Objective  Based on fundus color photography and optical coherence tomography angiography (OCTA) images, an artificial intelligence (AI) model was constructed to assess multiple tasks such as the grading of diabetes retinopathy (DR).

Methods  The study population comprised diabetic patients who received treatment at the Qixia Branch of Jiangsu Province Hospital, Jiangsu Province Hospital, or Yangzhou University Affiliated Hospital between January 2020 and December 2023. Fundus color photographs and OCTA images were collected from these patients and subjected to standardized pre-processing. Based on fundus color pho-tography and OCTA features, a predictive model was constructed using a multimodal con-volutional neural network. Three task branches were established: DR grading, prediction of proliferative diabetic retinopathy (PDR) progression, and assessment of systemic com-plication risk. Model performance was evaluated using receiver operating characteristic curves and their area under the curve (AUC), Fleiss' Kappa coefficient, sensitivity, specificity, and F1 score.

Results  A total of 482 diabetic patients were included, 340 with DR. The validation set and training set exhibited consistent trends in loss curves and classification accuracy curves with minimal divergence, indicating that the predictive model did not overfit. The training set achieved a DR five-level classification accuracy of 94.2%. Within the test set, the model achieved an AUC value exceeding 0.90 for all five DR severity classifications, demonstrating high concordance with ophthalmologist assessments (Fleiss' Kappa=0.78, P<0.001). The AUC for predicting progression from moderate non-PDR to PDR was 0.896; the AUCs for predicting future diabetic nephropathy, cardiovascular events, and neuropathy were 0.854, 0.839, and 0.812 respectively.

Conclusion  The AI-based multimodal retinal image analysis system can achieve accurate grading, future progression risk prediction of DR, and assessment of systemic diabetic complications based on retinal imaging. It holds potential for clinical implementation and primary-level screening  applications.

1. Jhaveri CD, Liu D, Maguire MG, et al. Risk factors for meeting criteria for switching from Bevacizumab to Aflibercept when treating eyes with diabetic macular edema and visual acuity of < 20/40[J]. Ophthalmology, 2024, 131(8): 967-974.

2. Chiavaroli L, Lee D, Ahmed A, et al. Effect of low glycaemic index or load dietary patterns on glycaemic control and cardiometabolic risk factors in diabetes: systematic review and Meta-analysis of randomised controlled trials[J]. BMJ, 2021, 374: n1651.

3. 李筱荣, 黎晓新, 许迅. 糖尿病视网膜病变社区筛查专家共识 [J]. 中华眼底病杂志, 2024, 40(2): 98-104. [Li XR, Li XX, Xu X. Expert consensus on community screening for diabetic retinopathy[J]. Chinese Journal of Ocular Fundus Diseases, 2024, 40(2): 98-104.]

4. 程泉, 查旭, 周丽琼, 等. 细胞焦亡在糖尿病视网膜病变发病机制中的作用[J]. 国际眼科纵览, 2025, 49(1): 45-50. [Chen Q, Zha X, Zhou LQ, et al. The role of pyroptosis in the pathogenesis of diabetic retinopathy[J]. International Review of Ophthalmology, 2025, 49(1): 45-50.]

5. 刘念恩, 龙泽, 徐小同, 等. 糖尿病视网膜病变房水细胞因子的研究进展[J]. 中华眼科杂志, 2025, 61(3): 221-229. [Liu NE, Long Z, Xu XT, et al. Research progress of aqueous humor cytokines in diabetic retinopathy[J]. Chinese Journal of Ophthalmology, 2025, 61(3): 221-229.]

6. 李青博, 邵彦. 代谢组学在糖尿病视网膜病变中的研究进展[J]. 中华眼底病杂志, 2023, 39(6): 510-514. [Li QB, Shao Y. Research progress of metabolomics in diabetic retinopathy[J]. Chinese Journal of Ocular Fundus Diseases, 2023, 39(6): 510-514.]

7. Yang Q, Bee YM, Lim CC, et al. Use of artificial intelligence with retinal imaging in screening for diabetes-associated complications: systematic review[J]. EClinicalMedicine, 2025, 81: 103089.

8. Tanabe H, Sato M, Miyake A, et al. Machine learning-based reproducible prediction of type 2 diabetes subtypes[J]. Diabetologia, 2024, 67(11): 2446-2458.

9. Hayati A, Abdol Homayuni MR, Sadeghi R, et al. Advancing diabetic retinopathy screening: a systematic review of artificial intelligence and optical coherence tomography angiography innovations[J]. Diagnostics (Basel), 2025, 15(6): 737.

10. Waheed NK, Rosen RB, Jia Y, et al. Optical coherence tomography angiography in diabetic retinopathy[J]. Prog Retin Eye Res, 2023, 97: 101206.

11. Dai L, Sheng B, Chen T, et al. A deep learning system for predicting time to progression of diabetic retinopathy[J]. Nat Med, 2024, 30(2): 584-594.

12. Ding X, Romano F, Garg I, et al. Expanded field OCT angiography biomarkers for predicting clinically significant outcomes in non-proliferative diabetic retinopathy[J]. Am J Ophthalmol, 2025, 270: 216-226.

13. Syed MG, Trucco E, Mookiah MRK, et al. Deep-learning prediction of cardiovascular outcomes from routine retinal images in individuals with type 2 diabetes[J]. Cardiovasc Diabetol, 2025, 24(1): 3.

14. Nabrdalik K, Irlik K, Meng Y, et al. Artificial intelligence-based classification of cardiac autonomic neuropathy from retinal fundus images in patients with diabetes: the Silesia Diabetes Heart Study[J]. Cardiovasc Diabetol, 2024, 23(1): 296.

15. Ghenciu LA, Dima M, Stoicescu ER, et al. Retinal imaging-based oculomics: artificial intelligence as a tool in the diagnosis of cardiovascular and metabolic diseases[J]. Biomedicines, 2024, 12(9): 2150.

16. Drinkwater JJ, Chen FK, Brooks AM, et al. Carotid disease and retinal optical coherence tomography angiography parameters in type 2 diabetes: the fremantle diabetes study phase II[J]. Diabetes Care, 2020, 43(12): 3034-3041.

17. Horton WB, Barrett EJ. Microvascular dysfunction in diabetes mellitus and cardiometabolic disease[J]. Endocr Rev, 2021, 42(1): 29-55.

18. DiKun KM, Gudas LJ. Vitamin A and retinoid signaling in the kidneys[J]. Pharmacol Ther, 2023, 248: 108481.

19. Shamsan A, Senan EM, Ahmad Shatnawi HS. Predicting of diabetic retinopathy development stages of fundus images using deep learning based on combined features[J]. PLoS One, 2023, 18(10): e0289555.

20. Heger KA, Waldstein SM. Artificial intelligence in retinal imaging: current status and future prospects[J]. Expert Rev Med Devices, 2023, 21(1-2): 73-89.

21. Sacks FM, Hermans MP, Fioretto P, et al. Association between plasma triglycerides and high-density lipoprotein cholesterol and microvascular kidney disease and retinopathy in type 2 diabetes mellitus: a global case-control study in 13 countries[J]. Circulation, 2014, 129(9): 999-1008.

22. Broe R, Rasmussen ML, Frydkjaer-Olsen U, et al. Retinal vascular fractals predict long-term microvascular complications in type 1 diabetes mellitus: the danish cohort of pediatric diabetes 1987 (DCPD1987)[J]. Diabetologia, 2014, 57(10): 2215-2221.

23. 王嘉鹏, 罗向霞, 庄家圆, 等. 氧化应激在糖尿病视网膜神经变性中的作用机制与潜在治疗的研究进展[J]. 中华眼底病杂志, 2024, 40(10): 813-818. [Wang JP, Luo XX, Zhuang JY, et al. Research progress on the mechanism and potential treatment of oxidative stress in diabetic retinal neurodegeneration[J]. Chinese Journal of Ocular Fundus Diseases, 2024, 40(10): 813-818.]