Multi-modal and Multi-model Interrogation of Large-Scale Functional Brain Networks

Castaldo, Páscoa Dos Santos et al. — NeuroImage 277, 120236, 2023.

Why I cared. Every whole-brain model seemed built for one signal — fMRI or MEG — as if the two came from different brains. I didn’t believe that. The same network dynamics should underlie both.

What I did. I took two models — Stuart–Landau and Wilson–Cowan — on the same structural connectome and asked one set of dynamics to reproduce features across modalities at once: functional connectivity, its dynamics (FCD), and metastable oscillatory modes, in both fMRI and MEG.

What I found. Both models captured MEG connectivity and metastable modes to a comparable degree, mostly by tuning global coupling and conduction delay (and, for Wilson–Cowan, the excitation–inhibition balance). Delays mattered enormously — removing them wrecked everything. But the optimal working point differed by modality, and no single parameter set matched empirical connectivity well across modalities. The shared dynamics are real; one tuning just can’t speak every language at once.

What it opened. If each modality wants its own working point, what is each one actually measuring — and is there a principled way to find a single model that stays multilingual across fMRI and MEG?