Automatic Segmentation and Quantitative Analysis of Gray Matter on MR Images of Patients with Epilepsy Based on Unsupervised Learning Methods

摘要

The quantitative analysis of volume information about gray matter (GM) on magnetic resonance (MR) images is important in both research and clinical diagnosis of patients with epilepsy. In this paper, a k-means method and an expectation maximization algorithm are implemented respectively to achieve segmentation of GM on MR images at the transverse and coronal plane. The experiments were performed on both multi-modal and mono-modal MR images and the similarity index values for the accuracy of automatic segmentation with manual segmentation were consistently high for patients with epilepsy (transverse plane: 0.806; coronal plane: 0.837). The results demonstrated that the automatic segmentation methods implemented in this paper are accurate and efficient to realize extraction of GM of patients with epilepsy in both transverse and coronal plane.