We use Kernel Canonical Correlation Analysis (KCCA) for detecting brain activity in function MRI by learning a semantic representation of fMRI brain scans and their associated time frequency. The semantic space provides a common representation and enables a comparison between the fMRI and time frequency. We compare the approach against Canonical Correlation Analysis (CCA) by localising brain regions that control finger movement and regions that are involved in mental calculation. We also compare the two approaches on a simulated null data set.
We hypothesis that once a link can be established between regions of the brain to task one could create a brain-computer interface were computer related tasks could be activated by brain "thought" activity.