Related Experiment Video
Updated: Mar 21, 2026
Experimental Autoimmune Uveitis: An Intraocular Inflammatory Mouse Model
Published on: January 12, 2022
Quantitative Assessment of Fluorescein Angiography Leakage via Deep Learning in Pediatric Uveitis: Correlation with
Yu Jin1, Jianguo Huang1, Yuhan Chang2
1National Clinical Research Center for Ocular Diseases, Eye Hospital, Wenzhou Medical University, Wenzhou, Zhejiang, China.
Purpose:
Develop a deep learning algorithm for automated segmentation of retinal vascular and macular edema (ME) leakage in fluorescein angiography (FA) of pediatric uveitis.
Design:
Model trained on internal FA images with external validation.
Participants:
One hundred eighty-three pediatric uveitis patients (3144 FA images) at the Eye Hospital of Wenzhou Medical University. External validation used 45 patients (396 FA images) from The First Affiliated Hospital of Soochow University and The Affiliated Eye Hospital of Nanjing Medical University.
Methods:
The model was trained, validated, and applied to 28 independent pediatric cases to correlate leakage area with clinical metrics and determine clinically significant change thresholds.
Main Outcome Measures:
Segmentation performance (Dice) and pixel-based leakage area for analysis.
Results:
In fivefold cross-validation on 2459 internal images (155 patients), the best mean Dice scores were 0.7241 (retinal leakage) and 0.7273 (macular leakage). On external data, the best mean Dice reached 0.7015 and 0.5918, respectively. In the cohort of 28 pediatric uveitis patients, vitreous cells (VCs) and central macular thickness (CMT) were the most consistent parameters associated with retinal leakage across multiple analyses. After multiple comparison correction, both VC and CMT remained significantly associated with total and posterior pole leakage areas at all time points. For ME leakage, CMT and best-corrected visual acuity showed consistent associations, while VC was significant only at follow-up, whereas other examined variables showed no consistent association after correction for multiple comparisons. Receiver operating characteristic analysis identified a -54.8% relative change in total visible retinal leakage area as the optimal cutoff, indicating quiescent inflammation.
Conclusions:
The deep learning model reliably segments leakage in pediatric uveitis FA, showing multicenter generalizability. Quantitative angiographic parameters correlate with clinical metrics, suggesting potential utility for informing disease management strategies.
Financial Disclosures:
The author has no/the authors have no proprietary or commercial interest in any materials discussed in this article.
