Segmentation and Centerline Extraction of the Coronary Arteries with GPU Processing

Coronary artery disease is one of the most common causes of death in Norway, and the world. While diagnosing this disease, and treating it is a common procedure at many hospitals, discovering and documenting a faster, safer and cheaper way to detect the disease could be critical for many patients. Our goal is to implement an algorithm that can automatically detect and produce a segmentation of the coronary arteries from Computed Tomography Angiography (CTA) images. In this thesis we explore and discuss several different methods and approaches for segmentation of the coronary arteries and vessels in general in relation to compatibility for parallel execution on a GPU, accuracy and practicality. A suitable method is chosen based on the discussion and implemented. The results show that the algorithm is able to segment large portions of the coronary arteries. But the produced results are vulnerable to noise, artifacts and irregular vessels. The most significant improvement would be a centerline selection addition, where the centerlines that are chosen to produce the segmentation is chosen based on their location in relation to the heart. Or some other method that is more reliable then the current bio-mechanical method. Our implementation is able to produce a segmentation and a centerline in around 10 minutes. The method implemented in this thesis is a general approach to vessel segmentation, but tuned for coronary artery segmentation. The approach could potentially be adapted and used for extraction of other tubular structures