Identifying schizophrenia subgroups using clustering and supervised learning.

Area of Science:

• Neuroscience
• Psychiatry
• Radiology

Background:

• Schizophrenia affects 1% globally, presenting heterogeneous positive, negative, and cognitive symptoms.
• Symptom severity and brain changes in schizophrenia are linked to prognosis.
• Individual differences in schizophrenia necessitate tailored prediction methods.

Purpose of the Study:

• To develop and validate a machine learning approach for predicting individual schizophrenia symptom profiles.
• To classify schizophrenia patients into distinct clinical subgroups based on neuroanatomical data.
• To enhance understanding of the relationship between brain structure and symptom heterogeneity.

Main Methods:

• Utilized symptomatic and magnetic resonance imaging (MRI) data from 167 schizophrenia subjects.
• Employed hierarchical clustering to define three clinical subgroups: high, positive-dominant, and mild symptom burden.
• Applied machine learning models (logistic regression, support vector machine, random forest) using cortical thickness and demographic data for subgroup classification.

Main Results:

• Random forest model achieved high accuracy (AUC: 0.81 for high symptom burden, 0.78 for mild symptom burden) in predicting subgroup membership.
• Classification performance surpassed baseline comparisons with the general schizophrenia population versus controls (AUC: 0.75).
• Feature importance analysis revealed consistencies with known regional brain impairments associated with schizophrenia symptoms.

Conclusions:

• Machine learning models, particularly random forest, can effectively predict schizophrenia clinical subgroups using MRI-derived neuroanatomical measures.
• This approach offers a promising tool for personalized prognosis and treatment stratification in schizophrenia.
• Findings underscore the link between specific neuroanatomical alterations and distinct symptom profiles in schizophrenia.