25 to 30 percent of the population suffer from sleep pathologies, which affect the quality of life and have important consequences on people's behaviour, vigilance and health. The early detection of those pathologies is essential because sleep pathologies have strong economic impact in terms of direct (healthcare), indirect (morbidity, absenteeism) and linked costs (accidents due to drowsiness).
Polysomnographic recording (PSG) is the reference tool for the diagnosis of sleep disorders. Traditionally, the analysis of PSG recordings is done according to international criteria through a 1 hour 30 minutes visual inspection. Many automated PSG treatment procedures have been developed and are quite reliable for healthy subjects. On the other hand, the existing systems are badly adapted to pathological subjects and require several corrections on the analysis of the EEG, pneumologic and muscular signals. The inefficiency of those systems in the pathological case is due to the sleep signals' instability, the abundance of artifacts, the difficulty to apply classification criteria, and the insufficiency of the the pathological sleep's macrostructure analysis (sleep stage analysis), which needs to be assisted by a microstructure analysis. To our knowledge, the use of video for detecting movements remains infrequent. Though video cameras are potentially very interesting because they detect body movements exclusively, contrary to the other signals. The objective in view is to use this discriminant property of video to make easier the fussy task of detecting movements during PSG recordings.
Classification rules for determining the sleep stages were published by Rechtschaffen and Kales in 1968, and are still in application today. The great variability of the recorded sleep signals makes the automatic application of those rules difficult. The project aims at solving this problem by using modern artificial intelligence techniques to simulate human quotation as well as possible. Clustering methods, possibly combined with neuron networks or fuzzy logic techniques, seem capable of carrying out this objective. The preliminary treatment of the recorded sleep signals is an essential step
for the quality of the classification. To extract the most distinctive features of the events we wish to detect from the electric signals, we plan to use recent signal processing methods (the wavelet analysis in particular) and the experience gained by the TCTS Lab in speech recognition - the speech signal being very similar to PSG signals.
Our partners' experience allows us to consider a direct integration of this software into the current commercial environment. Consequently, this project implies developing a user-friendly graphical interface, in addition to integrating video sensors. Moreover, the final application has to handle data stored in a standard format, include algorithms manageable by office computers, and be usable in any sleep laboratory and independent of the measuring equipments and the data-processing platform in use.
- S. Brousmiche et al. 2004 "Analysis and classification of sleep EEG using wavelets and radial-basis functions networks", 2004.
- J. Doucet , M. Kerkhofs 2004 "Exploration du sommeil chez l’adulte et l’adolescent", article in press, EMC-Psychiatrie, 2004.
- S. devuyst et al. 2004 "Une analyse automatique du sommeil? Vous ne rêvez pas...", Polytech-News, journal interne de la Faculté Polytechnique de Mons, 12ème année, No 26, Février-mars-avril 2004.
- S. devuyst et al. 2004 "Artifacts processing for sleep stage classification", URSI forum broadband communications, pp 53, 10 decembre 2004.
- A. de Smedt 2004 "Vers une analyse automatique du sommeil", Athena 206, Decembre 2004.
- S. Devuyst et al. "Automatic Sleep Spindle Detection in Patients with Sleep Disorders", Proceedings of the 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2006), FrEP5.5, New York Aug 30-Sept 3, 2006.