Using multivariate machine learning methods and structural MRI to classify childhood onset schizophrenia and healthy...

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Multivariate machine learning (ML) can classify schizophrenia patients and controls using structural magnetic resonance imaging (MRI).
• Previous ML applications have not been extended to clinical measures.
• This study hypothesized that brain measures would classify groups and relate to illness severity, developmental delays, and genetic risk.

Purpose of the Study:

• To classify childhood onset schizophrenia (COS) patients and healthy controls using structural MRI data.
• To investigate the relationship between brain-based probability of illness and clinical measures including illness severity, developmental delays, and genetic risk.

Main Methods:

• Utilized Random Forest (RF), a machine learning method, to classify 98 COS patients and 99 controls based on 74 brain MRI subregions.
• Estimated the probability of being classified as a patient using MRI measures.
• Explored associations between brain-based illness probability and symptoms, premorbid development, and copy number variation (CNV).

Main Results:

• Brain regions accurately classified COS patients and controls with 73.7% accuracy.
• Increased brain-based probability of illness correlated with worse functioning (p=0.0004) and fewer developmental delays (p=0.02).
• Copy number variation (CNV) presence was linked to a lower probability of being classified as schizophrenia (p=0.001).

Conclusions:

• Random Forest (RF) effectively classifies schizophrenia and control groups using anatomic brain MRI measures.
• Brain-based probability of illness shows a positive association with illness severity.
• Brain-based probability of illness demonstrates a negative relationship with developmental delays and copy number variation (CNV)-based risk.