Improving vascular segmentation by geometric fitting in level set evolution

Segmentation defines a way of splitting image into fragments establishing homogeneous and meaningful regions. Main purpose of segmentation is the reduction of information so that image contents can be investigated easily. State of the art developments in radiological imaging results in high spatial and temporal resolution leading to huge data cloud. Robust segmentation algorithms are required in diagnostic systems to extract and analyze targeted vasculature. This study aims to propose a quick semi-automated framework for extracting arteries from Magnetic Resonance Imaging (MRI) volumetric data by incorporating geometric shape features in level set evolution. Despite lots of efforts only partial success has been achieved towards automatic arterial segmentation due to intra-patient variation and complex structure of the arteries. The proposed method involves a blend of conventional threshold based techniques and level set formulation for achieving maximum accuracy along the vessel surface. Analytic shape fitting is used to approximate lumen at certain points to ensure accuracy. Shape fitting becomes effective for grabbing distal segments because the blood voxels concentration is generally decreased towards distal points. Segmented arterial tree appears promising and validates the model provided by the medical experts, however it is to be noted that no algorithm exist that performs absolute accurate segmentation due to the complex nature of arterial tree. Statistical quantification is the future work where this approach is to be extended for extraction of coronary arteries from CTA cardiac volumes.