Local Orientation Smoothness Prior for Vascular Segmentation of Angiography

Abstract

We present a new generic method for vascular segmentation of angiography. Angiography is used for the medical diagnosis of arterial diseases. To facilitate an effective and efficient review of the vascular information in the angiograms, segmentation is a first stage for other post-processing routines. The method we propose uses a novel a priori — local orientation smoothness prior — to enforce an adaptive regularization constraint for the vascular segmentation within the Bayes’ framework. It aspires to segment a variety of angiographies and is aimed at improving the quality of segmentation in low blood flow regions. Our algorithm is tested on numerical phantoms and clinical datasets. The experimental results show that our method produces better segmentations than the maximum likelihood estimation and the estimation with a multi-level logistic Markov random field model. Furthermore, the novel algorithm produces aneurysm segmentations comparable to the manual segmentations obtained from an experienced consultant radiologist.