Segmentation of vasculature for intravital microscopy using bridging vessel snake

Robustly segmenting vessels even with less sharp boundary is important for achieving accurate biological analysis of blood vessels regulation within organs such as liver. This process is crucial for microvaculature reconstruction which is necessary for red blood cells flow distribution regulation analysis. The vessels with sharp edges are often used during manual analysis skewing the data toward a certain group of vessels. In this paper, we propose bridging vessel snake (BVS) for segmenting a network of vessels (especially ones with less sharp boundary) in intravital microscopy images. Our method enables segmentation of vessels with varying diameter while imposing the structure of vessels by utilizing a ribbon snake and adding energies of width and region. The initialization achieved by the skeletonization is used to segment mostly sharper vessels. The "bridges" between the segmentation of sharper (thus higher confidence) vessels are used for hypothesizing less sharp vessels. A preliminary evaluation against a manual ground truth using a receiver operating characteristic (ROC) curve reveals that the algorithm was able to improve the area under ROC curve up to 20% on the vessels with lower sharpness and achieve the area under a ROC in all ground truthed vessels of 0.90