Evaluation of the sleep quality based on bed sensor signals: Time-variant analysis

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Signal Processing

Background:

• Assessing sleep quality outside of clinical settings is challenging.
• Developing non-invasive methods for sleep macrostructure analysis is crucial.
• Current sleep monitoring often relies on polysomnography in sleep centers.

Purpose of the Study:

• To evaluate the feasibility of automatic sleep macrostructure detection using bed sensor signals.
• To compare different feature extraction and classification methodologies for sleep staging.
• To determine the accuracy of out-of-center sleep quality assessment.

Main Methods:

• Utilized 17 polysomnography recordings from healthy subjects.
• Extracted features using Time-Variant Autoregressive Modeling (TVAM) and Wavelet Decomposition (WD).
• Compared K-Nearest Neighbor (KNN) and Feed Forward Neural Networks (FFNN) classifiers for sleep stage classification (Wake, NREM, REM).

Main Results:

• TVAM-LD achieved 71.95 ± 7.47% accuracy and 0.42 ± 0.10 kappa index for Wake-NREM-REM classification.
• WD-FFNN achieved 67.17 ± 11.88% accuracy and 0.39 ± 0.13 kappa index.
• TVAM-LD demonstrated superior performance compared to WD-FFNN.

Conclusions:

• Automatic sleep quality assessment using bed sensors is feasible.
• TVAM-LD shows promising results for accurate sleep macrostructure detection.
• This approach could enable wider access to sleep quality monitoring and benefit more individuals.