Functional Network Connectivity and SVM Classification of fMRI data using R Ariana Anderson, Ph.D., Mark S. Cohen, Ph.D. UCLA Center for Cognitive Neuroscience 760 Westwood Plaza, Suite 17-369 Los Angeles, CA 90095 E-mail: ariana82 @ ucla . edu March 15, 2011 Abstract We demonstrate and provide R code that can classify between groups of fMRI scans based on functional network connectivity differ- ences using the packages analyzefMRI, vegan, igraph, and e1071. We run ICA on fMRI data to establish functional networks, measure the functional connectivity between these networks using the temporal crosscorrelations between independent component, and then transform the connectivities of each scan into a graph structure. Connectivity properties of these graph structures are extracted, and used as a fea- ture matrix for an SVM classifier. Collectively, this manuscript pro- vides code that requires the user only to supply a list of filenames to be processed, the fMRI scans in Analyze or NIFTI format, and the scan labels, requiring only 4 lines of code to be altered. All other parameters are inferred from the data to make these routines adaptable for general usage. This code can also be altered to perform connectivity analysis and classification using ROI based methods by reading in distance ar- rays previously created. Collectively, this paper provides and explains both methods and code to perform functional network connectivity and fMRI SVM classification. The code is freely available on the NITRC website at http://www.nitrc.org/projects/fmriclassify/. 1 Introduction Functional Magnetic Resonance Imaging (fMRI) is a four-dimensional medi- cal imaging modality that captures changes in blood oxygenation over time, an indirect measure of neuronal activation. An increasing focus of interest is the classification between subject groups based on the fMRI signal vari- ations. One method of accomplishing this is through functional network connectivity (FNC), an established set of methods applied within fMRI to 1