Improved Multi-scale Opening Algorithm Using Fuzzy Distance Transform Based Geodesic Path Propagation

Vessel tree segmentation from CT scan angiogram images of the human brain is a challenging task. The complex geometry, interconnections, and fusion with soft tissues and bones make the segmentation process harder. The segmented cerebrovasculature plays a major role in the fast analysis of vascular geometry leading to an effective diagnosis of the diseased cerebrovascular segment. The present work proposes a geodesic path propagation based on fuzzy distance transform to improve the multi-scale opening algorithm for effective segmentation of carotid vessel with less user intervention. The geodesic path is estimated between a pair of vessel seeds given by the user. The points on the paths are used as the initial pure vessel seeds during the multi-scale opening of the vascular tree from other fused conjoint components (bone, soft tissue etc.) in a shared intensity space. We developed a 2D/3D user interface to mark user-specified vessel/bone seeds or separators on the input images. Experiments on three patients’ CTA images show significant qualitative improvement in segmentation results with much lesser user intervention.