Exploring the white matter disruptions for Schizophrenia based on convolutional ensemble kernel randomized network.

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Computational Neuroscience

Background:

• Schizophrenia (SZ) presents with cognitive deficits and structural brain abnormalities.
• Convolutional Neural Networks (CNNs) offer potential for identifying complex brain alterations.
• Structural Magnetic Resonance Imaging (sMRI) detects disruptions in white matter (WM), grey matter (GM), and cerebrospinal fluid (CSF).

Purpose of the Study:

• To develop and evaluate a CNN ensemble KRR-RVFL model for detecting WM disruptions in SZ.
• To compare the model's performance across different brain tissue types (WM, GM, CSF).
• To investigate the relationship between tissue volume and symptom severity in SZ.

Main Methods:

• An eight-layer CNN was integrated with five Kernel Ridge Regression-based Random Vector Functional Link (KRR-RVFL) classifiers.
• Ensemble averaging of classifier outputs was used for final classification.
• Correlation analysis was performed between tissue volumes and symptom scales.

Main Results:

• The proposed CNN ensemble KRR-RVFL achieved 97.33% accuracy in identifying WM disruptions.
• WM tissue volume showed greater reduction than GM in individuals with SZ.
• Significant correlations were found between tissue volumes and symptom severity.

Conclusions:

• The developed deep learning model effectively identifies WM disruptions associated with SZ.
• WM integrity is significantly compromised in SZ, more so than GM.
• This approach aids clinicians in the diagnosis of SZ by highlighting the role of WM alterations.