Independent component analysis of functional MRI data

Abstract

Functional Magnetic Resonance Imaging (fMRI) is a non-invasive technique that has been used by neuroscientists as a powerful tool to study human brain functions in response to stimuli. By generating high quality movies of the brain in action, it helps to determine which parts of human brain are activated by different task performances. The process can be modeled as a linear mixture of independent localized sources of oxygenation, where no a priori information is known about their properties. Here independent component analysis (ICA) is used to understand the brain functions and to explore spatiotemporal features in fMRI data. It has been especially successful to recover brain function related signals (task related and physiology related signals) from recorded mixtures of unrelated signals (noise). Due to the high dimensionality, high noise level and spikes (due to high sensitivity of MR scanners) analysis of fMRI data and order selection, i.e., estimation of independent component is critical. We have tried to find the independent components by a number of ICA algorithms from which Extended Efficient FastICA and Combi ICA are found to have better performance as they are robust to outliers (caused due to high sensitivity of MR scanners) and the accuracy in terms of Amari Performance Index is more as compared to others. In this paper we 1) describe fMRI data and its properties, 2) and show that the combi ICA faithfully separates the independent components from fMRI data.