fMRI Sonification & Brain Activity Prediction

The study of human brain functions has dramatically increased greatly due to the advent of functional Magnetic Resonance Imaging (fMRI), arguably the best technique for observing human brain activity that is currently available. However, fMRI techniques produce extremely high dimensional, sparse and noisy data which is dicult to visualize, monitor and analyze. In this document, we propose a sonication approach for exploratory fMRI data analysis. The goal of this tool is to allow the auditory identication of cognitive states produced by dierent stimuli. The system consists of a feature selection component and a sonication engine. We will explore dierent feature selection methods and sonication strategies. Moreover, we present a computational model which predicts the fMRI neural activation in humans produced by rhythm/no-rhythm auditory stimuli. The model was trained with acoustic features extracted from the auditory signals and the associated observed fMRI images. The obtained model is able to predict fMRI activation with high accuracy. This work represents a natural progression from building catalogues of patterns of fMRI activity associated with particular auditory stimuli to constructing computational models which predict the fMRI activity for auditory stimuli for which fMRI data are not available yet.