Vital signs reflect the most critical medical signs that imply the condition of the body’s vital functions or life-sustaining activities [1]. Clinically, vital signs can be used as indicators to judge or evaluate an individual’s health status. For example, vital signs usually include heart rate, respiratory rate, and body temperature.

leep-disordered breathing (SDB) is defined as a group of disorders characterized by repeated episodes of upper airway narrowing or obstruction during sleep [2]. It includes conditions of abnormal respiration or respiratory disturbance during sleep, ranging from simple snoring to obstructive sleep apnea (OSA). SDB is relatively common, and its prevalence tends to increase with age or weight gain [3]. There are various pathophysiological mechanisms related to symptoms and complications of SDB, including intermittent hypoxia, hypercapnia, intrathoracic pressure swings, hyperactivity of the sympathetic nervous system, and sleep fragmentation [4]. SDB may induce numerous symptoms such as excessive daytime sleepiness, loud or habitual snoring, pauses in respiration during sleep, morning headaches, poor concentration, and decreased quality of life [5]. If SDB is neglected or persists, serious complications, including hypertension, elevated insulin resistance, coronary artery disease, arrhythmia, and stroke, can occur [6]. Thus, it is very necessary and important to monitor vital signs in patients with SDB during sleep.

Conventionally, vital signs are measured through direct contact. However, with recent developments in medical science and technology, diverse non-contact or contactless methods have been developed to measure or monitor vital signs. Among these methods, ballistocardiography is a useful technique for detecting vital signs. Ballistocardiography is a technology that graphically represents the repeated movements of the human body caused by the sudden ejection of blood into large blood vessels with each heartbeat [7]. The mechanisms of ballistocardiography in the detection of body motion caused by cardiac activity are as follows [8]: 1) the heart beats and induces blood entry into the aortic vessels. 2) The blood that has pushed into the aortic vessels reacts to the blood flow, resulting in dilation and contraction. 3) This leads to weak, cyclic, involuntary body movements that can be measured by ballistocardiography. In addition, ballistocardiography can detect body motion caused by respiratory activity based on similar mechanisms. Moreover, sleep architecture can be analyzed by making an algorithm using cardiac and respiratory activity measured by ballistocardiography [9].

Many studies have been conducted to measure vital signs such as heart rate and respiratory rate using ballistocardiography in a non-contact manner [8,10,11]. However, studies that analyzed sleep architecture using ballistocardiography or studies that measured vital signs using a non-contact method for SDB patients are insufficient. Therefore, the aim of this study was to evaluate the accuracy of a contactless monitoring device for 1) heart rate, respiratory rate, and body temperature during sleep and 2) sleep architecture in adults with SDB.

A total of 30 subjects (mean age = 46.43 ± 12.9 years; 22 men and 8 women; BMI [kg/m 2 ] = 26.26 ± 3.37) with SDB were finally included in this study. Five subjects were excluded from this analysis due to inadequate data. The polysomnographic data from three subjects was excluded because breathing movement and respiratory signals could not be detected due to patient movements, and two patients were excluded because ballistocardiography data loss occurred in some period due to wireless communication failure and data quality issues. There were 24 patients with OSA (mild = 11, moderate = 6, and severe = 7) and 6 patients with simple snoring. The sleep scoring data, including arousal events, of all subjects were as follows; sleep efficiency (%), stage N1, N2, N3, R (% of total sleep time), and arousal index (events/hour of total sleep time) values: 85.26 ± 9.15, 18.68 ± 9.46, 56.41 ± 13.68, 2.21 ± 4.73, 17.12 ± 5.17, and 27.18 ± 13.06, respectively. Respiratory events, including snoring, of all subjects were as follows; apnea-hypopnea index (events/hour of total sleep time), oxygen desaturation index (events/hour of total sleep time), average arterial oxygen saturation (%), minimum arterial oxygen saturation (%), and snoring (%): 18.81 ± 15.53, 17.82 ± 14.81, 95.2 ± 1.57, 83.67 ± 6.74, and 44.52 ± 22.4, respectively. Baseline data are presented in Table 1.

This prospective validation study was conducted to investigate the agreement in vital signs and sleep stages between non-contact monitoring devices using ballistocardiography and contact measurements, including polysomnography. The results of the study indicated that there was good (almost perfect) agreement in heart rate, respiratory rate, and body temperature between contactless and contact measurements. In addition, moderate agreement in sleep architecture was found between contactless and contact measurements.

Heart rate or pulse rate is the number of heartbeats per unit of time and is generally presented as beats per minute (bpm). Heart rate can vary depending on age, physical activity or fitness level, psychological status, medications, and disease. In adults, the normal range of resting heart rate is 60–100 bpm. An unnaturally slow heart rate (less than 60 bpm) is called bradycardia and may be associated with various conditions such as athletic lifestyles, older ages, drug use, and heart disorders. On the other hand, an abnormally fast heart rate (> 100 bpm) is called tarchycardia and may be related to diverse conditions, including normal responses (due to pain, fever, and stress), substances, medications, and medical illness (e.g., anemia, hyperthyroidism, cardiopulmonary disease, sepsis, and so on). Heart rate is the most central component in ballistocardiograms, which have become more accurate than when first introduced in several studies [15,16]. Xie et al. [15] compared beat-to-beat heart rate detection from ballistocardiograms for smart home applications using ECG. Their results showed an almost perfect detection of 99% for beat-to-beat intervals. Albukhari et al. [16] also used a bed-embedded ballistocardiographic device to detect heart rate, and the overall accuracy was 83.9%. When patients laid on their left sides, it showed a higher accuracy of 91%.

Respiratory rate is the number of breaths (inspiration and expiration) per unit of time and is usually recorded as respirations per minute. Respiratory rate can change depending on age, exercise, emotional state, drugs, and medical conditions. The normal range of respiratory rate in adults is, on average, 16–20 breaths per minute. Abnormal respiratory rates such as bradypnea and tarchypnea may be clinically caused by pain, anxiety, stress, shock, medications, and medical conditions (e.g., cerebral lesions, pulmonary heart disease, neuromuscular disorders, and so on). Sadek et al. [17] used a microbend fiber optic sensor to collect ballistocardiogram data to monitor sleep apnea and vital signs. To quantify the performance of their study, normalized mean absolute error (NMAE), normalized root mean square error (NRMSE), and mean absolute percentage error (MAPE) proportions were adopted. On average, the NMAE, NRMSE, and MAPE proportions were 5.42%, 6.54%, and 5.41%, respectively, for heart rate estimation. They showed significant agreement on the respiratory rate between the microbend fiber optic sensor and polysomographic results (r = 0.41, p < 0.001). In the current study, a different type of sensor other than a microbend fiber optic sensor was used. In addition to home and medical environments, some researchers have applied ballistocardiograms in completely different environments, such as in vehicles and headsets, and have shown good results in terms of respiratory rate [18].

Sleep architecture is defined as a basic sleep structure in which the four sleep stages (stages N1, N2, N3, and R) change cyclically. Sleep stages are generally divided into rapid eye movement (REM) sleep and non-rapid eye movement (NREM) sleep, and NREM sleep is classified into stages N1, N2, and N3, indicating a continuity of relative depth. Sleep architecture can be affected by age, inappropriate sleep duration, drugs, and medical conditions. Abnormal sleep architecture may be clinically associated with short or long sleep, medications, and diverse sleep disorders (e.g., insomnia, OSA, periodic limb movement disorder, hypersomnia, parasomnia, and so on). Research adopting ballistocardiography as a way to measure detailed sleep stages began in the 2010s [9,19-21]. Early studies using ballistocardiographic signals showed a high accuracy of 82%, but most of these studies classified sleep stages briefly as sleep states and wake states or as NREM and REM stages [19]. Tuominen et al. [20] conducted a study using a commercial Beddit sleep tracker to monitor sleep architecture. Consensus in sleep stage classification between the Beddit sleep tracker and polysomnography was very poor when using a four-stage classification (wake, stage N1, N2, and N3 sleep). The authors concluded that it would be currently inapplicable to distinguish between sleep stages. Migliorini et al. [9] classified sleep stages as NREM, REM, and wake state and assessed automatic sleep classification based on heart rate variability, respiration, and movement signals using multi-channel electromechanical film technology (EMFIT) sensors. The agreement between their results and sleep stage classification using polysomnography was kappa = 0.55. They suggested that their proposed algorithm offers an interesting possibility for sleep monitoring at home. Kortelainen et al. [21] also used EMFIT sensors to evaluate sleep macrostructure using heartbeat intervals and movement activity data and analyzed 18 recordings from healthy subjects using reference polysomnography. When considering the wake, NREM, and REM stages, they obtained a total accuracy of 79% and kappa = 0.44. The results of their study indicated that bed sensors may be suitable for monitoring sleep macrostructure in an unobtrusive, low cost, and simple manner. When comparing the proposed algorithm with their algorithm, there is a time-variant autoregressive model for the feature extractor, and a hidden Markov model is used as the classifier. Their algorithm requires an additional movement activity signal and threshold value to classify the wake stage as an input; however, the proposed algorithm does not require an additional movement activity signal and threshold value to identify the wake stage due to the nature of its design.

This study had several limitations. First, although this study focused on detecting sleep stages using vital signs, blood pressure was not studied. There are reports that blood pressure is related to the sleep stages; however, it is not easy to measure in a contactless manner [22]. In fact, contactless methods for blood pressure measurement have been introduced, but these have not been widely commercialized [23]. Further clinical studies on the non-contact measurement of blood pressure and the association between blood pressure and sleep stages are needed. Second, the number of participants in this study was relatively small. In addition, five out of the 35 subjects were excluded from this analysis due to inadequate data. Therefore, in order to determine the reliability of the device and obtain accurate outcomes, it will be essential to increase the number of participants in future studies. Lastly, most participants in this study had OSA, which might have contributed to the moderate agreement seen for sleep architecture. Further studies with healthy controls will provide more reliable results and provide comparative data in patients with OSA.