Schizophrenia detection via lobe-wise and overall EEG features using VMD and bayesian-optimized machine learning...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Schizophrenia (SCH) is a severe mental disorder causing cognitive and neurophysiological impairments.
• Early diagnosis of SCH is difficult due to symptom manifestation over time.
• Electroencephalography (EEG) offers a promising non-invasive method for detecting brain activity changes.

Purpose of the Study:

• To develop and validate a machine learning framework for early schizophrenia detection using EEG signals.
• To assess the efficacy of Variational Mode Decomposition (VMD) combined with optimized machine learning (OML) classifiers.

Main Methods:

• EEG data from MHRC and RepOD datasets were segmented and decomposed into Intrinsic Mode Functions (IMFs) using VMD.
• Multi-domain features were extracted from IMFs and classified using various ML and OML models.
• A subject-wise Leave-One-Out Cross-Validation (LOOCV) strategy was implemented to prevent data leakage.

Main Results:

• The VMD + OML framework achieved high accuracies: 96.7% (MHRC) and 99.0% (RepOD).
• Optimizable KNN and Optimizable Ensemble classifiers showed superior performance for the respective datasets.
• Lobe-wise analysis indicated strong discriminative power in frontal and temporal regions, consistent with SCH pathophysiology.

Conclusions:

• The proposed VMD + OML framework provides a computationally efficient and clinically interpretable approach for early SCH detection.
• EEG-based analysis with advanced machine learning holds significant potential for improving schizophrenia diagnosis.
• Findings support the role of frontal-temporal dysconnectivity in schizophrenia and its detectability via EEG.