Development and evaluation of a continuous respiratory monitor for newborns in low-resource settings.

Introduction: Preterm birth interrupts physiological development, leaving newborns vulnerable to complications requiring respiratory rate (RR) monitoring. Complications include respiratory distress syndrome, a leading cause of death in preterm newborns, and apnea of prematurity, which affects 50% of preterm newborns and may be a presenting symptom of underlying conditions such as sepsis, hypoglycemia anemia, or seizures. When patient monitors are unavailable, the World Health Organization (WHO) recommends counting breaths for 60 seconds to measure neonatal RR. In low-resource settings, patient monitors are often prohibitively expensive, and manual counting is impractical due to high nurse-to-patient ratios. Rice360 is exploring methods for low-cost, continuous, RR/apnea monitoring for newborns in low-resource settings. This exploratory study evaluates video-based respiratory detection methods against manually counting breaths. The performance was compared to an existing patient monitor.

Materials and

Methods: Preterm newborns at risk for apnea or needing respiratory support at Queen Elizabeth Central Hospital (QECH) in Blantyre, Malawi were enrolled for respiratory monitoring. Participants received continuous monitoring from standard of care monitoring (impedance pneumography, SpO2, electrocardiogram, RR, heart rate, temperature) from a patient monitor (IntelliVue MP30, Philips). Subjects were also monitored by a novel respiratory rate monitor that uses two inertial measurement units (IMU) to detect breathing motion. Participants were additionally recorded by a webcam (WE800, Bonsaii); the video was used for automatic motion detection, automatic breath detection, and manual breath counting. Video block-matching in Matrix Laboratory (MATLAB, MathWorks) estimated inter-frame motion to automatically label participants were in motion or at rest throughout monitoring. Researchers manually counted breaths using a custom video annotation GUI (graphical user interface) from one-minute, non-overlapping epochs randomly selected from at-rest data for RR accuracy analysis. A respiratory signal was derived from each epoch using video processing methods. Pixels that were likely to contain respiratory motion were automatically identified, and the intensity changes of these pixels were used to construct a respiratory signal. Bland-Altman plots assessed inter-counter agreement as well as the video-based monitoring method and patient monitor’s error from the reference standard of manual counting. Rice University Institutional Review Board (IRB), the College of Medicine Research and Ethics Committee in Malawi, and Baylor College of Medicine IRB approved this study.

Results, Conclusions, and Discussions: Ten newborns at QECH averaging 33 weeks [32-34] gestational age and enrollment weight of 1563g [1030–2023g] were monitored for a total of 12 hours. On average, each participant was monitored for 72 minutes [49-92min]. Subjects spent an average of 50 min/hr [39-58min/hr] of monitoring time at rest with at-rest periods lasting 1.3±0.9 minutes. This amounted to a total of 10 hours of monitoring identified as at-rest. Ten epochs per subject, for a total of 100 epochs, were selected from at-rest periods for reference counting and RR Bland-Altman analysis. Inter-counter agreement was -0.11 bpm with -2.45, +2.24 bpm 95% limits of agreement (LOA). The video-based RR bias from manual counting RR was -0.82 bpm [-7.91, +6.27 bpm LOA]and patient monitor bias was +1.08 bpm [-10.82, +8.66 bpm LOA].
This preliminary respiratory rate accuracy analysis shows that our video-based method’s bias is within the optimal accuracy of ±2 bpm from the UNICEF and NEST360 Target Product Profile (TPP) for a RR/apnea monitor for low-resource settings.
Development and evaluation of our method is ongoing; we are working to design a device that can implement our algorithm in real time, as well as remain accurate under sub optimal conditions such as low lighting, during phototherapy treatment or throughout small movements.

Acknowledgements (Optional):