Artificial intelligence-based diagnostic model construction and quantitative analysis of retinal vein occlusion lesions using color fundus photographs.

Authors

Wenjuan Wu, Li Sun, Xingye Wang, Yan Liao, Jinrong Liu, Liwen He, Juan Liu, Jie Wang, Yilin Zhong, Hongyun Wu, Dennis S C Lam, Zhe Xu

Published in

Frontiers in cell and developmental biology. Volume 14. Pages 1888364. Epub Sep 29, 2026.

Abstract

Retinal vein occlusion (RVO) is the second most common retinal vascular disease worldwide, affecting over 7 million people in China. Currently, the clinical diagnosis and severity assessment of RVO largely rely on the subjective judgment of physicians, leading to high rates of missed diagnosis at an early stage, insufficient quantitative assessment, and poor consistency during follow-up. Most existing RVO-related AI models only support classification diagnosis and lack multi-dimensional quantitative assessment capabilities. This study evaluated an existing AI system (EVisionAI) for automatic segmentation and quantitative analysis of RVO lesions based on color fundus photographs, using a dataset of 106 RVO eyes and 60 normal control eyes. The system can simultaneously extract three core indicators: lesion characteristics, vascular morphology, and fundus tessellation density. The results showed that the Dice coefficients of the AI for the segmentation of hemorrhage, hard exudate, and cotton-wool spots were all above 0.78. After adjusting for demographic factors and systemic vascular risk factors in a multivariable GEE model, there were significant associations between the total area of hemorrhage, vascular fractal dimension, and vascular density for RVO (all P < 0.05), suggesting these metrics as promising objective biomarkers for RVO evaluation. A diagnostic model based on these multidimensional indicators achieved a cross-validated area under the curve (AUC) of 0.951 (SD: 0.018) for distinguishing RVO eyes from healthy controls. In summary, the proposed AI system enables precise, standardized, and reproducible multi-parametric quantitative evaluation of RVO. However, given the single-center cross-sectional design and the absence of external validation and disease controls, these findings are preliminary. Prospective multi-center studies with external validation are required to confirm its clinical utility.

PMID:
42871061
Bibliographic data and abstract were imported from PubMed on 14 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.