Discussion
Safe, noncontact techniques
The primary objective of this scoping review was to investigate the methods that have been studied for safe and noncontact monitoring of respiration in young children. The included studies explored various approaches, including bed-based methods, UWB radar, Doppler radar, video, IR cameras, garment-embedded sensors, and sound analysis. Among the 17 studies reviewed, the majority focus on monitoring either RR or apneas, while five studies aimed to do both. None of the studies reported safety risks. For very high-power UWB radar systems, there could be concerns for interference with other radiofrequency devices and safety, but such high powers are not reached in UWB radar in the medical field29. Liu et al.30 have outlined that RR measurement methods can be derived from other physiological signals, respiration movements or airflow. Notably, most of the techniques investigated relied on software for signal analysis, offering advantages such as faster and more objective analysis compared to manual assessment.
In the study conducted by Lee et al.18, ballistocardiography was employed to derive RR. Notably, this technique offers the potential to detect other physiological parameters that generate motion, including snoring and limb movements31,32. This capability opens up the possibility of assessing additional sleep characteristics such as REM sleep, as demonstrated by Kortelainen et al.33. However, it is important to note that research on these measurements has predominantly concentrated on adults.
Accuracy
The second aim of this review was to describe the accuracy of the techniques. Because the monitoring techniques would mostly function as a screening tool for SDB, sensitivity is an important measure of accuracy. Only Norman et al.14, Collaro et al.16 and Bani Amer et al.25provided the sensitivity of their techniques (82%, 85% and 94% respectively).
RR measurement accuracies found by Al-Naji and Chahl23, Kim et al.19 and Ranta et al.27 fall within an accuracy of ±2 breaths per minute. These researchers used different types of methods, namely video, UWB radar, and a garment-embedded sensor, respectively.
Not all studies explicitly described the impact on motion artifacts. However, the authors that did, were not able to monitor respiration. Interestingly, Bani Amer et al.25 claimed that their measurements with an IR sensor did not have any effect of motion artifacts. A conventional apnea monitor was used as a reference standard, but it was not specified what type of monitor this was. It is known that different respiration measurement techniques, such as methods that measure chest and abdominal wall movements, are also subject to motion artifacts.34 It would be interesting to see the performance validated against PSG.
Most authors removed data containing motion artifacts from the signal for apnea detection. Although this is not a problem for apnea detection in adults, children can experience apneas during body movements that last a long time13. Therefore, the ability of a technique to monitor respiration during movement would enable more accurate measurements. For Doppler radar, efforts have been made to fight the problem of motion artifacts. Li and Lin35researched a method using two radar sensors, and Gu et al.36 used a hybrid radar-camera system for random body movement cancellation.
When presenting data on a new monitoring technique, it is essential to provide comprehensive information on accuracy to establish its validity and reliability. This includes describing the validation methods used to assess the performance of the technique, such as comparisons with established reference standards or expert evaluations. Geographical validation is also crucial to ensure the generalizability of the results across different populations or settings. Additionally, reporting metrics such as receiver operating characteristics (ROC) curves, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score can further demonstrate the accuracy of the monitoring technique. These metrics allow for an in-depth analysis of the technique’s ability to correctly identify and distinguish between various physiological parameters or conditions. Furthermore, providing information on any potential limitations or sources of error associated with the monitoring technique is important for a comprehensive evaluation. This may include factors such as signal quality, potential interference, or specific conditions under which the technique may yield less accurate results.
Advantages and limitations of the techniques
Most researched methods were easy to set up, except those involving load cells in a bed frame, which would be more challenging in a domestic setting compared to a mattress or mattress overlay. Both bed-based methods and UWB radar demonstrated potential for movement tracking in addition to monitoring RR and detecting apneas. It is expected that all techniques, except sound analysis, would enable movement tracking, which is valuable for accurately measuring sleep duration, fragmentation, and assessing the impact of SDB. Combining sleep-wake classification with apnea detection allows a more precise estimation of the apnea-hypopnea index (AHI) compared to estimating sleep time based on time spent in bed. Arimoto et al.13, who took the time in bed as sleeping time, found an underestimation of 10-15% of AHI compared to PSG. Several studies have compared AHI calculations using UWB radar with PSG in adults. Zhou et al.37 found a sensitivity and specificity of the AHI measured with UWB radar were 0.95-1 and 0.96 – 1, respectively, depending on the AHI cut-off values. Kang et al.38 found that using UWB radar, three types of apnea (central, obstructive, and mixed) and hypopneas could be detected in adults. The authors concluded that UWB radar could be used as a screening tool for sleep apnea. If these results can be achieved in adults, there is reason to believe this is also possible in children.
Bed-based techniques and video analysis were found to be able to measure respiration in children in any sleeping position, allowing more accurate real-life measurements in children. Gramse et al.26, who integrated strain sensors in a pair of pajamas, found that more signals were out of range in prone and side positions compared to supine positions. Authors researching radar-based monitoring did not specify the effect of sleeping position. In a radar-based vital sign monitoring study by Turppa et al.39, larger errors in RR measurements in lateral sleeping positions were found compared to prone and supine positions. Not only the sleeping position influence the measurements, but the location and direction in which the radar is facing also play a significant role in measurement accuracy. Therefore, it would be valuable to explore various experimental setups in this context.
The added value of sound recording has been studied by Emoto et al.28 and Norman et al.15 Apart from using breathing sounds for the classification of obstructive, central and mixed apneas, Norman et al.15 found that measuring snoring and stertor allows assessment of partial upper airway obstruction, providing more information compared to identifying discrete obstructive apnea/hypopnea events. In addition to this, snoring and breathing sounds have recently been researched for the estimation of sleep-wake activity and sleep quality parameters in adults. A study by Dafna et al.40 found they could differentiate between sleep and wake based on breathing sounds, with a sensitivity of 92.2% and a specificity of 56.6% compared to PSG. In another study, by Akhter et al.41, breathing sounds were used to differentiate between REM and non-REM sleep, with a sensitivity of 92% and a specificity of 81% compared to PSG, indicating that these results could also be achievable in children.
Limitations
Three out of the 17 studies included in this review were retrieved from the reference lists of the other studies. This highlights the heterogeneity of terminology in this field of research. Four studies compared their method to the gold standard. Across the studies, different measures for accuracy were used, or no accuracies were determined. In addition to that, within groups of methods, different types of sensors and analytical algorithms were used, complicating the comparison between methods. The development of different sensors and algorithms, however, is not only logical in the current state of the art, but also essential for the field’s progress, reflecting the diversity of use cases, promoting innovation, and allowing for tailored solutions to meet the unique needs in various clinical scenarios.
This scoping review mainly focused on apnea detection and RR monitoring. It is important to acknowledge that SDB encompasses a larger spectrum, including hypoperfusion and decreasing saturations1. In order to properly diagnose SDB, these aspects should also be taken into account. Nonetheless, because OSA is the most common and clinically significant type of SDB in children42, this review is an important step towards contactless respiration monitoring and SDB screening in children.
Future research
To determine the most promising techniques for noncontact respiration monitoring in children, further research is necessary to compare these techniques against the gold standard PSG. Ideally, these studies should be openly accessible and encompass large cohorts of children across different age groups, encompassing both those with and without comorbidities. This approach will allow for the identification of the most effective techniques tailored to specific patient populations. Additionally, because respiration monitoring techniques can be used for various objectives, it is important to take into consideration the types of monitoring techniques for specific use. Common objectives include diagnosis or screening of SDB, long-term apnea detection, RR monitoring, or impact assessment of SDB treatment. Future research should focus on identifying both the appropriate monitoring techniques for each application and the adequate signal processing algorithms for those techniques. Techniques should be safe, accurate, reliable and flexible in terms of environment. Studies presenting new monitoring techniques should report important accuracy metrics such as ROC curves, sensitivity, specificity, positive predictive value, negative predictive value and, in case of machine learning algorithms, F1 score. Lastly, to ensure generalizability of the results, it is crucial to validate the techniques.