Mitigating activity mixing with personalized whole-brain modeling

Abstract

Mechanistic neuroimaging-based biomarkers based on localized brain activities or interactions are central to precision and personalized psychiatry. However, their accuracy may be limited by Activity Mixing, a novel construct indicating the entanglement of local neuronal activity with distributed network activities through long-range spatiotemporal correlations, which degrades brain–symptom localization. As a solution to mitigate this inverse problem, we developed a multi-objective fitting approach integrating synchronization and criticality metrics to estimate subject-specific local- and inter-areal control parameters from neuroimaging observables. In silico validation showed that fitted parameters improved brain–symptom correlation estimates by 30–85% and reduced false-negative rates by ~67% relative to conventional analyses. As in vivo proof-of-concept, resting-state MEG data from 230 patients with major depressive disorder demonstrated that alpha-band (11 Hz) criticality predicted disability (r = 0.236), which improved to 0.368 after model fitting with narrower anatomical localization. These findings suggest that model fitting can mitigate the effects of activity mixing, providing subject-specific control-parameter estimates that delineate potential biomarkers for brain disorders more accurately than raw brain imaging observables.