Automatic segmentation of vessels from angiogram sequences using adaptive feature transformation

Abstract

This paper proposes an efficient method for automatically segmenting vessels from angiogram sequences. The method includes two steps: extracting high-contrast angiograms and segmenting vessels. First, we select high-contrast angiograms automatically using vessel intensity distribution. Based on multiscale Hessian-based filtering, we propose an adaptive feature transformation function to improve the vesselness response. This method overcomes numerous problems, which exist in the X-ray angiograms by using the scale factors and transformed intensities. Various scales are established to mitigate variations of the intensity distribution. The transformed intensities are applied to coping with lower contrast and nonuniform intensity distribution. Finally, the connected component labeling method is used to extract the vessels. The proposed method can distinguish between the vessel and the background in a complex background. In our experiments, 20 angiogram sequences are used to evaluate the accuracy of the selected high-contrast angiogram. The accuracy of extracting high-contrast angiograms is 98%. For evaluating the accuracy of the segmentation results, 72 angiograms were selected. The accuracy of the proposed segmentation method is 96.3%. The Kappa value is 81.8%. After inspection by a cardiologist, the experimental results show that the proposed method can automatically and accurately segment vessels.