FPG-Seg: Annotation-free Retinal Vessel Topology Segmentation
FPG-Seg is an annotation-free method for segmenting retinal vessel topology. It combines classical vesselness filtering, Bayesian deep learning to create a probability map, and graph reconstruction to propose a network structure. This pipeline achieves high performance (86-98% clDice) without needing manual labels, demonstrating the power of combining different techniques for complex medical image analysis.