SDE-Driven Spatio-Temporal Hypergraph Neural Networks for Irregular Longitudinal fMRI Connectome Modeling in Alzheimer's Disease

Abstract

Longitudinal neuroimaging is essential for modeling disease progression in Alzheimer's disease (AD), yet irregular sampling and missing visits pose substantial challenges for learning reliable temporal representations. To address this challenge, we propose SDE-HGNN, a stochastic differential equation (SDE)-driven spatio-temporal hypergraph neural network for irregular longitudinal fMRI connectome modeling. The framework first employs an SDE-based reconstruction module to recover continuous latent trajectories from irregular observations. Based on these reconstructed representations, dynamic hypergraphs are constructed to capture higher-order interactions among brain regions over time. To further model temporal evolution, hypergraph convolution parameters evolve through SDE-controlled recurrent dynamics conditioned on inter-scan intervals, enabling disease-stage-adaptive connectivity modeling. We also incorporate a sparsity-based importance learning mechanism to identify salient brain regions and discriminative connectivity patterns. Extensive experiments on the OASIS-3 and ADNI cohorts demonstrate consistent improvements over state-of-the-art graph and hypergraph baselines in AD progression prediction. The source code is available at https://anonymous.4open.science/r/SDE-HGNN-017F.

Figures (4)

• Figure 1: Overview of SDE-HGNN framework. (a) Irregular longitudinal fMRI signals are reconstructed using a neural SDE, followed by hypergraph construction and SDE-driven spatio-temporal modeling with GRU refinement and aggregation. (b) Hypergraph construction from reconstructed ROI signals via pairwise distance computation and KNN-based hyperedge formation. (c) Downstream tasks include progression prediction, diagnosis classification and interpretation.
• Figure 2: Longitudinal visualization of salient ROIs in the progressive group across six follow-up visits. Color indicates regional importance (yellow: higher; red: lower) among the top-20 ROIs at each time point.
• Figure 3: Top-30 discriminative functional connections between stable and progressive groups across six longitudinal time points ($t_1$--$t_6$), identified via FDR-corrected two-sample $t$-tests ($p < 0.05$) on learned hyperedge importance probabilities. ROIs are grouped into seven large-scale neural systems based on the Schaefer100 (7-network) parcellation: visual (VIS), somatomotor (SMN), dorsal attention (DAN), ventral attention (VAN), limbic (LIM), frontoparietal control (CON), and default mode network (DMN). The intra-network connections are colored by their respective network and inter-network connections are colored in grey.
• Figure 4: Sensitivity analysis of three loss weighting hyperparameters.