Skip to content

From brain signals to meaningful networks

How can brain activity recordings reveal the interactions that distinguish subjects and characterise brain disorders?

A painted brain with sampled areas joined by grey candidate links. Dashed leaders connect five areas to activity traces; the three traces that share one rhythm turn blue, and a blue sub-network appears over washed brain areas.
From activity recorded at individual brain areas to a network of functional relationships: traces that share one rhythm single out a sub-network (blue) among the candidate links (grey). Areas, signals and links are schematic, not measured data or an atlas. Connections represent co-activation relationships, not anatomical fibres or causal pathways.
highlights

How can we reconstruct informative networks when we observe the activity of individual brain regions? Our work examines which collective patterns of co-activation carry useful information about subjects and conditions.

Functional brain networks reconstructed from fMRI usually rely on statistical pruning: each correlation between two regions of interest is thresholded independently, ignoring collective co-activation patterns within sub-networks. Functional pruning instead retains connections for their collective importance in distinguishing healthy from affected individuals.

We build multilayer network representations from a multi-frequency decomposition of the signals, learn them with geometric deep learning, and use an explainable-AI tool to identify the most discriminative sub-networks.

Selected papers

Marco Grassia, Valeria d’Andrea, Karolina Finc, Manlio De Domenico, Giuseppe Mangioni (2026).
Machine-enhanced reconstruction of functional connectomes unravels discriminative brain sub-systems in health and disease.
Scientific Reports.
DOI

Research topics: Learning on Graphs · Multilayer Networks and Mesoscale Structure