Repurposing electroencephalographic signal for automatic segmentation of intra-operative periods under general anesthesia

During general anesthesia, measuring the brain's response to the loss (LoC) and recovery of consciousness (RoC) provides valuable insights into cerebral fragility. However, accurately measuring these periods is challenging. Our study introduces an automatic method to identify LoC and RoC events from intraoperative electroencephalographic (EEG) recordings. Our dataset includes 289 intraoperative EEG from patients undergoing general anesthesia with propofol or sevoflurane. The LoC and RoC periods were labeled by a trained clinician. The EEG recordings were divided into 30-seconds epochs and analyzed using power spectral density as features. We employed a gradient-boosting algorithm with a 3-state label to classify states of unconsciousness. LoC and RoC were identified by fitting an activation function on the sequence of probabilities computed from the classifier. Our results show that the gradient-boosting algorithm has a test accuracy of 95.58% and a micro-average area under the curve (AUC) of 0.994. The LoC and RoC were predicted with a median error of 7.6 seconds [IQR: -76.3 to 56.8 sec] and -1.0 second [IQR: -45.4 to 41.0 sec], respectively. Our study demonstrates that the LoC and RoC during the perioperative period can be accurately identified using only intraoperative EEG recordings. This opens up new avenues for the study and the development of LoC/RoC-based digital biomarkers using large datasets.