Temporal Dynamic Synchronous Functional Brain Network for Schizophrenia Classification and Lateralization Analysis
IEEE Transactions on Medical Imaging
|June 25, 2024
Summary
This study introduces a novel deep learning model, Temporal-BCGCN, for analyzing brain activity in schizophrenia (SZ) using resting-state fMRI. The model reveals significant left-hemisphere dysfunction in SZ patients, particularly in perceptual and higher-order networks.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Artificial Intelligence in Medicine
Background:
- Dynamic functional connectivity in resting-state fMRI (rs-fMRI) offers insights into time-varying brain activity abnormalities.
- Schizophrenia (SZ) is associated with complex disruptions in brain network mechanisms.
- Existing methods may not fully capture the dynamic nature of these abnormalities.
Purpose of the Study:
- To develop and validate an advanced dynamic brain network analysis model for schizophrenia detection using rs-fMRI.
- To investigate hemispheric lateralization of brain dysfunction in schizophrenia.
- To introduce novel deep learning components for dynamic graph convolutional networks and pooling.
Main Methods:
- Developed the Temporal Brain Category Graph Convolutional Network (Temporal-BCGCN) model.
- Introduced a dynamic synchronization feature extraction module (DSF-BrainNet) and a novel graph convolution method (TemporalConv).
- Proposed a modular test tool (CategoryPool) for analyzing hemispherical lateralization in deep learning models.
Main Results:
- Achieved high classification accuracies (83.62% on COBRE, 89.71% on UCLA datasets), outperforming baseline and state-of-the-art methods.
- Ablation studies confirmed the superiority of TemporalConv and CategoryPool over traditional approaches.
- Identified more severe dysfunction in left-hemisphere lower-order perceptual and higher-order network regions in SZ patients compared to the right hemisphere.
Conclusions:
- The Temporal-BCGCN model effectively captures dynamic brain network abnormalities in schizophrenia using rs-fMRI.
- The study highlights significant left-hemisphere lateralization of dysfunction in schizophrenia, emphasizing the role of the medial superior frontal gyrus.
- The developed deep learning tools offer promising avenues for future research and clinical applications in psychiatric disorders.
