Non-Contact Automatic Vital Signs Monitoring of Infants in a Neonatal Intensive Care Unit Based on Neural Networks

Abstract

Infants with fragile skin are patients who would benefit from non-contact vital sign monitoring due to the avoidance of potentially harmful adhesive electrodes and cables. Non-contact vital signs monitoring has been studied in clinical settings in recent decades. However, studies on infants in the Neonatal Intensive Care Unit (NICU) are still limited. Therefore, we conducted a single-center study to remotely monitor the heart rate (HR) and respiratory rate (RR) of seven infants in NICU using a digital camera. The region of interest (ROI) was automatically selected using a convolutional neural network and signal decomposition was used to minimize the noise artefacts. The experimental results have been validated with the reference data obtained from an ECG monitor. They showed a strong correlation using the Pearson correlation coefficients (PCC) of 0.9864 and 0.9453 for HR and RR, respectively, and a lower error rate with RMSE 2.23 beats/min and 2.69 breaths/min between measured data and reference data. A Bland-Altman analysis of the data also presented a close correlation between measured data and reference data for both HR and RR. Therefore, this technique may be applicable in clinical environments as an economical, non-contact, and easily deployable monitoring system, and it also represents a potential application in home health monitoring.

Keywords: NICU; convolutional neural network; heart rate; respiratory rate; signal decomposition.

Figure 1

The experimental setup where the data recording was performed. A schematic of the setup is shown to the right of the image. A camera was mounted on a tripod closer to the infant to record the infant body’s heart rate and respiratory rate. Another camera was mounted on a tripod to capture vital signs’ ground truth (shown on the monitor). A schematic diagram on the left of the figure shows an overview of the setting.

Figure 2

The system framework consists of two branches to detect heart rate and respiratory rate. The input video was processed for automatic ROI detection, and the ROI was processed separately for heart rate and respiratory rate detection.

Figure 3

YOLO network architecture (adapted from [65]). The YOLO network has 24 layers followed by two fully connected layers.

Figure 4

Automatic ROI selection using the YOLO neural network. The detected ROIs were shown in green bounding boxes. (a) Infant under normal light, (b) infant under blue light.

Figure 5

The spectral analysis and band-pass filtering process.

Figure 6

An infant image with detected ROI is shown in (a). The corresponding ROI extracted from the original is shown in (b).

Figure 7

Raw cardiorespiratory signals for 300 frames are shown in the figure. (a) Raw cardiac signal, (b) raw respiratory signal.

Figure 8

IMF components of the raw cardiorespiratory signals using EEMD technique. (a) Cardiac signal (b) respiratory signal.

Figure 9

The frequency spectrum of decomposed IMF3, IMF4, IMF5 and IMF6. (a) Cardiac signal (b) respiratory signal.

Figure 10

The filtered cardiorespiratory signals are shown in the figure. The red colour markers indicate the peak locations of the filtered signal. (a) Filtered cardiac signal, (b) filtered respiratory signal.

Figure 11

Statistical measurement for HR. (a) Correlation Plot, (b) Bland–Altman plot.

Figure 12

Statistical measurement for RR. (a) Correlation Plot, (b) Bland–Altman plot.