Neurocircuitry-Inspired Hierarchical Graph Causal Attention Networks for Explainable Depression Identification

TL;DR

NH-GCAT addresses the challenge of explainable depression identification from rs-fMRI by integrating neurobiological priors into a graph-based model. It combines three components: RG-Fusion for temporal-BOLD and static FC fusion; HC-Pooling for hierarchical circuit encoding across DMN, SN, FPN, LN, and RN; and VLCA for variational latent causal attention to capture directed inter-circuit information flow, enabling counterfactual reasoning. The approach yields state-of-the-art performance on REST-meta-MDD, with interpretable, multi-scale insights into depression neurobiology, including frequency-specific dynamics, circuit hierarchy alterations, and causal circuit interactions, all framed within a clinically relevant predictive model. These contributions advance explainable neuropsychiatric prediction by linking predictive signals to known depression circuitry, supporting biomarker discovery and potential personalized treatment guidance.

Abstract

Major Depressive Disorder (MDD), affecting millions worldwide, exhibits complex pathophysiology manifested through disrupted brain network dynamics. Although graph neural networks that leverage neuroimaging data have shown promise in depression diagnosis, existing approaches are predominantly data-driven and operate largely as black-box models, lacking neurobiological interpretability. Here, we present NH-GCAT (Neurocircuitry-Inspired Hierarchical Graph Causal Attention Networks), a novel framework that bridges neuroscience domain knowledge with deep learning by explicitly and hierarchically modeling depression-specific mechanisms at different spatial scales. Our approach introduces three key technical contributions: (1) at the local brain regional level, we design a residual gated fusion module that integrates temporal blood oxygenation level dependent (BOLD) dynamics with functional connectivity patterns, specifically engineered to capture local depression-relevant low-frequency neural oscillations; (2) at the multi-regional circuit level, we propose a hierarchical circuit encoding scheme that aggregates regional node representations following established depression neurocircuitry organization, and (3) at the multi-circuit network level, we develop a variational latent causal attention mechanism that leverages a continuous probabilistic latent space to infer directed information flow among critical circuits, characterizing disease-altered whole-brain inter-circuit interactions. Rigorous leave-one-site-out cross-validation on the REST-meta-MDD dataset demonstrates NH-GCAT's state-of-the-art performance in depression classification, achieving a sample-size weighted-average accuracy of 73.3\% and an AUROC of 76.4\%, while simultaneously providing neurobiologically meaningful explanations.

Figures (6)

• Figure 1: Comparison between conventional data-driven and our proposed neurocircuitry-inspired approaches for MDD classification. AI: Artificial Intelligence; MDD: Major Depressive Disorder; HC: Healthy Control.
• Figure 2: The overall framework of the proposed NH-GCAT. BOLD: Blood Oxygenation Level Dependent; FC: Functional Connectivity; MDD: Major Depressive Disorder; HC: Healthy Control.
• Figure 3: Multi-scale interpretability analysis of NH-GCAT for MDD classification.
• Figure 4: Comprehensive performance evaluation of NH-GCAT across 5 cross-validation folds.
• Figure 5: Directional differences in hierarchical layer distributions between MDD and HC groups. Positive values (red) indicate higher proportions in MDD, negative values (blue) indicate higher proportions in HC. *$p < 0.05$, **$p < 0.01$.