Combining Laplacian Eigenmaps and Vesselness Filters for Vessel Segmentation in X-Ray Angiography

Abstract

Automatic vessel outline delineation from X-ray angiography is highly useful to cardiologists during interventional procedures, especially to measure clinical indices such as vessel diameters, perimeters and areas. The challenges of obtaining a fully automatic segmentation are plentiful: radiographic noise, irregular injection of contrast agent, vessel overlap, etc. While vesselness filters were proposed to detect probable vessel-like shapes, such techniques often fail to recover prominent vessels in a cluttered background, and may obtain irregular shapes when artifacts are present. In this study, we propose a novel approach to segment vessel-like structures, which combines vesselness filters and Laplacian eigenmaps. Our technique finds automatically a global optimum solution for the image segmentation problem. By using both vesselness and Laplacian features, this approach can recognize vessel-like shapes in the background, while preserving the regularity of the extracted shapes. A visual and quantitative evaluation of the proposed approach, on both simulated images and pediatric patient X-ray angiography data, demonstrates its usefulness and efficiency.