fMRI analysis using support vector machines

Abstract

Support Vector Machines (SVM) as a tool has become one of the most established techniques for analyzing functional magnetic resonance imaging (fMRI) data in recent years. The ability to deal with very high dimensions in the feature space as well as it’s robustness have played a crucial role in promoting SVM’s popularity among scientists in the field of neuroscience and related research. These data were acquired during an experiment conducted by the Max Planck Institue where 22 subjects were given an investment decision task with changing levels of uncertainty. Recent literature suggests that a lot of information about individual differences in decision making lies in the variability of the blood-oxygen-level dependent (BOLD) fMRI signals. Given the computed variability of the BOLD level following the stimuli I train an SVM to classify the subjects with respect to their risk attitude. By reducing the dimensions of the input to the areas of the brain previously ascertained as relevant for decision making under uncertainty I decrease the computation time without using time intensive dimension reduction techniques. I then compare my results with the results and technique presented by Mohr and H¨ardle et al. (2010).