Robust Independent Component Analysis for fMRI

Independent component analysis (ICA) is an effective exploratory tool for analyzing spatio-temporal data. It has been successfully applied in analyzing functional Magnetic Resonance Imaging (fMRI) data, to recover the interested source signals from different parts of the brain. Due to the high sensitivity of MR scanners, outliers are inevitable in acquiring fMRI datasets while they cause misleading effects for the analysis. In the current literature, no particular method exists yet to handle this problem. In this paper, we propose a robust ICA procedure that is less sensitive to outliers in fMRI analyses. Singular value decomposition (SVD) is commonly used prior to ICA for dimension reduction. We first motivate SVD from a low rank ∗email: [email protected] †email: [email protected] ‡email: [email protected]