A new data-driven approach to analyze functional-magnetic-resonance-image data of the brain

Functional MRI (fMRI) has been pivotal in exploring the relationship between structural and functional networks in human brains. To address the complex spatiotemporal data from these experiments, many studies are based on existing hypotheses with inherent assumptions about neural dynamics. Furthermore, the collective properties are usually undetected. Here a data driven approach is adopted to analyze whole-brain fMRI data from participants under four visual stimuli, fixation and two resting states. Using the maximum entropy principle and disregarding temporal data correlation, we simplified brain activity into a pairwise spin model, where each region is either ’silent’ or ’active’ (spin values 0 or 1, respectively). This approach enabled us to create hierarchical regional clusters based on interaction strength, revealing robust connections among visual, auditory, language, sensory-motor, and default mode networks. These networks collectively establish distinct communities during different conditions. Also uncovered are several long-range, strong negative or inhibitory-like couplings between the cerebral hemispheres, extending beyond previous models. Interestingly, the seven models all exhibit emergent properties of a critical state which closely aligns with prior studies of the mouse brain at the single-neuron level. This similarity, suggests new research directions to understand the collective property of the brain.