grading of DR.

Management of diabetic ocular complications: from cellular insights to community strategies.

The editorial outlines an integrated approach to managing diabetic ocular complications, combining advanced scientific research with practical public health strategies to improve the prevention, diagnosis, and treatment of diabetic retinopathy and macular edema globally.