Segmentation of Three-Dimensional Data of Brain Magnetic Resonance Angiography Based on Markov Random Field

In this paper, a full automatic method was proposed for the segmentation of brain magnetic resonance angiography (MRA) dataset, which improved the technologies of Markov random field (MRF). Existing 3D-MRF models generally faced some challenges including: (1) The parameter initialization of low level MRF model is not accurate; (2) The ordinary neighborhood system cannot deal with local fine vessel structure. Aiming to solve the two problems, the multiscale filtering with threshold analysis and a multi-pattern neighborhood system were proposed, respectively. Such method enabled the MRF model delineating vessels to be as small as two voxels in diameters. In the experiments, the parameters of the low level MRF model were estimated using the expectation maximization algorithm, while the parameters of the high level MRF models were estimated based on the maximum pseudo likelihood algorithm. A set of phantoms and some MRA clinical datasets were used to validate the algorithms, to yield smaller segmentation errors.