Recent Progress in Long-Term Sleep Monitoring Technology
长期睡眠监测技术的最新进展

1.2. Sleep Problems 1.2.睡眠问题

Inadequate, irregular, or poor-quality sleep is common in modern society. Factors contributing to sleep deprivation include occupation, social demands, mental illness, physical illness, sleep disorders, race, age, marital status, gender, and hospitalization [10,11]. Sleep deprivation or sleep disorders can lead to low cognition, poor alertness, poor mood, cardiovascular disease, diabetes, metabolic and immune disorders, and even death [12,13].
睡眠不足、不规律或质量差在现代社会很常见。导致睡眠不足的因素包括职业、社会需求、精神疾病、身体疾病、睡眠障碍、种族、年龄、婚姻状况、性别和住院治疗[ 10 , 11 ]。睡眠不足或睡眠障碍会导致认知低下、警觉性差、情绪不佳、心血管疾病、糖尿病、代谢和免疫紊乱，甚至死亡[ 12、13 ]。

For particular groups, sleep problems also have their own unique manifestations. Adolescents tend to sleep late, wake up early during school days, and catch up on sleep on weekends [14], leading to differences in their sleep on weekdays and days off [14,15]. Some older adults also experience sleep disturbances because the circadian system and sleep balance mechanisms become less robust with normal aging [16]. Finally, women with severe premenstrual syndrome (PMS) have poorer sleep quality, which may be related to altered melatonin rhythms [17]. There are also a variety of sleep disorders that may affect patients’ quality of life, such as obstructive sleep apnea, chronic insomnia, narcolepsy, delayed sleep–wake phase disorder, and Kleine–Levin syndrome [3].
对于特定群体来说，睡眠问题也有其独特的表现形式。青少年在上学期间往往晚睡早起，周末补觉[ 14 ]，导致工作日和休息日的睡眠情况存在差异[ 14 , 15 ]。一些老年人也会出现睡眠障碍，因为随着正常衰老，昼夜节律系统和睡眠平衡机制变得不那么健全[ 16 ]。最后，患有严重经前综合症（PMS）的女性睡眠质量较差，这可能与褪黑激素节律改变有关[ 17 ]。还有多种睡眠障碍可能影响患者的生活质量，如阻塞性睡眠呼吸暂停、慢性失眠、发作性睡病、睡眠-觉醒时相延迟障碍、克莱恩-莱文综合征等[ 3 ]。

For people with neurological and metabolic disorders, sleep quality is critical to health and even life. A classic example is people with depression. Antidepressant medications may affect sleep structure. Persistent sleep problems can, in turn, increase depression relapse or increased drug dependence and even potentially cause suicide in patients [18,19]. Attention to sleep problems can help determine the best medication regimen for depressed patients. In addition, for pregnant women, clinical pregnancy and live birth have occurred in 35% of women with sleep-disordered breathing (SDB) compared to 58% of women without SDB [20]. Sleep impairment is also a common comorbid and debilitating symptom for persons with opioid use disorder (OUD). Research into underlying mechanisms and efficacious treatment interventions for OUD-related sleep problems requires both precise and physiologic measurements of sleep-related outcomes and impairment [21].
对于患有神经和代谢疾病的人来说，睡眠质量对健康乃至生命至关重要。一个典型的例子是抑郁症患者。抗抑郁药物可能会影响睡眠结构。持续的睡眠问题反过来会增加抑郁症复发或增加药物依赖性，甚至可能导致患者自杀[ 18 , 19 ]。关注睡眠问题有助于确定抑郁症患者的最佳药物治疗方案。此外，对于孕妇来说，患有睡眠呼吸障碍 (SDB) 的女性有 35% 发生临床妊娠和活产，而没有 SDB 的女性这一比例为 58% [ 20 ]。睡眠障碍也是阿片类药物使用障碍 (OUD) 患者常见的合并症和衰弱症状。对 OUD 相关睡眠问题的潜在机制和有效治疗干预的研究需要对睡眠相关结果和损害进行精确和生理测量 [ 21 ]。

2.2. Sleep Cycle 2.2.睡眠周期

Human sleep is a complex physiological behavior that is complicated to evaluate comprehensively. However, a common evaluation criterion is whether a night’s sleep is characterized by multiple complete and healthy sleep cycles.
人类睡眠是一种复杂的生理行为，难以全面评价。然而，一个常见的评价标准是一晚的睡眠是否具有多个完整且健康的睡眠周期的特征。

Human consciousness can be divided into three states: wakefulness, non-rapid eye movement (NREM or non-REM) sleep (NREMS or non-REMS), and rapid eye movement (REM) sleep (REMS) [40]. NREMS can be further divided into three or four different stages. These stages alternate throughout the night, a phenomenon known as the sleep cycle. A cycle is roughly 2–3 h (Figure 3) [41]. Sleep with several complete sleep cycles is healthy.
人类意识可分为三种状态：清醒状态、非快速眼动睡眠（NREM或non-REM）睡眠（NREMS或非REMS）、快速眼动睡眠（REMS）[ 40 ]。 NREMS可以进一步分为三个或四个不同的阶段。这些阶段在整个晚上交替出现，这种现象称为睡眠周期。一个周期大约为2-3小时（图3 ）[ 41 ]。睡眠有几个完整的睡眠周期是健康的。

Open in a separate window

Figure 3 图3

(a) Sleep cycles and REM sleep [41]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2022 by the authors, published by MDPI. (b) EEG signals in different stages of sleep [42]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2016 by the authors, published by MDPI.
( a ) 睡眠周期和快速眼动睡眠 [ 41 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2022，由 MDPI 发布。 ( b )不同睡眠阶段的脑电图信号[ 42 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2016，由 MDPI 发布。

The wakefulness period is the first stage of sleep when the person is still conscious. The EEG is a low-amplitude mixed-frequency signal with relatively high muscle tone, and the predominant EEG frequency is the alpha rhythm in the wakefulness period [40]. The eyes may move in response to the person’s consciousness.
清醒期是人仍然有意识的睡眠的第一阶段。脑电图是一种低幅度的混频信号，肌张力相对较高，脑电图的主要频率是清醒期的α节律[ 40 ]。眼睛可能会随着人的意识而移动。

When the alpha wave disappears, the person enters NREM sleep, in which the eye moves more slowly. This stage can be further subdivided into three stages according to the depth of sleep. The first one is when the person just enters the sleep state from the waking state when the sleep is very light and can be easily awakened. The second one is the longest and takes up about half of a person’s total sleep time. The third one is the deepest sleep, which has a large number of low-frequency delta waves in the brain waves [43]. This standard was published by the American Academy of Sleep Medicine (AASM) in 2007. The R&K criteria, widely used before that, was proposed in 1968 [44]. In the latter, using the slow wave percentage as a criterion, the stages of deep sleep are further split into S3 (20–50%) and S4 (50%). The difference can be clearly seen in the EEG images, so this classification method is still followed in many studies (in Figure 4) [42].
当阿尔法波消失时，人进入 NREM 睡眠，此时眼睛移动得更慢。此阶段根据睡眠深度又可进一步细分为三个阶段。第一个是人从清醒状态刚刚进入睡眠状态时，睡眠很浅，很容易被惊醒。第二个时间最长，大约占据一个人总睡眠时间的一半。第三种是最深睡眠，脑电波中有大量低频δ波[ 43 ]。该标准由美国睡眠医学会（AASM）于2007年发布。此前广泛使用的R&K标准于1968年提出[ 44 ]。后者以慢波百分比为标准，将深度睡眠阶段进一步分为S3（20-50%）和S4（50%）。在脑电图图像中可以清楚地看到差异，因此许多研究仍然遵循这种分类方法（图4 ）[ 42 ]。

In the monitoring of the sleep cycle, REM is a very important stage. It accounts for 20–25% of nighttime sleep in healthy adults [45]. During this sleep stage, the brain is so excited that it is difficult to distinguish the EEG from waking hours while the muscles are most relaxed. This is why REM sleep is also called “paradoxical sleep” [43].
在睡眠周期的监测中，REM是一个非常重要的阶段。它占健康成年人夜间睡眠的 20-25% [ 45 ]。在这个睡眠阶段，大脑非常兴奋，以至于很难将脑电图与清醒时肌肉最放松的时间区分开来。这就是为什么快速眼动睡眠也被称为“矛盾睡眠”[ 43 ]。

REM is considered to be the most unstable period of respiratory and cardiac sleep. Patients with diaphragmatic dysfunction may be particularly at risk due to the reduced tone of the accessory respiratory muscles. In addition, almost all antidepressants inhibit REM sleep [46]. The suppression of REM sleep in depressed patients may be one of the reasons for their poor sleep quality. A significant coupling of REM sleep cycles was observed when couples slept in the same bed. REM sleep may contain feedback to the surrounding environment [47]. Therefore, it is of great significance to monitor the physiological information during REM sleep.
REM 被认为是呼吸和心脏睡眠最不稳定的时期。由于辅助呼吸肌张力减弱，膈肌功能障碍患者可能面临特别高的风险。此外，几乎所有抗抑郁药都会抑制快速眼动睡眠[ 46 ]。抑郁症患者快速眼动睡眠的抑制可能是其睡眠质量差的原因之一。当夫妻睡在同一张床上时，可以观察到快速眼动睡眠周期的显着耦合。快速眼动睡眠可能包含对周围环境的反馈[ 47 ]。因此，监测REM睡眠期间的生理信息具有重要意义。

The technique of performing the classification of sleep stages in clinical practice is well established. Awakening is with high muscle tone, targeted eye movements, and activated cerebral cortex. Non-REM sleep is characterized by moderate muscle tone, no eye movements, and slow EEG waves. REM sleep is characterized by low muscle tone, eye movements, and active cerebral cortex [43]. Thus, polysomnography is the standard gold method for measuring sleep cycles, but it is not convenient. As a result, the vast majority of patients do not receive effective diagnosis and treatment [48].
在临床实践中进行睡眠阶段分类的技术已经很成熟。觉醒需要高肌张力、有针对性的眼球运动和激活的大脑皮层。非快速眼动睡眠的特点是肌张力适中、没有眼球运动、脑电图波缓慢。快速眼动睡眠的特点是低肌张力、眼球运动和活跃的大脑皮层[ 43 ]。因此，多导睡眠图是测量睡眠周期的标准黄金方法，但它并不方便。结果，绝大多数患者没有得到有效的诊断和治疗[ 48 ]。

3.1. Electroencephalography
3.1.脑电图

EEG has long been an important part of sleep monitoring. The changes in brain waves during the various stages of the sleep state have been described in Section 2.2, and this has become a crucial item in sleep monitoring. Brain-wave characteristics are the gold standard for sleep cycle classification. The most accurate sleep staging analysis system is based on EEG, the only single sensing modality capable of identifying all sleep stages [67].
脑电图长期以来一直是睡眠监测的重要组成部分。睡眠状态各阶段脑电波的变化已在2.2节中描述，这已成为睡眠监测中至关重要的项目。脑电波特征是睡眠周期分类的黄金标准。最准确的睡眠分期分析系统基于脑电图，这是唯一能够识别所有睡眠阶段的单一传感方式[ 67 ]。

The brain, as the most important nerve center in the body, has very distinct characteristics during all stages of the sleep cycle and has good results as a sleep monitoring indicator. As shown in Figure 6, the performance of brain waves varies greatly from stage to stage and between different genders [68,69,70], which is why the EEG has become the gold standard for sleep cycle identification. EEG can reflect the effects of previous nights of sleep, over-the-counter and prescription drugs, and even illicit drugs on brain activity during sleep [71]. In 2006, Guilleminault et al. studied the effects of different levels of sound stimulation on human brain waves during sleep and on the performance of sleepiness the next day, examining the analysis of the quality of disturbed sleep at the level of brain waves [72]. In addition to this, many studies have been reported on related detection devices due to the importance of EEG signals in the diagnosis of Alzheimer’s disease, Parkinson’s disease, epilepsy, etc. [73,74].
大脑作为人体最重要的神经中枢，在睡眠周期的各个阶段都有非常鲜明的特征，作为睡眠监测指标有很好的效果。如图6所示，不同阶段、不同性别的脑电波表现差异很大[ 68,69,70 ]，这就是为什么脑电图成为识别睡眠周期的金标准。脑电图可以反映前一天晚上的睡眠、非处方药和处方药、甚至非法药物对睡眠期间大脑活动的影响[ 71 ]。 2006 年，Guilleminault 等人。研究了不同水平的声音刺激对人类睡眠时脑电波以及第二天困倦表现的影响，检验了脑电波水平上睡眠不安质量的分析[ 72 ]。除此之外，由于脑电信号在阿尔茨海默病、帕金森病、癫痫等诊断中的重要性，相关检测装置的研究也有很多报道[ 73 , 74 ]。

Open in a separate window

Figure 6 图6

(a) Topographic distribution of absolute sleep EEG power (µV2) across derivations for the beta 1 to gamma two bands during REM sleep. The first and the second column depict the values averaged across females and males on the same scale, while the third column depicts the difference between females and males. Minimum and maximum values are shown in the upper right corner of each topographic map. The fourth and the fifth column show the F-values from the analysis of variance (ANOVA) for the factors of age and pubertal status. In the third to fifth columns, significant electrodes are shown in white (p < 0.05), gray (p < 0.01), and magenta (p < 0.001). In these columns, warm colors represent increased activity in females, while cool colors represent increased activity in males [69]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2020 by the authors, published by Springer Nature. (b) EEG signals of C4-Fz channels under different sleep stages [70]. Reproduced under the CC-BY-NC-ND license, Copyright 2022 by International Federation of Clinical Neurophysiology, Published by Elsevier B.V.
( a ) 快速眼动睡眠期间 β 1 至 γ 2 频带的绝对睡眠 EEG 功率 (μV2) 的地形分布。第一列和第二列描述了相同范围内女性和男性的平均值，而第三列描述了女性和男性之间的差异。最小值和最大值显示在每个地形图的右上角。第四列和第五列显示年龄和青春期状态因素的方差分析 (ANOVA) 的 F 值。在第三至第五列中，重要电极以白色 ( p < 0.05)、灰色 ( p < 0.01) 和洋红色 ( p < 0.001) 显示。在这些栏中，暖色代表女性活动增加，而冷色代表男性活动增加[ 69 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2020，由 Springer Nature 出版。 ( b )不同睡眠阶段C4-Fz通道的EEG信号[ 70 ]。根据 CC-BY-NC-ND 许可复制，版权所有 2022，国际临床神经生理学联合会，由 Elsevier BV 出版

Miniaturization of traditional electrodes, or home use, allows for wearable EEG monitoring, but the principle remains that electrodes attached to the skin’s surface can pick up electrical signals of neural activity [56]. With the arrival of the new coronary epidemic, many people are less willing to go to the hospital, and home healthcare has become a healthcare trend. In 2020, Arnal et al. fabricated EEG sensors integrated into a headband. The mean percentage error of the EEG signal obtained with PSG monitoring α was 15 ± 3.5%, β was 16 ± 4.3%, λ was 16 ± 6.1%, and theta frequency during sleep was 10 ± 1.4% [75]. In 2021, Hsieh et al. developed a real-time EEG acquisition system for home use and used a deep-learning model that allowed the average absolute error of the wearable device to measure sleep efficiency to be reduced to 1.68% [76]. Studies using machine-learning algorithms for brain-wave recognition and analysis based on the same sensors are beyond the focus of this paper, but these studies are a good example of the significance of brain-wave sensors [77].
传统电极或家用电极的小型化允许可穿戴脑电图监测，但原理仍然是附着在皮肤表面的电极可以拾取神经活动的电信号[ 56 ]。随着新冠疫情的到来，很多人不太愿意去医院，家庭医疗保健已经成为一种医疗保健趋势。 2020 年，Arnal 等人。将脑电图传感器集成到头带中。 PSG监测获得的EEG信号的平均百分比误差α为15±3.5%，β为16±4.3%，λ为16±6.1%，睡眠期间的θ频率为10±1.4%[ 75 ]。 2021 年，Hsieh 等人。开发了家用实时脑电图采集系统，并使用深度学习模型，使可穿戴设备测量睡眠效率的平均绝对误差降低至 1.68% [ 76 ]。基于相同传感器使用机器学习算法进行脑电波识别和分析的研究超出了本文的重点，但这些研究是脑电波传感器重要性的一个很好的例子[ 77 ]。

EEG is the most demanding for signal quality in bioelectrical signal sensing. Very often, there is a balance between wearing comfort and signal quality. Conventional EEG uses patch electrodes that have good signal quality but are not breathable and may cause skin swelling (see Figure 5c). To minimize the effects of contact resistance, Li et al. designed an array of microneedles that can be pierced into the skin and prepared the apparatus on a flexible substrate. The electrodes have record low skin–electrode contact resistance, 1/250th that of conventional electrodes.
脑电图是生物电信号传感中对信号质量要求最高的。很多时候，佩戴舒适度和信号质量之间需要取得平衡。传统脑电图使用的贴片电极具有良好的信号质量，但不透气，可能导致皮肤肿胀（见图5 c）。为了最大限度地减少接触电阻的影响，Li 等人。设计了一系列可以刺入皮肤的微针，并在柔性基板上准备了该装置。该电极具有创纪录的低皮肤电极接触电阻，仅为传统电极的 1/250。

From another improvement perspective, many researchers are exploring more user-friendly forms of wearable sensors. In 2019, Shustak et al. prepared soft, non-gel flexible electrodes with printed electrode technology to improve the comfort of brain-wave detection [61]. In 2017, Nakamura et al. designed the acquisition of brain-wave signals in the ear, which also achieved good results compared to the acquisition of brain-wave signals in the scalp patch [78]. In 2021, da Silva et al. designed flexible printed electrode sensors in the ear using graphene electrodes and combined them with a smartphone for recording and analysis [79].
从另一个改进的角度来看，许多研究人员正在探索更加用户友好的可穿戴传感器形式。 2019 年，舒斯塔克等人。利用印刷电极技术制备了柔软的非凝胶柔性电极，以提高脑电波检测的舒适度[ 61 ]。 2017 年，中村等人。设计了在耳朵中采集脑电波信号，与在头皮贴片中采集脑电波信号相比也取得了很好的效果[ 78 ]。 2021 年，达席尔瓦等人。使用石墨烯电极设计了耳朵中的柔性印刷电极传感器，并将其与智能手机结合起来进行记录和分析[ 79 ]。

Finally, in recent years, implantable brain–computer interfaces have enabled stable monitoring of signals [80,81,82,83]. Although its main application area is to assist people with motor impairments to control assistive devices [80,81], sensors on EEG signals may further advance the development of sleep monitoring technology in the future. Topchiy et al. studied in vivo implanted electrodes to monitor sleep in mice. They experimented with sleep monitoring with implanted electrodes and telemetry and found they could classify sleep stages more effectively than in vitro monitoring devices [84]. As the technology of implanted electrodes matures, this may also be a future technology that can strike a good balance between contact resistance and non-sensory use.
最后，近年来，植入式脑机接口已经实现了信号的稳定监测[ 80、81、82、83 ]。虽然其主要应用领域是帮助运动障碍人士控制辅助设备[ 80 , 81 ]，但脑电图信号传感器未来可能会进一步推动睡眠监测技术的发展。托普奇等人。研究了体内植入电极来监测小鼠的睡眠。他们用植入电极和遥测技术进行睡眠监测实验，发现它们可以比体外监测设备更有效地对睡眠阶段进行分类[ 84 ]。随着植入电极技术的成熟，这也可能是未来能够在接触电阻和非感知使用之间取得良好平衡的技术。

3.2. Electrocardiography 3.2.心电图

ECG is an important physiological examination closely related to sleep cycles and sleep apnea [85]. As shown in Figure 7a, the periodic movement of the heart will show different electrical signals and form regular ECG curves [85]. In hospitals, ECGs are collected through specialized equipment with the help of professional staff, but self-monitoring by patients is hardly up to this standard [86]. The fit of wearable device contacts is also a common problem. For this reason, many studies have expanded the relevant algorithms and databases so that testing devices can be adapted to self-testing using devices such as wearables to improve signal-to-noise ratios outside the hospital, exclude motion artifacts [87,88], and more accurately determine the occurrence of phenomena such as arrhythmias [89].
心电图是与睡眠周期和睡眠呼吸暂停密切相关的重要生理检查[ 85 ]。如图7a所示，心脏的周期性运动会表现出不同的电信号并形成规则的心电图曲线[ 85 ]。在医院，心电图是在专业人员的帮助下通过专门设备收集的，但患者的自我监测很难达到这个标准[ 86 ]。可穿戴设备触点的适配也是一个常见问题。为此，许多研究扩展了相关算法和数据库，使测试设备能够适应使用可穿戴设备等设备进行自我测试，以提高医院外的信噪比，排除运动伪影[ 87 , 88 ]，并更准确地判断心律失常等现象的发生[ 89 ]。

Open in a separate window

Figure 7 图7

(a) The different phases of muscle movement during the heartbeat produce different parts of the ECG curve. The depolarization and repolarization of the heart for the generation of a P-wave, QRS complex, and T-wave [85]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2022 by the authors, published by MDPI. (b) A flexible skin electrode sensor. Monitoring ECG at the wrist [90]. Reproduced under the Creative Commons Attribution 4.0 International License, Copyright 2022 by the authors, published by Nature Portfolio.
( a ) 心跳期间肌肉运动的不同阶段产生心电图曲线的不同部分。心脏的去极化和复极化产生 P 波、QRS 波群和 T 波 [ 85 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2022，由 MDPI 发布。 ( b ) 柔性皮肤电极传感器。在手腕处监测心电图 [ 90 ]。作者根据 Creative Commons Attribution 4.0 International License 复制，版权所有 2022，由 Nature Portfolio 出版。

Wearable ECG allows the detection of sleep apnea. The classification accuracy obtained from the ECG belt has a sensitivity of 70% and a specificity of 74%, while the patched ECG has a sensitivity of 88% [91]. Single-lead ECG, worn on the abdomen, can also be good for detecting sleep apnea index and abnormal breathing [92]. In 2019, Hammour et al. studied in-ear ECG. The delay was reduced by up to 88% [93].
可穿戴心电图可以检测睡眠呼吸暂停。从心电图带获得的分类准确性具有 70% 的敏感性和 74% 的特异性，而补丁心电图的敏感性为 88% [ 91 ]。佩戴在腹部的单导联心电图也有助于检测睡眠呼吸暂停指数和呼吸异常[ 92 ]。 2019 年，Hammour 等人。研究了耳内心电图。延迟减少了高达 88% [ 93 ]。

The electrodes can be kept naturally close to the skin compared to watch-type and headphone-type ECG measurement devices. ECGs on the torso often require patch electrodes, and many electrode materials can be irritating to the body. ECG sensing can be integrated into clothing and localized to locations with good signal-to-noise ratios [94] (Figure 7b).
与手表式和耳机式心电图测量设备相比，电极可以自然地靠近皮肤。躯干心电图通常需要贴片电极，并且许多电极材料会对身体产生刺激。 ECG 传感可以集成到衣服中并定位到具有良好信噪比的位置 [ 94 ]（图 7b ）。

Compared to EEG, ECG requires less signal quality, so many studies can use non-wearable, skin-tight electrodes. Lim et al. arranged electrodes on a mattress [95]. Won Kyu Lee et al. integrated flexible electrodes into the mattress to collect ECG signals from the skin’s surface after a person lies on it [62]. This avoids the need to wear a dedicated device and is well-suited for sleep scenarios. In 2020, Klum et al. used multimodal ECG and analyzed the effects of different sleeping positions [96]. Left ventricular ejection time and pre-ejection period estimation errors were 10% and 21%.
与脑电图相比，心电图对信号质量的要求较低，因此许多研究可以使用非穿戴式、紧贴皮肤的电极。林等人。将电极布置在床垫上[ 95 ]。李元奎等人。将柔性电极集成到床垫中，以在人躺在床垫上后收集皮肤表面的心电图信号[ 62 ]。这样就无需佩戴专用设备，非常适合睡眠场景。 2020 年，克鲁姆等人。使用多模态心电图并分析不同睡眠姿势的影响[ 96 ]。左心室射血时间和射血前期估计误差分别为10%和21%。

3.3. Electromyography and Electrooculography
3.3.肌电图和眼电图

EMG can conveniently reflect human muscles’ tension and limb activity and assist in measuring sleep cycles. While the monitoring of limb movements is challenged by mechanical sensors or camera sensing (in Section 3), EMG has an irreplaceable role in many fields. For example, it is difficult to monitor eye movements outside of the body because the eyelids obscure them in the human sleep state; nocturnal muscle tensions, such as changes in neck muscle tone, do not manifest as obvious changes in limb position. On the other hand, such muscle behaviors have electrical signals that can penetrate the tissues and be measured on the skin’s surface with good accuracy [97,98]. A variety of new methods of collecting electrical signals on the skin’s surface in Figure 8 enable convenient daily measurements of EMG.
肌电图可以方便地反映人体肌肉的张力和肢体活动，并辅助测量睡眠周期。虽然肢体运动的监测受到机械传感器或摄像头传感（第3节）的挑战，但肌电图在许多领域具有不可替代的作用。例如，在人体睡眠状态下，由于眼睑遮挡，很难监测体外的眼球运动；夜间肌肉紧张，例如颈部肌张力的变化，并不表现为肢体位置的明显变化。另一方面，这种肌肉行为具有可以穿透组织并在皮肤表面上进行高精度测量的电信号[ 97 , 98 ]。图 8中收集皮肤表面电信号的各种新方法可以方便地进行日常肌电图测量。

Open in a separate window

Figure 8 图8

Method for collecting electrical signals on the skin’s surface: (a) gold electronic tattoo [99]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2023 by the authors, published by MDPI. (b) Multi-layer electronic tattoo [100]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2020 by the authors, published by Springer Nature. (c) Conductive fabrics doped with graphene [101]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2022 American Chemical Society. (d) Microneedle array electrodes on a flexible substrate [102]. Reproduced under Creative Commons Attribution 4.0 International License, Copyright 2022 by the authors, published by Shanghai Jiao Tong University Press.
采集皮肤表面电信号的方法：（ a ）金电子纹身[ 99 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2023，由 MDPI 发布。 ( b )多层电子纹身[ 100 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2020，由 Springer Nature 出版。 ( c )掺杂石墨烯的导电织物[ 101 ]。根据 CC-BY 知识共享归属许可条款复制，版权所有 2022 美国化学会。 ( d )柔性基板上的微针阵列电极[ 102 ]。作者根据知识共享署名 4.0 国际许可复制，版权所有 2022，上海交通大学出版社出版。

In 2007, Magosso et al. used electrooculography to assess the sleep cycle, which proved to be very reliable, addressing the high labor cost and inconsistency of previous manual scoring [103]. Eye movement is tracked by several muscles. Skin electrodes affixed to the corners of the eye can pick up electrical signals and thus determine whether eye movement is occurring. Beach et al. achieved comfortable wear of eye movement detection devices by integrating EOG sensors in an eye patch through fabric sensor electrodes made of nylon and graphene. Though, with EOG alone, the accuracy of sleep time calculation is only about 70% [104]. However, EOG can be included in the sleep cycle analysis as an important item in sleep polysomnography.
2007 年，Magosso 等人。使用眼电图来评估睡眠周期，事实证明这是非常可靠的，解决了以前人工评分的高劳动力成本和不一致的问题[ 103 ]。眼球运动由几块肌肉跟踪。贴在眼角的皮肤电极可以接收电信号，从而确定眼球是否发生运动。海滩等人。通过由尼龙和石墨烯制成的织物传感器电极将 EOG 传感器集成在眼罩中，实现了眼动检测设备的舒适佩戴。然而，仅使用 EOG，睡眠时间计算的准确度仅为 70% 左右[ 104 ]。然而，EOG可以作为睡眠多导睡眠图的一个重要项目纳入睡眠周期分析中。

Iranzo et al. used polysomnography to analyze EMG analysis of REM sleep in patients with REM sleep behavior disorders. These patients need more accurate monitoring of their REM sleep. EMG of the cardiac, flexor superficial, and extensor profundus muscles can help in the identification of REM [105]. Maeda used single-channel EMG, which also enabled sleep mydriasis detection with 100% sensitivity and specificity under some conditions, demonstrating that single-channel EMG signals can also be of good monitoring value [106].
伊朗佐等人。使用多导睡眠图对 REM 睡眠行为障碍患者的 REM 睡眠进行肌电图分析。这些患者需要更准确地监测他们的快速眼动睡眠。心脏、浅屈肌和深伸肌的肌电图可以帮助识别 REM [ 105 ]。 Maeda使用单通道肌电图，在某些条件下也能够实现100%灵敏度和特异性的睡眠瞳孔散大检测，证明单通道肌电图信号也可以具有良好的监测价值[ 106 ]。

In 2018, Beniczky et al. used wearable EMG signals to capture the evolution of TCS-related signals on the human surface for the detection of muscle rigidity and epilepsy occurring during sleep [66]. In 2022, Yeung et al. completed the diagnosis of obstructive sleep through muscle electrical signals in the tongue and epiglottis to epiglottal pressure and nasal airflow and then through EMG at the level of muscle movement [107]. The diagnosis of apnea was made by Rebelo after collecting the EMG signals generated by the apical muscles of the tongue and generating electrical signals to stimulate the apical muscles of the tongue to terminate the respiratory obstruction when sleep apnea was detected [108]. In 2018, Yamaguchi et al. designed a wearable miniature EMG system weighing 9 g, including the battery, to assess the occurrence of nocturnal teeth grinding [109] (sleep bruxism is described in detail in Section 3.3). In 2019, Prasad et al. connected the EMG device to a smartphone and used it to assist in monitoring teething behavior [110].
2018 年，贝尼茨基等人。使用可穿戴式 EMG 信号来捕获人体表面 TCS 相关信号的演变，以检测睡眠期间发生的肌肉僵硬和癫痫[ 66 ]。 2022 年，Yeung 等人。通过舌头和会厌中的肌肉电信号到会厌压力和鼻气流，然后通过肌肉运动水平的肌电图完成了阻塞性睡眠的诊断[ 107 ]。 Rebelo在检测到睡眠呼吸暂停时收集舌尖肌产生的肌电信号并产生电信号刺激舌尖肌终止呼吸阻塞后做出呼吸暂停的诊断[ 108 ]。 2018 年，山口等人。设计了一个重 9 g 的可穿戴微型肌电图系统，包括电池，用于评估夜间磨牙的发生情况[ 109 ]（睡眠磨牙症在第 3.3 节中有详细描述）。 2019 年，普拉萨德等人。将肌电图设备连接到智能手机并用它来协助监测出牙行为[ 110 ]。

3.4. Electroretinography 3.4.视网膜电图

Among the various types of electrical signals, the study of retinal electrical signals was the latest to begin and has the least application in sleep monitoring. The retinal electrical signal expressed the perception of light by the retina and was first used for the diagnosis of eye diseases [111].
在各类电信号中，视网膜电信号的研究起步最晚，在睡眠监测中的应用最少。视网膜电信号表达了视网膜对光的感知，最初用于眼部疾病的诊断[ 111 ]。

With the development of basic research demonstrating the influence of the light environment on human circadian rhythms, the response of the human nervous system to light became an item in sleep monitoring. In 1994, Galambos et al. found that ERG amplitude during slow-wave sleep was more than twice as high as during wakefulness. Moreover, ERG patterns during REM sleep were different from those during slow-wave sleep. Galambos confirmed that ERG signals are also associated with the sleep cycle [112]. In 2016, Liguori et al. demonstrated that ERGs could differ in patients with obstructive sleep apnea [63]. However, the relationship between fundus disease and sleep needs to be further explored [113] (Figure 9c).
随着基础研究的发展证明光环境对人体昼夜节律的影响，人体神经系统对光的反应成为睡眠监测的一个项目。 1994 年，Galambos 等人。发现慢波睡眠期间的 ERG 幅度是清醒期间的两倍多。此外，快速眼动睡眠期间的 ERG 模式与慢波睡眠期间的 ERG 模式不同。 Galambos 证实 ERG 信号也与睡眠周期相关[ 112 ]。 2016 年，Liguori 等人。证明阻塞性睡眠呼吸暂停患者的 ERG 可能有所不同 [ 63 ]。然而，眼底疾病与睡眠之间的关系还需要进一步探讨[ 113 ]（图9c ）。

Open in a separate window

Figure 9 图9

(a) Schematic diagram of ERG sensing [114]. The retina produces an electrical signal when light enters the human eye. A wire needs to be introduced to the surface of the eye to read the signal. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2021 by the authors, published by Springer Nature. (b) Conventional ERG monitoring requires leads from the surface of the eye, making it difficult to use for long periods of time. The new sensing method allows ERG signals to be collected on the skin around the eye, and all-night monitoring is possible [115]. Reproduced under the CC-BY-NC License, Copyright 2021 by the Korean Ophthalmological Society, published by Korean Ophthalmological Society. (c) A skin electrode was placed on the lower lid of each eye. The contralateral eye was not covered. A gold-cup electrode was placed on the right earlobe as the ground electrode [116]. Reproduced under the Creative Commons Attribution License, Copyright 2023 by the author(s), published by Public Library Science.
( a ) ERG传感示意图[ 114 ]。当光线进入人眼时，视网膜会产生电信号。需要将一根电线引入眼睛表面才能读取信号。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2021，由 Springer Nature 出版。 ( b )传统的ERG监测需要从眼睛表面引出导线，因此难以长时间使用。新的传感方法允许在眼睛周围的皮肤上收集ERG信号，并且可以进行整夜监测[ 115 ]。根据 CC-BY-NC 许可复制，版权所有 2021，韩国眼科学会，由韩国眼科学会出版。 ( c ) 将皮肤电极放置在每只眼睛的下眼睑上。未遮盖对侧眼睛。将金杯电极放置在右耳垂作为接地电极[ 116 ]。作者根据知识共享署名许可复制，版权所有 2023，由公共图书馆科学出版。

Since ERG often requires electrodes placed on the inner eyelid (Figure 9a), it is more difficult and device-demanding to use than tests such as EOG, which can be applied to the skin. Research is also underway to attach electrodes to the skin around the eye to monitor ERG [115]. However, the signal quality is still not as good as the intraocular type. In addition, the need for ERG signals in sleep monitoring needs to be supported by more studies.
由于 ERG 通常需要将电极放置在内眼睑上（图 9a ），因此与可应用于皮肤的 EOG 等测试相比，使用起来更加困难且对设备要求更高。将电极附着在眼睛周围的皮肤上以监测 ERG 的研究也在进行中 [ 115 ]。然而，信号质量仍然不如眼内式。此外，睡眠监测中ERG信号的需求还需要更多研究的支持。

3.5. Passive Bioelectricity Detection
3.5.被动生物电检测

The electrical signals generated by the nervous system during activity are voltages actively generated by the body. The four previous vignettes are based on this. However, the human body can also be considered as a load consisting of resistance and capacitance, and passive bioelectrical detection is achieved by applying voltage and an electric field.
神经系统在活动过程中产生的电信号是身体主动产生的电压。前面的四个小插曲都是以此为基础的。然而，人体也可以被认为是由电阻和电容组成的负载，通过施加电压和电场来实现被动生物电检测。

Blood pressure (BP) causes changes in the diameter of the blood vessels, which in turn affects the impedance of the tissues. Kireev et al. used graphene electronic skin to detect the impedance between electrodes at different locations on the skin [117]. The signal of impedance change can be detected as blood flow pulses move through the blood vessels. BP was calculated with an accuracy of 0.2 ± 4.5 mm Hg for diastolic pressures and 0.2 ± 5.8 mm Hg for systolic pressures.
血压 (BP) 会导致血管直径发生变化，进而影响组织的阻抗。基里耶夫等人。使用石墨烯电子皮肤来检测皮肤上不同位置的电极之间的阻抗[ 117 ]。当血流脉冲穿过血管时，可以检测到阻抗变化的信号。血压的计算精度为：舒张压为 0.2 ± 4.5 mm Hg，收缩压为 0.2 ± 5.8 mm Hg。

Changes in human posture and position in the external electric field will lead to different polarization responses. By arranging the Wi-Fi device in the room after the placement design, the receiver can detect the human body’s activities. Epilepsy detection with 100% sensitivity is achieved without wearing a human device [118]. This method can also achieve 92% accuracy in the recognition of rhythmic movement disorders [119]. The use of multi-antenna arrays allows for the acquisition of richer electric field information. Yu et al. achieved 81.8% classification of sleep stages and breath detection with an average error of 0.23 bpm based on a multi-antenna Wi-Fi receiver [120].
人体在外部电场中的姿势和位置的变化会导致不同的极化响应。通过放置设计后将Wi-Fi设备放置在房间内，接收器可以检测到人体的活动。无需佩戴人体设备即可实现 100% 灵敏度的癫痫检测 [ 118 ]。该方法在节律性运动障碍的识别上也能达到92%的准确率[ 119 ]。多天线阵列的使用可以获取更丰富的电场信息。于等人。基于多天线 Wi-Fi 接收器，实现了 81.8% 的睡眠阶段分类和呼吸检测，平均误差为 0.23 bpm [ 120 ]。

In these two examples, mechanical changes in the human body cause changes in electrical properties. They are placed in Section 2, as the sensor collects electrical signals directly. More research will be reported in Section 3 on BP and motion.
在这两个例子中，人体的机械变化导致电气特性的变化。它们被放置在第 2 部分，因为传感器直接收集电信号。更多关于 BP 和运动的研究将在第 3 节中报告。

4.1. Motion Detection 4.1.运动检测

Limb movement is an important concomitant behavior during sleep; many people experience vigorous limb movement. Based on motion sensors on the wrist, Chun et al. monitored how often people with dermatitis may itch at night, demonstrating the effects of pruritus on sleep [128]. In addition to the common sleep onset tests and sleep stage divisions, some sleep disorders are also reflected in body movements. The most typical one is restless legs syndrome. In 2022, Brooks et al. used conductive fabric to form a capacitance with a person’s body, and the magnitude of this virtual capacitance changed after a change in the person’s posture, which in turn was detected. The potential improvement in diagnostic accuracy for assessing sleep disturbances associated with restless legs syndrome using this method can be estimated at approximately 68.1%, far exceeding the diagnosis of measuring anterior tibial EMG signals [129].
肢体运动是睡眠期间重要的伴随行为；许多人经历过剧烈的肢体运动。 Chun 等人基于手腕上的运动传感器。监测皮炎患者夜间瘙痒的频率，证明瘙痒对睡眠的影响[ 128 ]。除了常见的入睡测试和睡眠阶段划分外，一些睡眠障碍还体现在身体动作上。最典型的就是不宁腿综合症。 2022 年，布鲁克斯等人。利用导电织物与人的身体形成电容，这种虚拟电容的大小随着人的姿势的变化而变化，进而被检测到。使用这种方法评估与不宁腿综合征相关的睡眠障碍的诊断准确性的潜在改进可估计约为 68.1%，远远超过测量胫骨前肌电图信号的诊断[ 129 ]。

Wristband motion sensors are the most common form of sleep monitoring. Accelerometers based on a micro-electro-mechanical system (MEMS) can be combined with everyday wearable items such as watches for wearability (Figure 10a,b). Nomoto et al. analyzed the wearer’s sleep quality and tracked various types of phenomena that affect sleep quality with a wristwatch-based motion sensor worn for a long period [130]. In 2019, Yeom et al. integrated sensors on watches that can analyze sleep apnea and send results in real time to a cell phone [131]. In 2022, Katori et al. analyzed over 100,000 data sets and analyzed the classification of 16 sleep problems [132].
腕带运动传感器是最常见的睡眠监测形式。基于微机电系统 (MEMS) 的加速度计可以与日常可穿戴物品（例如手表）结合使用，以实现可穿戴性（图 10 a、b）。野本等人。分析了佩戴者的睡眠质量，并通过长时间佩戴的基于手表的运动传感器来跟踪影响睡眠质量的各种类型的现象[ 130 ]。 2019 年，Yeom 等人。手表上的集成传感器可以分析睡眠呼吸暂停并将结果实时发送到手机[ 131 ]。 2022 年，Katori 等人。分析了超过 100,000 个数据集并分析了 16 种睡眠问题的分类 [ 132 ]。

Open in a separate window

Figure 10 图10

(a) Acceleration sensors made using MEMS technology, the mass block (6 L-beams L1 to L6) is displaced by the acceleration [133]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2019 by the authors, published by MDPI. (b) The displacement of the object block brings about a change in capacitance [134]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2020 by the authors, published by IOP Publishing Ltd. (c) A pressure sensor made of a multi-layer structure monitors the changes in received pressure [34]. Information such as human movement and heartbeat can be resolved. Reproduced under the terms of the CC-BY License, Copyright 2022 by the authors, published by MDPI.
( a ) 使用 MEMS 技术制造的加速度传感器，质量块（6 个 L 型梁 L1 至 L6）因加速度而位移 [ 133 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2019，由 MDPI 发布。 ( b )物体块的位移引起电容的变化[ 134 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2020，由 IOP Publishing Ltd 出版。 ( c ) 由多层结构制成的压力传感器监测接收压力的变化 [ 34 ]。可以解析人体运动、心跳等信息。根据 CC-BY 许可条款复制，作者版权所有 2022，由 MDPI 发布。

Although there are multiple joints between the wrist and torso in sleep staging and the human body, restoring the overall body posture with the wrist is difficult. In 2019, Trevenen et al. attempted to improve the accuracy of recognition with machine-learning algorithms [135]. In the same year, Walch et al. also attempted to analyze raw acceleration data from Apple Watch to analyze sleep, but the specificity was not satisfactory [136]. In 2022, Ode et al. achieved relatively high sensitivity and specificity by designing an acceleration-based long-term sleep–wake cycle classification and estimation algorithm (ACCEL) based on simple arm acceleration sensor results [137]. It shows that recognition rates can be improved by algorithms when sensors can provide limited information, but it is not feasible to compensate only by algorithms hoping to achieve the effect of more sensors.
虽然睡眠分期中手腕和躯干与人体之间存在多个关节，但用手腕恢复整体身体姿势是很困难的。 2019 年，Trevenen 等人。尝试通过机器学习算法提高识别的准确性[ 135 ]。同年，Walch 等人。还尝试分析Apple Watch的原始加速度数据来分析睡眠，但特异性并不令人满意[ 136 ]。 2022 年，Ode 等人。通过根据简单的手臂加速度传感器结果设计基于加速度的长期睡眠-觉醒周期分类和估计算法（ACCEL），实现了相对较高的灵敏度和特异性[ 137 ]。这表明，当传感器能够提供的信息有限时，可以通过算法来提高识别率，但希望通过更多传感器来达到的效果，仅通过算法进行补偿是不可行的。

In addition to the wrist, the chest is also a common location for placing accelerometers to more effectively reflect the human torso’s motion and detect mechanical signals of respiration and heartbeat (described in detail in Section 4.3). In 2017, Razjouyan et al. also demonstrated that a single chest accelerometer for sleep analysis was closer to the polysomnography results than a wrist sensor [138]. In 2021, Chen et al. built a detection system with temporal memory using long- and short-term memory (LSTM) networks after enriching the sensor data types and also achieved good results. The behavioral sensor on the wrist was able to identify sleep data with 92% accuracy [139].
除了手腕之外，胸部也是放置加速度计的常见位置，以更有效地反映人体躯干的运动并检测呼吸和心跳的机械信号（详细描述见4.3节）。 2017 年，Razjouyan 等人。还证明，用于睡眠分析的单个胸部加速度计比手腕传感器更接近多导睡眠图结果[ 138 ]。 2021 年，陈等人。在丰富传感器数据类型后，利用长短期记忆（LSTM）网络构建了具有时间记忆的检测系统，也取得了良好的效果。手腕上的行为传感器能够以 92% 的准确度识别睡眠数据 [ 139 ]。

In addition to the two broad categories mentioned above, the types of sensors for detecting posture and movement are actually very rich in various combinations [140]. Sunderam et al. incorporated MEMS accelerometers in a wearable detector for the head, which aided the training set for partitioning different sleep stages and can potentially be used for neuroprosthetic applications for movement disorders and seizures [141]. Yoshihi et al. achieved a higher accuracy sleep stage analysis based on a single 3D accelerometer of the head [142]. However, the accuracy for sleep stage recognition was only 74.6%. For REM sleep, the accuracy was only 52.7%. Therefore, the wrist and torso are still ideal locations for sensor placement.
除了上面提到的两大类之外，用于检测姿势和运动的传感器类型实际上非常丰富，各种组合[ 140 ]。桑德拉姆等人。将 MEMS 加速度计纳入头部可穿戴探测器中，这有助于训练集划分不同的睡眠阶段，并有可能用于治疗运动障碍和癫痫发作的神经假体应用[ 141 ]。吉日等人。基于头部的单个 3D 加速度计实现了更高精度的睡眠阶段分析 [ 142 ]。然而，睡眠阶段识别的准确率仅为74.6%。对于 REM 睡眠，准确度仅为 52.7%。因此，手腕和躯干仍然是传感器放置的理想位置。

When a person moves, vibrations are transmitted to the bedding. So, it is also common to prevent mechanical sensors in bed sheets, pillows, and other locations (Figure 10c). Umetani et al. integrated an IoT system in a comforter that can measure the person’s movement and the bedding to improve sleep quality and prevent accidents during sleep [143]. Xin et al. used a flexible piezoelectric material, polyvinylidene fluoride, to create a flexible piezoelectric film that was placed on a pillow to convert the human force on the pillow into an electrical signal. These methods avoid the occlusion of the quilt in optical methods [144]. Xu et al. integrated a piezoelectric film in the mattress, using PVDF material, with a sensing area of 0.114 m² and a thickness of only 0.28 mm [145]. It can detect motion signals in a large area and instantly alert the elderly in case of abnormal sleep.
当人移动时，振动会传递到床上用品。因此，在床单、枕头和其他位置防止机械传感器也很常见（图 10 c）。梅塔尼等人。在被子中集成了物联网系统，可以测量人的运动和床上用品，以改善睡眠质量并防止睡眠期间发生事故[ 143 ]。辛等人。使用柔性压电材料聚偏二氟乙烯制造出柔性压电薄膜，将其放置在枕头上，将人体作用在枕头上的力转换为电信号。这些方法避免了光学方法中被子的遮挡[ 144 ]。徐等人。在床垫中集成了压电薄膜，采用PVDF材料，感应面积为0.114 m ² ，厚度仅为0.28 mm[ 145 ]。它可以大范围检测运动信号，一旦出现睡眠异常，立即提醒老人。

4.2. Posture Detection 4.2.姿势检测

Sleep position also has a great impact on sleep quality. A person’s tongue may fall under the influence of gravity when relaxed, obstructing the airway in some positions or under specific conditions, which can lead to snoring or even sleep apnea. Since accelerometers can sense the direction of gravity, many of the studies mentioned in the previous section have detected motion along with pose (Figure 11b) [142]. In 2007, Kishimoto et al. placed accelerometers on the user’s chest to accurately distinguish whether the user was in a supine, prone, or lateral sleeping position compared to sensors on the wrist or lateral sleeping position and could analyze the user’s sleep and wake times based on movement [87]. In 2015, Heenam et al. used patch accelerometers and achieved an average agreement of 99.16% for sleep position assessment [146]. Research on pose detection alone also has important implications in Figure 11.
睡眠姿势对睡眠质量也有很大影响。人的舌头在放松时可能会受到重力的影响而下垂，在某些姿势或特定条件下会阻塞气道，从而导致打鼾甚至睡眠呼吸暂停。由于加速度计可以感知重力方向，因此上一节中提到的许多研究都检测到了运动和姿势（图11b ）[ 142 ]。 2007 年，岸本等人。将加速度计放置在用户胸部，与手腕上的传感器或侧睡位置相比，可以准确地区分用户是否处于仰卧、俯卧或侧睡位置，并可以根据运动分析用户的睡眠和醒来时间[ 87 ]。 2015 年，Heenam 等人。使用贴片加速度计进行睡眠姿势评估，平均一致性达到 99.16% [ 146 ]。仅对姿势检测的研究在图 11中也具有重要意义。

Open in a separate window

Figure 11 图11

(a) Camera images used to analyze sleep posture [5]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2019 by the authors, published by Inst Electrical Electronics Engineers INC. (b) The acceleration sensor realizes pose analysis by detecting the direction of gravity [147]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2021 by the authors, published by MDPI. (c) Stress sensor stuck to the position vulnerable to pressure when sleeping [148]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2021 by the authors, published by Springer Nature.
( a )用于分析睡眠姿势的相机图像[ 5 ]。根据 CC-BY 知识共享署名许可条款转载，作者版权所有 2019，Inst Electrical Electronics Engineers INC 出版。( b ) 加速度传感器通过检测重力方向实现姿态分析[ 147 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2021，由 MDPI 发布。 ( c )压力传感器粘在睡觉时容易受到压力的位置[ 148 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2021，由 Springer Nature 出版。

An infrared camera is an ideal method to analyze human posture. An infrared camera can record human posture without a visible light source, and infrared light has good penetration. Insung et al. used an infrared camera to analyze the effect of sleeping posture improvement on sleep apnea [149]. Cheung et al. used an infrared sensor to monitor the movement of the sleeping elderly and alert the healthcare personnel in times when there is bad activity [150]. Non-contact sleep monitoring based on infrared cameras differed from sleep monitoring devices in the identification of sleep quality by only 4.7%. Infrared array sensors under laboratory conditions are more than 95% accurate in sleep detection [151]. Using infrared sensors together with microwave sensors, the overall accuracy of sleep cycle measurements can be as high as 98% [152].
红外热像仪是分析人体姿势的理想方法。红外摄像机无需可见光源即可记录人体姿态，且红外光具有良好的穿透力。仁成等人。使用红外热像仪分析睡姿改善对睡眠呼吸暂停的影响[ 149 ]。张等人。使用红外传感器来监测熟睡的老年人的运动，并在出现不良活动时提醒医护人员[ 150 ]。基于红外摄像头的非接触式睡眠监测与睡眠监测设备在睡眠质量的识别上仅相差4.7%。实验室条件下的红外阵列传感器睡眠检测准确率超过 95% [ 151 ]。将红外传感器与微波传感器结合使用，睡眠周期测量的总体准确度可高达 98% [ 152 ]。

Force sensors also have many roles in this area. As shown in Figure 11c, a pressure sensor made of a multilayer piezoelectric structure can detect which part of the body is touching the bed and subjected to body gravity [148]. In 2022, Zhang et al. prepared resistive flexible angle sensors using metal foil foils, and the accuracy of flexible wearable sleep posture monitoring devices exceeded 90% [153]. Zhou et al. also achieved sensing of human posture through a bed sheet made of ultra-thin conductive fabric. The sensors can divide the bed into a total of 60 zones and detect in which zones the body’s pressure is located [154].
力传感器在这一领域也发挥着许多作用。如图11c所示，由多层压电结构制成的压力传感器可以检测身体的哪个部分正在接触床并受到身体重力[ 148 ]。 2022 年，Zhang 等人。利用金属箔制备电阻式柔性角度传感器，柔性可穿戴睡眠姿势监测设备的准确度超过90%[ 153 ]。周等人。还通过超薄导电织物制成的床单实现了人体姿势的传感。传感器可以将床分为总共 60 个区域，并检测身体压力位于哪些区域 [ 154 ]。

4.3. Sleep Bruxism Detection
4.3.睡眠磨牙症检测

During sleep, 8% of the population has reported awareness of tooth grinding [155]. Sleep bruxism also represents the third most frequent parasomnia [155]. People in a state of high mental tension and psychological stress may maintain an excited state of the occlusal muscles at night and experience nocturnal teeth-grinding symptoms [155,156,157,158]. Subjects with obstructive sleep apnea syndrome, loud snorers, subjects with moderate daytime sleepiness, heavy alcohol drinkers, caffeine drinkers, and smokers are at higher risk of reporting sleep bruxism [156]. In 2021, Lee et al. integrated sensors such as accelerometers and gyroscopes in a jaw advancement device used to improve sleep apnea to help monitor the occurrence of sleep apnea and teeth grinding and to improve the effectiveness of related treatment devices [159].
在睡眠期间，8% 的人表示有磨牙的意识 [ 155 ]。睡眠磨牙症也是第三种最常见的异态睡眠[ 155 ]。精神高度紧张、心理压力较大的人，夜间可能会保持咬合肌的兴奋状态，出现夜间磨牙症状[ 155、156、157、158 ]。患有阻塞性睡眠呼吸暂停综合征、打鼾者、白天中度嗜睡的受试者、大量饮酒者、咖啡因饮用者和吸烟者出现睡眠磨牙症的风险较高[ 156 ]。 2021 年，Lee 等人。在用于改善睡眠呼吸暂停的下颌推进装置中集成加速度计和陀螺仪等传感器，以帮助监测睡眠呼吸暂停和磨牙的发生，并提高相关治疗装置的有效性[ 159 ]。

Muscle electrical signals can show the activity of local muscles, but mechanical methods can more truly and directly detect the occlusion method of teeth, as shown in Figure 12. D’Addona et al. measured stresses in the human mouth using a Wheatstone bridge to detect changes in resistance and force output from a miniature strain gauge. When a person undergoes nocturnal teeth grinding, the strain gauges are stressed. The resistance changes and is amplified by the Wheatstone bridge into a voltage signal that can be collected [160]. In 2022, Coimbra et al. used light Bragg grating sensors, encapsulated in a PDMS, to also make wearable pressure sensors that detect different signals from a person biting a splint during an episode of teeth grinding [161]. In 2021, Jucevicius et al. used permanent magnets and a triaxial magnetometer. The magnetic field generated by the permanent magnets at the magnetometer changes when the spatial position relationship between the mandible and maxilla changes, based on which the movement of the jaw joint can be monitored [162]. In 2022, O’Hare et al. used pressure sensors to detect the deformation of the occlusal muscles, which are smaller than myoelectric sensors that are smaller and more accurate in the analysis of occlusal forces. The results of these works are difficult to achieve with myoelectric signals and exemplify the need for mechanical sensors [163].
肌肉电信号可以显示局部肌肉的活动情况，但机械方法可以更真实、更直接地检测牙齿的咬合情况，如图12所示。达多纳等人。使用惠斯通电桥测量人体口腔中的应力，以检测微型应变计的阻力和力输出的变化。当一个人在夜间磨牙时，应变计会受到压力。电阻发生变化并被惠斯通电桥放大成可以收集的电压信号[ 160 ]。 2022 年，科英布拉等人。使用封装在 PDMS 中的光布拉格光栅传感器来制造可穿戴压力传感器，该传感器可以检测在磨牙期间咬夹板的人发出的不同信号[ 161 ]。 2021 年，Jucevicius 等人。使用永磁体和三轴磁力计。当下颌和上颌之间的空间位置关系发生变化时，磁力计上的永磁体产生的磁场发生变化，据此可以监测颌关节的运动[ 162 ]。 2022 年，奥黑尔等人。使用压力传感器来检测咬合肌肉的变形，该传感器比肌电传感器更小，在咬合力的分析中更准确。这些工作的结果很难用肌电信号来实现，并且例证了对机械传感器的需求[ 163 ]。

Open in a separate window

Figure 12 图12

Different methods for detecting bruxism. (a) Magnetic field based molar detection [162]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2021 by the authors, published by MDPI. (b) The system, composed of strain gauge and microprocessor, is packaged in the tooth socket to form a bite force sensor. The results are sent to the mobile app [164]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2019 by the authors, published by MDPI.
检测磨牙症的不同方法。 ( a )基于磁场的摩尔检测[ 162 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2021，由 MDPI 发布。 ( b ) 该系统由应变片和微处理器组成，封装在牙槽中形成咬合力传感器。结果被发送到移动应用程序[ 164 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2019，由 MDPI 发布。

4.4. Mechanical Breath Detection
4.4.机械呼吸检测

In addition to the extreme case of sleep apnea, breathing rate and lung capacity are also important physiological information. In 2003, Atanasov et al. found a strong link between nasal cycles and sleep cycles. It is due to the regular circulation of nasal airflow through the nostrils caused by nasal congestion and congestion. In REM sleep, the nasal cycle is synchronized with the sleep cycle [165]. In 2006, Kohler et al. analyzed the effects of different drugs on nocturnal breathing and sleep by detecting airflow and nasal pressure in the right and left nasal passages with three sensors, respectively. Sleep apnea is a common sleep-related disorder that poses a significant threat to the life and health of patients [166]. There have been many studies to detect the occurrence of sleep apnea by heart rate, blood oxygen, and electromyographic signals, which are described in the corresponding subsection. However, the detection of respiratory airflow remains the essential method. Since mechanical signals are important physiological indicators of the respiratory system, relevant sensing is important for both sleep quality monitoring and guidance of treatment [167].
除了睡眠呼吸暂停这种极端情况外，呼吸频率和肺活量也是重要的生理信息。 2003 年，阿塔纳索夫等人。发现鼻周期和睡眠周期之间存在密切联系。它是由于鼻气流经鼻孔有规律的循环而引起鼻塞、充血。在快速眼动睡眠中，鼻周期与睡眠周期同步[ 165 ]。 2006 年，科勒等人。通过用三个传感器分别检测左右鼻道的气流和鼻压，分析了不同药物对夜间呼吸和睡眠的影响。睡眠呼吸暂停是一种常见的睡眠相关疾病，对患者的生命和健康构成重大威胁[ 166 ]。已有许多研究通过心率、血氧和肌电信号来检测睡眠呼吸暂停的发生，相关小节对此进行了描述。然而，呼吸气流的检测仍然是重要的方法。由于机械信号是呼吸系统的重要生理指标，因此相关传感对于睡眠质量监测和治疗指导都很重要[ 167 ]。

Teichtahl et al. used thermistors and nasal pressure sensors to analyze whether humans were breathing and to assess sleep apnea well [168]. In 2021, Moshizi et al. prepared nano-complex airflow sensors by growing graphene nanosheets on the surface of PDMS with good sensitivity and linearity [169]. NP thermistor is a common respiratory detection sensor. The detection principle is that the air pressure in the nasal cavity and the temperature at the nostril changes when a person is breathing. In 2018, Jiang et al. combined a respiratory monitor with a motion sensor to more comprehensively screen for sleep apnea [170]. The respiratory sensor designed by Vernon et al. takes advantage of the fact that changes in temperature and humidity in the mouth and nose during human breathing affect the signal of the acoustic surface wave sensor, as shown in Figure 13a [171]. Sleep apnea can be captured sensitively.
泰希塔尔等人。使用热敏电阻和鼻压传感器来分析人类是否在呼吸并很好地评估睡眠呼吸暂停[ 168 ]。 2021 年，Moshizi 等人。通过在PDMS表面生长石墨烯纳米片制备了纳米复合气流传感器，具有良好的灵敏度和线性度[ 169 ]。 NP热敏电阻是一种常见的呼吸检测传感器。检测原理是人呼吸时鼻腔内的气压和鼻孔处的温度发生变化。 2018 年，Jiang 等人。将呼吸监测器与运动传感器相结合，可以更全面地筛查睡眠呼吸暂停[ 170 ]。 Vernon 等人设计的呼吸传感器。利用了人类呼吸过程中口鼻温度和湿度的变化影响声表面波传感器的信号，如图13a [ 171 ]所示。可以灵敏地捕获睡眠呼吸暂停。

Open in a separate window

Figure 13 图13

(a) Surface acoustic wave sensors [171]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2022 by the authors, published by AIP Publishing. (b) A method for measuring thoracic volume using chest marker loci [176]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2022 by the authors, published by Springer Nature.
( a )表面声波传感器[ 171 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2022，由 AIP Publishing 出版。 ( b )使用胸部标记位点测量胸廓体积的方法[ 176 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2022，由 Springer Nature 出版。

In addition, breathing behavior can be reflected by detecting thoracic motion. In 2011, Dehkordi et al. obtained signals about breathing by fixing an acceleration sensor on the sternum. It is also possible to correlate the sleeping position and screen for sleep apnea [172]. Jortberg et al. fixed accelerometers on the chest and measured the respiratory rate (RR) with an average error of 1.84 breaths per minute [173]. In 2020, Yuzer et al. placed accelerometers on the diaphragm, which vibrate a motor on the wristband when the respiratory movement of the diaphragm is detected to stop, stimulating the patient to change the sleeping position until breathing resumes [174]. In 2021, Ghahjaverestan et al. measured the range of motion of the abdominal and thoracic cavities with acceleration and position sensors, respectively, to restore the respiratory signal more precisely [175]. Stubbe et al. fixed 12 markers on the user’s thorax, used an infrared camera to locate the markers, and calculated the volume of the thorax, shown in Figure 13b [176]. This method is known as optoelectronic plethysmography (OEP). The spirometry measured by this method has an error of only 0.4% with the spirometer, and the mechanics sensor provides richer information than the bioelectrical signal that can only measure respiratory rate.
此外，呼吸行为可以通过检测胸部运动来反映。 2011 年，Dehkordi 等人。通过在胸骨上固定加速度传感器来获取有关呼吸的信号。还可以将睡眠姿势与睡眠呼吸暂停筛查相关联[ 172 ]。约特伯格等人。将加速度计固定在胸部并测量呼吸频率（RR），平均误差为每分钟 1.84 次呼吸 [ 173 ]。 2020 年，Yuzer 等人。将加速度计放置在隔膜上，当检测到隔膜的呼吸运动停止时，加速度计会振动腕带上的电机，刺激患者改变睡姿直到呼吸恢复[ 174 ]。 2021 年，Ghahjaverestan 等人。分别用加速度和位置传感器测量腹腔和胸腔的运动范围，以更精确地恢复呼吸信号[ 175 ]。斯图贝等人。在用户的胸部固定了12个标记，使用红外相机定位标记，并计算出胸部的体积，如图13b [ 176 ]所示。这种方法被称为光电体积描记法（OEP）。这种方法测量的肺活量与肺活量计的误差仅为0.4%，而且力学传感器比只能测量呼吸频率的生物电信号提供了更丰富的信息。

4.5. Blood Flow Detection
4.5.血流检测

BP and HR vary with circadian rhythms. Prolonged stress may lead to increased HR during sleep [177]. Although heart rate information can also be obtained from ECG, obtaining heart rate in the scheme without bioelectrical sensing is still meaningful. Moreover, continuous blood pressure detection, including sleep time, is relevant for diagnosing and treating hypertension [178]. Blood pressure also rises at the end of obstructive episodes in patients with sleep apnea, which is undoubtedly dangerous for patients with both obstructive sleep apnea and vascular disease such as hypertension [179]. In 2000, Dimsdale et al. studied the effect of continuous positive airway pressure therapy on blood pressure in patients with obstructive sleep apnea through nocturnal blood pressure monitoring. The significance of nocturnal blood pressure monitoring was demonstrated [180]. However, with an automated device, early detection techniques were based on measuring the patient’s blood pressure every 15 min. A traditional arm band balloon blood pressure detection method was used. Such a method does not allow continuous monitoring, and the sudden working of the air pump and the squeezed arm can affect the patient’s sleep quality.
血压和心率随昼夜节律变化。长时间的压力可能会导致睡眠期间心率增加[ 177 ]。虽然也可以从心电图中获取心率信息，但在没有生物电传感的方案中获取心率仍然是有意义的。此外，连续血压检测，包括睡眠时间，与诊断和治疗高血压相关[ 178 ]。睡眠呼吸暂停患者在阻塞性发作结束时血压也会升高，这对于同时患有阻塞性睡眠呼吸暂停和高血压等血管疾病的患者来说无疑是危险的[ 179 ]。 2000 年，迪姆斯代尔等人。通过夜间血压监测，研究了持续气道正压通气治疗对阻塞性睡眠呼吸暂停患者血压的影响。夜间血压监测的重要性已得到证实[ 180 ]。然而，使用自动化设备时，早期检测技术是基于每 15 分钟测量一次患者的血压。采用传统的臂带气球血压检测方法。这种方法不能进行连续监测，并且气泵的突然工作和挤压手臂会影响患者的睡眠质量。

In 2006, Kaniusas et al. used a magnetoelastic skin curvature sensor to measure carotid blood pressure, enabling continuous blood pressure measurement at night. However, this method has limited accuracy, with a correlation coefficient of less than 0.9 between measured and reference values [181]. Such stress-based blood pressure sensors can cause discomfort from local compression when worn for long periods, both in the fingers and in the arm. Muscle movement and postural changes can alter the mechanical environment to affect measurements [182]. A more non-sensitive blood pressure monitoring that people can wear for long periods at night and does not interfere with their sleep quality is the only truly usable technology for nighttime blood pressure monitoring.
2006 年，卡尼萨斯等人。使用磁弹性皮肤曲率传感器测量颈动脉血压，实现夜间连续血压测量。然而，该方法的精度有限，测量值与参考值之间的相关系数小于0.9[ 181 ]。这种基于压力的血压传感器在长时间佩戴时，可能会因手指和手臂的局部压迫而引起不适。肌肉运动和姿势变化可以改变机械环境，从而影响测量结果[ 182 ]。一种更加不敏感的血压监测装置，人们可以在夜间长时间佩戴，并且不会干扰他们的睡眠质量，这是夜间血压监测唯一真正可用的技术。

Heart rate and blood flow sensing based on optical signals have the advantages of being non-invasive and having low wearing requirements. Photodensitometry, a technique widely used to monitor blood volume changes according to the Lambert–Beer law, is more suitable for home use [183]. This can be used for heart rate monitoring, allowing sleep staging. A type of blood volume sensor—a technology known as photoplethysmography (PPG)—is also widely used [184]. From this, the heart rate and other physiological parameters can be extracted to inform about user activity, fitness, sleep, and health, as shown in Figure 14. However, sleep staging based solely on heart rate measured by PPG waves has a low classification accuracy of between 55% and 78%. Sleep staging combining exercise and heart rate also has an accuracy of 78.2% and cannot replace sleep polysomnography [185,186]. The recognition accuracy can be improved to more than 93% by machine-learning techniques such as data augmentation and convolutional neural networks [187].
基于光信号的心率和血流传感具有非侵入性、佩戴要求低的优点。光密度测定法是一种根据兰伯特-比尔定律广泛用于监测血容量变化的技术，更适合家庭使用[ 183 ]​​。这可用于心率监测，从而实现睡眠分级。一种血量传感器——一种称为光电体积描记法（PPG）的技术——也被广泛使用[ 184 ]。由此，可以提取心率和其他生理参数以了解用户活动、健身、睡眠和健康状况，如图14所示。然而，仅根据 PPG 波测量的心率进行睡眠分期的分类准确度较低，在 55% 至 78% 之间。结合运动和心率的睡眠分期也具有 78.2% 的准确度，并且不能取代睡眠多导睡眠图 [ 185 , 186 ]。通过数据增强和卷积神经网络等机器学习技术可以将识别准确率提高到93％以上[ 187 ]。

Open in a separate window

Figure 14 图14

(a) Graphical definition of fiducial points detected from photoplethysmogram (PPG), velocity plethysmogram (VPG), and acceleration plethysmogram (APG) signals [188]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2022 by the authors, published by MDPI. (b) Optical sensor for detecting blood flow in fingers [189]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2021 by the authors, published by Springer Nature.
（ a ）从光电体积描记图（PPG）、速度体积描记图（VPG）和加速度体积描记图（APG）信号检测到的基准点的图形定义[ 188 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2022，由 MDPI 发布。 ( b )用于检测手指血流的光学传感器[ 189 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2021，由 Springer Nature 出版。

Although less accurate independently as a basis for sleep staging, several works have reported the detection of blood flow under the skin at night by light sensing and recording the time of pulse passage (PTT), which can restore blood pressure information by biomechanical models. Blood pressure can be calculated by measuring the time the pulse wave travels through the blood vessels [178]. Measurements can be made using the difference in the speed of propagation of the ECG signal and the pulse wave signal. Shahrbabaki et al., based on fixing sensors at the fingertips [190], and Kireev et al., based on electronic skin [117], respectively, achieved reliable nocturnal measuring. Zadi et al. performed the error of blood pressure measurement by PTT in different states, and the mean value of the model residuals was considered to be less than 3.2 mm Hg during both normal breathing and breath-holding maneuvers [191]. Krefting, based on this technique, successfully observed elevated blood pressure due to different postures at night [192]. Carek et al. integrated sensors in underpants, detected the signal when blood pulses flowed through the legs, and correlated with cuffed sensors [193]. PTT-based blood pressure functions have been integrated into smart watches [194].
虽然独立作为睡眠分期的基础不太准确，但有几篇论文报道了通过光传感和记录脉搏通过时间（PTT）来检测夜间皮下血流量，这可以通过生物力学模型恢复血压信息。血压可以通过测量脉搏波穿过血管的时间来计算[ 178 ]。可以利用心电图信号和脉搏波信号的传播速度的差异来进行测量。 Shahrbabaki等人基于将传感器固定在指尖[ 190 ]，Kireev等人基于电子皮肤[ 117 ]，分别实现了可靠的夜间测量。扎迪等人。等人在不同状态下进行 PTT 血压测量的误差，正常呼吸和屏气动作时模型残差的平均值被认为小于 3.2 mm Hg [ 191 ]。 Krefting基于该技术成功观察到夜间不同姿势导致的血压升高[ 192 ]。卡雷克等人。在内裤中集成传感器，检测血液脉冲流经腿部时的信号，并与袖带传感器相关联[ 193 ]。基于 PTT 的血压功能已集成到智能手表中[ 194 ]。

PTT measurements can also be performed by two mechanical sensors located in different parts of the body. Thin-film piezoelectric sensors give the possibility of wearability of this sensing. Xin et al. achieved blood pressure monitoring with a standard deviation of only 1.7 mm Hg using two flexible sensors made of PVDF piezoelectric material. Resonant amplifier circuits were designed, too [195]. PTT measurements were also implemented by Fan et al. using textile electronics [196].

Mechanics signals can also be used for heartbeat detection. Heart rate measurements were performed by Sanchez et al. by collecting vibration signals in the chest cavity [197]. Xin et al. collected signals from pulse vibrations using thin films made of flexible piezoelectric materials [144]. The ultra-thin conductive textile bed sheet made by Zhou et al. has a wide operating frequency bandwidth range of 0 Hz to 40 Hz, good mechanical durability, and washability [154]. Its high sensitivity allows the sensor to obtain heart rate measurements with an error of only 1.33% without wearing any wearable device specifically. However, HR detection by mechanical methods is not common because optical sensors have been able to monitor HR effectively and accurately.

In the last method, Van et al. monitored the heart rate of users based on an infrared camera, using three different frequencies of infrared sensors and a broad spectrum of infrared light sources to achieve transmission of the occlusion [198]. The heart rate detection accuracy was 92%. In addition, because the camera does not touch the body at all, it is not disconnected by the person’s movement, as is the case with ECG. After the person’s posture changes, the camera can also be positioned again in time and continue to measure heart rate.

4.6. Acoustic Detection

The human body has many mechano-acoustic (MA) signals [199]. Snoring is a common phenomenon in sleep, which is caused by the poor ventilation of the airway and has a significant impact on the quality of sleep and in some cases may lead to sleep apnea or even asphyxia. Dafna et al. analyzed the sleep of patients with obstructive sleep apnea by recording their sleep sounds [200]. The accuracy of snoring frequency monitoring can be improved from 81% to 89% by presetting age and gender information [201].

Human sleep is also susceptible to the effects of environmental sounds. Many sleep monitoring systems now integrate ambient sound acquisition to help people analyze the causes of poor sleep. Some sounds may prevent a person from falling asleep, while white noise has shown sleep-aiding effects in many studies [202]. Chen et al. integrated ambient sound sensors into the sleep monitoring system, which automatically plays white noise to mask the noise affecting people’s sleep when ambient noise is detected, and the body sensors in the system monitor sleep quality in real time [203].

If just collecting the sound in the room, perhaps no special equipment is needed. Snoring signals have strong penetration, so sensing these can often be implemented directly with the help of a smartphone [202]. Xin et al. prepared acoustic sensors using flexible piezoelectric films to detect snoring, but there is no significant advantage over existing integrated microphones [144,204].

However, in some cases, acoustic sensors close to the skin can give more in vivo information. Ghahjaverestan et al. analyzed respiratory airflow by measuring the sound of tracheal airflow through a microphone close to the skin and used it for the assessment of sleep apnea [175]. Li et al. designed a sensor that can be attached to the chest [199]. It is very sensitive to snoring, and the measured frequency information can be interpreted by human anatomy.

4.7. Other Mechanical Detection

In men, penile engorgement and erection occur during REM sleep and are important physiological phenomena accompanying the sleep cycle [205]. There is also an association between erectile dysfunction and sleep disorders [206,207]. In 2021, Krkovich et al. enabled monitoring of erections during sleep by recording the diameter of the user’s penis. The results obtained made it possible to determine reference values for qualitative and quantitative indicators of PNT in healthy male volunteers [208]. In 2022, Edgar proposed several forms of sensors that could be used to monitor nocturnal sleep erections: a penile arterial pulse for measuring plethysmograph, a displacement sensor to measure axial length, a strain gauge to measure radial stiffness and circumference, and a temperature sensor to measure skin and cavernosal temperature [209]. In 2022, Heo et al. used an electronic fabric strain sensor to replace the bulky and heavy Rigiscan device in Figure 15. By immobilizing carbon nanotubes on the fabric, resistive length, perimeter, and curvature measurements were achieved. The result shows a 1.44% error rate and a cavity radius of 110 to 300.

Open in a separate window

Figure 15

(a) A film that detects strain. (b) A combination of strain sensors that detects changes in penis length and diameter [210]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2021 by the authors, published by MDPI.

Rapid eye movements are a hallmark feature of REM sleep and can reflect how neurologically active a person is under that stage of sleep. Although, the occlusion of the eyelids makes eye movements less easily observable, so the EOG in the previous section is a more common way to detect eye movements. Many studies have reported direct mechanical measurements of eye movements. In 2020, Wu et al. designed a smart eye mask [211]. The hydrogel sensor was integrated with a sleep mask for real-time monitoring of human sleep. Compared to the sleep recorded using a popular sleep monitoring mobile app that measures sleep only based on body movements and voices, the sleeping process measured using the smart sleep mask shows much higher reliability for the recognition of REM sleep. In 2021, Dang et al. used an infrared optical sensor integrated with an eyecup to detect eyelid-surface-shaped edges caused by eye movements and used an array of four sensors to each detecting motion in two vertical degrees of freedom [212].

The last mechanical signal is the intraocular pressure (IOP). Continuous monitoring of IOP, especially during sleep, remains a great challenge for glaucoma care. Zhang et al. designed contact lenses with integrated strain sensors and induction coils that allow continuous IOP monitoring at night but with some discomfort [213]. However, very mature, formally usable studies have not been reported. The important difficulties are the passive wireless devices and the lack of oxygen caused by wearing the device all night.

5.1. O₂ Level Detection

Blood oxygenation is a common physiological indicator. In 2013, Elizur et al. identified the effects of hypoxemia on glucose metabolism during REM sleep [221]. In 2014, the patterns of brain tissue oxygen content changes in adults and adolescents during different sleep stages were revealed by NIRS [222]. In 2022, Elmenhorst et al. used blood oxygen sensors to analyze the sleep quality of long-haul flight crews at high altitudes and investigated the effect of a hypoxic environment on sleep at high altitudes, which has important implications for civil aviation safety [223,224,225,226].

The oxygen content of the different parts of human vasculature is different. Easily measured and well-referenced is the percutaneous arterial oxygen saturation (SpO₂). Cakmak et al. detected obstructive sleep apnea with the help of an optical blood oxygen sensing device. This relies on the different absorption rates of light by hemoglobin in the human body before and after binding oxygen, which in turn monitors the oxygen content within the blood through a reflected light sensor [227]. The finger is rich in capillaries, which is a common location for blood oxygen detection [226]. The dual-channel continuous oxygen saturation sensor designed by Zhang et al. explored different types of fingers as well as different wearing positions. It ended up with a root-mean-square error of only 1.8 [228]. The correlation coefficient tested by Tran et al. reached 0.93, with a 95% agreement limit of ±2.5% [229].

Because optical signals can easily detect oxygen levels in the blood at any capillary, various types of wearable sensing devices can be used. Earlobe sensors [230] and ear canal sensors [231] can be developed without being limited to common locations such as the wrist, as shown in Figure 17. Among them, brain tissue oxygen saturation is very important for the quality of sleep, and Metz et al. designed measurement of this using near-infrared spectroscopy to detect the oxygen saturation of brain tissue before and after human sleep [222,232], since the oxygen content of human arterial and venous vessels is not the same, and the veins at the arms may interfere with the results. Capillary-based monitoring in areas such as between the fingers, which is 2–3% higher than the armed vessel oxygen content test, is more suitable as a measure of sleep apnea [233]. Nabavi et al. used intraoral photoplethysmography and showed more than 96% accuracy in estimating physiological characteristics such as SpO₂ compared to conventional monitoring techniques [234].

Open in a separate window

Figure 17

Positioning of the pulse oximeter sensor. (a) Forehead [224]. Reproduced with permission, Copyright 2004 Blackwell Publishing Ltd. (b) In ear [225]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2020 by the authors, published by MDPI. (c) Finger [226]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2021 by the authors, published by MDPI.

In 2021, Van et al. used a broad-spectrum infrared light source to enhance the infrared signal and used three infrared cameras with different frequencies to measure the reflected light signal and calculate the blood oxygen concentration. Based on this, non-contact infrared measurements were achieved, avoiding the detachment of the sensing device from the body due to motion. The non-contact oximeter estimated blood oxygen values with an 89% time error within four 4% [198].

5.2. CO₂ Level Detection
5.2. CO ₂水平检测

The carbon dioxide level is another important test. Insufficient sleep breathing at night may cause hypercapnia, leading to respiratory failure. This process occurs mainly due to increased carbon dioxide levels in blood vessels, such as arteries, due to hypoventilation, in relation to the person’s height, body mass index, and the degree of obstruction of the upper airway [235]. In cardiac patients, the partial pressure of carbon dioxide at night alters their pathophysiology during the day and night.
二氧化碳水平是另一个重要的测试。夜间睡眠呼吸不足可能会引起高碳酸血症，导致呼吸衰竭。这一过程的发生主要是由于通气不足导致血管（例如动脉）中二氧化碳水平增加，与人的身高、体重指数和上呼吸道阻塞程度有关[ 235 ]。对于心脏病患者，夜间二氧化碳分压会改变他们白天和夜间的病理生理学。

Ramos et al. measured carbon dioxide levels, volatile organic compounds (VOCs), and air temperature in indoor environments using low-cost gas sensors [236]. Rauhala et al. used electromechanical film sensors, a flexible material that can analyze carbon dioxide concentrations in blood vessels on the skin’s surface through differences in electrical signals of the skin’s surface and analyze increased carbon dioxide concentrations due to sleep apnea [237]. Kang et al. used gas sensors to measure the concentration of breathing gases in human bodies [238]. Chhajed et al. used an earlobe carbon dioxide sensor to monitor nocturnal carbon dioxide concentrations and monitored the effectiveness of positive pressure ventilation for chronic hypercapnia. This took advantage of the fact that small carbon dioxide molecules have high tissue solubility and can diffuse rapidly through the skin [239]. Tipparaju et al. improved the accuracy of a transdermal continuous carbon dioxide sensor by solving humidity interference with a miniature non-dispersive sensor through a hydrophobic membrane, as shown in Figure 18.
拉莫斯等人。使用低成本气体传感器测量室内环境中的二氧化碳水平、挥发性有机化合物（VOC）和空气温度[ 236 ]。劳哈拉等人。使用机电薄膜传感器，这是一种柔性材料，可以通过皮肤表面电信号的差异来分析皮肤表面血管中的二氧化碳浓度，并分析由于睡眠呼吸暂停而增加的二氧化碳浓度[ 237 ]。康等人。使用气体传感器来测量人体呼吸气体的浓度[ 238 ]。查吉德等人。使用耳垂二氧化碳传感器监测夜间二氧化碳浓度，并监测正压通气对慢性高碳酸血症的有效性。这是利用了二氧化碳小分子具有高组织溶解度并且可以通过皮肤快速扩散的事实[ 239 ]。蒂帕拉朱等人。通过疏水膜解决微型非分散传感器的湿度干扰，提高了透皮连续二氧化碳传感器的精度，如图18所示。

Open in a separate window

Figure 18 图18

Wearable transcutaneous CO₂ monitor based on miniaturized non-dispersive infrared sensor. (a) Image of the integrated wristband, ø20.9 mm × 18.1 mm. (b) The zoom-in shows cross-section view of the gas chamber, sealing O-ring, and wristband body. (c) Miniaturized Cozir NDIR CO₂ sensor [218]. Reproduced under the terms of the CC-BY Creative Commons Attribution License, Copyright 2021 by the authors, published by Inst Electrical Electronics Engineers INC.
基于小型化非色散红外传感器的可穿戴经皮CO ₂监测仪。 ( a ) 集成腕带的图像，ø20.9 mm × 18.1 mm。 ( b ) 放大显示气室、密封 O 形圈和腕带主体的横截面图。 ( c )小型化Cozir NDIR CO ₂传感器[ 218 ]。根据 CC-BY 知识共享署名许可条款复制，作者版权所有 2021，由 Inst Electrical Electronics Engineers INC 出版。

5.3. Hormone Detection 5.3.激素检测

The human body has many hormones that play a regulatory role in the sleep process. For example, melatonin plays an important role in regulating circadian rhythms in humans. Through feedback to light, the accumulation of melatonin in the body will make people sleepy [240,241,242]. Melatonin is also now available to treat some insomnias [243]. In addition, diseases such as adrenal hyperplasia or a prolonged state of emergency may lead to an overproduction of hormones such as adrenal hormone, norepinephrine, and adrenocortical. These hormones can put a person in a hyperactive state and cause difficulty falling asleep, poor sleep quality, and easy awakening [244,245].
人体有多种激素在睡眠过程中发挥调节作用。例如，褪黑激素在调节人类昼夜节律方面发挥着重要作用。通过对光的反馈，体内褪黑激素的积累会使人犯困[ 240,241,242 ]。褪黑激素现在也可用于治疗某些失眠症[ 243 ]。此外，肾上腺增生等疾病或长期处于紧急状态，可能会导致肾上腺素、去甲肾上腺素、肾上腺皮质等激素分泌过多。这些激素会使人处于过度活跃的状态，导致入睡困难、睡眠质量差、容易惊醒[ 244 , 245 ]。

Multi-hormones can be detected in human saliva through secretory glands and thus enable non-invasive sensing. Previously, the level of hormone detection in non-invasive samples was mainly performed by sending the samples to the laboratory and detecting them by fluorescent probe method without real time. Massey et al. prepared a non-invasive body fluid sensor based on EG-FET to monitor the cortisol hormone concentration in saliva samples. The detectable cortisol concentration range is currently identified as 27.3 pM–27.3 μM [246]. Shahub et al., on a nanoporous matrix by electrochemical impedance spectroscopy, measured cortisol concentrations in sweat with 100% accuracy and 0% false negatives, with a dynamic range of 8–140 ng/mL [247].
可以通过分泌腺检测人类唾液中的多种激素，从而实现非侵入性传感。此前，无创样本中激素水平的检测主要是将样本送到实验室，采用荧光探针法进行检测，无法实时进行。梅西等人。制备了基于EG-FET的非侵入性体液传感器来监测唾液样本中的皮质醇激素浓度。目前可检测的皮质醇浓度范围为 27.3 pM–27.3 μM [ 246 ]。 Shahub 等人通过电化学阻抗谱在纳米多孔基质上测量了汗液中的皮质醇浓度，准确度为 100%，假阴性率为 0%，动态范围为 8–140 ng/mL [ 247 ]。

Sensor research at the hormone level is less compared to the rest of the field, but with the maturation of wearable chemical sensor technology, the future promises to provide more health information at the secretory system and drug therapy level.
与其他领域相比，激素层面的传感器研究较少，但随着可穿戴化学传感器技术的成熟，未来有望在分泌系统和药物治疗层面提供更多健康信息。

6.1. Multi-Signal, Single Physiological Information

The first category uses multiple sensors of different types to jointly analyze a particular physiological activity, such as REM sleep detection, sleep apnea detection, etc. Models that analyze sleep based on individual sensor data may vary with the underlying conditions of different individuals. For example, the EEG sensor and model designed by Sharma et al. achieved an accuracy of 83% in the sleep stage classification test for healthy individuals [256]. However, the same method yielded an accuracy of only 72% for patients with an REM disorder. Different types of sleep-related disorders have an impact on the results. Deep-learning methods have limited effectiveness in solving this problem [35]. In this case, introducing the rest of the sensors for auxiliary classification is often necessary. Another method is to use multiple signals for sleep stage classification.

Sleep apnea, sleep stage division, and other physiologies with multiple signal presentations are the most common. A relatively simple monitoring such as respiratory rate can also be boosted with a multi-channel signal. Combining different broad categories of signals helps better complement each other but also adds more cost and invariance to the use. For the identification of sleep apnea, the best studies have increased the accuracy to 100%. Respiratory rate monitoring can also have an error of less than one per minute. The complete information is provided in Table 3. The future direction of this type of research will be comfortable to wear. In addition, there is still room for improvement in the truth rate of some behavioral monitoring.
睡眠呼吸暂停、睡眠阶段划分和其他具有多种信号呈现的生理现象是最常见的。相对简单的监测（例如呼吸频率）也可以通过多通道信号来增强。组合不同的大类信号有助于更好地相互补充，但也增加了使用的成本和不变性。对于睡眠呼吸暂停的识别，最好的研究已将准确性提高到 100%。呼吸频率监测的误差也可能小于每分钟一次。表 3提供了完整的信息。此类研究的未来方向将是佩戴舒适。此外，部分行为监测的真实率仍有提升空间。

6.2. Single Sensor, Multiple Physiological Information
6.2.单一传感器，多种生理信息

The second category is the implementation of different tests based on single-sensor hardware with different usages. Sometimes the sensor’s response may be related to several different mechanical, biochemical, etc., signals. For example, the PPG sensor can measure blood pressure and heart rate separately when measuring different values. Another example is the limb acceleration signal, which is a superposition of multiple signals such as posture, movement, and respiration and thus can be interpreted with multiple information. So, some studies, although based on different physiological signals, may ultimately be achieved using the same hardware in different ways.
第二类是基于不同用途的单传感器硬件实施不同的测试。有时，传感器的响应可能与几种不同的机械、生化等信号相关。例如，PPG传感器在测量不同值时可以分别测量血压和心率。再比如肢体加速度信号，它是姿势、运动、呼吸等多种信号的叠加，因此可以用多种信息来解释。因此，一些研究虽然基于不同的生理信号，但最终可能会使用相同的硬件以不同的方式实现。

The accuracy of heart rate and blood oxygen monitored by a single sensor has been reported in many studies. However, no new protocols have emerged from such studies, and several major protocols are relatively well-established. More accurate measurements or more comprehensive analyses require multi-bed sensor combinations.
许多研究都报道了通过单个传感器监测心率和血氧的准确性。然而，此类研究尚未出现新的方案，并且几个主要方案已经相对完善。更准确的测量或更全面的分析需要多床传感器组合。

The paper is divided into sections according to physiological signal categories, and this subsection adds results that are somewhat classified from a sensor perspective. Achieving multiple monitoring through a single sensor can significantly reduce costs. The complete information is provided in Table 4. Here, we can see studies that use a single sensor to acquire multiple signals and enhance the correlation between different signals.
本文根据生理信号类别分为几个部分，本小节添加了从传感器角度进行分类的结果。通过单个传感器实现多重监测可以显着降低成本。表 4提供了完整的信息。在这里，我们可以看到使用单个传感器获取多个信号并增强不同信号之间的相关性的研究。

6.3. Integrated Sleep Monitoring
6.3.综合睡眠监测

The third category is using multi-signal sensors for a comprehensive sleep quality analysis. It is similar to a PSG system for home use.
第三类是利用多信号传感器进行全面的睡眠质量分析。它类似于家用 PSG 系统。

The ideal product provides professional PSG monitoring under in-home conditions. It includes accurate sleep stage classification and disease diagnosis. It should also be easy to use, inexpensive, and not interfere with sleep. There is much room for improvement in integrated sensing solutions. Some existing commercial solutions appear in the table, with advances in the laboratory.
理想的产品可在家庭条件下提供专业的 PSG 监控。它包括准确的睡眠阶段分类和疾病诊断。它还应该易于使用、价格便宜并且不会干扰睡眠。集成传感解决方案还有很大的改进空间。表中列出了一些现有的商业解决方案，其中包括实验室的进展。

The combinations of signals and sensing modalities are very diverse and difficult to exhaust. The ability to achieve the best-integrated sleep monitoring under home conditions with the lowest cost and most user-friendly combination of sensors is the core research goal. Many researchers have investigated how to improve this combination. The complete information is provided in Table 5. In the table, there is an example of the conclusion that some techniques can detect anomalies earlier than others. Many studies have also been conducted on special populations or commercial devices.
信号和传感方式的组合非常多样化且难以穷尽。能够在家庭条件下以最低的成本和最人性化的传感器组合实现最佳集成的睡眠监测是核心研究目标。许多研究人员研究了如何改进这种组合。表 5提供了完整的信息。表中的一个示例表明某些技术可以比其他技术更早地检测到异常。还针对特殊人群或商业设备进行了许多研究。