Fusing speed and phase information for vascular segmentation in phase contrast MR angiograms

This paper presents a statistical approach to aggregating speed and phase (directional) information for vascular segmentation in phase contrast magnetic resonance angiograms (PC-MRA), and proposes a Maxwell-Gaussian finite mixture distribution to model the background noise distribution. In this paper, we extend our previous work [6] to the segmentation of phase-difference PC-MRA speed images. We demonstrate that, rather than relying on speed information alone, as done by others [12,14,15], including phase information as a priori knowledge in a Markov random field (MRF) model can improve the quality of segmentation, especially the region within an aneurysm where there is a heterogeneous intensity pattern and significant vascular signal loss. Mixture model parameters are estimated by the Expectation-Maximization (EM) algorithm [3]. In addition, it is shown that a Maxwell-Gaussian finite mixture distribution models the background noise more accurately than a Maxwell distribution and exhibits a better fit to clinical data.