Computer-aided Segmentation of Foveal Avascular Zone in OCT-A Images

Abstract Precise segmentation of the foveal avascular zone (FAZ) is critical because FAZ size and integrity are significant predictors of retinal health and visual acuity. FAZ area measurement holds substantial research value, as it provides additional information that can be used to develop new treatment modalities. Quantitative evaluation of the FAZ can detect early retinal microvascular alterations that are not yet symptomatic. In this paper, a computationally light deep neural network structure is built for the segmentation of FAZ in OCT-A images. The model should be computationally light and have high accuracy, thus being appropriately applied in clinical scenarios and deployed in environments with limited resources. Our structure outperforms traditional structures in both speed and segmentation accuracy when evaluated on a retrospective OCT-A image dataset.