arxiv: 2604.02448 · v1 · submitted 2026-04-02 · 📡 eess.IV · cs.AI· cs.CV

Recognition: no theorem link

Managing Diabetic Retinopathy with Deep Learning: A Data Centric Overview

Shramana Dey , Zahir Khan , T. A. PramodKumar , B. Uma Shankar , Ashis K. Dhara , Ramachandran Rajalakshmi , Rajiv Raman , Sushmita Mitra

Authors on Pith no claims yet

Pith reviewed 2026-05-13 20:53 UTC · model grok-4.3

classification 📡 eess.IV cs.AIcs.CV

keywords diabetic retinopathyfundus imagesdeep learningdatasetsannotationslesion localizationscreeningclinical reliability

0 comments

The pith

Existing diabetic retinopathy datasets limit deep learning reliability due to narrow geography, limited samples, inconsistent annotations, and variable image quality.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper reviews fundus image datasets for deep learning detection and grading of diabetic retinopathy. It shows that most available collections come from limited regions, include relatively few images, and suffer from inconsistent labels or uneven quality, which reduces how well models trained on them can work in actual clinics. The review sorts datasets by the tasks they support, such as simple classification, full severity grading, spotting specific lesions, and screening for multiple diseases at once. It flags missing elements like standardized lesion markings and follow-up data over time. Recommendations focus on building better datasets to enable more trustworthy and explainable automated screening tools.

Core claim

The central claim is that existing DR fundus datasets are too geographically narrow, too small in scale, and too inconsistent in annotations and image quality to support clinically reliable deep learning models, and that closing these gaps through better curation and standardization is required for progress.

What carries the argument

Comparative analysis that categorizes fundus datasets by size, accessibility, annotation level (image-level, lesion-level, multi-disease), and suitability for binary classification, severity grading, lesion localization, and multi-disease tasks.

If this is right

Standardized lesion-level annotations would enable more explainable deep learning models that highlight specific disease features rather than black-box predictions.
Inclusion of longitudinal data would support models that track disease progression over time instead of single-visit snapshots.
Broader geographic coverage would reduce bias and improve model performance when deployed in new populations.
Better datasets would allow reliable multi-disease screening alongside diabetic retinopathy grading.
Future curation following the outlined recommendations would make automated DR tools more suitable for routine clinical use.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Models trained on improved datasets could meaningfully reduce the screening workload for ophthalmologists in high-volume clinics.
Public release of more diverse datasets might accelerate collaboration across research groups working on explainable AI for eye disease.
Addressing annotation inconsistencies could lead to benchmark challenges that compare methods on equal footing rather than dataset-specific quirks.

Load-bearing premise

The reviewed datasets and task groupings are representative enough to reveal the main limitations and direct future dataset improvements.

What would settle it

Release of a large, multi-region dataset with standardized lesion-level annotations, consistent image quality, and longitudinal records that trains deep learning models showing high clinical performance across diverse populations would test whether the identified gaps are real and persistent.

Figures

Figures reproduced from arXiv: 2604.02448 by Ashis K. Dhara, B. Uma Shankar, Rajiv Raman, Ramachandran Rajalakshmi, Shramana Dey, Sushmita Mitra, T. A. PramodKumar, Zahir Khan.

**Figure 2.** Figure 2: Representative fundus images, illustrating the severity grades of DR. (a) Healthy case, and the (b)–(e) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Key challenges in developing high-quality, robust and standardized datasets for DR screening. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Figure 4: Chronological overview of major DR fundus image datasets (from 2003 to 2025), showing year of release, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: Data distribution in SaNMoD, in terms of (a) the DR severity classes (grades), and (b) samples in the different [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗

**Figure 6.** Figure 6: Representative Grad-CAM visualization, highlighting lesion-specific activation for (a) DR, (b) RDR, and (c) [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗

read the original abstract

Diabetic Retinopathy (DR) is a serious microvascular complication of diabetes, and one of the leading causes of vision loss worldwide. Although automated detection and grading, with Deep Learning (DL), can reduce the burden on ophthalmologists, it is constrained by the limited availability of high-quality datasets. Existing repositories often remain geographically narrow, contain limited samples, and exhibit inconsistent annotations or variable image quality; thereby, restricting their clinical reliability. This paper presents a comprehensive review and comparative analysis of fundus image datasets used in the management of DR. The study evaluates their usability across key tasks, including binary classification, severity grading, lesion localization, and multi-disease screening. It also categorizes the datasets by size, accessibility, and annotation type (such as image-level, lesion-level, and multi-disease). Finally, a recently published dataset is presented as a case study to illustrate broader challenges in dataset curation and usage. The review consolidates current knowledge while highlighting persistent gaps such as the lack of standardized lesion-level annotations and longitudinal data. It also outlines recommendations for future dataset development to support clinically reliable and explainable solutions in DR screening.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

A solid but unoriginal survey of DR datasets that flags real gaps in annotations and diversity, though its dataset selection method is not clearly documented.

read the letter

This paper is a review that gathers existing fundus image datasets for diabetic retinopathy work and sorts them by size, access, annotation level, and suitability for tasks like binary classification or lesion localization. It also walks through one recent dataset as a case study and lists common shortcomings such as narrow geography, small sample counts, inconsistent labels, and missing longitudinal data. That consolidation is the main value: it gives someone starting out a single place to compare options instead of hunting through scattered papers. The abstract lays out clear criteria for evaluating usability, and the recommendations for future datasets (standardized lesion annotations, better multi-disease coverage) track with what the field already complains about. Nothing here is new in method or theory; it is synthesis. The soft spot is the lack of any stated search protocol or inclusion rules. Without that, it is difficult to judge whether the reviewed set is representative or whether stronger counter-examples were simply not included. The strongest claim about persistent gaps therefore rests on an unshown foundation. The paper is aimed at researchers who need a quick map of available DR data rather than experts already deep in the literature. It is coherent on its own terms and shows straightforward engagement with the sources it cites. A serious editor could send it for review if the authors add a methods paragraph on how the datasets were chosen; otherwise it risks looking like an incomplete list. I would bring it to a reading group only if the group is specifically looking for dataset resources.

Referee Report

2 major / 2 minor

Summary. The manuscript is a literature review surveying fundus image datasets for diabetic retinopathy (DR) management with deep learning. It evaluates dataset usability for tasks including binary classification, severity grading, lesion localization, and multi-disease screening; categorizes datasets by size, accessibility, and annotation type (image-level, lesion-level, multi-disease); presents a recent dataset as a case study; and identifies persistent gaps such as lack of standardized lesion-level annotations and longitudinal data while offering recommendations for future dataset development. The central claim is that existing repositories are often geographically narrow, limited in samples, and suffer from inconsistent annotations or variable quality, restricting clinical reliability.

Significance. If the comparative analysis accurately captures the state of DR datasets, the review could meaningfully guide development of more diverse, standardized, and clinically reliable datasets, helping address a key bottleneck for deployable DL models in DR screening.

major comments (2)

[Abstract] Abstract and implied Methods: The claim that 'existing repositories often remain geographically narrow, contain limited samples, and exhibit inconsistent annotations or variable image quality' is load-bearing for the paper's conclusions on persistent gaps, yet no explicit search strategy, inclusion/exclusion criteria, or PRISMA-style flow is described. This leaves open the possibility that large multi-ethnic or high-quality annotated datasets were omitted, weakening the representativeness of the reviewed set.
[Case study] Case study section: The choice of the 'recently published dataset' as illustrative is not justified by explicit criteria linking it to the broader comparative analysis (e.g., how its curation challenges differ quantitatively from those already tabulated for other repositories), making it unclear whether the case study adds new insight or merely restates prior points.

minor comments (2)

[Tables] Tables summarizing dataset characteristics would benefit from consistent column ordering and explicit definitions of 'accessibility' and 'annotation type' to aid quick comparison.
[Recommendations] The recommendations section lists high-level suggestions (e.g., need for longitudinal data) without concrete examples of existing partial solutions or proposed annotation standards that could be adopted.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful and constructive comments on our review of diabetic retinopathy fundus datasets. We address each major comment point by point below, providing clarifications and committing to revisions that strengthen the manuscript's transparency and rigor without altering its core findings.

read point-by-point responses

Referee: [Abstract] Abstract and implied Methods: The claim that 'existing repositories often remain geographically narrow, contain limited samples, and exhibit inconsistent annotations or variable image quality' is load-bearing for the paper's conclusions on persistent gaps, yet no explicit search strategy, inclusion/exclusion criteria, or PRISMA-style flow is described. This leaves open the possibility that large multi-ethnic or high-quality annotated datasets were omitted, weakening the representativeness of the reviewed set.

Authors: We agree that an explicit description of the literature search process would improve transparency and address potential concerns about completeness. The datasets were identified through systematic searches of PubMed, Google Scholar, IEEE Xplore, and arXiv using keywords including 'diabetic retinopathy dataset', 'fundus image dataset DR', 'public DR fundus repository', and 'deep learning diabetic retinopathy data' with a cutoff of December 2023. Inclusion criteria required publicly available fundus image datasets used in at least one peer-reviewed DL study for DR tasks (classification, grading, or localization); exclusion criteria removed non-fundus modalities, private datasets, or those without any DL validation. We will add a new 'Search Strategy and Dataset Selection' subsection (with a PRISMA-style flow diagram) in the revised manuscript to document this process fully. We believe the reviewed set captures the major publicly referenced datasets in the DL-DR literature, but we acknowledge that formalizing the method strengthens the claim. revision: yes
Referee: [Case study] Case study section: The choice of the 'recently published dataset' as illustrative is not justified by explicit criteria linking it to the broader comparative analysis (e.g., how its curation challenges differ quantitatively from those already tabulated for other repositories), making it unclear whether the case study adds new insight or merely restates prior points.

Authors: The case study was included to move beyond tabular summaries by providing a concrete, real-world example of curation challenges (such as achieving multi-ethnic balance and consistent lesion-level annotations) that the comparative analysis identifies as persistent gaps. To make the linkage explicit, we will revise the opening paragraph of the case study section to state the selection criteria: (1) publication within the last 24 months, (2) sample size exceeding the median of tabulated datasets, and (3) inclusion of annotation types and geographic diversity not fully quantified in the earlier tables. This will clarify how the case study supplies quantitative and qualitative depth (e.g., specific annotation inconsistency rates observed during curation) that the aggregate tables cannot convey, thereby adding distinct insight rather than restating prior points. revision: yes

Circularity Check

0 steps flagged

Literature review contains no derivations, equations, or self-referential predictions

full rationale

This paper is a literature review of existing fundus image datasets for diabetic retinopathy. It contains no mathematical derivations, fitted parameters, equations, or predictive models that could reduce to prior results by construction. All claims about dataset limitations (geographic narrowness, sample size, annotation inconsistency) rest on descriptions of external repositories rather than any internal reduction or self-citation chain. The single minor self-citation risk noted by the reader does not load-bear any central claim, satisfying the criteria for a score of 0 with no circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a literature review paper with no mathematical derivations, free parameters, axioms, or invented entities; it synthesizes and categorizes prior published datasets.

pith-pipeline@v0.9.0 · 5537 in / 960 out tokens · 41800 ms · 2026-05-13T20:53:32.751266+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

59 extracted references · 59 canonical work pages · 1 internal anchor

[1]

Samia Akhtar, Shabib Aftab, andet al.A deep learning based model for Diabetic Retinopathy grading.Scientific Reports, 15:3763, 2025

work page 2025
[2]

Rajiv Raman, Joana C Vasconcelos, andet al.Prevalence of Diabetic Retinopathy in India stratified by known and undiagnosed diabetes, urban-rural locations, and socioeconomic indices: Results from the SMART India population-based cross-sectional screening study.The Lancet Global Health, 10:e1764–e1773, 2022

work page 2022
[3]

Ranjit Mohan Anjana, Ranjit Unnikrishnan, andet al.Metabolic non-communicable disease health report of India: The ICMR-INDIAB national cross-sectional study (ICMR-INDIAB-17).The Lancet Diabetes and Endocrinology, 11:474–489, 2023

work page 2023
[4]

Rajiv Raman, Padmaja Kumari Rani, andet al.Prevalence of Diabetic Retinopathy in India: Sankara Nethralaya Diabetic Retinopathy Epidemiology and Molecular Genetics study report 2.Ophthalmology, 116:311–318, 2009

work page 2009
[5]

Salil S Gadkari, Quresh B Maskati, andet al.Prevalence of Diabetic Retinopathy in India: The All India Ophthalmological Society Diabetic Retinopathy Eye Screening Study 2014.Indian Journal of Ophthalmology, 64:38–44, 2016

work page 2014
[6]

Mohan Rema, Sundaram Premkumar, andet al.Prevalence of Diabetic Retinopathy in urban India: The Chennai Urban Rural Epidemiology Study (CURES) eye study - I.Investigative Ophthalmology and Visual Science, 46:2328–2333, 2005

work page 2005
[7]

Praveen K Nirmalan, Joanne Katz, andet al.Prevalence of vitreoretinal disorders in a rural population of Southern-India: The Aravind comprehensive eye study.Archives of Ophthalmology, 122:581–586, 2004

work page 2004
[8]

Charles P Wilkinson, Frederick L Ferris III, andet al.Proposed international clinical Diabetic Retinopathy and Diabetic Macular Edema disease severity scales.Ophthalmology, 110:1677–1682, 2003

work page 2003
[9]

Convolutional neural networks in diabetic eye disease detection: A survey on retinopathy and macular edema.Edelweiss Applied Science and Technology, 9:1207–1218, 2025

Preethi Kulkarni et al. Convolutional neural networks in diabetic eye disease detection: A survey on retinopathy and macular edema.Edelweiss Applied Science and Technology, 9:1207–1218, 2025

work page 2025
[10]

Elif Setenay Cutur and Neslihan Gokmen Inan. Multi-class classification of retinal eye diseases from ophthal- moscopy images using transfer learning-based vision transformers.Journal of Imaging Informatics in Medicine, 25:1–15, 2025

work page 2025
[11]

Dinggang Shen, Guorong Wu, andet al.Deep learning in medical image analysis.Annual Review of Biomedical Engineering, 19:221–248, 2017

work page 2017
[12]

Archana Senapati, Hrudaya Kumar Tripathy, andet al.Artificial Intelligence for Diabetic Retinopathy detection: A systematic review.Informatics in Medicine Unlocked, 45:101445, 2024

work page 2024
[13]

Shramana Dey, Pallabi Dutta, andet al.Multi-scale Deep Supervised Attention network for Red Lesion seg- mentation.in Proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI), pages 1–4, 2023

work page 2023
[14]

Siying Teng, Bo Wang, andet al.MediDRNet: Tackling category imbalance in Diabetic Retinopathy classification with dual-branch learning and prototypical Contrastive Learning.Computer Methods and Programs in Biomedicine, 253:108230, 2024

work page 2024
[15]

Shramana Dey, Pallabi Dutta, andet al.Adaptive Class Learning to screen Diabetic disorders in fundus images of eye.in Proceedings of the International Conference on Pattern Recognition (ICPR), pages 124–137, 2024

work page 2024
[16]

Very Deep Convolutional Networks for Large-Scale Image Recognition

Karen Simonyan and Andrew Zisserman. Very Deep Convolutional Networks for Large-Scale Image Recognition. in Proceedings of the International Conference on Learning Representations (ICLR), 2015

work page 2015
[17]

Kaiming He, Xiangyu Zhang, andet al.Deep residual learning for image recognition.in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016

work page 2016
[18]

in Proceedings of the International Conference on Computer Vision and Pattern Recognition (CVPR), pages 2818–2826, 2016

Christian Szegedy, Vincent Vanhoucke, andet al.Rethinking the Inception architecture for computer vision. in Proceedings of the International Conference on Computer Vision and Pattern Recognition (CVPR), pages 2818–2826, 2016

work page 2016
[19]

EfficientNet: Rethinking model scaling for convolutional neural networks.in Proceedings of the International Conference on Machine Learning (ICML), 97:6105–6114, 2019

Mingxing Tan and Quoc Le. EfficientNet: Rethinking model scaling for convolutional neural networks.in Proceedings of the International Conference on Machine Learning (ICML), 97:6105–6114, 2019

work page 2019
[20]

Gao Huang, Zhuang Liu, andet al.Densely connected convolutional networks.in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 4700–4708, 2017

work page 2017
[21]

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Alexey Dosovitskiy, Lucas Beyer, andet al.An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020. 15 Managing Diabetic Retinopathy with Deep Learning: A Data Centric Overview

work page internal anchor Pith review Pith/arXiv arXiv 2010
[22]

https://kaggle.com/competitions/ diabetic-retinopathy-detection, 2015

Emma Dugas, Jared, andet al.Diabetic Retinopathy Detection. https://kaggle.com/competitions/ diabetic-retinopathy-detection, 2015. Kaggle (accessed on March 15, 2025)

work page 2015
[23]

Etienne Decencière, Xiwei Zhang, andet al.Feedback on a publicly distributed image database: The Messidor Database.Image Analysis and Stereology, 33:231–234, 2014

work page 2014
[24]

Tao Li, Yingqi Gao, andet al.Diagnostic assessment of deep learning algorithms for Diabetic Retinopathy screening.Information Sciences, 501:511–522, 2019

work page 2019
[25]

Prasanna Porwal, Samiksha Pachade, andet al.Indian Diabetic Retinopathy image Dataset (IDRiD): A database for Diabetic Retinopathy screening research.Data, 3:25, 2018

work page 2018
[26]

Jia Deng, Wei Dong, andet al.ImageNet: A large-scale hierarchical image database.in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 248–255, 2009

work page 2009
[27]

APTOS 2019 blindness detection

Maggie Karthik and Sohier Dane. APTOS 2019 blindness detection. https://kaggle.com/competitions/ aptos2019-blindness-detection, 2019. Kaggle (accessed on March 12, 2025)

work page 2019
[28]

Mahesh S Patil, Satyadhyan Chickerur, andet al.Effective deep learning data augmentation techniques for Diabetic Retinopathy classification.Procedia Computer Science, 218:1156–1165, 2023

work page 2023
[29]

Mark Sandler, Andrew Howard, andet al.MobileNetV2: Inverted residuals and linear bottlenecks.in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 4510–4520, 2018

work page 2018
[30]

Ramazan ˙Incir and Ferhat Bozkurt. A study on effective data preprocessing and augmentation method in Diabetic Retinopathy classification using pre-trained deep learning approaches.Multimedia Tools and Applications, 83:12185–12208, 2024

work page 2024
[31]

Expert Systems with Applications, 213:118835, 2023

Yinghua Fu, Xin Lu, andet al.Automatic grading of Diabetic Macular Edema based on end-to-end network. Expert Systems with Applications, 213:118835, 2023

work page 2023
[32]

Juan Cao, Jiaran Chen, andet al.Diabetic Retinopathy classification based on dense connectivity and asymmetric convolutional neural network.Neural Computing and Applications, 37:7527–7540, 2025

work page 2025
[33]

CRA-Net: Transformer guided category-relation attention network for Diabetic Retinopa- thy grading.Computers in Biology and Medicine, 170:107993, 2024

Feng Zang and Hui Ma. CRA-Net: Transformer guided category-relation attention network for Diabetic Retinopa- thy grading.Computers in Biology and Medicine, 170:107993, 2024

work page 2024
[34]

https://www.nature.com/articles/s41598-025-91941-w

Ramachandran Rajalakshmi, Thyparambil Aravindakshan PramodKumar, andet al.Creating a retinal image database to develop an automated screening tool for Diabetic Retinopathy in india.Scientific Reports, 15, 2025. https://www.nature.com/articles/s41598-025-91941-w

work page 2025
[35]

P Saranya, R Pranati, and Sneha Shruti Patro. Detection and classification of red lesions from retinal images for Diabetic Retinopathy detection using deep learning models.Multimedia Tools and Applications, 82:39327–39347, 2023

work page 2023
[36]

Shiqi Huang, Jianan Li, andet al.RTNet: Relation transformer network for Diabetic Retinopathy multi-lesion segmentation.IEEE Transactions on Medical Imaging, 41:1596–1607, 2022

work page 2022
[37]

Yijin Huang, Li Lin, andet al.Automated hemorrhage detection from coarsely annotated fundus images in Diabetic Retinopathy.in Proceedings of the International Symposium on Biomedical Imaging (ISBI), pages 1369–1372, 2020

work page 2020
[38]

Detection of red lesions in retinal fundus images using YOLO V3.Curr

Prayas Pal, Swagata Kundu, and Ashis Kumar Dhara. Detection of red lesions in retinal fundus images using YOLO V3.Curr. Indian Eye Res. J. Ophthalmic Res. Group, 7:49, 2020

work page 2020
[39]

Carlos Santos, Marilton Aguiar, andet al.A new approach for detecting fundus lesions using image processing and deep neural network architecture based on YOLO model.Sensors, 22:6441, 2022

work page 2022
[40]

Atul Kumar, Divya Agarwal, andet al.Diabetic Retinopathy screening and management in India: Challenges and possible solutions.Indian Journal of Ophthalmology, 69:479–481, 2021

work page 2021
[41]

Early treatment Diabetic Retinopathy study design and baseline patient characteristics: ETDRS report number 7.Ophthalmology, 98:741–756, 1991

Early Treatment Diabetic Retinopathy Study Research Group et al. Early treatment Diabetic Retinopathy study design and baseline patient characteristics: ETDRS report number 7.Ophthalmology, 98:741–756, 1991

work page 1991
[42]

Etienne Decenciere, Guy Cazuguel, andet al.Teleophta: Machine Learning and Image Processing methods for teleophthalmology.IRBM, 34:196–203, 2013

work page 2013
[43]

Moorfields Diabetic Retinopathy Dataset 005

INSIGHT Health Data Hub. Moorfields Diabetic Retinopathy Dataset 005. Health Data Research UK Gateway,

work page
[44]

Accessed: 08 September, 2025

Version 1.0.0; modified 8 Oct 2024. Accessed: 08 September, 2025

work page 2024
[45]

16 Managing Diabetic Retinopathy with Deep Learning: A Data Centric Overview

Tomi Kauppi, Valentina Kalesnykiene, andet al.DIARETDB0: Evaluation Database and Methodology for Diabetic Retinopathy Algorithms.Machine Vision and Pattern Recognition Research Group, Lappeenranta University of Technology, Finland, 73:1–17, 2006. 16 Managing Diabetic Retinopathy with Deep Learning: A Data Centric Overview

work page 2006
[46]

DIARETDB1 Diabetic Retinopathy database and evaluation protocol.in Proceedings of the International Conference on Medical Image Understanding and Analysis, 2007:61, 2007

RVJPH Kälviäinen and H Uusitalo. DIARETDB1 Diabetic Retinopathy database and evaluation protocol.in Proceedings of the International Conference on Medical Image Understanding and Analysis, 2007:61, 2007

work page 2007
[47]

Meindert Niemeijer, Bram Van Ginneken, andet al.Retinopathy Online Challenge: Automatic detection of microaneurysms in digital color fundus photographs.IEEE Transactions on Medical Imaging, 29:185–195, 2009

work page 2009
[48]

Pavle Prentaši´c, Sven Lonˇcari´c, andet al.Diabetic Retinopathy image DataBase (dridb): A new database for Diabetic Retinopathy screening programs research.in Proceedings of the International Symposium on Image and Signal Processing and Analysis (ISPA), pages 711–716, 2013

work page 2013
[49]

Ning Li, Tao Li, andet al.A benchmark of Ocular disease intelligent recognition: One shot for multi-disease detection.in Proceedings of the International Symposium on Benchmarking, Measuring and Optimization, pages 177–193, 2020

work page 2020
[50]

Ruhan Liu, Xiangning Wang, andet al.DeepDRiD: Diabetic Retinopathy grading and image quality estimation challenge.Patterns, page 100512, 2022

work page 2022
[51]

Li Lin, Meng Li, andet al.The SUSTech-SYSU dataset for automated exudate detection and Diabetic Retinopathy grading.Scientific Data, 7:409, 2020

work page 2020
[52]

Yi Zhou, Boyang Wang, andet al.A benchmark for studying Diabetic Retinopathy: segmentation, grading, and transferability.IEEE Transactions on Medical Imaging, 40:818–828, 2021

work page 2021
[53]

Samiksha Pachade, Prasanna Porwal, andet al.Retinal Fundus Multi-Disease image Dataset (RFMiD): A dataset for multi-disease detection research.Data, 6:14, 2021

work page 2021
[54]

Gabriel Lepetit-Aimon, Clément Playout, andet al.MAPLES-DR: MESSIDOR anatomical and pathological labels for explainable screening of Diabetic Retinopathy.Scientific Data, 11:914, 2024

work page 2024
[55]

PhysioNet, 13026:1–16, 2023

Luis Filipe Nakayama, Mariana Goncalves, andet al.A BRazilian multilabel ophthalmological dataSET (BRSET). PhysioNet, 13026:1–16, 2023

work page 2023
[56]

org/10.13026/QXPD-1Y65, 2024

Luis Filipe Nakayama, L Zago Ribeiro, andet al.mBRSET, a mobile Bazilian Retinal dataSET.PhysioNet https://doi. org/10.13026/QXPD-1Y65, 2024

work page doi:10.13026/qxpd-1y65 2024
[57]

Decoupled Weight Decay Regularization

Ilya Loshchilov and Frank Hutter. Decoupled Weight Decay Regularization. Inin Proceedings of the International Conference on Learning Representations (ICLR). ICLR, 2018

work page 2018
[58]

Fabian Pedregosa, Gaël Varoquaux, andet al.Scikit-Learn: Machine Learning in Python.The Journal of Machine Learning Research (JMLR), 12:2825–2830, 2011

work page 2011
[59]

Ramprasaath R Selvaraju, Michael Cogswell, andet al.Grad-CAM: Visual explanations from deep networks via gradient-based localization.in Proceedings of the International Conference on Computer Vision (ICCV), pages 618–626, 2017. 17

work page 2017