Peripapillary Atrophy Segmentation Based on ASM Loss

Abstract

Peripapillary atrophy (PPA) is a clinical finding that reflects atrophy of the retinal layer and retinal pigment epithelium and it is a symptom related to many diseases. The shape and region area indicate the pathologic severity of myopia and glaucoma. So it is important to segment PPA area to analyze the progression of these diseases. The shape of PPA area is mostly crescent. In order to combine the prior knowledge, a PPA segmentation method is proposed with a novel active shape model (ASM) loss. The shape constraint is introduced to improve the segmentation accuracy. It is realized by the proposed ASM loss module, which contains three parts, θ predictor, b predictor and an affine transformation module. Our approach is evaluated on a clinical dataset. Extensive experiments demonstrate that our method provides good performance both qualitatively and quantitatively.