Diagnosis of Schizophrenia: A Comprehensive Evaluation

Area of Science:

• Neuroimaging
• Computational Psychiatry
• Machine Learning

Background:

• Machine learning models are increasingly used for Schizophrenia diagnosis.
• The influence of specific classification models and feature selection techniques on diagnostic accuracy remains underexplored.

Purpose of the Study:

• To evaluate the performance of various classification models and feature selection techniques for Schizophrenia diagnosis using structural MRI data.
• To determine the optimal combination of models and feature selection for improved diagnostic accuracy.

Main Methods:

• Utilized structural magnetic resonance imaging (MRI) data from 72 Schizophrenia patients and 74 healthy controls.
• Evaluated classification algorithms including Support Vector Machine (SVM), random forest, kernel ridge regression, and randomized neural networks.
• Assessed feature selection techniques such as T-Test, ROC, Wilcoxon, entropy, Bhattacharyya, MRMR, and NCA.

Main Results:

• Support Vector Machine (SVM) models with a Gaussian kernel demonstrated superior performance over other classification models.
• Wilcoxon feature selection emerged as the most effective feature selection method.
• Integrating both grey matter and white matter data yielded better diagnostic performance than using either modality individually.

Conclusions:

• The choice of classification algorithms and feature selection techniques significantly impacts the accuracy of Schizophrenia diagnosis.
• Optimizing feature selection and model choice can enhance diagnostic capabilities for Schizophrenia using neuroimaging data.