H-Net: Heterogeneous Neural Network for Multi-Classification of Neuropsychiatric Disorders.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Structural MRI (sMRI) and functional MRI (fMRI) are linked to neuropsychiatric disorders (NDs).
• Previous research focused on binary classification of NDs using integrated MRI modalities.
• Classifying multiple ND subclasses remains difficult due to disease complexity.

Purpose of the Study:

• To develop a novel heterogeneous neural network (H-Net) for multi-class classification of NDs.
• To integrate sMRI and fMRI data effectively for improved diagnostic accuracy.
• To address the challenge of differentiating complex ND subclasses.

Main Methods:

• Developed H-Net, a heterogeneous neural network integrating sMRI and fMRI data.
• Employed MLP-based and GAT-based encoders for feature extraction from sMRI and fMRI, respectively.
• Utilized a cross-modality transformer block with MLP-mixer and cross-modality alignment for feature interaction and improved classification.

Main Results:

• H-Net achieved 90% classification accuracy on a public dataset (CNP) for multi-class ND diagnosis.
• Demonstrated the complementary benefits of combining sMRI and fMRI for enhanced ND identification.
• Visual and statistical analyses confirmed distinct differences between ND subclasses.

Conclusions:

• H-Net offers a powerful tool for accurate multi-class classification of neuropsychiatric disorders.
• Integrating sMRI and fMRI data through H-Net significantly improves diagnostic performance.
• The study underscores the potential of advanced AI models in neuroimaging for understanding complex brain disorders.