Related Experiment Video
Updated: Mar 29, 2026
12:21
Probing the Brain in Autism Using fMRI and Diffusion Tensor Imaging
Published on: September 12, 2011
26.0K
A Dual-Stream Transformer with Self-Supervised Contrastive Training for fMRI-Based Autism Spectrum Disorder
1Research Institute of Electronic Science and Technology, University of Electronic Science and Technology of China, Chengdu 611731, China.
Brain Sciences
|March 27, 2026
Summary
A new dual-stream Transformer, TwoTST, enhances Autism Spectrum Disorder (ASD) identification by integrating dynamic and global brain connectivity. This method improves classification accuracy, offering a reliable neuroimaging approach.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Imaging
Background:
- Autism Spectrum Disorder (ASD) diagnosis is challenging due to its heterogeneous nature.
- Existing Time-series Transformer (TST) models struggle to simultaneously analyze dynamic and global brain connectivity, impacting Autism Spectrum Disorder classification accuracy.
Purpose of the Study:
- To develop an advanced deep learning model for improved Autism Spectrum Disorder (ASD) classification.
- To integrate both dynamic and global brain connectivity features for enhanced diagnostic capabilities.
Main Methods:
- Proposed TwoTST, a dual-stream Transformer architecture combining raw Region of Interest (ROI) time series and Pearson Correlation Matrices (PCC).
- Employed self-supervised learning for pre-training, contrastive learning for feature alignment, and evaluated five fusion strategies.
- Utilized the ABIDE I dataset with the CC200 atlas for experimental validation.
Main Results:
- The dual-stream structure, contrastive learning, and pre-training significantly improved mean Area Under the Curve (AUC) by 3-4%, 3-6%, and 3-7% respectively.
- Attention Pooling emerged as the optimal fusion strategy for integrating connectivity data.
- Relative parameter changes indicated effective learning within TST modules and contrastive projection heads.
Conclusions:
- TwoTST effectively integrates dynamic and global brain connectivity for improved Autism Spectrum Disorder identification.
- The proposed dual-stream approach surpasses single-stream models in neuroimaging-based disorder classification.
- This methodology offers a reliable framework for analyzing brain connectivity in neurological disorders.
