Abstract
Purpose :
The purpose of this study is to advance differential Artery–Vein (AV) analysis in optical coherence tomography angiography (OCTA) and evaluate its impact on the stage classification of diabetic retinopathy (DR).
Methods :
212 OCTA images were obtained from 146 patients encompassing 28 control, 36 NoDR, and 82 DR patients with different stages. A convolutional neural network (CNN) AVA-Net was used to segment arterial-venous areas (Fig. 1B) within OCTA (Fig. 1A). Overlaying OCTA image and the AVA map results in the OCTA-AV map (Fig. 1C). After applying the Frangi filter, binarization, and skeletonization, various quantitative features, including perfusion intensity density (PID), blood vessel density (BVD), vessel area flux (VAF), blood vessel caliber (BVC), blood vessel tortuosity (BVT), and vessel perimeter index (VPI) were derived for total vasculature (Fig. 1A), arterial and venous areas (Fig. 1C). Using a sequential forward selection (SFS) algorithm in conjunction with support vector machine (SVM) classifiers, quantitative OCTA features before and after AV analysis were used for the binary and multiclass classification of DR stages.
Results :
Table 1 showcases performance comparison of binary and multiclass classifications before and after AV analysis. For binary classification, AV feature integration improved mean accuracy to 87.63%, compared to 78.86% accuracy with total vasculature features without AV differentiation. For multiclass classification, the AV analysis resulted in an accuracy increase to 85.66%, surpassing the previous 79.62% accuracy without AV analysis. These findings underscore the efficacy of AV differentiation in robust DR stage classification.
Conclusions :
By extracting features specific to arterial and venous areas, we assessed the efficacy of AV analysis for binary and multiclass classification of DR stages. Differential AV analysis in OCTA significantly enhances DR detection and classification.
This abstract was presented at the 2024 ARVO Imaging in the Eye Conference, held in Seattle, WA, May 4, 2024.