Toward Accurate Extraction of Respiratory Frequency From the Photoplethysmogram: Effect of Measurement Site

Abstract

Background: It is known that the respiration-modulated photoplethysmographic (PPG) signals could be used to derive respiratory frequency (RF) and that PPG signals could be measured from different body sites. However, the accuracy of RF derived from PPG signals of different body sites has not been comprehensively investigated. Objective: This study aims to investigate the difference in the accuracy of PPG-derived RFs between measurements from different body sites, respectively, for normal and deep breathing conditions. Methods: Under normal and deep breathing patterns, the PPG signals were recorded sequentially in a randomized order from six body sites [finger, wrist under (anatomically volar), wrist upper (dorsal), earlobe, and forehead] of 36 healthy subjects. Simultaneously, the reference respiratory signal was measured by a respiratory belt on the chest. Using the frequency demodulation approach, respiratory signals were extracted from PPG signals for calculating RF by power spectral density. The bias between PPG-derived and reference RFs was then analyzed statistically using analysis of variance and non-parametric tests, Bland-Altman analysis, and linear regression to investigate the difference in RF bias between different sites. Results: The RF bias was significantly influenced by the breathing pattern and measurement site (both p < 0.001). Under normal breathing, the RF bias was insignificant in the arm, forehead, and wrist under (all p > 0.05) and significant in the other sites (all p < 0.05). Significant linear relationship between PPG-derived and reference RFs existed at all the sites ( p < 0.001) except the wrist upper ( p > 0.05). The linearity between PPG-derived and reference RFs was highest at the forehead (slope of best-fit line: 0.90, R 2 : 0.64), followed by the earlobe, finger, arm, and wrist under (slope: 0.71, R 2 : 0.40). Under deep breathing, there was no significant RF bias in all the measurement sites ( p > 0.05) except forehead ( p = 0.048). The effect of measurement site on RF bias was not significant ( p > 0.05). The finger had the smallest RF bias and the narrowest limits of agreement. Conclusion: This study has demonstrated that the accuracy of PPG-derived RF depends on the measurement site and breathing pattern. The best sites are the forehead and finger, respectively, for normal and deep breathing patterns.

Derivation of Respiratory Metrics in Health and Asthma

“…Another approach to reduce variability would be to assess the quality of the PPG signal and exclude windows with poor quality [ 10 , 15 ], though this would mean loss of information during, e.g., noisy periods. Nevertheless, the bias and standard deviation presented in this work fall in line with our previous findings [ 8 ] and perform better in terms of bias and 95% LoAs for respiratory rate in other studies employing larger datasets [ 10 , 16 , 17 , 18 , 19 ]. The high variability may limit applicability in the clinical setting, but the performance may be adequate for general long-term monitoring of breathing rate for day-to-day use.…”

Wearable monitoring of sleep-disordered breathing: estimation of the apnea–hypopnea index using wrist-worn reflective photoplethysmography

“…However, the different physiological nature of these modalities may cause differences such as HRV mismatches 14 , 24 , hampering the direct application of algorithms developed for one sensor to another. A similar reasoning also applies to transmissive PPG-based methods 13 , 25 , 26 . In addition, most of the OSA monitoring methods using transmissive PPG employ the derived blood oxygen saturation measurement as input 20 ; while the green light-rPPG usually embedded in wrist-worn devices cannot measure saturation 23 .…”

Influences of Sensor Placement Site and Subject Posture on Measurement of Respiratory Frequency Using Triaxial Accelerometers

“…The DC-filtered signals were firstly processed using a wavelet algorithm. For frequency domain analysis (e.g., PSD), the variation of RF during the measurement can cause estimation error ( Hartmann et al, 2019 ). To improve the reliability of RF estimation we used wavelet analysis which discloses the properties of the signals in time-frequency domain.…”