A30-06 Sleep Scoring of Apnea and Hypopnea Events Using Touchless Monitoring

Abstract Background Wired connections between physiological sensors and recording devices are common in sleep studies but disrupt sleep and cause discomfort. Attaching sensors is also time-consuming, expensive, and inconvenient for patients. Consequently, there is growing interest in the development of wire-free and sensor-free alternatives for sleep study monitoring. We have developed a non-contact (‘touchless’) respiration monitoring technology based on a depth sensing camera (Intel RealsenseTM, Santa Clara, CA). This generates a high fidelity, non-contact flow signal, NCMflow, which does not require sensors attached to the patient. Following on from our previous work 1, we performed a proof-of-concept study to assess touchless monitoring’s potential to identify respiratory disturbances and derive an apnea-hypopnea index (AHI) value. Methods Data was acquired using a prototype system (Figure 1a) from a cohort of 24 patients attending the sleep clinic for assessment of sleep disorders, comprising 170 hours of recorded data. The system provides a real time view of patient breathing and an associated waveform (Figure 1b). An automated sleep scoring algorithm was developed for NCMflow based on the AASM guidelines for respiratory disturbance scoring. For the pilot study we performed a score matching analysis where the scoring algorithm was developed to match PSG scoring: this used the touchless respiratory waveform and co-collected scored desaturation data (for hypopneas). The resulting AHIs were assessed for accuracy by comparing with PSG scoring. Results A total of 2,038 AH events were detected by the touchless scoring algorithm compared to 2,103 events in the manually scored PSG reference. Assessment of the scored events resulted in a correlation coefficient of 0.99 between the touchless and PSG-derived scores, with a corresponding root mean square deviation (RMSD) of 3.8 events/hour (bias = -1.1). (Figure 1c.) Further, stratification of AHI classifications into normal / mild / moderate / severe resulted in only a single misclassification (Figure 1d) corresponding to an accuracy of 95.8%. Conclusions The results provide further evidence for scoring respiratory disturbances during sleep without the need for wires or probes. We used a direct score matching algorithm to determine patient AHI’s based on touchless respiratory monitoring technology. In future work we aim to provide a more rigorous analysis of the data using a leave-one-out cross validation paradigm to mitigate potential algorithm overtraining. Reference: 1 Addison, P., et al (2025). Hourly Scoring of Apnea-Hypopnea Events Using a Novel Touchless Monitoring System. Am. J. Resp. Critical Care Med., 211, A1055-A1055. This abstract is funded by: Medtronic