Background/Objectives: Retinal vein occlusion (RVO) commonly causes vision loss from macular edema (ME). OCT biomarkers (IRF, SRF, HRF, and ELM/EZ disruption) inform prognosis and treatment but are rarely quantified routinely due to time burden and interobserver variability. We aimed to validate a deep-learning algorithm for automated detection and quantification of key OCT biomarkers in RVO-ME versus expert assessment. Methods: In this retrospective single-center study, 93 eyes with RVO-ME imaged with spectral-domain OCT were analyzed. The AI quantified IRF/SRF volumes, ELM/EZ interruption, and HRF counts. Two masked expert clinicians provided reference evaluations. Performance and agreement were assessed using ROC AUC, Cohen’s kappa, intraclass correlation coefficient (ICC), Pearson correlation, and Bland–Altman analysis. Image-quality metrics (foveal centration and retinal layer segmentation) were recorded. Results: The AI showed high diagnostic performance (AUC: SRF 0.969; ELM 0.871; EZ 0.958) and substantial-to-almost-perfect agreement (kappa: SRF 0.807; ELM 0.788; EZ 0.914). HRF quantification correlated strongly with experts (r = 0.89, p < 0.001), with very good agreement (ICC = 0.87) and minimal bias. Image-quality accuracy exceeded 98% for foveal centration and layer segmentation. Conclusions: This AI software enables reliable, rapid automated assessment of major OCT biomarkers in RVO-ME, supporting streamlined personalized management; prospective studies should confirm longitudinal monitoring and treatment-guidance value.
Carnevali et al. (2026) studied this question.