Respiratory Rate Estimation Using Respiratory
Sinus Arrhythmia from Photoplethysmography
Walter Karlen*, Member, IEEE, Christopher J. Brouse, Erin Cooke,
J. Mark Ansermino, and Guy A. Dumont, Fellow, IEEE
Abstract—Respiratory rate (RR) is an important measurement
for ambulatory care and there is high interest in its detection
using unobtrusive mobile devices. For this study, we investigated
the estimation of RR from a photoplethysmography (PPG) signal
that originated from a pulse oximeter sensor and had a sub-
optimal sampling rate. We explored the possibility of estimating
RR by extracting respiratory sinus arrhythmia (RSA) from the
PPG-derived heart rate variability (HRV) measurement using
real-time algorithms. Data from 29 children and 13 adults
undergoing general anesthesia were analyzed. We compared the
RSA power derived from electrocardiography (ECG) with PPG
at the reference RR derived from capnography. The power of
the PPG was significantly higher than that of the ECG (182.42
 36.75 dB vs. 162.30  43.66 dB). Further, the mean RR
error for PPG was lower than ECG. Both PPG and ECG RR
estimation techniques were more powerful and reliable in cases of
spontaneous ventilation than when pressure controlled ventilation
was used. The analysis of cases containing artifacts in the PPG
revealed a significant increase in RR error, a trend that was less
pronounced for controlled ventilation. These results indicate that
the estimation of RR from the sub-optimally sampled PPG signal
is possible and more reliable than from the ECG.
Index Terms—photo-plethysmogram, respiratory rate, heart
rate variability, pulse oximeter, anesthesia, respiratory sinus
arrhythmia
I. INTRODUCTION
Mobile health technology is a rapidly advancing field that
holds great promise for improving medical services and chang-
ing the way that health care is delivered. A common theme in
this area is the use of general purpose consumer devices, in
particular smart phones. An increasing number of health care
applications use these mobile platforms to interface directly to
biomedical sensors, such as blood pressure cuffs, actigraphs, or
pulse oximeters. This reduces or eliminates the cost of custom
embedded hardware and facilitates the measurement process.
However, features such as increased noise level, limited battery
and computational resources, and the requirement for real-
time processing challenge the accurate, real-time detection of
physiological parameters.
Respiratory rate (RR) is an important measurement for
diagnosing chronic illnesses, such as sleep apnea. The detec-
tion of RR using mobile or wearable sensors is not trivial
W. Karlen, C. Brouse and G. Dumont are with the Electrical and
Computer Engineering in Medicine group (ECEM), University of British
Columbia (UBC), Vancouver, Canada; *Corresponding author, e-mail: wal-
ter.karlen@ieee.org
J.M. Ansermino and E. Cooke are with the Department of Anesthesiology,
Pharmacology Therapeutics, University of British Columbia (UBC), Vancou-
ver, Canada;
because gold standard methods such as spirometry or cap-
nometry are too obtrusive and impractical. Other, less direct
methods for estimating RR exist. Respiration modulates the
heart rate (HR). When subjects are breathing spontaneously,
HR decreases on expiration and increases on inspiration. This
phenomenon is called respiratory sinus arrhythmia (RSA).
RSA is regulated by mechanical effects and changes in vagal
and sympathetic tone. [1]. Under positive pressure ventilation,
this phenomenon can present large phase shifts and variations
[2]. Respiration also modulates blood pressure. This effect can
be observed in the photoplethysmogram (PPG) recorded with a
pulse oximeter and is more pronounced under positive pressure
ventilation [3].
In this paper, we consider the estimation of RR from PPG
using the RSA phenomenon. We explore the possibility of
estimating RR from its heart rate variability (HRV) measure-
ment. HRV is the time variation between heart beats derived
from either the PPG or the electrocardiogram (ECG) signal.
Comparison of RR estimation by extracting RSA from the
HRV measurement of PPG and ECG to the gold standard RR
estimation from capnometry is performed. The primary goal of
this work is to identify potential limitations of estimating RR
from PPG due to temporal smearing of the distal pulses in the
PPG compared to the ECG. Further, we want to assess possible
pitfalls when extracting RSA from sub-optimally sampled
signals.
A. Background and Related Work
The most common method for assessing HRV is the analysis
of ECG signals sampled at high rates (>250 Hz) [4]. The
measurement of ECG is standard in the clinical environment,
but other methods, such as PPG measurement, are easier
to perform and are less obtrusive in the ambulatory setting.
Because of this, numerous research groups have tried to
assess whether ECG can be replaced with the measurement
of PPG [5]–[9] for the estimation of HRV. The major focus
of these studies has been to compare the outcome of the
HRV over the full frequency spectrum of interest. RR was
kept constant using a metronome [5] or ignored [6], [9]. In
[8], it was observed that the RSA components of HRV were
more pronounced when recorded with PPG compared to ECG.
While this positive bias towards RSA power was considered
a disturbance for HRV analysis by the authors, it might be
beneficial for RR estimation. Other research groups have used
inter-beat interval as one of many components to compute RR.
For example, an artificial neural network for RR estimation
978-1-4244-4122-8/11/$26.00 ©2011 IEEE 1201
33rd Annual International Conference of the IEEE EMBS
Boston, Massachusetts USA, August 30 - September 3, 2011