Causal Markov random field for brain MR image segmentation.

Publication Type	Academic Article
Authors	Razlighi Q, Orekhov A, Laine A, Stern Y
Journal	Annu Int Conf IEEE Eng Med Biol Soc
Volume	2012
Pagination	3203-6
Date Published	01/01/2012
ISSN	2694-0604
Keywords	Brain
Abstract	We propose a new Bayesian classifier, based on the recently introduced causal Markov random field (MRF) model, Quadrilateral MRF (QMRF). We use a second order inhomogeneous anisotropic QMRF to model the prior and likelihood probabilities in the maximum a posteriori (MAP) classifier, named here as MAP-QMRF. The joint distribution of QMRF is given in terms of the product of two dimensional clique distributions existing in its neighboring structure. 20 manually labeled human brain MR images are used to train and assess the MAP-QMRF classifier using the jackknife validation method. Comparing the results of the proposed classifier and FreeSurfer on the Dice overlap measure shows an average gain of 1.8%. We have performed a power analysis to demonstrate that this increase in segmentation accuracy substantially reduces the number of samples required to detect a 5% change in volume of a brain region.
DOI	10.1109/EMBC.2012.6346646
PubMed ID	23366607
PubMed Central ID	PMC3771086