Identification and reconnection of interrupted vessels in retinal vessel segmentation

The morphology of retinal blood vessels contains valuable information for the diagnosis of retinal dysfunctions. The vessels can be segmented from color fundus images but the connectivity of the segmented vessels is not always preserved because of low contrast, imaging noise and artifacts. If a continuous vessel is interpreted as multiple disjoint vessel segments, the morphological measurements such as tortuosity may not be representative of true properties of retinal vessels. We describe an algorithm to identify the vessel segment interruptions based on connected component analysis and then reconnect them using a graph based approach. The proposed method was evaluated on a dataset of 25 vessel segmentation images resulting into a reconnection performance measure of 81.63% compared to the gold standard obtained by the manual reconnection process. Our approach has allowed the complete vessel tree to be connected, and has potential in providing improved morphological measurements.