Unconstrained deep learning-based sleep stage classification using cardiorespiratory and body movement activities in...

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Accurate sleep stage classification is crucial for diagnosing sleep disorders.
• Traditional polysomnography (PSG) is resource-intensive and not suitable for widespread home monitoring.
• Non-invasive methods using unobtrusive sensors are needed for scalable sleep analysis.

Purpose of the Study:

• To evaluate the feasibility of deep learning for unconstrained sleep stage classification.
• To utilize cardiorespiratory and body movement data from piezoelectric under-mattress sensors.
• To assess the performance of a bidirectional long short-term memory (LSTM) network for sleep staging.

Main Methods:

• Simultaneous acquisition of cardiorespiratory signals, variability, coupling, and body movement via PSG and under-mattress sensors.
• Development and application of a bidirectional LSTM network for predicting five sleep stages.
• Testing five different input feature combinations for model optimization.

Main Results:

• The best performing model, using six parameters including cardiorespiratory variability, achieved a balanced accuracy of 0.70 ± 0.05.
• Performance metrics included Cohen's κ of 0.40 ± 0.12 and an F1 score of 0.62 ± 0.08.
• Deming regression and Bland-Altman analyses revealed significant correlations (r = 0.426-0.695) and low bias between model estimates and PSG-determined sleep parameters.

Conclusions:

• Deep learning-based sleep stage classification using under-mattress sensors is feasible and effective.
• The proposed approach demonstrates potential for expanding in-home sleep monitoring capabilities.
• This technology could facilitate more accessible and widespread sleep disorder screening and management.