Exploring Brain Structural and Functional Biomarkers in Schizophrenia via Brain-Network-Constrained Multi-View SCCA

Area of Science:

• Neuroimaging
• Psychiatric Disorders
• Computational Neuroscience

Background:

• Schizophrenia research increasingly uses dynamic functional magnetic resonance imaging (d-fMRI) measures like dynamic fractional amplitude of low-frequency fluctuation (d-fALFF) to understand brain dynamics.
• Unimodal imaging features are insufficient for capturing complex brain deficits in schizophrenia, necessitating integrated functional and structural analyses.
• Investigating the coupling between neural function and structure is crucial for identifying schizophrenia biomarkers and understanding its biological underpinnings.

Purpose of the Study:

• To propose and validate a novel brain-network-constrained multi-view sparse canonical correlation analysis (BN-MSCCA) method.
• To explore intrinsic associations between brain structure and dynamic brain function in schizophrenia.
• To identify interpretable and biologically meaningful biomarkers for schizophrenia.

Main Methods:

• Acquired dynamic fractional amplitude of low-frequency fluctuation (d-fALFF) using a sliding window method and gray matter maps via voxel-based morphometry.
• Extracted and selected region-of-interest (ROI)-based features using multi-view sparse canonical correlation analysis integrated with diagnosis information.
• Incorporated brain-network-based structural constraints to enhance biomarker interpretability.

Main Results:

• The BN-MSCCA method successfully identified critical ROIs with sparse canonical weights, corresponding to specific brain networks.
• The method demonstrated higher canonical correlation coefficients on testing data, indicating robust association identification.
• Biomarkers derived from BN-MSCCA improved schizophrenia classification accuracy by 5-8% and AUC by 6-10%.
• Two top biomarkers showed significant negative correlation with positive symptom scores on the PANSS.

Conclusions:

• The BN-MSCCA method effectively reveals associations between brain structure and dynamic function in schizophrenia.
• The identified biomarkers are biologically meaningful and show potential for clinical applications in diagnosis and understanding disease mechanisms.
• This approach enhances our comprehension of schizophrenia's neural basis by integrating multi-modal imaging data.