Adapting SAM for Foveal Avascular Zone Segmentation in Retinal Optical Coherence Tomography Angiography Images

The foveal avascular zone (FAZ) is a crucial indicator for retinal disease detection. Accurate FAZ localization and segmentation are the two main steps in designing an automatic screening system for retinopathy. In this paper, we propose a novel approach for detecting the FAZ based on saliency object detection and a modified adaptive Segment Anything Model (SAM). It consists of two stages: in the first stage, the saliency detection technique is introduced to the enhanced retinal image to locate the FAZ region. In the second stage, an adaptive module is designed to integrate the semantic prior information of the FAZ region into the SAM encoder segmentation network to extract the FAZ boundary. The performance of our proposed approach is tested on the public OCTA-500 database. Compared to the state-of-the-art approaches, the experimental results show the advantages and effectiveness of the proposed approach.