Detecting groups of coherent voxels in functional MRI data using spectral analysis and replicator dynamics

Abstract

Purpose: To investigate the relationship between functional MRI (fMRI) time series in the human brain, combining fMRI spectral analysis and replicator dynamics.

Materials and methods: Simulated and real fMRI time courses were investigated using the bivariate spectral coherence. Coherence values were placed in coherence matrices encoding the relationship between the time courses. Groups of maximally coherent voxels were detected using replicator dynamics. Results were compared to a former approach called number of coherent voxels (NCV).

Results: NCV critically depends on a threshold that has to be chosen in advance. The lower this threshold, the larger the detected group. Using higher NCV thresholds in our simulations, the method did not detect all voxels that were constructed to have a high coherence among each other. In contrast, the replicator process found the whole group in all simulations.

Conclusion: The application of replicator dynamics to spectral matrices is a reliable method for detecting groups of maximally coherent voxels. A replicator process is able to determine groups of voxels with the property that each voxel in the group exhibits a high coherence with every other group member. In contrast to the NCV approach, this method is parameter-free and does not require the a priori selection of a reference voxel.