Home Monitoring: Breathing Rate from PPG and ECG David A. Clifton, David Meredith, Mauricio Villarroel, Lionel Tarassenko Institute of Biomedical Engineering, University of Oxford

1: The PPG is amplitude modulated by breathing in a manner

that is not fully understood; see our review article,

Meredith et al. (2011), J. Med. Eng. & Tech. 36(1), 2012, pp. 60-66

2: The heart rate is increased or decreased with breathing due to

action of the parasympathetic nervous system, termed respiratory sinus

arrythmia (RSA), an effect which decreases with increasing age

3: The ECG is amplitude-modulated by breathing due to the

mechanical movement of the chest, which changes the orientation of

the heart with respect to the axis between the ECG electrodes

Pulse oximeters are worn on the finger, and acquire the plethysmogram (PPG) at 75 Hz; the PPG

is a time-series of light intensities that varies according to the quantity of oxygenated blood in the tissue

A chest-worn Hidalgo Equivital band measures the mechanical action of

the torso at 25.6 Hz as breathing occurs – this is the “gold standard”

The Hidalgo Equivital system also measures the electrocardiogram (ECG)

at 256 Hz using electrodes fitted within the belt

30s of PPG, showing a period of artefactual data that was automatically identified from the waveform morphology 30s of ECG, showing automatically-identified artefact from the waveform morphology

The ECG is amplitude-modulated by breathing3;

(upper plot) the ECG filtered to retain breathing frequencies and

(lower plot) resampled at 4Hz

30s of belt-derived breathing waveform

The ECG is frequency-modulated by breathing due to RSA2

(upper plot) the r-peaks of the ECG give instantaneous heart rate

(lower plot) resampled at 4Hz

The PPG is amplitude-modulated by breathing1;

(upper plot) the PPG filtered to retain breathing frequencies and

(lower plot) resampled at 4Hz

The PPG is frequency-modulated by breathing due to RSA2

(upper plot) the peaks of the PPG give instantaneous heart rate

(lower plot) resampled at 4Hz

Estimation method 2: peak detection is performed to count the number

of peaks occurring within 60s

Estimation method 1: the Fourier transform is used to determine

the frequency spectrum of the 4Hz waveforms, in which the peak

frequency component is taken to be the breathing rate

Estimation method 3: an autoregressive (AR) model is fitted to

the 4Hz waveform; the dominant pole within the range of possible

breathing rates gives the estimate

Estimation method 4: AR poles from amplitude- and frequency-modulated

waveforms are plotted in blue and red, respectively; the dominant pole within the

possible range of breathing rates (shown in grey) gives the “fused” estimate

Data

acquis

itio

n

Pre

pro

cessin

g

Bre

ath

ing

rate

estim

ation

Objective: breathing rate is one of the vital signs that is most indicative of physiological deterioration (Hodgetts et al., 2002; Cretikos et al., 2007), and yet the majority of patients in hospital only have

infrequent, manual measurements taken by clinical staff counting movements of the chest wall; the breathing rate of patients at home is typically not monitored. There is therefore a need for robust

algorithms that can estimate respiration rate from the lightweight sensors that can be worn by that majority of patients in hospital who are not connected to bedside monitors, and to patients in the home;

such sensors include a mobile pulse oximeter and a “smart sticking plaster”, which acquires single-lead ECG. All clinical investigations undertaken to date have been small studies (of typically around 20

subjects) of patients with healthy physiology. We undertook a large-scale clinical study to investigate the efficacy of estimating breathing rate from lightweight sensors, using a large cohort of patients

undergoing haemodialysis. This group closely represents the “unwell” class of patients for whom automatic measurement of breathing rate is most necessary.

Results: the study resulted in 2,749 hours of monitoring from 500 sessions of dialysis (a cohort of 77 patients), each session of which was between 4 and 7 hours duration. Mean error in breathing-rate

estimation (with error s.d.) over the whole dataset for methods 1, 2, 3, and 4, respectively, was 3.2 (4.7) bpm, 5.2 (8.4) bpm, 3.8 (4.3) bpm, and 2.0 (1.9) bpm, indicating that the new “fused” estimate was most successful.

We conclude that the difficulty of performing breathing-rate estimation in unwell, elderly patients is significant, and that previously-existing methods, which were trained and validated using data from small numbers of

healthy patients, are not suitable for clinical use with “real” patient populations; our proposed method estimates breathing-rate sufficiently successfully that it can be used in clinical practice for monitoring home patients.

References:

Cretikos, M.A., Chen, J., Hillman, K., Bellomo, R., Finfer, S., and Flabouris, A.: The objective medical emergency team

activation criteria: a case control study, Resuscitation 73 (2007), pp. 62-72

Hodgetts, T.J., Kenward, G., Vlachonikolis, I.G., Payne, S., and Castle, N.,: The identification of risk factors for cardiac

arrest and formulation of activation criteria to alert a medical emergency team, Resuscitation 54 (2002), pp. 125-131