Bifurcation detection in 3D vascular images using novel features and random forest

Bifurcation detection is important in medical image analysis for mainly two reasons: 1) plaques are easy to accumulate at artery bifurcations, which leads to atherosclerosis and strokes; 2) for quantification (e.g. branch length, thickness, tortuosity), visualization, and blood flow simulation, it's necessary to extract all the branches and their connectivity in a vessel tree, which makes bifurcation localization crucial. In this paper, several novel features are designed for classifying bifurcations in 3D vascular images using random forest. Encouraging results with both synthetic and real datasets are obtained.