Automatic segmentation of coronary arteries using Bayesian driven implicit surfaces

In this paper, we propose a hybrid approach for the automatic three-dimensional segmentation of coronary arteries using multi-scale vessel filtering and a Bayesian probabilistic approach in a level set image segmentation framework. The initial surface of the coronaries is obtained from the multiscale vessel filter response, and the surface then evolves to capture the exact boundary of the coronaries according to an improved evolution model of implicit surfaces. In our model, the image force and the propagation terms are re-defined using posterior probabilities obtained via Bayes' rule in order for the surface to approach to the boundaries faster and stop at the boundaries more accurately. The proposed method is tested on seven CT angiography (CTA) data-sets of left and right coronary arteries, and the quantitative comparison of our result against manually delineated contours on two of the data-sets yields a mean error of 0.37mm.