Drug induced hypovigilance identification by means of an EEG Fractal Dimension Classifier

Abstract

Event Abstract Back to Event Drug induced hypovigilance identification by means of an EEG Fractal Dimension Classifier Emmanouil Michail1*, Ioanna Chouvarda1 and Nicos Maglaveras1 1 Magdeburg University, Germany Aim: This work aims at identifying drug induced hypovigilance caused by lorazepam administration, a psychoactive drug with hypnotic effects (1). The main objective is to classify subjects who have taken 2.5 mg of lorazepam or placebo based on the brain activity changes that drug administration induces. Methods: EEG data were recorded from 14 healthy male subjects aged between 18 and 40 years. For each subject two different conditions were examined: lorazepam 2.5 mg single administration (verum state) and placebo administration (placebo state) early in the morning. For each of these two conditions EEG was recorded with eyes closed during 3 minutes in resting condition at a sampling frequency of 256Hz. 20 EEG channels were recorded at different times during the first and the second day of the experiment. After artefact removal and filtering, a single Fractal Dimension value was extracted for each 3-minute recording. Higuchi’s method (2) was used, with maximum time interval kmax=15. A linear classifier was applied to FD values of the same time period and EEG channel, in order to automatically classify between placebo and verum state. “Leave one out” cross validation approach was adopted. Results: In almost all subjects, FD has a higher value in verum case than in placebo case at the first 8 hours after drug intake. Some leftover effects are still noticeable 12 hrs after drug administration but not during next morning. Best discriminant results occur 2 hours after drug administration. At that time, classifier reaches 100% accuracy for channels F3, Fz and P4. At the same time, mean specificity over all channels was 97.5% and maximum specificity was 100% whereas, mean sensitivity over all channels was 85% and maximum sensitivity was 92.8571%. Conclusion: The practical application of these findings would be an automatic system able to detect if a subject is under the influence of psychoactive drugs that cause sleepiness. Acknowledgments: Special thanks to the research group of FORENAP for making available the dataset used in this study.