Vector-field classification in magnetic-resonance angiography

Phase-contrast magnetic-resonance angiography (PC MRA) generates images of vascular structures as three-dimensional maps of the blood-flow velocity in a volume of interest. To improve visualization methods for PC MRA, radiologists can benefit from image-processing algorithms that can classify flow and stationary tissue. In this paper, I describe a vector-difference distribution (VDD): a statistical model of noisy PC MRA that allows us to compute a measure of probability of flow for each voxel, based on the expected mixed distribution of flow and background samples. The estimates of flow probability form an image that can be used as a mask with, or as a surrogate for, the standard images for further processing and display. The implementation demonstrates that VDD (1) can classify probabilistically PC MRA images into flow and stationary tissue, and (2) can extract reliably first- and second-order statistical measures for flow and noise (background). A comparison of MIP images with and without a VDD-based probability mask demonstrates a 30-to-56-percent improvement in contrast-to-noise ratio.