Regularization of voxelwise autoregressive model for analysis of functional magnetic resonance imaging data

Abstract

This study reports the construction of a simple and elegant method to improve the detection of brain regions with increased neuronal activity in functional magnetic resonance imaging (fMRI). The main advantages of the method are regularization of voxelwise autoregressive parameters and accurate estimation of parameters and statistical inferences as compared to the conventional restricted maximum likelihood (ReML) method, where each voxel uses the two voxel-wide (global) hyperparameters. The model is tested for its accuracy and efficiency in analysis of fMRI data for stimulated and real brains of healthy individuals. Receiver operating characteristic curves are used to show that the proposed method is superior to the conventional ReML method. The proposed method will improve the diagnostic accuracy of fMRI studies in the human brain. © 2011 Wiley Periodicals, Inc..