Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

BP: Fachverband Biologische Physik

BP 11: Posters: Statistical Physics in Biological Systems (joint with DY)

BP 11.24: Poster

Montag, 11. März 2013, 17:30–19:30, Poster B2

Approach for automated sleep stage classification from spectral data — •Stephan Volkland and Jens Christian Claussen — INB, University of Lübeck, Germany

Manual scoring of sleep stages according to the Rechtschaffen-Kales rule catalogue (or the only simplified AAMS rules) is done on 30s epochs and requires extensive manual labor. Further the inter-rater and intra-rater reliabilities are not fully satisfactory for subsequent quantitative analysis. Hence it would be desired to develop and use automatic methods: attempts in this direction have been tried numerously but not paved their way to clinical practice as the EEG signatures of the sleep stages differ between subjects even in a qualitative way. Furthermore it would be desired to detect and resolve sub-stages as well as acheive a higher time resolution; both is unfeasible for manual scoring which would then be slower than real time. Here we refer and compare to an approach by [1] where three quantitative indexes are derived from EEG and EMG and broken down to 8 possible states. Our approach also is based on EMG and EEG, whereby we also use EOG, and use full information from five physiological EEG bands as input for unsupervised clustering (k-means) of data where clusters then are assigned to sleep stages [2]. We observe that the known difficulties to reliably distinguish S1, REM and wake stages persist. For the range between stages S2 and S4, we however can obtain a reliable interpolation between the sleep stages even for 16s or shorter time intervals.
[1] B. Müller, W.D. Gäbelein, H. Schulz, Sleep 29, 967 (2006).
[2] Stephan Volkland, BA thesis, INB, Lübeck (2012).