Sleep state classification is essential for managing and comprehending sleep patterns, and it is usually the first step in identifying sleep disorders. Polysomnography (PSG), the gold standard, is intrusive and inconvenient for regular/long-term sleep monitoring. Many sleep-monitoring techniques have recently seen a resurgence as a result of the rise of neural networks and advanced computing. Ballistocardiography (BCG) is an example of such a technique , in which vitals are monitored in a contactless and unobtrusive manner by measuring the body's reaction to cardiac ejection forces. A Multi-Headed Deep Neural Network is proposed in this study to accurately classify sleep-wake state and predict sleep-wake time using BCG sensors. This method achieves a 95.5% sleep-wake classification score. Two studies were conducted in a controlled and uncontrolled environment to assess the accuracy of sleep-awake time prediction. Sleep-awake time prediction achieved an accuracy score of 94.16% in a controlled environment on 115 subjects and 94.90% in an uncontrolled environment on 350 subjects. The high accuracy and contactless nature make this proposed system a convenient method for long-term monitoring of sleep states, and it may also aid in identifying sleep stages and other sleep-related disorders.
Clinical Relevance
Current sleep-wake state classification methods, such as actigraphy and polysomnography, necessitate patient contact and a high level of patient compliance. The proposed BCG method was found to be comparable to the gold standard PSG and most wearable actigraphy techniques, and also represents an effective method of contactless sleep monitoring. As a result, clinicians can use it to easily screen for sleep disorders such as dyssomnia and sleep apnea, even from the comfort of one's own home.
Introduction
Sleep-state detection refers to identifying sleep and non- sleep episodes for a subject. Estimating sleep states is an initial and primary step in analysing and addressing more serious sleep disorders. In the general population, the majority of sleep disorders go undiagnosed. [1]. Many of these sleep disorders are detectable through sleep state estimation [2]. Although polysomnography (PSG) is the current gold standard in high-resolution sleep monitoring, it is obtrusive, costly, and requires recording in an unfamiliar and controlled setting [3]. Individuals must spend the night in a sleep laboratory, which is a controlled environment where they are constantly monitored by a sleep technician. For decades, techniques for generating fresh solutions for effective sleep-state estimation have been a primary emphasis in sleep medicine [4]. Researchers have begun investigating wearables for detecting sleep-state with the rise of computational efficiencies [5]. Though these devices are simpler to operate than a PSG, [6] they are susceptible to errors and lack essential confirmation, such as device settings standardisation. In addition, they need to be worn all the time, which can be uncomfortable, making continuous monitoring difficult.
One such promising technique is ballistocardiography (BCG). While it was originally intended for recording cardiac and cardiovascular-related mechanical motions, the phenomenon can also detect any other physiological functions that produce a motion – including breathing, snoring and limb movements (4). In addition, it works in a contactless manner provided there is a medium to carry the vibrations, for e.g., if a person is lying on a mattress and the BCG data is recorded from underneath it. Despite its benefits, BCG's practical adoption has been hampered by two major challenges: (a) it captures other forces produced by the human body, such as movements, respiration, and snoring, which affect the accuracy and detection rate of cardiac and respiratory measurements, and (b) the measurements can vary depending on the setup and placement of the sensors, as well as body postures. Turtle Shell Technologies Pvt. Ltd. has developed a BCG based Dozee device for monitoring of vitals. The Dozee device has addressed the two major challenges mentioned above, with a novel unsupervised clustering algorithm [5,6] that is effectively able to isolate cardiac contractions and respiratory events from the unconstrained BCG signal. Dozee is designed to be a remote patient monitoring system for vitals and sleep monitoring. The data collected can be accessed by the patient/clinician/caretaker/user via the mobile application or a web-based platform (Figure 1).
Non-contact ballistocardiography (BCG) is a non-invasive method of assessing cardiovascular functions. Ballistocardiography, unlike PSG, does not require the attachment of external electrodes and or direct contact with a subject, thereby avoiding patient discomfort. This system is ideal for discrete long-term continuous data monitoring [7,8]. It can also be used in almost any sleep setting (including the subject's own bed at home), not just a hospital sleep lab. This paper proposes an efficient and contactless method based on BCG and deep learning to monitor sleep states. The sleep-awake time is calculated using a multi-head 1D-CNN architecture and a prediction algorithm for sleep state classification. Two independent studies were conducted in a controlled and an uncontrolled setting to assess the accuracy of sleep and wake-up time prediction.
