Abstract 1792 Poster Session I, Saturday, 5/1 (poster 28)

Purpose: To demonstrate the feasibility of a non-contact sensor for respiratory movement.

Methods: We developed a non-contact sensor for respiratory and body motion. It consists of a transmitter that emits a continuous beam of ultrasound (40kHz) toward the infant from 0.3-0.6m, and receives a reflected signal with a transducer placed next to the transmitter. A proprietary phase detection circuit determines the phase between the reflected and transmitted signals in real time, so that, in conjunction with software, the motion of the infant can be tracked over a range of many centimeters, with a resolution of tens of microns. Since the reflected signal is comprised of many reflections, digital signal processing was used to differentiate body and respiratory motions.

Analysis: We developed two algorithms to detect apnea from the respiratory signal. The first defined breaths with a zero-crossing detector whose threshold was determined by the maximum change of the signal in a 3-second window centered on the point of detection. This algorithm minimized the detection of non-respiratory motion. The second algorithm used a Fourier transform on a 15s window moved through the data in 1s increments to assess the spectral power of the signal in 5 bands of frequencies between 0.2 to 10Hz that could differentiate respiratory, cardiac, and body movements and background noise. A neural network was trained to differentiate infant motion from background noise. We trained the network on 12% of recorded data and tested it on the remaining 88%.

Protocol: To test the device, we made 36 recordings of 10-33min duration during sleep in 16 term infants. Infants were lightly clothed or covered with two layers of blankets during the recordings. To simulate apnea, we removed the infant from the ultrasound beam for 2-minute periods during the recordings. Both algorithms discriminated the presence or absence motion for each second of the recording, based on the previous 15s data window. False negative and positive discriminations are expressed as a percentage of the total number of seconds of recordings.

Results: See Table below. The false positives for the Fourier-based algorithm were nearly totally eliminated by applying a 10s averaging window to the algorithm output. These results were not affected by the infants' clothing or blankets.

Table 1 No caption available
Full size table

Conclusions: We conclude that it is feasible to detect respiratory and body motion with this non-contact sensor, which may prove useful for respiratory monitoring in infants.

Michael J. Treadaway is funded by, and has equity in Jaycor.