Methods for Snoring and Heart Pulse Detection Using Unobtrusive Mattress Sensor

In 2014, the lack of sleep was regarded as a public health epidemic by the Centers for Disease Control and Prevention. Lack of and poor-quality sleep has been related to obesity, diabetes, heart disease, dementia, and depression. Sleep disorders are commonly caused by sleep-disordered breathing (SDB) events. The most common events include complete obstructions (apneas), partial obstructions (hypopneas), snoring, and prolonged increased breathing efforts. The most well-known SBD is obstructive sleep apnea (OSA). In 2019, its prevalence was estimated to exceed 50% in some countries. However, patients suffering from other breathing disorders, such as prolonged partial obstruction (PPO), may suffer from similar symptoms but remain undiagnosed and untreated. Snoring and prolonged increased breathing efforts often co-occur with breathing disorders, and monitoring is essential as it might indicate the presence of SDB.

Nowadays, sleep diagnosis remains cumbersome, expensive, and time-consuming. EMFiT (Electromechanical Film Transducer) mattress sensors might bring new characteristics to decrease the cost and intrusiveness of these tests. In Finland, EMFiT supplements polysomnography (PSG) and cardiorespiratory polygraphy, facilitating clinical diagnoses in hospital and ambulatory settings. Additionally, these sensors can be used for continuous home monitoring, aiding in follow-up or preliminary screenings. This research focuses on understanding the EMFiT signal under different physiological processes. In particular, careful analysis, signal processing, and classification techniques were applied to the EMFiT signal to characterize and detect snoring, PPO, and EMFiT heart pulse (EHP).

The first investigation targeted snoring (SN). Spectral characterization of snoring epochs using the EMFiT mattress sensor was conducted and compared with normal breathing (NB). In addition, mono-sound source separation techniques were employed to separate snoring from breathing and heart signals, aiming to isolate repetitive SN events and obtain their spectral profile. Characterization results were used to train a support vector machine (SVM) and compared with state-of-the-art event detection algorithms using Deep Learning. The second investigation targeted the characterization and detection of EMFiT signal spikes, which are related to the increased respiratory effort present in PPO. The third investigation targeted isolating the heart signals from EMFiT, using a spectral source separation method to obtain a photopletysmography-like signal. The resulting signal, the EMFiT heart pulse (EP), was evaluated against an electrocardiogram (ECG). The study also visualized systolic and diastolic differences in the spectrogram. This study is significant in the sense that the signal appears the same independently of the subject's position.

In conclusion, this research presents methods to detect snoring using the EMFiT mattress as well as new findings. Second, it explores the time characteristics of PPO and proposed breathing asymmetry parameters. Finally, it defines a novel method to obtain the heart pulse independent of position and evaluate its usefulness in estimating heart rate.