Dynamic Functional Connectivity Features for Brain State Classification: Insights from the Human Connectome Project

TL;DR

This work shows that linear classifiers can robustly decode 14 brain states from HCP fMRI data by using mean regional activations across 379 regions. It demonstrates state-specific, interpretable neurofunctional signatures and reveals that temporal dynamics critically shape functional connectivity, as evidenced by comparisons to shuffled data and KS tests. The study provides a principled method for feature selection, revealing unique sets of regions per state and showing that high-accuracy states are characterized by focused regional patterns. These findings support functional specialization in cortical and subcortical networks and offer a transparent framework for neurobiological interpretation of brain-state decoding.

Abstract

We analyze functional magnetic resonance imaging (fMRI) data from the Human Connectome Project (HCP) to match brain activities during a range of cognitive tasks. Our findings demonstrate that even basic linear machine learning models can effectively classify brain states and achieve state-of-the-art accuracy, particularly for tasks related to motor functions and language processing. Feature importance ranking allows to identify distinct sets of brain regions whose activation patterns are uniquely associated with specific cognitive functions. These discriminative features provide strong support for the hypothesis of functional specialization across cortical and subcortical areas of the human brain. Additionally, we investigate the temporal dynamics of the identified brain regions, demonstrating that the time-dependent structure of fMRI signals are essential for shaping functional connectivity between regions: uncorrelated areas are least important for classification. This temporal perspective provides deeper insights into the formation and modulation of brain neural networks involved in cognitive processing.

Figures (4)

• Figure 1: Accuracy plot depending on the number of features.
• Figure 2: The graphs of cardinality of the top features sets.
• Figure 3: The Jaccard coefficient matrix $\mathbf{J}$ heatmap.
• Figure 4: Distributions of time series correlation of top and least features.