108 — A Novel Brain Decoding Method: a Correlation Network Framework for Revealing Brain Connections

This paper proposes a CorrNet-based approach to develop correlation graphs between individual voxels of a functional brain scan (fMRI). In other words, this project aims to further refine fMRI data to “learn” which voxels of the scan (i.e. brain regions) are functionally connected, both in space and time.

The authors first determine topological correlation: This is a measure of a few features such as distance between two voxels, and the regularity with which the two neighborhoods’ activity patterns in the fMRI BOLD signal overlap. This accomplishes both anatomical and topological correlation, because it avoids boundary conditions where two brain regions are anatomically close but are not topological neighbors (for example, brain regions on either side of a fissure; they appear to touch in a low-resolution brain scan, but are not physically touching in vivo).

Next, signal strength data can be overlaid in order to enrich the existing correlations. This adds a higher resolution to the dataset, which can now incorporate relationships between brain regions that fire with different (but reliable) degrees of activity to a stimulus.

This information, the authors propose, can be used to direct our understanding of higher-level cortical brain areas.