A level set based predictor-corrector algorithm for vessel segmentation

Abstract

Vessel segmentation is an essential task in many computer-aided medical systems. However, the topology complexity of vascular structures and the intensity inhomogeneity of angiogram make it a challenging problem. We propose a level set based predictor-corrector algorithm to meet these challenges. In the predictor step, the overall contour of vessel structures is delineated by piecewise constant (PC) model, which is insensitive to the initial contour and adaptive to the complex morphological variations of vessel structures. In the corrector step, the segmented results are refined by an improved local binary fitting (LBF) model, which can efficiently deal with intensity inhomogeneity in the angiogram, especially in the distal part of the vessels. Compared to original LBF model, our approach can avoid the emergence of new contour in non-vascular regions. The proposed algorithm takes both global and local information into consideration and combines the advantages of PC model and LBF model. Experimental results on MRA images demonstrate the feasibility of our algorithm.