2014 IEEE Global High Tech Congress on Electronics (GHTCE 2014)

124 / 1971-01-01 00:00:00

Brain Magnetic Resonance Imaging Segmentation Using Scale-space Based Expectation Maximum Initial Method

5133,5134,5135

Abstract—Expectation Maximum (EM) algorithm is well-known in medical image segmentation. But it has disadvantage that it is very sensitive to initial values. In this paper, we propose a method in EM initial process based on Gaussian mixture models (GMM) and scale-space filtering by which we can get the fingerprint of kernelized density from original images. Compared our segmentation results with others in the similarity to segmentation groundtruth, ours can achieve better results than other EM variants.
Keywords—Expectation Maximum (EM) algorithm; medical image segmentation; initial values; Gaussian mixture models (GMM); space-scale filtering; segmentation groundtruth