BOF-Net: A Retinal Branch Vein Occlusion Segmentation Diagnostic Network for Fundus Fluorescein Angiography Images

Branch Retinal Vein Occlusion (BRVO) is a common retinal vascular disease in clinical practice, typically associated with retinal arteriosclerosis. Currently, doctors can only perform subtype diagnosis of patients by visually assessing the size ratio between non-perfused areas and the optic disc in fundus fluorescein angiography (FFA). In recent years, deep learning technology has been widely applied in medical imaging. However, due to the scarcity of retinal vein occlusion imaging data and the lack of publicly available datasets, applying deep learning network models to segment and diagnose the occlusion areas remains challenging. Our goal is to propose a segmentation network model, BOF-Net, based on BRVO imaging data, to segment the non-perfused areas and the optic disc regions in the images and calculate their respective areas, thereby assisting doctors in more accurately performing subtype diagnosis. BOF-Net achieved a Dice coefficient of 85.26% and an accuracy of 93.99% on the BRVO dataset. Compared to other segmentation network models, BOF-Net shows greater potential and effectiveness in the auxiliary diagnosis of BRVO.