Optimization of Time-Variant Autoregressive Models for tracking REM - non REM transitions during sleep | IEEE Conference Publication | IEEE Xplore

Optimization of Time-Variant Autoregressive Models for tracking REM - non REM transitions during sleep


Abstract:

The aim of this study was the optimization of Time-Variant Autoregressive Models (TVAM) for tracking REM - non REM transitions during sleep, through the analysis of spect...Show More

Abstract:

The aim of this study was the optimization of Time-Variant Autoregressive Models (TVAM) for tracking REM - non REM transitions during sleep, through the analysis of spectral indexes extracted from tachograms. A first improvement of TVAM was achieved by choosing the best typology of forgetting factor in the analysis of a tachogram obtained during a sitting-to-standing test; then, a method for improving robustness of AR recursive identification with respect to outliers was selected by analyzing a tachogram with an ectopic beat. A variable forgetting factor according to the Fortescue method and a specific condition on the prediction error for recursive AR identification gave the best performances. The optimized TVAM was then employed in the analysis of tachograms derived from ECGs recorded during a whole night, through a sensorized T-shirt, from 9 healthy subjects. The spectral indexes (power of tachogram in the LF and HF bands, LF/HF ratio and the absolute value of the spectrum pole in the HF band) were computed from the estimated AR parameters on a beat-to-beat basis. A two groups T-test aimed at comparing values assumed by each spectral index in REM and non-REM sleep epochs was performed. Significant statistical differences (p-value <; 0.05) were found in three of the four spectral indexes computed. In conclusion, the combination of the Fortescue variant and of the robustness method based on the prediction error in the TVAM seems to be helpful in the differentiation between REM and non-REM sleep stages.
Date of Conference: 28 August 2012 - 01 September 2012
Date Added to IEEE Xplore: 10 November 2012
ISBN Information:

ISSN Information:

PubMed ID: 23366368
Conference Location: San Diego, CA, USA

Contact IEEE to Subscribe

References

References is not available for this document.