| Paper: | PS-2B.73 | ||
| Session: | Poster Session 2B | ||
| Location: | H Fläche 1.OG | ||
| Session Time: | Sunday, September 15, 17:15 - 20:15 | ||
| Presentation Time: | Sunday, September 15, 17:15 - 20:15 | ||
| Presentation: | Poster | ||
| Publication: | 2019 Conference on Cognitive Computational Neuroscience, 13-16 September 2019, Berlin, Germany | ||
| Paper Title: | Do sleep and anesthesia share common multifractal EEG dynamics? Insights from adversarial domain adaptation | ||
| Manuscript: | Click here to view manuscript | ||
| License: |  This work is licensed under a Creative Commons Attribution 3.0 Unported License. |
||
| DOI: | https://doi.org/10.32470/CCN.2019.1172-0 | ||
| Authors: | Louis Leconte, Ecole Normale Supérieure Paris-Saclay, France; Tarek Lajnef, Thomas Thiery, Université de Montréal, Canada; George Mashour, University of Michigan, United States; Stefanie Blain-Moraes, McGill University, Canada; Perrine Ruby, Université Claude Bernard Lyon 1, Université de Lyon, France; Karim Jerbi, Université de Montréal, Canada | ||
| Abstract: | The brain displays scale-free and multifractal dynamics that change across different states of consciousness. Whether the multifractal properties of EEG data change in a similar way when shifting from a conscious to an unconscious state compared to shifting from wakefulness to sleep is still largely unknown. To address this we ask a slightly different question: How well can we use a classifier trained on sleep EEG multifractality data to correctly discriminate conscious and unconscious states. To this end, we used a Domain Adversarial Neural Network (DANN) framework geared towards discriminating neural signals recorded during conscious vs unconscious states (target domain), based on classification of brain signals recorded during wakefulness vs sleep (source domain). We compare results obtained with naïve transfer learning (no domain adaptation), with supervised and unsupervised domain adaptation. The input data consisted of multifractal parameters computed from EEG recordings. This paper reports two important findings: First, our analyses provide evidence for the feasibility of creating a DANN architecture that can learn to discriminate consciousness from anesthetic-induced unconsciousness via adversarial adaptation of sleep/wakefulness discrimination. Second, by exploring the topographies of the successful classification rates across the EEG array, we were able to identify functional similarities of EEG multifractality patterns across sleep and anesthesia. | ||