Towards Continuous Camera-Based Respiration Monitoring in Infants

Abstract

Aiming at continuous unobtrusive respiration monitoring, motion robustness is paramount. However, some types of motion can completely hide the respiration information and the detection of these events is required to avoid incorrect rate estimations. Therefore, this work proposes a motion detector optimized to specifically detect severe motion of infants combined with a respiration rate detection strategy based on automatic pixels selection, which proved to be robust to motion of the infants involving head and limbs. A dataset including both thermal and RGB (Red Green Blue) videos was used amounting to a total of 43 h acquired on 17 infants. The method was successfully applied to both RGB and thermal videos and compared to the chest impedance signal. The Mean Absolute Error (MAE) in segments where some motion is present was 1.16 and 1.97 breaths/min higher than the MAE in the ideal moments where the infants were still for testing and validation set, respectively. Overall, the average MAE on the testing and validation set are 3.31 breaths/min and 5.36 breaths/min, using 64.00% and 69.65% of the included video segments (segments containing events such as interventions were excluded based on a manual annotation), respectively. Moreover, we highlight challenges that need to be overcome for continuous camera-based respiration monitoring. The method can be applied to different camera modalities, does not require skin visibility, and is robust to some motion of the infants.

Keywords: NICU; camera; infants; infrared; non-nutritive sucking; respiration; thermal camera; thermography; unobtrusive; vital signs.

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Figure 1

Results of the manual annotation: the breakdown of the included class into the subcategories for the training and testing, and the validation set.

Figure 2

Main blocks of the processing chain and an example of the results.

Figure 3

Example of features obtained during a type 2 motion, i.e., arm motion. In (a,d) the merged thermal images are presented, the circle indicates the position of the baby’s arm where the type 2 motion is happening. The images in (b,c,e) show the three features. While in this case, pseudo-periodicity and gradient are sensitive to the presence of type 2 motion, Respiration Rate (RR) clusters are not, this is due to the RR^ matrix shown in (f) where the arm area can have frequencies equal to zero.

Figure 4

Receiver Operating Characteristics (ROC) curve obtained with the nine folds of the cross-validation by using all the parameters combinations.

Figure 5

Bland-Altman and correlation plot: (a) training and testing set, (b) validation set. RRCI and RRVideo are in BPM.

Figure 6

Example of the Short Time Fourier Transform (STFT) obtained using the camera and the Chest Impedance (CI) reference. The noisiness of the reference’s spectrum during type 1 motion shows the sensitivity of the reference to this type of artifact. The excluded segments are due to camera motion.

Figure 7

Example of results showing the RR estimated using our cameras and algorithm, and the reference one. The difference in the manual annotation of type 1 motion and the detected one is visible in the bottom plot. Examples of frames during the type 1 motion (infant crying) are also shown.

Figure 8

Example of results with periodic breathing. The sudden changes in RR can be seen in the STFTs close to the breathing pauses (indicated using the rectangular boxes with width of 8 s).