An accurate and efficient Bayesian method for automatic segmentation of brain MRI

Abstract

Automatic three-dimensional (3-D) segmentation of the brain from magnetic resonance (MR) scans is a challenging problem that has received an enormous amount of attention lately. Of the techniques reported in the literature, very few are fully automatic. In this paper, we present an efficient and accurate, fully automatic 3-D segmentation procedure for brain MR scans. It has several salient features; namely, the following. 1) Instead of a single multiplicative bias field that affects all tissue intensities, separate parametric smooth models are used for the intensity of each class. 2) A brain atlas is used in conjunction with a robust registration procedure to find a nonrigid transformation that maps the standard brain to the specimen to be segmented. This transformation is then used to: segment the brain from nonbrain tissue; compute prior probabilities for each class at each voxel location and find an appropriate automatic initialization. 3) Finally, a novel algorithm is presented which is a variant of the expectation-maximization procedure, that incorporates a fast and accurate way to find optimal segmentations, given the intensity models along with the spatial coherence assumption. Experimental results with both synthetic and real data are included, as well as comparisons of the performance of our algorithm with that of other published methods.

Fast and robust parameter estimation for statistical partial volume models in brain MRI

“…Moreover, parameter estimation based on segmented image was demonstrated to be much faster than competing EM style algorithms (Noe and Gee, 2001;Van Leemput et al, 2003). Also, as demonstrated with the IBSR data set, our fast PV estimation routine produced hard segmentations of a similar quality than a state of the art algorithm for this task (Marroquin et al, 2002). In summary, we recommend the use of the TMCD estimator based on hard labeling for fast and reliable parameter estimation for a statistical PVE model.…”

“…Rajapakse and Krugge (1998) compared several algorithms and the best TC values were obtained by Adaptive MAP method, and they were 0.567 for the WM and 0.564 for the GM. A more recent method by Marroquin et al (2002) achieved TCs 0.683 for the WM and 0.662 for the GM. The average running time for this algorithm was 19.2 min with the IBSR data set, which is similar to our methods, but no PV estimation can be obtained with this method.…”

“…Furthermore, the initial hard classification and correction of intensity nonuniformity can be performed with other methods than those applied in this study. The state of the art algorithms for these tasks (such as Marroquin et al, 2002;Van Leemput et al, 1999;Zhang et al, 2001) are however more time consuming than the methods applied in this study. Moreover, the robustness in parameter estimation compensates also for classification errors as was shown with misregistered Brainweb images.…”

Closed-Form Relaxation for MRF-MAP Tissue Classification Using Discrete Laplace Equations

“…A method that closely relates with graph cuts is the random walker (RW) algorithm of Grady [13,14], which was previously exposed in a different and less general form by Marroquin et al [15]. In this work, we show how to re-tune RW to yield a powerful MRF-MAP relaxation method that boils down to solving a sparse linear system.…”

Image Segmentation with Implicit Color Standardization Using Spatially Constrained Expectation Maximization: Detection of Nuclei

“…In an effort to maintain tractability, authors have invoked simplifying assumptions; unfortunately, these assumptions tended to reduce the generality of the solutions. For example, Comer and Delp [4] demanded that the Markov prior remain constant during the EM iteration; Nikou et al [5] imposed Gaussianity constraints; Marroquin [6] tied the implementation to a predetermined cost function, thus linking estimation (i.e. model fitting) and classification.…”