(50-4) 18 * << * >> * Russian * English * Content * All Issues

Deep Learning for Diabetic Retinopathy Screening: An Adaptation Methodology for Real-World Clinical Environments
E.V. Kozlov1, D.D. Lysukhin1, A.P. Pershina-Milyutina1, E.V. Kovaleva1, A.V. Aredov1, V.K. Aleksandrova1, T.A. Chistyakov1, A.A. Tolkacheva1, N.G. Shebardina1, E.A. Korchuganova1, N.G. Mokrysheva1

1 I.I. Dedov National Medical Research Center for Endocrinology, Ministry of Health of the Russian Federation,
Dmitry Ulyanov Str. 11, Moscow, 117292, Russia

  Full text (PDF)

DOI: 10.18287/COJ1839

Article ID: 1839

Language: Russian

Abstract:
The deployment of deep learning systems for diabetic retinopathy screening in real-world clinical settings is limited by performance degradation when algorithms are applied to novel data sources. This study aims to develop and validate a methodology for adapting pre-trained models to the heterogeneity inherent in medical imaging data. The research analyzes the impact of dataset disparities on classification performance. It is demonstrated that the direct transfer of a model trained on public databases to a local clinical dataset (1,123 images, 302 patients), characterized by distinct technical parameters and pronounced class imbalance, results in a significant decline in the macro-F1 score from 0.86 to 0.38.

To ensure robust knowledge transfer and minimize training noise, a critical stage of this work involved constructing a high-quality reference local dataset. This was achieved through a multi-stage annotation protocol and an assessment of inter-rater agreement among expert physicians. Utilizing the curated data, a two-stage transfer learning strategy was implemented using the EfficientNet-B0 architecture. Despite the limited sample size, this annotation scheme, combined with the proposed fine-tuning approach, improved the macro-F1 score on the validation set to 0.63. The results confirm that the creation of annotated local datasets is a critical step for the successful adaptation of diagnostic systems to practical healthcare settings.

Keywords:
diabetic retinopathy, deep learning, transfer learning, limited data, automated screening.

Citation:
Kozlov EV, Lysukhin DD, Pershina-Milyutina AP, Kovaleva EV, Aredov AV, Aleksandrova VK, Chistyakov TA, Tolkacheva AA, Shebardina NG, Korchuganova EA, Mokrysheva NG. Deep Learning for Diabetic Retinopathy Screening: An Adaptation Methodology for Real-World Clinical Environments. Computer Optics 2026; 50(4): 1839. doi: 10.18287/COJ1839.

References:

  1. Federal State Budgetary Institution "Central Research Institute for Organization and Informatization of Health Care" of the Ministry of Health of the Russian Federation (FSBI "CNIOIZ" of the Ministry of Health of Russia). Medical statistics [in Russian]. 2025. Available at: https://mednet.ru/mediczinskaya-statistika.
  2. Dedov II, Shestakova MV, Vikulova OK, Zheleznyakova AV, Isakov MA, Eremina IA, Galstyan GR. Diabetes mellitus in the Russian Federation: Dynamics of epidemiological indicators according to the Federal Register of Diabetes Mellitus for the period 2010-2022 [in Russian]. Diabetes Mellitus 2023; 26(2): 104-123. DOI: 10.14341/DM13035.
  3. Ribeiro L, Oliveira CM, Neves C, Ramos JD, Ferreira H, Cunha-Vaz J. Screening for diabetic retinopathy in the central region of Portugal: Added value of automated "disease/no disease" grading. Ophthalmologica 2015; 233(2): 96-103. DOI: 10.1159/000368426.
  4. Bhaskaranand M, Ramachandra C, Bhat S, Cuadros J, Nittala MG, Sadda S, Solanki K. The value of automated diabetic retinopathy screening with the EyeArt system: A study of more than 100,000 consecutive encounters from people with diabetes. Diabetes Technology & Therapeutics 2019; 21(11): 635-643. DOI: 10.1089/dia.2019.0164.
  5. Abramoff MD, Lavin PT, Birch M, Shah N, Folk JC. Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices. NPJ Digital Medicine 2018; 1: 39. DOI: 10.1038/s41746-018-0040-6.
  6. Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, Venugopalan S, Widner K, Madams T, Cuadros J, Kim R, Raman R, Nelson PC, Mega JL, Webster DR. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 2016; 316(22): 2402-2410. DOI: 10.1001/jama.2016.17216.
  7. Walton OB, Garoon RB, Weng CY, Gross J, Young AK, Camero KA, et al. Evaluation of an automated teleretinal screening program for diabetic retinopathy. JAMA Ophthalmology 2016; 134(2): 204-209. DOI: 10.1001/jamaophthalmol.2015.5083.
  8. Ting DSW, Cheung CYL, Lim G, Tan GSW, Quang ND, Gan A, Hamzah H, Garcia-Franco R, Yeo IYS, Lee SY, Wong EYL, Sabanayagam C, Baskaran M, Ibrahim F, Tan NCS, Finkelstein EA, Lamoureux EL, Wong IYH, Bressler NM, Sivaprasad S, Varma R, Jonas JB, He MG, Cheng CY, Cheung GCM, Aung T, Hsu W, Lee ML, Wong TY. Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes. JAMA 2017; 318(22): 2211-2223. DOI: 10.1001/jama.2017.18152.
  9. Zhang Y, Wang L, Wu X, Chen H, Li Y, Wang S. A cross-domain weakly supervised diabetic retinopathy lesion identification method based on multiple instance learning and domain adaptation. Bioengineering 2023; 10(9): 1100. DOI: 10.3390/bioengineering10091100.
  10. Wang Z, Yin Y, Shi J, Fang W, Li H. Multi-model domain adaptation for diabetic retinopathy classification. Frontiers in Physiology 2022; 13: 918929. DOI: 10.3389/fphys.2022.918929.
  11. Benouaich X, Soares M, Azrak C, Hajjar C, Weber M, Kodjikian L. Accuracy of autonomous artificial intelligence-based diabetic retinopathy screening in real-life clinical practice. Journal of Clinical Medicine 2024; 13(16): 4776. DOI: 10.3390/jcm13164776.
  12. Li H, Liu P, Guo J, Wang Y, Liang D, Chen H. Cross-modality transfer learning with knowledge infusion for diabetic retinopathy grading. Frontiers in Medicine (Lausanne) 2024; 11: 1400137. DOI: 10.3389/fmed.2024.1400137.
  13. Zhang K, Liu X, Xu J, Wang H, Li Y, Zhao S. Generalizing to unseen domains in diabetic retinopathy with disentangled representations. arXiv preprint arXiv:2406.06384; 2024. Available at: https://arxiv.org/abs/2406.06384.
  14. Sarki R, Ahmed K, Wang H, Zhang Y, Wang K. A systematic review of transfer learning-based approaches for diabetic retinopathy detection. Artificial Intelligence in Medicine 2022; 134: 102513. DOI: 10.1016/j.artmed.2022.102513.
  15. Oltu B, Karaca BK, Erdem H, Özgür A. A systematic review of transfer learning-based approaches for diabetic retinopathy detection. Gazi University Journal of Science 2023; 36(3): 1140-1157. DOI: 10.35378/gujs.1081546.computeroptics+1
  16. Sharma S, Gupta S, Gupta D, Arora C, Vyas R. Learning self-supervised representations for label efficient cross-domain knowledge transfer on diabetic retinopathy fundus images. arXiv preprint arXiv:2304.11168; 2023. Available at: https://arxiv.org/abs/2304.11168.
  17. Su W, Tang S, Liu X, Zhang Y, Chen H. Domain adaptive diabetic retinopathy grading with model absence and flowing data. arXiv preprint arXiv:2412.01203; 2022. Available at: https://arxiv.org/abs/2412.01203.
  18. Neroev VV, Bragin AA, Zaitseva OV. Development of a prototype service for diagnosing diabetic retinopathy based on fundus images using artificial intelligence methods [in Russian]. National Health Care (Russia) 2021; 2(2): 64-72. DOI: 10.47093/2713-069X.2021.2.2.64-72.
  19. APTOS 2019 blindness detection. Kaggle competition; The Asia Pacific Tele-Ophthalmology Society; 2019. Available at: https://www.kaggle.com/competitions/aptos2019-blindness-detection.
  20. Lipatov DV, Bessmertnaya EG, Kuzmin AG, Tolkacheva AA, Chistyakov TA. Atlas of diabetic retinopathy: A practical guide for physicians [in Russian]. In: Dedov II, Shestakova MV, eds. Moscow: Medical Information Agency; 2017. ISBN: 978-5-9986-0275-8.
  21. Buslaev A, Iglovikov VI, Khvedchenya E, Parinov A, Druzhinin M, Kalinin AA. Albumentations: Fast and flexible image augmentations. Information 2020; 11(2): 125. DOI: 10.3390/info11020125.
  22. Tan M, Le QV. EfficientNet: Rethinking model scaling for convolutional neural networks. In: Proceedings of the 36th International Conference on Machine Learning (ICML). Proceedings of Machine Learning Research 2019; 97: 6105-6114.
  23. Glorot X, Bengio Y. Understanding the difficulty of training deep feedforward neural networks. In: Proceedings of the 13th International Conference on Artificial Intelligence and Statistics (AISTATS). Journal of Machine Learning Research - Workshop and Conference Proceedings 2010; 9: 249-256.

151, Molodogvardeiskaya str., Samara, 443001, Russia; E-mail: journal@computeroptics.ru; Tel: +7 (846) 242-41-24 (Executive secretary), +7 (846) 332-56-22 (Issuing editor), Fax: +7 (846) 332-56-20