An EEG-based machine learning method to screen alcohol use disorder.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Screening for alcohol use disorder (AUD) is often subjective and lacks objective measures.
• There is a need for automated and reliable methods to identify individuals with AUD.

Purpose of the Study:

• To develop and validate a machine learning model for objective AUD screening.
• To utilize resting-state electroencephalography (EEG)-derived features for classifying AUD patients and healthy controls.

Main Methods:

• Resting-state EEG data were recorded from 30 AUD patients and 15 healthy controls under eyes-closed and eyes-open conditions.
• Inter-hemispheric coherences and spectral power in delta, theta, alpha, beta, and gamma bands were computed from 19 scalp locations.
• A rank-based feature selection method using receiver operating characteristic curves identified the most discriminant EEG features.

Main Results:

• Inter-hemispheric coherences showed significant differences between groups, yielding high classification efficiency (Accuracy: 80.8%).
• Spectral power features provided significant classification results (Accuracy: 86.6%).
• Integrating both feature types achieved the highest classification performance (Accuracy: 89.3%, Sensitivity: 88.5%, Specificity: 91%, F-Measure: 0.90).

Conclusions:

• EEG-derived features, particularly theta, beta, and gamma band power along with inter-hemispheric coherence, can serve as objective markers for AUD screening.
• The proposed machine learning approach demonstrates potential for automating the screening of AUD patients.
• Objective EEG markers can significantly improve the diagnostic process for alcohol use disorder.