Statistical Cerebrovascular Segmentation for Phase-Contrast MRA Data

In this paper, we present a statistically based segmentation algorithm to extract vascular tree from Phase Contrast Magnetic Resonance Angiography, “PCMRA”. Classification is based on the intensity distribution of the tissues in the data volume, where voxels are classified as either vessels or non-vessels. Our algorithm is adaptive to reduce the bias field effect in the images by estimating the model parameters for each slice of the data volume using the Expectation Maximization, “EM” algorithm for finite mixture of densities. A connectivity filter is designed taking into account the topology of the vascular tree to remove nonvessel tissues that appear after statistical segmentation in the form of small islands. We also implemented the Maximum Intensity Projection algorithm, “MIP” to validate the results by projecting both the segmented and original data volumes at different angles. Hence, the resultant images from both volumes can be compared side by side, which facilitates the validation process. Finally, the segmented tree is visualized in 3D using Visualization Toolkit, “VTK”, which can be viewed on a stereo graphics workstation or in a virtual reality environment. The results are validated by our medical research team and successfully showed small MRA vessels down to the limit of the scanner resolution.