Topological vascular tree segmentation for retinal images using shortest path connection

This paper presents a novel algorithm for vascular tree segmentation based on shortest path connection. The connected vascular tree provides topological features that are instrumental in image-aided diagnosis. The proposed method can enforce the connectivity as well as remove the false detection at same time. Multi-scale ridge detector is employed that can locate vessels with different widths. To connect the isolated ridge, the single-source shortest path algorithm is tailored that searches the optimal path. The path metric is defined in terms of probability of pixel belong to foreground and background. This mechanism enables that the false detection could be removed via hypothesis testing. The topological vascular tree with 1-pixel width and fully-connection is segmented from the retinal image. The simplicity and efficiency of the proposed method make it practical to be employed in image-aided diagnosis system readily.