Gaussian mixture models for brain activation detection from fmri data

Abstract

Gaussian Mixture Model (GMM) based clustering has been successfully used in various types of medical and image data analysis, because of its robustness and stability under high noise levels. GMMs are employed in this work to extract the activation patterns from functional Magnetic Resonance Imaging (fMRI) data. The highly correlated time-series obtained with a given stimulus has been used to find the voxels contributing to the Blood Oxygenation Level Dependent (BOLD) activation regions. GMM clustering has been used for modeling of various activation patterns considering the strength, delay and duration of the epochs. A synthetic dataset and a real dataset provided by the Wellcome Trust Centre for Neuroimaging, University College London, UK are used to demonstrate the superiority of this approach in automating the process of identifying activated brain regions.