Spectral Graph Neural Networks for Cognitive Task Classification in fMRI Connectomes

TL;DR

This work targets cognitive task classification from fMRI-derived functional connectomes by modeling brain regions as graphs and applying spectral graph neural networks. SpectralBrainGNN performs exact graph Fourier transforms using the normalized Laplacian $L = I_N - D^{-1/2} A D^{-1/2}$ with eigen-decomposition $L = U \Lambda U^T$, enabling learnable frequency filters $g(\lambda)$ (via an MLP) in the frequency domain, followed by inverse transforms and a channel-wise linear projection. An attention-based readout then yields a graph-level representation for final classification, trained end-to-end with cross-entropy. On the HCPTask dataset, SpectralBrainGNN achieves 96.25% accuracy, outperforming strong baselines and displaying robust, low-variance performance, with a statistically significant improvement (p = 0.028). The method emphasizes frequency-specific connectivity motifs and functional activation pathways in brain networks, offering improved interpretability and potential clinical relevance, while noting scalability considerations related to eigendecomposition and outlining avenues for dynamic and multimodal extensions.

Abstract

Cognitive task classification using machine learning plays a central role in decoding brain states from neuroimaging data. By integrating machine learning with brain network analysis, complex connectivity patterns can be extracted from functional magnetic resonance imaging connectomes. This process transforms raw blood-oxygen-level-dependent (BOLD) signals into interpretable representations of cognitive processes. Graph neural networks (GNNs) further advance this paradigm by modeling brain regions as nodes and functional connections as edges, capturing topological dependencies and multi-scale interactions that are often missed by conventional approaches. Our proposed SpectralBrainGNN model, a spectral convolution framework based on graph Fourier transforms (GFT) computed via normalized Laplacian eigendecomposition. Experiments on the Human Connectome Project-Task (HCPTask) dataset demonstrate the effectiveness of the proposed approach, achieving a classification accuracy of 96.25\%. The implementation is publicly available at https://github.com/gnnplayground/SpectralBrainGNN to support reproducibility and future research.

Figures (2)

• Figure 1: Schematic diagram of SpectralBrainGNN model for Cognitive Task Classification.
• Figure 2: Confusion matrix of HCPTask classification task