Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

AS3: A framework for automatic sleep stage scoring

 
: Schlüter, T.; Kißels, T.; Conrad, S.

Palma dos Reis, A. ; International Association for Development of the Information Society -IADIS-:
IADIS International Conference Intelligent Systems and Agents 2010 and European Conference on Data Mining 2010. Proceedings : Part of the MCCSIS 2010, Freiburg, Germany, 28-31 July 2010
Lisbon: IADIS Press, 2010
ISBN: 978-972-8939-23-6
S.83-91
International Conference Intelligent Systems and Agents (ISA) <2010, Freiburg>
European Conference on Data Mining (DM) <2010, Freiburg>
Multi Conference on Computer Science and Information Systems (MCCSIS) <2010, Freiburg/Brsg.>
Englisch
Konferenzbeitrag
Fraunhofer IAIS ()

Abstract
The analysis of human sleep has attracted much attention during the last decades, due to its nature mostly in the area of medicine. Results of this research are applied all about, even as app on the iPhone (Sleep Cyle alarm clock). There are several challenging problems, which have not been solved adequately. One of them is sleep stage scoring, i.e the analysis of recorded biophysical signals from a patient during his sleep, with the aim of classifying all contained epochs (i.e. small time intervals) into predefined sleep stages. Sleep stage scoring is an important basis for the detection of sleep disturbances, for example for sleep apnea. Up to now, reliable sleep stage scoring is done mostly by experts, who have to classify every 30 second lasting epoch of sleep recordings (that contain usually 8 hours and more) by hand. In this paper we present AS3, a framework for Automatic Sleep Stage Scoring. By means of several combined techniques (i.e. Fourier transform, wavelet transform, Derivative Dynamic Time Warping [1] and waveform recognition [2]) AS3 extracts several features from biophysical signals recorded during sleep, like frequencies and patterns, and verifies certain criteria for sleep stage scoring. It uses machine learning techniques (decision tree induction and k-nearest neighbor classification) to create a classifier for the EEG data, which classifies every epoch into its sleep stage. We tested and evaluated AS3 on a large set of real life EEG, EOG and EMG data from PhysioBank [3], and achieved very good result (with an accuracy of more than 90%), which are presented and discussed in this paper as well.

: http://publica.fraunhofer.de/dokumente/N-188460.html