Detecting Abnormal Brain Regions in Schizophrenia Using Structural MRI via Machine Learning.

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatric Disorders

Background:

• Schizophrenia diagnosis often relies on clinical observation, lacking objective biomarkers.
• Neuroimaging and machine learning (ML) offer potential for objective differentiation of schizophrenia (SZ) patients from normal controls (NCs).
• Previous ML approaches often focused on group-level differences, limiting individual diagnostic application.

Purpose of the Study:

• To develop and validate a novel ML framework for accurate individual-level diagnosis of schizophrenia using structural magnetic resonance images (sMRIs).
• To identify brain regions (gray matter and white matter) associated with schizophrenia.
• To establish a robust method for distinguishing SZ patients from NCs with high accuracy.

Main Methods:

• A coarse-to-fine feature selection ML framework was proposed.
• Two-sample t-tests identified initial group differences in sMRIs.
• Recursive feature elimination (RFE) removed irrelevant/redundant features.
• Support vector machine (SVM) classified patients using selected gray matter (GM) and white matter (WM) features.

Main Results:

• The proposed ML framework successfully distinguished SZ patients from NCs.
• The highest classification accuracy achieved was over 85%.
• Identified neuroimaging biomarkers are consistent with existing literature.

Conclusions:

• The developed ML framework provides an effective tool for individual-level schizophrenia diagnosis.
• This approach offers higher recognition rates compared to previous methods.
• The framework's universality allows for potential extension to diagnosing other neurological and psychiatric disorders.