Towards an Automated Detection of Alcohol Dependence Using EEG Spectral Power Estimates

Abstract

The study utilized spectral power estimates evaluated from the Electroencephalogram (EEG) of alcoholics and control participants to attempt an automatic detection of individuals suffering alcohol dependence. Power estimates were obtained for non-overlapping consecutive EEG segments of 0.5-second duration while using a 5th order Burg Autoregressive estimator. EEG power was averaged within δ (1-4 Hz), θ (4-8 Hz), α1 (8-10 Hz), α2 (10-12 Hz), β1 (12-20 Hz), β2 (20-30 Hz), γ1 (30-40 Hz), and γ2 (40-50 Hz) rhythms and used as features in the "k nearest neighbors" classifier. A leave-one-out cross-validation procedure was implemented to evaluate the classification performance. The highest classification accuracy was observed for power estimates for α1 and α2 EEG rhythms. Depending on the number of neighbors included into classification, Sensitivity of the classifier was ranging between 90.91% and 98.70%, while Specificity was between 91.11% and 95.56% for these rhythms. Compared to other reported classification approaches, present work utilizes simpler and more robust data analysis techniques that, perhaps, may be preferred for practical applications. We conclude that it is possible to detect (with reasonably high accuracy) the individuals, who suffer alcohol dependence by analyzing their EEG.

Keywords

• alcohol dependence
• autoregressive spectral estimates
• k-nn classifier
• eeg