Brain Tissue Classification of MR Images Using Fast Fourier Transform Based Expectation- Maximization Gaussian Mixture Model

Abstract

This paper proposes MR image segmentation based on Fast Fourier Transform based Expectation and Maximization Gaussian Mixture Model algorithm (GMM). No spatial correlation exists when classifying tissue type by using GMM and it also assumes that each class of the tissues is described by one Gaussian distribution but these assumptions lead to poor performance. It fails to utilize strong spatial correlation between neighboring pixels when used for the classification of tissues. The FFT based EM-GMM algorithm improves the classification accuracy as it takes into account of spatial correlation of neighboring pixels and as the segmentation done in Fourier domain instead of spatial domain. The solution via FFT is significantly faster compared to the classical solution in spatial domain — it is just O(N log 2N) instead of O(N^2) and therefore enables the use EM-GMM for high-throughput and real-time applications.