OCTA Retinal Vessel Segmentation Based on Vessel Thickness Inconsistency Loss

Optical coherence tomography angiography (OCTA) technology has been applied to retinal examination for clinical diagnosis. OCTA images reveal important details of eye diseases such as diabetic retinopathy (DR), glaucoma and age-related macular degeneration (AMD). DR and AMD are the leading causes of blindness in these diseases. Quantitative analysis of retinal vessel can help doctors diagnose retinal diseases and track the progression of these diseases. In this paper, we propose a new OCTA vessel segmentation framework based on variable vessel thickness. Specifically, to guide the network to adapt to the scale changes of vessels, we construct a vessel structure attention module. It can better capture the vessel structure by guiding the network to pay attention to the vessel edge information and help the network to establish a good context dependency. By assigning corresponding weights to different pixels, it helps the network to better learn vessels of different thicknesses and segment a more complete vessel structure. Finally, the framework is evaluated on the OCTA500 dataset, and experimental results demonstrate the effectiveness of the proposed segmentation framework.