Nonsubsampled Contourlet Transform based Expectation Maximization Method with Adaptive Mean Shift for Automatic Segmentation of MR Brain Images

An automatic method of MR brain image segmentation into three classes White Matter, Gray Matter and Cerebrospinal fluid is presented. The intensity non uniformity or bias field and noise present in the MR brain images pose major limitations to the accuracy of traditional EM segmentation algorithm. To overcome these drawbacks, Nonsubsampled Contourlet Transform low pass filter is used as preprocessing step. Since the bias field is found to be smoothly varying, it is proposed and applied that the GMM is preserved locally in the image blocks of appropriate size. Hence the image is divided into blocks and then EM segmentation is applied. To ensure smoothness among the segmentation output of the successive blocks, an adaptive mean shift followed by pixel stretching is proposed. The algorithm is evaluated on T1 weighted simulated brain MR images and 20 normal T1-weighted 3-D brain MR images from IBSR database. Results ensure that there is around 4% improvement in accuracy in Gray Matter Segmentation for 3-D brain MR images compared to fuzzy local Gaussian mixture model. Also the computational costs are reduced in this method.