Knowledge-based 3D segmentation and reconstruction of coronary arteries using CT images.

An approach for the 3D segmentation and reconstruction of human left coronary arteries using angio-CT images is presented in This work. Each voxel in the 3D dataset is assumed to belong to one of the three homogeneous regions: blood, myocardium, and lung. A priori knowledge of the regions is introduced via Bayes' rule. Posterior probabilities obtained using Bayes' rule are anisotropically smoothed, and the 3D segmentation is obtained via MAP classifications of the smoothed posteriors. An active contour model is then applied to extract the coronary arteries from the rest of the volumetric data with subvoxel accuracy. The geometric model of the left coronary arteries obtained in this work may be used to provide accurate boundary conditions for hemodynamic simulations, or to provide objective measurements of clinically relevant parameters such as lumen sizes in a 3D sense.