Dual Graph Strategy with Diffusion Tensor Imaging for Autism Spectrum Disorder Diagnosis.

Area of Science:

• Neuroimaging
• Machine Learning
• Biomedical Engineering

Background:

• Diffusion Tensor Imaging (DTI) is a key MRI technique for analyzing brain connectivity.
• Existing DTI research often focuses on single data types, potentially missing complementary information for complex neurological disorders like Autism Spectrum Disorder (ASD).

Purpose of the Study:

• To develop a novel graph neural network (GNN) framework utilizing a dual graph strategy with DTI data for improved ASD diagnosis.
• To overcome limitations of small datasets and class imbalance in neuroimaging research.

Main Methods:

• Employed data augmentation techniques (sample replication, mixup) to enhance dataset diversity and address class imbalance.
• Integrated a threshold-based physical connectivity matrix with a microstructure adjacency matrix derived from node features.
• Developed a Multi-Layer Pooling Fusion (MLPF) method for capturing multi-layered feature representations.

Main Results:

• The proposed dual graph GNN framework achieved 75.24% accuracy and 73.12% AUC in five-fold cross-validation on 198 subjects.
• Outperformed multiple existing methods, demonstrating the efficacy of the dual graph strategy.

Conclusions:

• DTI is vital for identifying connectivity abnormalities in ASD.
• The proposed framework offers a more efficient, objective, and reliable approach for ASD diagnosis using DTI.
• This study presents a valuable reference for applying DTI data in neurological disorder research.