U-Net Framework for Simultaneous Segmentation of Foveal Avascular Zone and Retinal Vessels Across Different Retinal Plexuses

Exploring and analyzing the foveal avascular zone (FAZ) and surrounding retinal vessels (RV) is crucial for early ocular disease detection. The arrangement of RV influences FAZ characteristics, yet previous research often segmented these biomarkers independently, ignoring their intrinsic relationship. This paper introduces a novel framework for simultaneously extracting FAZ and RV from distinct retinal plexuses in OCTA images using a U-Net architecture-based network. The network incorporates squeeze-and-excitation (SE) blocks, spatial attention, and channel attention mechanisms. It is trained and validated on the OCTA-500 dataset, which includes data from 500 subjects, covering 3mm and 6mm fields of view (FOVs) and three projection maps: Full, ILM-OPL, and OPL-BM. Performance evaluation uses metrics like F1 Score, Jaccard index (IoU), Precision, Recall, and Accuracy. Comparative analysis shows the proposed network outperforms the standard U-Net. Additionally, the alignment of quantitative findings with automated segmentation maps verifies method efficacy.