Detection of respiratory activity in newborn infants using a noncontact vision-based monitor

To examine the effectiveness of a noncontact vision-based infrared respiratory monitor (IRM) in the detection of authentic respiratory motion in newborn infants.

Observational study in a neonatal intensive care unit.

Eligible infants lay supine with torso exposed under the IRM's infrared depth-map camera and torso images were recorded at 30 frames/s. Respiratory motion waveforms were subsequently derived from upper (IRM_upper ) and lower (IRM_lower ) torso region images and compared with contemporaneous impedance pneumography (IP) and capsule pneumography (CP). Waveforms, in 15 s investigative epochs, were scanned with an 8 s sliding window for authentic respiratory waveform (spectral purity index [SPI] ≥ 0.75, minimum five complete breaths). Maximum SPI and frequency of occurrence of authentic respiratory waveform in 15 s epochs were compared between monitoring modalities in pooled and per patient data (Friedman ANOVA).

Recordings comprised 532 min of images from 35 infants, yielding 2131 investigative epochs, with authentic respiratory motion detected in all infants. For CP, IP, IRM_upper , and IRM_lower , the proportion of epochs containing authentic respiratory motion in pooled data were 65%, 50%, 36%, and 48%, with median SPI_max of 0.79, 0.75, 0.70, and 0.74, respectively. Per-patient average SPI_max was 0.79, 0.75, 0.69, and 0.74 for CP, IP, IRM_upper , and IRM_lower with proportion of authentic respiratory motion being 64%, 50%, 29%, and 49%, respectively.

An IRM focused on the lower torso detected authentic respiratory motion with comparable performance to IP in newborn infants in intensive care and deserves further investigation.