Investigating White Matter Abnormalities Associated with Schizophrenia Using Deep Learning Model and Voxel-Based...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Psychiatry

Background:

• Schizophrenia (SCZ) presents significant diagnostic challenges.
• Accurate and timely diagnosis is critical for patient outcomes.
• Deep learning (DL) offers potential for improved SCZ prediction and diagnosis.

Purpose of the Study:

• To classify individuals with SCZ from healthy controls using structural MRI.
• To identify specific brain regions associated with SCZ abnormalities.
• To evaluate the efficacy of DL models in SCZ diagnosis.

Main Methods:

• Extracted Gray Matter (GM), Cerebrospinal Fluid (CSF), and White Matter (WM) from 99 MRI scans.
• Utilized a pretrained ResNet-50 network for feature extraction.
• Employed an ensemble deep random vector functional link (edRVFL) network for classification.
• Performed voxel-based morphometry (VBM) analysis to examine tissue volumes.

Main Results:

• The edRVFL model achieved a classification accuracy of 96.5% using WM features.
• edRVFL outperformed traditional algorithms in SCZ classification.
• VBM analysis revealed 1363 significant voxels in the WM region of SCZ patients (T-value=6.90, Z-value=6.21).

Conclusions:

• Deep learning, particularly the edRVFL model, demonstrates high accuracy in diagnosing SCZ from MRI data.
• White Matter (WM) alterations are strongly associated with structural changes in SCZ.
• DL models combined with VBM analysis can effectively identify SCZ-related brain abnormalities.