Robust brain activation detection in functional MRI

Functional Magnetic Resonance Imaging (MRI) is today one of the most important non-invasive tools to study the brain from a functional point of view. The blood-oxygenation-level-dependent (BOLD) signal is used to detect the activated regions based on the assumption that in these regions the metabolic activity increases. The normal procedure is the application of known sequences of stimulus and find out the brain regions whose activation sequence is correlated with the applied stimulus. This inference problem is difficult because the BOLD signal is very week and noisy. The underlying information is embedded in a large number of other signal related with the normal brain activity and in the noise introduced by the MRI scanner. Furthermore, the hemodynamic impulse response function (HRF), needed to know the expected BOLD response to a given stimulus, is usually unknown and is not constant across the whole brain. In this paper a robust Bayesian algorithm is proposed to detect regions where the activation patterns are correlated with the applied stimulus. The activation process is modeled by using binary explicative variables and the HRF is estimated at each location according to a physiological model proposed by the authors in [1]. Monte Carlo tests using synthetic data are performed to evaluate the performance of the algorithm and results with real data are compared with the ones obtained by a neurologist with the commercial package BrainVoyager.