Perioperative electroencephalography to predict postoperative delirium (POD)

Details

Supervisors

Mouraux, André ; Standaert, François-Xavier

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil biomédical, à finalité spécialisée

Abstract

Postoperative delirium (or POD) is a potentially life-threatening neurocognitive disorder that regularly occurs after sternum opening heart surgery. The study of EEG signals, biological signals from the patient's scalp during surgery will be investigated. Indeed, according to the literature, in the intraoperative frontal EEG, there are significant differences between the spectral power of the alpha-band between 2 groups of different cognitive natures. Another study underlines the differences in the temporal evolution of emergence trajectories with hypnograms, marked by the power of the frequency subbands through time between two groups, healthy and delirious. For our thesis, we will, thanks to our set of 32 intraoperative EEG channels covering the whole scalp, try to classify our patients, but also to compare the results of the literature and their frontal EEG with our results including the study of EEGs in the frequency and time-frequency domain, from EEGs covering the whole scalp. The goal is to determine whether extensions of the EEG montage (i.e. parietal and occipital electrodes) reveal important attributes not captured by the studies. In this thesis, we have tried, thanks to features extracted in the Wavelet space (frequency-time domain) and in the frequency space with the Welch method, to classify our data from a single measurement per patient, with a set of 116 patients, and compare the results for frontal and 32-channels EEG. Using statistical and learning tools to obtain information, we implemented a classification model to know if this patient will develop delirium or not after the operation, and thus take better care of him and adapt his treatment. The result of our study shows that the addition of information on the occipital and parietal area of the EEG gives us better classification results than with a simple frontal EEG for frequency features, and that our results remain consistent with the literature, having common features. However, the results for Wavelet features give worse classification metrics than frequency features, as well as the non-significant difference in performance between frontal EEG and 32-channel EEG classification.