Bifurcation Localization in 3D Images via Evolutionary Geometric Deformable Templates

Abstract

Given the importance of studying bifurcations in 3D anatomical trees (e.g. vasculature and airway), we propose a bifurcation detector that operates by fitting a parametric geometric deformable model to 3D medical images. A fitness function is designed to integrate features along the model skeletons, surfaces and internal areas. To overcome local optima while detecting multiple bifurcations in a single image, we adopt genetic algorithm with a tribes niching technique. Results on both VascuSynth data and clinical CT data demonstrate not only high bifurcation detection accuracy and stability, but the ability to locate parent and children branch directions and vessel wall locations simultaneously.