An isotropic minimal path based framework for segmentation and quantification of vascular networks

Abstract

Minimal path approaches for image analysis aim to extract curves minimizing an energy functional. The energy of a path corresponds to its weighted curve length according to a relevant metric function. In this study, we design a binary isotropic metric model with the use of a Hessian-based vascular enhancement filter in order to extract geometrical features from vascular networks. We introduce a constrained keypoint search method able to extract subpixel vessel centrelines, diameters and bifurcations. Experiments on retinal images demonstrated that the proposed framework achieves similar even better segmentation performances as compared with methods using more sophisticated metric designs.