B54-36 Touchless Respiratory Monitoring in the Anesthesia Setting: Challenges With Comparing to a Reference

Background We have demonstrated accurate non-contact monitoring of respiratory rate (RR) in the anesthesia setting using a depth sensing camera [1]. Due to the challenging nature of the environment, it is difficult to obtain a stable and accurate reference signal for comparison with the touchless RR. Hence, a combined reference, RRref, was used, defined as the average of the two reference RRs (capnograph and an impedance-based respiratory monitor) and performance assessed only when the two devices were within ±3 breaths/min of each other for at least a 15 second period. Using these criteria, we achieved a performance of 1.92 breaths/min [1]. Here we explore the effect of variations in these criteria in the apparent performance of the touchless system. Methods Data was collected opportunistically from 34 healthy volunteers from two IRB-approved studies, which had the primary objective of investigating the relationship between depth of anesthesia monitoring and anesthetic agents (inhaled and intravenous) across a wide range of anesthetic concentrations and hypnotic states. (clinicaltrials.gov: NCT04466384 & NCT04602546). We measured RR across a wide range of anesthetic levels. Depth information was acquired using an Intel D415 RealSenseTM and processed to extract frame-by-frame depth changes within the subject’s torso region corresponding to respiratory activity. A respiratory rate (RRdepth) was calculated and compared to RRref. Allowable differences between the reference devices were varied from ±1 to ± 5 breaths/min over minimum agreement periods from 5 to 30 seconds. Results Figure 1 contains the results of the analysis. The original result [1] is shown circled. As can be seen from the figure, increasing the agreement period and decreasing the allowable difference in RR between the two reference signals improves the apparent performance, as we would expect. Interestingly, all results are below an RMSD of 3 breaths/min which is the performance criteria we aim for in our technology. Conclusions These results demonstrate the effect that changing the reference signal truth criteria has on the resulting performance metric. However, we find that even the least stringent criteria, allowing a wide variation in the reference devices over a relatively short period of time (±5 breaths/min over five seconds) resulted in a performance RMSD of less than 3 breaths/min, which we believe is a reasonable performance accuracy for use in practice. References [1] MacLeod, D.B. et al., 2025, Non-contact (touchless) monitoring of respiratory rate in a challenging anesthesia setting using a depth camera, Journal of Clinical Monitoring and Computing, 1-10. This abstract is funded by: Medtronic