Relevance vector machine analysis of functional neuroimages

We propose the use of the relevance vector machine (RVM) regression framework for statistical analysis of PET or fMRI data sets in a two state ("on-off") activation study. According to this approach the shape of the activations is a superposition of kernel functions, one at each pixel of the image, of unknown amplitude and a hierarchical Bayesian model is employed which imposes a sparse representation. This allows accurate estimation of the activation locations when correlated noise is present even at low signal-to-noise ratios. We tested this method using an artificial phantom derived from a previous neuroimaging study. This proposed approach compared favorably with previous approaches.