Publication detail

Respiratory Rate Estimation Using the Photoplethysmogram: Towards the Implementation in Wearables

KOZUMPLÍK, J. SMITAL, L. NĚMCOVÁ, A. RONZHINA, M. KRÁLÍK, M. SMÍŠEK, R. VÍTEK, M. ŠACLOVÁ, L.

Original Title

Respiratory Rate Estimation Using the Photoplethysmogram: Towards the Implementation in Wearables

English Title

Respiratory Rate Estimation Using the Photoplethysmogram: Towards the Implementation in Wearables

Type

conference paper

Language

en

Original Abstract

Respiratory rate (RR) is one of the most important physiological parameters. In recent years, the RR estimation from PPGs widely used in smart devices has been promoted. The effect of respiration on PPGs manifests in three ways: BW (intensity variation), AM (amplitude variation), FM (frequency variation). In addition to sophisticated RR estimation methods, reliable results can be achieved with simple and efficient methods implementable in wearables. The BW signal (respiratory signal estimation, RS) can be obtained by linear filtering of the PPG. The RR estimation is based on BW extremes (sBW), BW autocorrelation extremes ( aBW) and their spectra (SBW, ABW). Estimation of the AM RS requires PPG extremes detection and interpolation. The RR estimation is based on extremes of the AM signal (sAM), its autocorrelation ( aAM) and their spectra (SAM, AAM). The fusion of RR estimates leads to more robust results. To test the algorithms, the annotated BIDMC and CapnoBase Datasets were used. RR estimates were made for 60 s sections. The simplest and the most accurate method for both datasets is the RR estimation based on sBW (RsBW). The median absolute error was 0.40 (0.16-1.09 interquartile range 25-75th) bpm for the 60s window, mean absolute error was 1.42 bpm.

English abstract

Respiratory rate (RR) is one of the most important physiological parameters. In recent years, the RR estimation from PPGs widely used in smart devices has been promoted. The effect of respiration on PPGs manifests in three ways: BW (intensity variation), AM (amplitude variation), FM (frequency variation). In addition to sophisticated RR estimation methods, reliable results can be achieved with simple and efficient methods implementable in wearables. The BW signal (respiratory signal estimation, RS) can be obtained by linear filtering of the PPG. The RR estimation is based on BW extremes (sBW), BW autocorrelation extremes ( aBW) and their spectra (SBW, ABW). Estimation of the AM RS requires PPG extremes detection and interpolation. The RR estimation is based on extremes of the AM signal (sAM), its autocorrelation ( aAM) and their spectra (SAM, AAM). The fusion of RR estimates leads to more robust results. To test the algorithms, the annotated BIDMC and CapnoBase Datasets were used. RR estimates were made for 60 s sections. The simplest and the most accurate method for both datasets is the RR estimation based on sBW (RsBW). The median absolute error was 0.40 (0.16-1.09 interquartile range 25-75th) bpm for the 60s window, mean absolute error was 1.42 bpm.

Keywords

Respiration rate, PPG signal

Released

12.09.2021

Publisher

IEEE

Location

Brno, Czech Republic

ISBN

2325-887X

Periodical

Computing in Cardiology

Year of study

48

Number

1

State

US

Pages from

1

Pages to

4

Pages count

4

URL

https://www.cinc.org/archives/2021/pdf/CinC2021-015.pdf

Documents

BibTex 


@inproceedings{BUT174969,
  author="Jiří {Kozumplík} and Lukáš {Smital} and Andrea {Němcová} and Marina {Filipenská} and Martin {Králík} and Radovan {Smíšek} and Martin {Vítek} and Lucie {Šaclová}",
  title="Respiratory Rate Estimation Using the Photoplethysmogram: Towards the Implementation in Wearables",
  annote="Respiratory rate (RR) is one of the most important physiological parameters. In recent years, the RR estimation from PPGs widely used in smart devices has been promoted. The effect of respiration on PPGs manifests in three ways: BW (intensity variation), AM (amplitude variation), FM (frequency variation). In addition to sophisticated RR estimation methods, reliable results can be achieved with simple and efficient methods implementable in wearables. The BW signal (respiratory signal estimation, RS) can be obtained by linear filtering of the PPG. The RR estimation is based on BW extremes (sBW), BW autocorrelation extremes ( aBW) and their spectra (SBW, ABW). Estimation of the AM RS requires PPG extremes detection and interpolation. The RR estimation is based on extremes of the AM signal (sAM), its autocorrelation ( aAM) and their spectra (SAM, AAM). The fusion of RR estimates leads to more robust results. To test the algorithms, the annotated BIDMC and CapnoBase Datasets were used. RR estimates were made for 60 s sections. The simplest and the most accurate method for both datasets is the RR estimation based on sBW (RsBW). The median absolute error was 0.40 (0.16-1.09 interquartile range 25-75th) bpm for the 60s window, mean absolute error was 1.42 bpm.",
  address="IEEE",
  booktitle="Computing in Cardiology 2021",
  chapter="174969",
  doi="10.22489/CinC.2021.015",
  howpublished="online",
  institution="IEEE",
  number="1",
  year="2021",
  month="september",
  pages="1--4",
  publisher="IEEE",
  type="conference paper"
}