Differential Capillary-Large Vessel Analysis in OCT Angiography for Predicting DME Treatment Outcomes

This study aims to evaluate the impact of differential capillary-large vessel (CLV) analysis in optical coherence tomography angiography (OCTA) on predicting anti-VEGF treatment outcomes in diabetic macular edema (DME). In this retrospective study, quantitative vascular features were extracted from OCTA images of DME patients before and after deep learning-based CLV segmentation. Seven biomarkers—vessel skeleton density (VSD), blood vessel density (BVD), perfusion intensity density (PID), vessel perimeter index (VPI), vessel tortuosity (BVT), blood vessel caliber (BVC), and vessel area flux (VAF)—were computed for total vasculature, capillaries, and large vessels, along with feature ratios. Two support vector machine (SVM) classification models were trained to predict anatomical (CRT) and functional (BCVA) treatment responses. Our comparative evaluation demonstrated that the integration of separate CLV features substantially improved prediction metrics. Specifically, the classification accuracy for BCVA enhancement increased from 71.43% to 88.1%, alongside an AUC boost from 73.32% to 94.4%. For CRT prediction, accuracy rose from 69.05% to 78.57%, with a corresponding AUC gain from 81.1% to 83.3%. Analyzing capillary and large vessel features independently through CLV analysis substantially improves the ability to predict responses to anti-VEGF in DME. By capturing complementary vascular characteristics and inter-vessel relationships, this approach supports more tailored therapeutic planning and facilitates refined clinical evaluations.