Machine learning for prediction of schizophrenia based on identifying the primary and interaction effects of minor...

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Minor physical anomalies (MPAs) are recognized as potential neurodevelopmental markers for schizophrenia.
• Previous research has not explored the interactive effects of MPAs in schizophrenia risk assessment.
• Understanding these markers can enhance early detection and intervention strategies for schizophrenia.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting schizophrenia risk based on MPAs.
• To investigate the primary and interaction effects of specific MPAs in relation to schizophrenia.
• To compare the predictive performance of classical statistical models with machine learning approaches.

Main Methods:

• Utilized a dataset of 470 schizophrenia patients and 354 healthy controls.
• Employed Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF) models.
• Analyzed primary MPA measurements and their two- and three-item interactions.

Main Results:

• Random Forest (RF) demonstrated the highest predictive power, achieving an AUC of 0.97 (full training) and 0.75 (cross-validation).
• Identified significant primary MPAs including mouth region, high palate, furrowed tongue, skull height, and mouth width.
• Discovered significant two-item interactions (e.g., skull height and mouth width) and a weak three-item interaction (high palate, skull height, mouth width) associated with schizophrenia risk.

Conclusions:

• Machine learning models, particularly Random Forest, show strong predictive capabilities for schizophrenia risk assessment using MPAs.
• Both primary and interaction effects of MPAs are crucial indicators in predicting schizophrenia.
• The study highlights the potential of quantitative MPA analysis for identifying individuals at higher risk for schizophrenia.