Abstract
INTRODUCTION
Diagnosing the level of consciousness in patients suffering from severe brain lesions is still a major challenge. EEG-based systems can help discriminate conscious from unconscious patients. This study aims to confront the results from two of the most reliable methods: the Perturbational Complexity Index (PCI) which is based on Transcranial Magnetic Stimulation (TMS-EEG), and a recent machine learning approach using EEG-extracted markers from a standardized oddball auditory stimulation paradigm (EEG-ERP).
METHODS
Patients presenting either an unresponsive wakefulness syndrome (UWS), a minimally conscious state (MCS) or an emergence of MCS (EMCS) underwent both TMS-EEG and EEG-ERP. We computed PCI value by compressing the spatiotemporal pattern of cortical responses to the perturbation of the cortex with TMS. For EEG-ERP, we extracted 60 markers corresponding to quantification of power spectrum and complexity in individual EEG sensors and information sharing between them. Using machine-learning, we predicted the individual probability of being (minimally) conscious.
RESULTS
PCI and EEG markers, when considered categorically (i.e. UWS vs MCS), were consistent for all UWS and EMCS patients, whereas the results for MCS patients showed less consistency. Nevertheless, we found a significant correlation between PCI values and the probability of being conscious with the multivariate classifier.
CONCLUSION
PCI correlated positively with the combination of EEG markers in severely brain-injured patients. These findings imply that EEG signatures of consciousness can be reliably extracted from different contexts and combined into coherent predictive models, encouraging future efforts in large-scale data-driven clinical neuroscience.
Peer reviewed