Ensemble learning techniques reveals multidimensional EEG feature alterations in pediatric schizophrenia.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Schizophrenia (SCZ) diagnosis is challenging due to reliance on subjective assessments.
• Objective diagnostic tools for pediatric SCZ are critically needed.
• Understanding neural alterations in pediatric SCZ is crucial for early intervention.

Purpose of the Study:

• To develop a machine learning framework for auxiliary diagnosis of pediatric SCZ.
• To utilize multi-dimensional electroencephalogram (EEG) features for diagnosis.
• To investigate underlying neural alterations in pediatric SCZ patients.

Main Methods:

• Collected resting-state EEG data from pediatric SCZ patients and healthy controls.
• Extracted relative power (RP), fuzzy entropy (FuzEn), and functional connectivity (FC) EEG features.
• Employed ensemble learning models and Recursive Feature Elimination for feature selection and classification.

Main Results:

• Identified 212 highly discriminative EEG features (RP, FuzEn, FC).
• Achieved 99.60% classification accuracy using the Categorical Boosting model with selected features.
• Revealed altered EEG patterns, including fronto-parietal FC changes and reduced FuzEn in pediatric SCZ.

Conclusions:

• Machine learning with multi-dimensional EEG features shows high potential for pediatric SCZ diagnosis.
• Identified specific neural alterations (FC, RP, FuzEn) provide insights into SCZ pathophysiology.
• The developed framework can aid in the development of automatic diagnostic systems for pediatric SCZ.