MHNet: Multi-view High-Order Network for Diagnosing Neurodevelopmental Disorders Using Resting-State fMRI

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning models show potential for diagnosing neurodevelopmental disorders (NDD) like ASD and ADHD.
• Existing models often overlook high-order features in brain functional networks (BFNs) derived from rs-fMRI data, limiting diagnostic accuracy.
• Graph neural networks (GNNs) and spatial convolution are commonly used but have limitations in capturing complex network hierarchies.

Purpose of the Study:

• To introduce a novel Multi-view High-order Network (MHNet) for enhanced prediction of neurodevelopmental disorders (NDD).
• To capture hierarchical and high-order features from multi-view BFNs derived from rs-fMRI data.
• To improve NDD classification by integrating features from both Euclidean and non-Euclidean spaces.

Main Methods:

• Developed MHNet with two branches: Euclidean Space Features Extraction (ESFE) and Non-Euclidean Space Features Extraction (Non-ESFE).
• ESFE utilizes Functional Connectivity Generation (FCG) and High-order Convolutional Neural Network (HCNN) modules.
• Non-ESFE employs Generic Internet-like Brain Hierarchical Network Generation (G-IBHN-G) and High-order Graph Neural Network (HGNN) modules, followed by Feature Fusion-based Classification (FFC).

Main Results:

• MHNet demonstrated superior performance compared to state-of-the-art methods on three public datasets using AAL1 and Brainnetome Atlas templates.
• Ablation studies confirmed the effectiveness of multi-view fMRI information and high-order features in MHNet.
• The study identified key brain regions associated with NDD and provided atlas options for network construction.

Conclusions:

• MHNet effectively leverages multi-view feature learning from both Euclidean and non-Euclidean spaces.
• Incorporating high-order information from BFNs significantly enhances NDD classification performance.
• The proposed method offers a promising approach for improving the diagnosis of neurodevelopmental disorders.