Local maximum detection for fully automatic classification of EM algorithm

Abstract

In this paper, we proposed a method for fully-automatic EM segmentation on brain MR images without a priori knowledge. Instead of manually predetermination on number of tissue classes, the proposed method automatically find mean intensities of distinct tissues from the histogram. The brain MR images were chosen to test our proposed method, but our method can, in fact, be general for other MR segmentations using EM with which the Gaussian mixture distribution of an image histogram holds. The results from our method suggested that a fully automatic segmentation using EM can be achieved with no significant difference in segmentation accuracy compared to the conventional EM.