Interpretable Neuropsychiatric Diagnosis via Concept-Guided Graph Neural Networks

TL;DR

The paper tackles the interpretability challenge in rs-fMRI-based neuropsychiatric diagnosis by introducing ConceptNeuro, a framework that combines LLM-driven generation of clinically grounded connectivity concepts with a concept bottleneck GNN classifier. Concepts are encoded as structured subgraphs of functional connections and mediate predictions, enabling explanations aligned with neurobiological knowledge. Across ABCD and HCP-D datasets, ConceptNeuro improves predictive accuracy over vanilla GNNs and demonstrates strong alignment with expert concepts, while ablation studies confirm the importance of diverse, high-quality concepts. The work offers a pathway to clinically trustworthy, domain-informed diagnostic tools that can generate testable hypotheses about disorder-specific brain connectivity patterns.

Abstract

Nearly one in five adolescents currently live with a diagnosed mental or behavioral health condition, such as anxiety, depression, or conduct disorder, underscoring the urgency of developing accurate and interpretable diagnostic tools. Resting-state functional magnetic resonance imaging (rs-fMRI) provides a powerful lens into large-scale functional connectivity, where brain regions are modeled as nodes and inter-regional synchrony as edges, offering clinically relevant biomarkers for psychiatric disorders. While prior works use graph neural network (GNN) approaches for disorder prediction, they remain complex black-boxes, limiting their reliability and clinical translation. In this work, we propose CONCEPTNEURO, a concept-based diagnosis framework that leverages large language models (LLMs) and neurobiological domain knowledge to automatically generate, filter, and encode interpretable functional connectivity concepts. Each concept is represented as a structured subgraph linking specific brain regions, which are then passed through a concept classifier. Our design ensures predictions through clinically meaningful connectivity patterns, enabling both interpretability and strong predictive performance. Extensive experiments across multiple psychiatric disorder datasets demonstrate that CONCEPTNEURO-augmented GNNs consistently outperform their vanilla counterparts, improving accuracy while providing transparent, clinically aligned explanations. Furthermore, concept analyses highlight disorder-specific connectivity patterns that align with expert knowledge and suggest new hypotheses for future investigation, establishing CONCEPTNEURO as an interpretable, domain-informed framework for psychiatric disorder diagnosis.

Figures (3)

• Figure 1: Overview of the Proposed Framework.(Left) We first prompt LLMs to generate disorder-specific functional connectivity concepts, which are refined through filtering to remove irrelevant or redundant concepts, resulting in a compact set of $N_c$ concepts. (Center) For each subject, we extract subgraphs corresponding to these concepts and encode them together with the subject’s input functional connectivity graph. (Right) Finally, we compute concept scores, which are then passed through a concept bottleneck classifier to perform disorder prediction in an interpretable manner.
• Figure 2: The distribution of similarity scores.
• Figure 3: Ablation study results for two architectures.