Analysis of anaesthesia EEG recordings with machine learning techniques: prediction of postoperative delirium

Details

Supervisors

Momeni, Mona ; Verleysen, Michel

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique, à finalité spécialisée

Abstract

Nowadays, Postoperative Delirium is a disorder that affects a large number of people after a surgical operation under general anesthesia. Over the years, evidence seems to suggest that anesthesia is closely related to Postoperative Delirium. Among several findings, one under study shows a link between the frequency space of electroencephalographic signals and postoperative delirium, in particular the frequencies corresponding to the alpha rhythm. The goal of this master thesis is to find Machine Learning models capable of determining the prevalence of a patient to develop Postoperative Delirium. In addition to the alpha frequency band, the beta frequency band proves to be important in the predictive ability of the models for our data set. Among the models and the reconstructed datasets tested, the best model was the Support Vector Machine on a dataset with 13 electrodes located in the front of the brain. It achieves a Fbeta score (beta = 1.5) of 0.70 with a 95\% CI of [0.49, 0.72], a recall of 0.85, a precision of 0.50, a AUC score of 0.75 and a specificity of 0.64.