Non-Intrusive Sleep Staging with Integration of Ballistocardiography and Audio Signals.

This paper investigates the use of audio analysis alongside non-intrusive sensors to measure cardiorespiratory parameters for sleep stage classification. A ballistocardiogram (BCG) detected cardiorespiratory signals through vibrations on the mattress, while an Apple Watch Ultra 2 used optical sensors on the wrist for measurement. The cardiorespiratory signals were processed with non-linear methods and autocorrelation functions to create a dataset from the entire system, which was then trained in a long short-term memory (LSTM) model to classify sleep stages. The output was validated using a polysomnography (PSG) system, showing an agreement of 81% for REM and non-REM sleep stages. Good results were also obtained with audio and BCG, with approximately 76% sleep stage agreement, demonstrating a fully non-contact environment viable for sleep detection. Furthermore, incorporating audio features proved effective in evaluating sleep stages with a non-intrusive setup, ensuring minimum disturbance to participants' natural sleep cycle.Clinical Relevance-Evaluating sleep quality is an inconvenient process for patients due to the extensive setup and discomfort of a clinical PSG system. A more convenient, minimal contact system that effectively assesses sleep quality without disrupting the patient's natural sleep could provide clinicians with measurements that better reflect real-world sleep patterns. This would enable more accurate diagnoses of sleep disorders without inducing symptoms associated with the intensive clinical study experience.