A fast algorithm for extracting the breathing rate from PPG signal Davide Locatelli (1,*), Alessandra Fusco (1,†), Francesco Onorati (2) and Marco D. Santambrogio (1) Abstract — In this contribution, an algorithm for breathing rate extraction with low computational complexity is proposed. The here presented approach is based on EMD method and it proves to be robust and accurate, even in presence of noisy epochs.
can be noted that the errors in the estimation are relatively small and no bias nor offset is caused by the algorithm. This is also confirmed by the high correlation between the ground truth values and our algorithm estimates.
I. INTRODUCTION Extraction of breathing rate (BR) from photoplethysmographic (PPG) signals is a current topic in the scientific community. Although several approaches addressing PPG-derived BR have shown good performances [1], algorithm complexity and computational requirements have prevented real-time applications [1, 2]. Within this context, the here presented algorithm was designed as a trade-off between performances and computational cost. II. METHODS Madhav et al. [4] proposed a method based on the Empirical Mode Decomposition (EMD) to estimate BR from PPG signal. The key point was to identify the oscillatory modes at different time scales, and then to decompose the signal accordingly. When decomposing the PPG signal via EMD, we have observed that two principal components can fully describe the PPG signal dynamic, and the component corresponding to the lowest frequency is an estimate of the BR. Hence, in our approach the EMD is stopped at the first step, reducing algorithm complexity, and thus the computational burden. The respiratory component was used to estimate BR taking the dominant frequency peak of its power spectral density (PSD). Recording sessions from Physiobank MIMIC II Waveform Database archive [4] were selected for assessing the accuracy of the proposed approach. All the sessions include PPG and respiration signals. The latter is used as “ground truth” in the validation of the algorithm. Sixty (60) one-minute epochs, not affected by missing data nor signal saturation artifacts, were selected from different sessions.
Figure 1 - Bland-Altman plot showing the distribution of the difference between the BR estimated with the EMD-based method and the ground truth.
I. CONCLUSIONS In this late contribution we showed that it is possible to track BR from PPG signal with high accuracy at a low computational cost. Despite the relatively simple structure of the algorithm, the results indicate a strong correlation with the ground truth. Analysis of the PPG signal offers an alternative way of monitoring BR; this indirect estimation can be extremely useful when only PPG signal is available. In particular, in the field of wearable devices, this kind of fast and robust algorithms are crucial for real-time applications. IV. ACKNOWLEDGMENTS This work was developed in and supported by Empatica Inc.
III. RESULTS AND DISCUSSIONS This approach shows good performances in estimating BR from PPG signal. The Mean Absolute Error (MAE) is 0.0044 Hz, corresponding to 0.26 breaths per minute; the Spearman's correlation coefficient (ρs) is 0.991. Although Madhav et al. method seems to outperform our results, they analyzed only five (5) one-minute epochs by using a complex and computationally heavy iterative procedure. By contrast, ou simplified approach proves to be robust on a more extended dataset even when processing noisy epochs or analyzing recordings with abnormal BR (i.e., 0.7 Hz). Figure 1 shows the Bland-Altman plot of the difference between the groud truth and the EMD-based BR estimates; it (1)
DEIB, Politecnico di Milano, Milan, Italy; (2) Empatica Inc. (*)
[email protected]; (†)
[email protected]
REFERENCES [1] [2]
[3]
[4]
Fleming, S. G., and Lionel T. “A comparison of signal processing techniques for the extraction of breathing rate from the photoplethysmogram”. Int J Biol Med Sci 2(4), 232-6, 2007. Leonard, P. A., Douglas, J. G., Grubb, N. R., Clifton, D., Addison, P. S., Watson, J. N. “A fully automated algorithm for the determination of respiratory rate from the photoplethysmogram.” J Clin Monitor Comp, 20(1), 33-36, 2006. Goldberger A. L., Amaral L. A. N., Glass L., Hausdorff J. M., Ivanov P. Ch., Mark R. G., Mietus J. E., Moody G. B., Peng C.-K., Stanley H. E. “PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation 101(23), e215-e220, 2000. Madhav, K. V., Ram, M. R., Krishna, E. H., Komalla, N. R., & Reddy, K. A. “Estimation of respiration rate from ECG, BP and PPG signals using empirical mode decomposition”. In: Instrumentation and Measurement Technology Conference (I2MTC), 2011 IEEE. IEEE, 1-4, 2011.
