Can the Coronary Artery Centerline Extraction in Computed Tomography Images Be Improved by Use of a Partial Volume Model?

摘要

We propose the use of a statistical partial volume (PV) model to improve coronary artery tracking in 3D cardiac computed tomography images, combined with a modified centerline extraction algorithm. PV effect is a challenge when trying to separate arteries from blood-filled cardiac cavities, causing leakage and erroneous segmentations. We include a Markov Random Field with a modified weighting scheme. First, synthetic phantoms were used to evaluate the robustness and accuracy of PV detection, as well as to determine the best settings. Average Dice similarity index obtained for PV voxels was 86%. Then cardiac images from eight patients were used to evaluate the usefulness of PV detection to separate real arteries from cavities, compared to Fuzzy C-means classification. Our PV detection scheme reduced approximately by half the number of leakages between artery and cavity. The new version of artery centerline extraction algorithm takes advantage of the PV detection capacity to separate arteries from cavities and to retrieve low-signal small vessels. We show some preliminary qualitative results of the complete method.