Measuring Uncertainty During Respiratory Rate Estimation Using Pressure-Sensitive Mats

Shermeen Nizami, Amente Bekele, Mohamed Hozayen, Kim Greenwood, JoAnn Harrold, James R. GreenIEEE Trans. Instrumentation & Measurement 2018 · DOI · OpenReview

Introduction

Monitoring the respiratory rate (RR) of neonates in the Intensive Care Unit (NICU) is a critical task where accuracy can be a matter of life and death. Traditional methods, such as thoracic impedance electrodes, require adhesive contact with the infant's skin. Given the extreme fragility of neonatal skin, these "gold standard" sensors often cause irritation or injury. Furthermore, they are highly susceptible to motion artifacts. Pressure-sensitive mats (PSMs) represent a promising non-contact alternative, yet they introduce engineering challenges related to sensor drift, creep, and the overwhelming noise generated by patient movement.

Experimental Methodology

To address these challenges, researchers conducted a series of controlled bench tests using high-fidelity clinical-grade neonatal simulators: "SimNewB" (2.8 kg) and "Premature Anne" (0.9 kg). These simulators were placed on PSMs to evaluate performance across various clinical scenarios, including different mattress types (soft overhead warmer vs. firm crib), patient positions (supine vs. prone), and breathing patterns (normal vs. grunting).

Figure 1: Experimental setup showing a neonatal simulator on a pressure-sensitive mat with corresponding pressure heatmaps for the head, thorax, pelvis, and right foot.

The study employed the "Guide to the Expression of Uncertainty in Measurement" (GUM) to quantify random errors (drift) and systematic errors (creep). A primary focus was the development of a robust RR estimation algorithm capable of performing in the presence of motion, which was identified as the most significant factor degrading sensor accuracy.

Signal Processing and Motion Remediation

The core innovation of this research is a modified frequency-domain algorithm designed to isolate the subtle breathing signal from larger motion artifacts. Standard Fast Fourier Transform (FFT) methods often fail when motion frequencies overlap with respiratory frequencies.

The researchers introduced a time-domain pre-processing step:

  1. Motion Identification: A moving average filter with a 1.5-second window is applied to the raw pressure data to isolate low-frequency motion artifacts.
  2. Signal Subtraction: This isolated motion signal is subtracted from the raw data, effectively "cleaning" the signal and highlighting the higher-frequency oscillations corresponding to chest movement.
  3. Frequency Estimation: An FFT is then performed on the remediated signal to determine the dominant respiratory frequency.
Figure 2: Visualization of the motion remediation process. The top plot shows the raw average pressure signal; the middle plot isolates the low-frequency motion component; the bottom plot shows the isolated high-frequency breathing signal.

Results and Discussion

The implementation of the motion remediation algorithm yielded significant improvements in accuracy. The original RR estimation algorithm exhibited a mean bias of 5.38 breaths per minute (bpm). With the modified approach, this bias was reduced to 0.56 bpm, well within the clinical requirement of ±5 bpm.

Statistical analysis using mixed-effects Limits of Agreement (LoA) confirmed the robustness of the method. The Pearson correlation coefficient between the estimated RR and the simulator's ground truth increased from 0.53 to 0.99. Interestingly, while mattress firmness affected the absolute pressure values and sensor drift, the algorithmic remediation remained effective across different physical interfaces.

Conclusion

This study demonstrates that pressure-sensitive mats, combined with intelligent signal processing, can provide a clinically viable, non-contact method for neonatal respiratory monitoring. By heuristically subtracting low-frequency motion noise in the time domain, the researchers achieved high-fidelity RR estimation that meets the rigorous standards of the NICU.

Future work will focus on long-term sensor characterization to account for "creep" over extended monitoring periods (14+ hours) and the development of automated motion detectors to trigger remediation only when necessary. Ultimately, clinical validation with live patients will be required to confirm these findings in the unpredictable environment of a real-world NICU.

Summary updated

Abstract

We develop and evaluate a respiratory rate (RR) estimation algorithm that utilizes data from the pressure-sensitive mat (PSM) technology for continuous patient monitoring in neonatal intensive care units. An analysis of the random effect of drift and systematic effect of creep in the PSM data is presented, showing that these are essentially dependent on the applied load and contact surface. Uncertainty measurements are pivotal when estimating physiologic parameters. The standard uncertainty in the PSM data is here represented by the percent drift. Next, we evaluate the applicability of the PSM technology to estimate RR in neonatal patient simulator trials under five mixed effects including internally and externally induced motion, mattress type, grunting, laying position, and different breathing rates. We analyze the limits of agreement on the mixed effects model to derive the uncertainty in the estimated RR obtained through two estimation techniques. In comparison with the gold standard RR values, we achieved a mean bias of 0.56 breaths per minute (bpm) with an error bounded by a 95% confidence interval of [-2.26, 3.37] bpm. These results meet the clinical accuracy requirements of RR within ±5 bpm.
BibTeX
@article{nizami2018measuring,
  author = {Shermeen Nizami and Amente Bekele and Mohamed Hozayen and Kim Greenwood and JoAnn Harrold and James R. Green},
  title = {{Measuring Uncertainty During Respiratory Rate Estimation Using Pressure-Sensitive Mats}},
  journal = {IEEE Trans. Instrumentation \& Measurement},
  year = {2018},
  doi = {10.1109/tim.2018.2805154},
}
Figure