Neuro-CXG: A Graph Neural Network Framework for Brain Disease Classification Based on Causal Inference and Explainable AI

Neuro-CXG is an open-source Python framework that combines Graph Neural Networks (GNN), causal inference, and explainable AI technologies to analyze fMRI/DTI brain imaging data for intelligent classification of brain diseases and diagnostic assistance. The project uses the Apache 2.0 license and is developed by GitHub user Nidszxh, providing tool support for neuroscience research and clinical diagnosis.

Traditional brain disease classification has three major limitations: voxel-based methods ignore brain region connections; feature engineering relies on expert manual design; deep learning black-box models lack interpretability. The brain is a complex network structure—GNNs are naturally suitable for modeling brain region connections (nodes = brain regions, edges = connection strength); causal inference can distinguish between causal relationships among brain regions and confounding factors; explainable AI provides feature importance, etc., enhancing clinical credibility.

The framework includes three modules:

1. Data preprocessing: supports fMRI (time series extraction, functional connectivity matrix) and DTI (fiber tract tracking, structural connectivity matrix) processing;
2. GNN models: implements GCN (spectral graph convolution), GAT (attention mechanism), and causal GNN (causal discovery, intervention simulation, counterfactual reasoning);
3. Explainable module: integrates Grad-CAM for Graphs, attention visualization, SHAP value analysis, causal effect estimation.

Application scenarios:

• Alzheimer's disease: extract functional connectivity features from fMRI to identify early abnormal brain networks;
• Schizophrenia: analyze connection differences between patients and controls to assist in efficacy evaluation;
• Autism: discover changes in small-world properties of brain networks to support personalized interventions;
• Brain tumors: combine DTI to track fiber tracts, assess tumor impact on networks, assist in surgical planning.

Scientific research value: provides a standardized GNN framework, promotes the combination of causal inference and brain networks, and the application of explainable AI in neuroscience; Clinical potential: assists doctors in diagnosis, identifies drug targets, supports precision medicine; Open-source contribution: lowers entry barriers, promotes academic-industry collaboration, provides a platform for algorithm innovation.

Participation methods:

1. Installation: pip install or clone the GitHub repository;
2. Contribution: submit PRs to improve algorithms/add features;
3. Feedback: report bugs or suggestions via the Issue system;
4. Sharing: contribute use cases and tutorials.

Neuro-CXG represents the cutting-edge application of AI in neuroscience, integrating three major technologies to provide a credible tool for brain disease classification. With the accumulation of data and improvement of computing power, such frameworks will play a more important role in precision neuromedicine, and relevant researchers are advised to explore in depth.