Task Based Representation of fMRI Signals Using Spatiotemporal Characterization

Functional task evoke extrinsic responses in the brain that is expected to capture the cognitive and motor interaction, specific to the task. However, such responses in brain regions can vary between individuals and are superimposed with ongoing background activities. Thus deriving a typical and gross signature of such response in terms of neuronal activation over the entire duration of the task becomes challenging. In this paper, we attempt for a task representation to find a group of regions whose activation levels are consistent among multiple subjects. Functional Magnetic Response Imaging (fMRI) data of four cognitive tasks namely, Visual word and object recognition, Working memory, Simon task and Tone counting are used to derive the representations. Hammersmith Atlas, consisting of 83 brain regions, is used as a mask to study the fMRI data. Considering the challenges due to inherent noise in the fMRI data, functional overlaps between the tasks and differences between individual responses, a band of activation values is used instead of the absolute ones. For spatiotemporal representation of a task initially, every voxel is binarized for an activation with a predefined band. Next, for each brain region, the fraction of activated voxels in that region is partitioned into three equal percentiles. Then, the temporal summary of each regional activation is obtained using the proportion of time for which it is in each of the three percentiles. Finally, the representation is obtained using group of regions, which are clustered based on the temporal summary and consistent for multiple subjects. The representation is able to identify few important brain regions whose functionalities match with the task objectives and the findings of Neuroscience research. Such representations formed using healthy subjects will be useful to create the models to enable the quantification of the deviations due to neurological diseases.