Colloquium: Jan Bernard Marsman

Support Vector Machines in the Frequency Discrimination Task

Traditionally, fMRI analysis uses a univariate method for statistical testing, such as the General Linear Model. In the past few years multivariate approaches have been developed also taking spatial relations between voxels into account, including Support Vector Machines (SVMs). An SVM is a binary classifier which calculates the optimal hyperplane (decision boundary) between two classes in a training set, which separates them best. This hyperplane is then used to predict the class of the items in a test set.

The goal of this study is to inform the question how vibrotactile stimuli are represented in the somatosensory cortex. We have conducted an experiment where the subject has to make a discrimination between vibrotactile stimuli with different frequencies applied to the index finger. The performances of human frequency discrimination have been compared with the performance of the SVM on the corresponding captured fMRI images, in order to find correlations. These correlations will provide information about how frequency in vibrotactile sensation is represented in the brain.

During my presentation I will discuss the set-up, methods and results of this experiment conducted at the Magnetic Resonance Imaging and Analysis Research Center, University of Liverpool