Unsupervised segmentation of 3-D brain MR images

In this paper, we propose an algorithm for unsupervised segmentation of 3-dimensional sagittal brain MR images. Three-dimensional images consist of sequences of two dimensional images. We start the three dimensional segmentation from mid-sagittal brain MR images. Once these mid-sagittal images are successfully segmented, we use the resulting images to simplify the processing of the more lateral sagittal slices. In order to segment mid-sagittal brain MR images, we first apply thresholding to obtain binary images. Then we find some landmarks in the binary images. The landmarks and anatomical information are used to preprocess the binary images. The preprocessing includes eliminating small regions and removing the skull, which substantially simplifies the subsequent operations. The strategy is to perform segmentation in the binary image as much as possible and then return to the original gray scale image to solve problematic areas. Once we accomplish the segmentation of the mid-sagittal brain MR image, the segmented brain area is used as a mask for adjacent slices. Experiments show promising results.