MRI brain tissues segmentation using non-parametric technique

This paper presents a fully-automatic and robust MRI segmentation method for brain tissues. The proposed method classifies the brain MRI volume to 4 classes: white matter tissue (WM), gray matter tissue (GM), cerebrospinal fluid (CSF), and the remaining tissues as non-brain tissues (NBT). We utilize the pre-segmented volumes to determine statistically the prior probability for each class, prior information of the spatial locations of the voxels in the class, and also the intensity of each voxel. Parzen window is used to estimate non-parametrically the PDF of the prior information. Bayes rule is used to find the maximum posterior probability for each voxel. Experiments on real and simulated data demonstrate the advantages of the method over the recent methods. Several experimental results are reported.