Partial directed coherence analysis of resting-state EEG signals for alcohol use disorder detection using machine learning.

Introduction: Excessive alcohol consumption negatively impacts physical and psychiatric health, lifestyle, and societal interactions. Chronic alcohol abuse alters brain structure, leading to alcohol use disorder (AUD), a condition requiring early diagnosis for effective management. Current diagnostic methods, primarily reliant on subjective questionnaires, could benefit from objective measures. Method: The study proposes a novel EEG-based classification approach, focusing on effective connectivity (EC) derived from resting-state EEG signals in combination with support vector machine (SVM) algorithms. EC estimation is performed using the partial directed coherence (PDC) technique. The analysis is conducted on an EEG dataset comprising 35 individuals with AUD and 35 healthy controls (HCs). The methodology evaluates the efficacy of connectivity features in distinguishing between AUD and HC and subsequently develops and assesses an EEG classification technique using EC matrices and SVM. Result: The proposed methodology demonstrated promising performance, achieving a peak accuracy of 94.5% and an area under the curve (AUC) of 0.988, specifically using frequency bands 29, 36, 45, 46, and 52. Additionally, feature reduction techniques applied to the PDC adjacency matrices in the gamma band further improved classification outcomes. The SVM-based classification achieved an accuracy of 96.37 ± 0.45%, showcasing enhanced performance through the utilization of reduced PDC adjacency matrices. Discussion: These results highlight the potential of the developed algorithm as a robust diagnostic tool for AUD detection, enhancing precision beyond subjective methods. Incorporating EC features derived from EEG signals can inform tailored treatment strategies, contributing to improved management of AUD.