AG Kommunikationstheorie


Automatic and unobtrusive identification of sleep and wake conditions based on multiparameter analysis


Reducing the personal burden and the societal cost associated with sleep disturbances has become one of the major challenges in the last decades. The system which is presented in this work addresses this issue and aims at automatically detecting wake and sleep stages using unobtrusive in-bed sensors, a textile ECG and a ferro-electret foil. Hence the core of this thesis is a pattern recognition task with two classes: "wake"' and "sleep"'. Data are first collected from healthy subjects and sleep lab patients. After preprocessing the ECG and the foil signals, 33 features are extracted. Their discriminatory capabilities are tested using statistical hypothesis tests and feature subset selection algorithms: they reveal which features are indeed relevant for healthy subjects on the one hand, and for sleep lab patients on the other hand. The best feature subsets identified, two classifier designs which seem to be relevant for our classification task are presented and evaluated: the Bayesian and the SVM. In each case, the classifier parameters are tuned regarding either accuracy, sensitivity and specificity, or sleep efficiency accuracy. The performances of the classifiers are also assessed in particular cases, e.g. when one of the two signals is missing. After analysis of the results, some suggestions to improve the performances are made.

zurück zur Terminübersicht