Activity-State Entropy: A novel brain entropy measure based on spatial patterns of activity

Background: Brain entropy is a measure of the complexity of brain activity that has been linked to various cognitive abilities. The measure is based on Shannon Entropy, a measure from Information Theory that quantifies the information capacity of a system from the probability distribution of its states. Most fMRI studies measure brain entropy at the voxel level as time-series entropy and assume that entropic time-series indicate complex large-scale spatiotemporal patterns of activity.

New method: We developed a novel measure of brain entropy called Activity-State Entropy. The method quantifies entropy based on underlying patterns of coactivation identified using Principal Components Analysis. These patterns, termed eigenactivity states, combine in time-varying proportions.

Results: We showed that Activity-State Entropy is sensitive to the complexity of the spatiotemporal patterns of activity in simulated fMRI data. We then applied this measure to real resting-state fMRI data and found that the eigenactivity states that explained the most variance in the data were comprised of large clusters of coactivating voxels, including clusters within Default Mode Network regions. More entropic brains were increasingly influenced by eigenactivity states comprised of smaller and more sparsely distributed clusters.

Comparison to existing methods: We compared Activity-State Entropy to Sample Entropy and Dispersion Entropy, two time-series entropy measures commonly used in neuroimaging research, and found all three measures were positively correlated.

Conclusions: Activity-State Entropy provides a measure of the spatiotemporal complexity of brain activity that complements time-series based measures of brain entropy.