Machine learning algorithms and artificial neural networks for predicting schizophrenia using orbital parameters.

Area of Science:

• Neuroimaging
• Artificial Intelligence in Psychiatry
• Medical Diagnostics

Background:

• Schizophrenia is a complex mental illness with unclear origins.
• Current diagnostic methods can be lengthy and subjective.
• Investigating novel biomarkers for schizophrenia is crucial.

Purpose of the Study:

• To explore the potential of using computed tomography (CT) scans of the orbit to diagnose schizophrenia.
• To apply artificial neural networks (ANNs) and machine learning (ML) algorithms for schizophrenia detection based on orbital morphometry.

Main Methods:

• Retrospective analysis of CT scans from 180 individuals (90 healthy, 90 with schizophrenia).
• Measurement of various orbital and skull parameters, including orbital width, aperture area, and optic nerve sheath width.
• Application of ML algorithms (e.g., Extra Tree Classifier) and ANNs (e.g., Multilayer Perceptron Classifier) for classification.

Main Results:

• Significant differences in orbital and skull measurements were found between healthy individuals and those with schizophrenia.
• The Extra Tree Classifier achieved the highest accuracy (0.78), and the Multilayer Perceptron Classifier achieved 0.75.
• Left orbital width was identified as the most influential feature for diagnosis using the SHAP analyzer.

Conclusions:

• AI models analyzing orbital morphometry show potential as diagnostic tools for schizophrenia.
• Orbital measurements derived from CT scans can serve as objective biomarkers for schizophrenia.
• This study highlights the growing role of machine learning in psychiatric diagnostics.