Mining for associations between text and brain activation in a functional neuroimaging database

We describe a method for mining a neuroimaging database for associations between text and brain locations. The objective is to discover association rules between words indicative of cognitive function as described in abstracts of neuroscience papers and sets of reported stereotactic Talairach coordinates. We invoke a simple probabilistic framework in which kernel density estimates are used to model distributions of brain activation foci conditioned on words in a given abstract. The principal associations are found in the joint probability density between words and voxels. We show that the statistically motivated associations are well aligned with general neuroscientific knowledge.