Multimodal Graph Neural Networks for Prognostic Modeling of Brain Network Reorganization

TL;DR

This paper introduces a multimodal graph neural network framework that models brain network reorganization over time using structural, diffusion, and functional MRI data, providing accurate and interpretable prognostic biomarkers.

Contribution

It presents a novel multimodal graph neural network approach incorporating stochastic dynamics and attention mechanisms for modeling brain network evolution and predicting clinical outcomes.

Findings

Achieves high predictive accuracy on longitudinal neuroimaging data.

Provides interpretable biomarkers like network entropy and graph curvature.

Demonstrates potential for clinical prognostics without new data collection.

Abstract

Understanding the dynamic reorganization of brain networks is critical for predicting cognitive decline, neurological progression, and individual variability in clinical outcomes. This work proposes a multimodal graph neural network framework that integrates structural MRI, diffusion tensor imaging, and functional MRI to model spatiotemporal brain network reorganization. Brain regions are represented as nodes and structural and functional connectivity as edges, forming longitudinal brain graphs for each subject. Temporal evolution is captured via fractional stochastic differential operators embedded within graph-based recurrent networks, enabling the modeling of long-term dependencies and stochastic fluctuations in network dynamics. Attention mechanisms fuse multimodal information and generate interpretable biomarkers, including network energy entropy, graph curvature, fractional memory indices, and modality-specific attention scores. These biomarkers are combined into a composite prognostic index to quantify individual risk of network instability or cognitive decline. Experiments on longitudinal neuroimaging datasets demonstrate both predictive accuracy and interpretability. The results highlight the potential of mathematically rigorous, multimodal graph-based approaches for deriving clinically meaningful biomarkers from existing imaging data without requiring new data collection.