ON THE INDEPENDENT COMPONENTS OF FUNCTIONAL NEUROIMAGES

K.S. Petersen, L.K. Hansen, T. Kolenda

We discuss the application of ICA procedures to fMRI (functional Magnetic Resonance Imaging) sequences. While principal component analysis can identify acti- vation patterns that are uncorrelated in both space and time ICA can identify events that are independent in either time or space. We show that the activation re- lated components found by either spatial or temporal independency are consistent, hence robust to choice of spatial or temporal separation and to choice of ICA ap- proach. We discuss these issues in the context of three ICA algorithms applied to an fMRI visual activation study.