Journal of Medical Engineering & Technology, Vol. 32, No. 6, November/December 2008, 479–484
`
`Assessment of heart rate variability derived from finger-tip
`photoplethysmography as compared to electrocardiography
`
`N. SELVARAJ{, A. JARYAL{, J. SANTHOSHx, K. K. DEEPAK{ and S. ANAND*{
`
`{Centre for Biomedical Engineering, Indian Institute of Technology-Delhi, New Delhi 110 016, India
`{Department of Physiology, All India Institute of Medical Sciences, New Delhi 110 029, India
`xComputer Services Centre, Indian Institute of Technology-Delhi, New Delhi 110 016, India
`
`Heart rate variability (HRV) is traditionally derived from RR interval time series of
`electrocardiography (ECG). Photoplethysmography (PPG) also reflects the cardiac
`rhythm since the mechanical activity of the heart is coupled to its electrical activity. Thus,
`theoretically, PPG can be used for determining the interval between successive heartbeats
`and heart rate variability. However, the PPG wave lags behind the ECG signal by the
`time required for transmission of pulse wave. In this study, finger-tip PPG and standard
`lead II ECG were recorded for five minutes from 10 healthy subjects at rest. The results
`showed a high correlation (median¼ 0.97) between the ECG-derived RR intervals and
`PPG-derived peak-to-peak (PP) intervals. PP variability was accurate (0.1 ms) as
`compared to RR variability. The time domain, frequency domain and Poincare´ plot
`HRV parameters computed using RR interval method and PP interval method showed
`no significant differences (p 5 0.05). The error analysis also showed insignificant
`differences between the HRV indices obtained by the two methods. Bland-Altman
`analysis showed high degree of agreement between the two methods for all the
`parameters of HRV. Thus, HRV can also be reliably estimated from the PPG based PP
`interval method.
`
`Keywords: Photoplethysmography; Electrocardiography; Heart rate variability
`
`1. Introduction
`
`Heart rate variability (HRV) analysis is a common practice
`for assessment of the cardiovascular autonomic nervous
`system. The oscillations present in the beat-to-beat pacing
`intervals of heart rate are of clinical relevance for both
`diagnostic and prognostic purposes as they are directly
`influenced by the sympathetic and parasympathetic systems
`[1]. Traditionally, time and frequency domain and non-
`linear methods have been used to interpret the physiolo-
`gical
`information embedded in the HRV signal. The
`classical spectral analysis of the HRV signal, generally
`determined from the sequences of successive inter-beat
`RR intervals of the ECG signal, enables separation of
`power distribution in different frequency bands. The low
`
`*Corresponding author. Email: snandaa@gmail.com
`
`frequency (LF) band corresponds mainly to sympathetic
`activity, whereas the high frequency (HF) band is related to
`respiratory sinus arrhythmia mediated by parasympathetic
`activity. The LF/HF ratio and LF and HF band powers are
`good indicators for the assessment of alterations in the
`nervous system behavior [1,2].
`Photoplethysmography (PPG) is a noninvasive optical
`technique for monitoring beat-to-beat
`relative blood
`volume changes in the microvascular bed of peripheral
`tissues. The PPG pulsatile waveform reflects the fluctuations
`in finger blood volume and vasculature characteristics. The
`autonomic influences on spontaneous fluctuations in finger
`blood volume have been assessed by spectral analysis of the
`PPG signal [3,4]. The PPG waveform characteristics such as
`amplitude, baseline and cycle period are useful for the study
`
`Journal of Medical Engineering & Technology
`ISSN 0309-1902 print/ISSN 1464-522X online ª 2008 Informa Healthcare USA, Inc.
`http://www.tandf.co.uk/journals
`DOI: 10.1080/03091900701781317
`
`J Med Eng Technol Downloaded from informahealthcare.com by University of Guelph on 08/26/13
`
`For personal use only.
`
`APPLE 1014
`
`1
`
`

`

`480
`
`N. Selvaraj et al.
`
`of autonomic control of the peripheral vascular tone [5 – 8].
`As the pulse period of blood volume pulse is directly related
`to the cardiac activity,
`the physiological
`information
`derived from RR intervals of ECG can also be derived
`from the pulse period of PPG (figure 1). The PP interval
`variability of the PPG signal was proven to be reasonably
`accurate compared to RR interval variability of ECG
`signal with high correlation [9,10]. The authors suggested
`the possibility of determining HRV parameters using PPG
`technique. Selvaraj et al. [11] and Bolanas et al. [12] have
`supported the same idea.
`Electrical activity of the heart (ECG) is followed by
`spread of the pulsatile wave of blood to the periphery. The
`pulse travel time shows very minor (a few milliseconds)
`beat-to-beat fluctuations [13 – 15], such that heartbeat
`intervals derived from ECG and PPG are very similar but
`they are not exactly the same. Such small variations in the
`heartbeat interval between the two methods do not appear
`to be significant in the time domain analysis but may
`potentially and significantly affect the frequency domain
`and nonlinear analysis of HRV. This issue has not been
`addressed in previous studies. In the present study, a
`comprehensive and systematic analysis of PPG based HRV
`as compared to ECG based HRV has been conducted to
`demonstrate the feasibility and reliability of deriving all the
`traditional HRV parameters from the PPG based method.
`
`2. Materials and methods
`
`Ten healthy subjects (age 21 – 28 years, nine males and one
`female) with no history of cardiovascular diseases and
`hypertension were included in this study. The experimental
`
`Figure 1. The RR interval (RRI) and PP interval (PPI)
`representing cardiac beat-to-beat interval extracted from
`the simultaneously recorded ECG and PPG signals
`respectively.
`
`protocol was explained to the subjects and written consent
`was obtained. The protocol was approved by the institu-
`tional ethics committee, All India Institute of Medical
`Sciences, New Delhi, India. The experiments were con-
`ducted at Autonomic Function Laboratory, All India
`Institute of Medical Sciences, New Delhi, India. The
`subjects were given a resting period of 15 minutes prior
`to the study.
`
`2.1. Experimental set-up
`
`The photoplethysmograph transducer TSD200 (BIOPAC
`Systems, Inc., CA, USA), a reflectance type sensor pri-
`marily designed for finger attachment, consisted of a
`matched infrared emitter of wavelength 860 + 6 nm and a
`photo diode to detect variation in infrared reflectance
`resulting from blood flow variation. This transducer was
`strapped on to the right middle finger of the subject and
`connected to the PPG amplifier (PPG100C) through a
`shielded cable to record the blood volume pulse (BVP)
`waveform with band-pass cut-off frequencies of 0.05 –
`10 Hz and gain of 100. Disposable Ag-AgCl electrodes
`were used to record standard lead II ECG signal. An
`ECG amplifier (ECG100C) was used to amplify the ECG
`signal with band-pass cut-off frequencies of 0.05 – 35 Hz
`and gain of 1000. The MP 150 (BIOPAC Systems), a
`computer-based data acquisition system with software
`1
`3.8.2, was used to acquire the standard
`AcqKnowledge
`lead II ECG and PPG data simultaneously at a sampling
`rate of 1 kHz. The data were recorded for a duration of
`five minutes under relaxed supine condition. The signal
`processing techniques were implemented offline with
`1
`7.0 (The Mathworks, Natick, MA, USA).
`Matlab
`
`2.2. Data analysis
`
`The recorded ECG and PPG signals were extracted
`separately. The short-term recordings were free of ectopic
`beats, missing data and noise. The R wave peaks of the
`ECG signal were identified and an RR tachogram
`representing heartbeat
`intervals was constructed. The
`following HRV measures were computed from the RR
`tachogram.
`
`1. Time domain HRV measures: mean normal-to-normal
`(NN) interval, mean HR, standard deviation of NN
`interval (SDNN), the square root of the mean squared
`differences of
`successive NN intervals
`(RMSSD),
`standard deviation of differences between adjacent
`NN interval (SDSD), the number of interval differences
`of successive NN intervals greater than 50 ms (NN50)
`and the proportion derived by dividing NN50 by the
`total number of NN intervals (pNN50).
`2. Frequency domain measures:
`total band power,
`normalized very low frequency (VLF), low frequency
`
`J Med Eng Technol Downloaded from informahealthcare.com by University of Guelph on 08/26/13
`
`For personal use only.
`
`2
`
`

`

`Heart rate variability using photoplethysmography
`
`481
`
`(LF) and high frequency (HF) band powers and LF/
`HF ratio.
`3. Poincare´ plot measures: short-term HRV (SD1), long-
`term HRV (SD2) and SD ratio (SD1/SD2).
`
`In frequency domain analysis, the series of RR intervals
`was cubic interpolated and re-sampled at 4 Hz [2], as RR
`intervals were non-uniformly spaced according to the
`heartbeat intervals. The DC component of the HRV signal
`was removed and then the power spectrum was obtained
`using discrete Fourier transform (DFT). The HRV power
`spectrum was divided into three bands: VLF (0.003 –
`0.04 Hz), LF (0.04 – 0.15 Hz) and HF (0.15 – 0.4 Hz) to
`evaluate the sympathetic and parasympathetic activities of
`the ANS [1]. The Poincare´ plot, one of the most accepted
`techniques of nonlinear HRV analysis, is a diagram which
`plots each RR interval against the previous interval. The
`standard deviations of the distances of the HR intervals to
`the lines y¼ x and y¼ 7xþ 2 6 mean (RR intervals) were
`quantified as SD1 and SD2 respectively [16].
`The systolic peaks of the PPG signal were identified and
`a PP tachogram was constructed from the time difference
`between successive systolic peak instances of PPG signal.
`The time domain and frequency domain and Poincare´ plot
`parameters were obtained for PP tachogram by the above
`mentioned procedures used for ECG based HRV analysis.
`The Pearson correlation coefficient was determined to
`correlate the beat-to-beat changes of RR interval and PP
`interval derived from ECG and PPG respectively.
`The absolute difference (actual error) between the values
`of each HRV parameter derived by two methods was
`calculated for each individual. Then, the overall actual
`error for each parameter was calculated as mean + SD
`from the group of 10 subjects. The agreement between two
`methods for every derived HRV parameter was assessed
`using Bland-Altman technique by GraphPad Prism version
`4.00 for Windows (GraphPad Software, USA). Further,
`paired t-test was used to test any significant difference
`between each parameter derived from two methods.
`
`3. Results
`
`Figures 2 – 5 show the HRV parameters derived from ECG
`and PPG based methods for a representative subject (male,
`22 years). A high correlation (0.998) was found between
`beat-to-beat RR intervals and PP intervals (figure 2).
`Figure 3 shows a good match between PP tachogram
`and RR tachogram. Similarly, the HRV power spectrum
`(figure 4) and Poincare´ plot (figure 5) were well matched
`between the two methods.
`For 10 subjects, the mean correlation between RR and
`PP intervals was 0.87 + 0.19 with median of 0.97. Table 1
`shows the various HRV parameters derived by the two
`methods. The mean NN interval derived from PP
`error *0.1 ms
`variability was accurate with actual
`
`Figure 2. The Pearson correlation coefficient between beat-
`to-beat changes of ECG-derived RR intervals and PPG-
`derived PP intervals for a representative male subject of
`age 22.
`
`Figure 3. Comparison of tachograms derived from RR
`intervals (RRI) of ECG signal (top) and PP intervals (PPI)
`of PPG (bottom) represented in figure 1 after cubic
`interpolation at 4 Hz. The PP variability matched RR
`variability all along the time scale.
`
`compared to RR variability. The error analysis also showed
`insignificant differences between all
`the HRV indices
`obtained by the two methods. More over, all the HRV
`parameters showed no significant difference (p 5 0.05)
`between the two methods.
`The degree of agreement between two methods was
`assessed using Bland-Altman analysis (figure 6). The Bland-
`Altman plot showed the mean difference of NN interval,
`LF/HF ratio and SD ratio as 70.02 ms, 70.02 and
`0.02 respectively and their corresponding 95% limits of
`
`J Med Eng Technol Downloaded from informahealthcare.com by University of Guelph on 08/26/13
`
`For personal use only.
`
`3
`
`

`

`482
`
`N. Selvaraj et al.
`
`Figure 4. The comparison of discrete Fourier transform
`based HRV power spectra of RR (top) and PP (bottom)
`tachograms represented in figure 2. The frequency char-
`acteristics of RR and PP tachograms were found to be
`similar.
`
`agreement were 70.41 to 0.38 ms, 70.12 to 0.09 ms and
`70.04 to 0.07 ms, respectively. Similarly, high degree of
`agreement was found for all the parameters derived by two
`methods of HRV analysis.
`
`4. Discussion
`
`The results clearly demonstrate an excellent correspon-
`dence between the HRV parameters derived by ECG and
`PPG based methods. A high correlation was found between
`the beat-to-beat RR interval and PP intervals. This is in
`agreement with Teng and Zhang [9] and Johnson and
`Mendelson [10], who reported similar correlations between
`heart rate intervals derived from ECG and PPG. However,
`the authors did not derive and compare HRV parameters
`from the PPG signal with the ECG method. In the present
`study, a complete HRV analysis as recommended by the
`task force has been performed and the comparison of
`derived HRV measures showed insignificant actual errors
`between the two methods. Further, the Bland-Altman
`technique demonstrated good agreement between the two
`methods.
`The PP intervals were calculated from systolic peak of
`PPG waveform. Teng and Zhang [9] used similar method to
`compute the PP interval. Nitzan et al. [5] and Bolanos et al.
`[12] used end-diastolic point and dicrotic notch, respec-
`tively, of the PPG waveform to compute the pulse period.
`From our point of view, changes in the baseline can
`potentially affect the identification of end-diastolic point
`and dicrotic notch may not be sharply demarcated in all the
`subjects. Therefore, a method based on identification of
`systolic peak is practical and accurate, as shown in the
`results.
`
`Figure 5. The Poincare´ plot of RR intervals (a) and PP
`intervals (b) represented in figure 1. SD1 and SD2 are
`referred to short term and long term HRV respectively.
`
`Table 1. The comparison of time domain, frequency domain
`and Poincare´ plot parameters derived from ECG and PPG for
`group of subjects (n¼ 10) using error analysis.
`
`Parameters
`
`ECG
`
`PPG
`
`Actual error
`
`Time domain
`Mean NN (ms)
`SDNN (ms)
`RMSSD (ms)
`SDSD (ms)
`NN50 (count)
`
`Frequency domain
`LF nu*
`HF nu
`LF/HF ratio
`
`Poincare´ plot
`SD1 (ms)
`SD2 (ms)
`SD ratio
`
`800.0 + 94.7
`52.0 + 15.8
`49.0 + 25.0
`49.1 + 25.0
`45.1 + 24.5
`
`800.0 + 94.7
`52.4 + 15.4
`50.3 + 23.1
`50.4 + 23.1
`46.2 + 24.1
`
`44.72 + 13.17
`55.28 + 13.17
`0.92 + 0.54
`
`44.39 + 12.43
`55.61 + 12.43
`0.90 + 0.51
`
`34.7 + 17.7
`64.4 + 16.3
`0.51 + 0.15
`
`35.6 + 16.4
`64.6 + 16.1
`0.53 + 0.13
`
`*nu—normalized unit; all the values are mean + SD.
`
`0.1 + 0.1
`0.9 + 0.5
`2.4 + 1.4
`2.4 + 1.4
`3.9 + 2.4
`
`1.93 + 2.02
`1.93 + 2.02
`0.04 + 0.04
`
`1.7 + 1.0
`1.1 + 0.8
`0.03 + 0.02
`
`J Med Eng Technol Downloaded from informahealthcare.com by University of Guelph on 08/26/13
`
`For personal use only.
`
`4
`
`

`

`Heart rate variability using photoplethysmography
`
`483
`
`lag time is called the pulse
`tissue. This
`peripheral
`transit time (PTT). The PTT shows beat-to-beat fluctua-
`tions [13 – 15] and thus it can potentially introduce errors in
`the estimation of pulse period and HRV parameters by
`PPG method. The present results show that the errors are
`negligible in all the HRV parameters of the PPG method
`during resting conditions.
`Methods other than ECG and PPG have also been used
`to derive HRV parameters. McKinley et al.
`[17] and
`Giardino et al.
`[18] used the arterial blood pressure
`waveform to compute HRV parameters and found it to be
`reliable. PPG is a simple and highly cost-effective method
`and it is not subjected to electrical interference and drying
`and dropping-off of electrodes. Hence, it can be used for
`long-term recordings with minimal discomfort to patients.
`Apart from HRV estimation, PPG has been shown to be of
`clinical utility in measurement of oxygen saturation, cardiac
`output, arterial compliance, endothelial functions, etc. [19].
`Although the ECG based method is the gold standard for
`estimation of HRV, the PPG based method can also be
`employed for HRV estimation when it is in use for other
`applications with patients as single recording procedure.
`In conclusion, PPG waveforms can easily be recorded
`from finger and digitized to compute reliable estimates of
`HRV. The correspondence between the ECG and PPG
`derived HRV parameters suggests that PPG based method
`can be used for estimation of short term HRV and long term
`monitoring of patients for diagnostic and prognostic
`purposes.
`
`References
`
`[1] Task Force of the European Society of Cardiology and The North
`American Society of Pacing and Electrophysiology, 1996, Heart rate
`variability—Standards of measurement, physiological interpretation
`and clinical use. European Heart Journal, 17, 354 – 381.
`[2] Carvalho, J.L.A., Rocha, A.F., Junqueira, L.F. Jr, Neto, J.S., Santos, I.
`and Nascimento, F.A.O., 2003, A tool for time-frequency analysis of
`heart rate variability. Proceedings of the 25th Annual International Con-
`ference of the IEEE EMBS, Cancun, Mexico, Sept 17 – 21, 2574 – 2577.
`[3] Bernadi, L., Radaelli, A., Solda, P.L., Coats, A.J., Reeder, M.,
`Calciati, A., Garrard, C.S. and Sleight, P., 1996, Autonomic control of
`skin microvessels: assessment by power spectrum of photoplethysmo-
`graphic waves. Clinical Science (London), 90, 345 – 355.
`[4] Murthy, V.S., Ramamoorthy, S., Srinivasan, N., Rajagopal, S. and
`Rao, M.M., 2001, Analysis of photoplethysmographic signals of
`the 23rd Annual EMBS
`cardiovascular patients. Proceedings of
`International Conference, Istanbul, Turkey, Oct 25 – 28, 2204 – 2207.
`[5] Nitzan, M., Babchenko, A., Khanokh, B. and Landau, D., 1998, The
`variability of the photoplethysmographic signal—a potential method
`for the evaluation of the autonomic nervous system. Physiological
`Measurement, 19, 93 – 102.
`[6] Khanokh, B., Slovik, Y., Landau and D., Nitzan, M., 2004,
`Sympathetically induced spontaneous fluctuations of
`the photo-
`plethysmographic signal. Medical & Biological Engineering & Comput-
`ing, 42, 80 – 85.
`[7] Tanaka, G. and Sawada, Y., 2003, Examination of normalized pulse
`volume-blood volume relationship: toward a more valid estimation of
`the finger sympathetic tone. International Journal of Psychophysiology,
`48, 293 – 306.
`
`Figure 6. Bland-Altman agreement analysis between ECG
`and PPG methods showed higher agreement of HRV
`parameters: heart rate interval (a), LF/HF ratio (b) and SD
`ratio (c). Solid lines indicate mean difference and dotted
`lines refer to boundaries of 95% limits of agreement.
`
`The PPG pulse is generated due to contractility of the
`heart and flow of blood from the heart to the peripheral
`tissue. Thus, the PPG waveform lags behind the R wave of
`ECG by a time period required for the mechanical activity
`of the heart and time required for transit of blood to the
`
`J Med Eng Technol Downloaded from informahealthcare.com by University of Guelph on 08/26/13
`
`For personal use only.
`
`5
`
`

`

`484
`
`N. Selvaraj et al.
`
`[8] Gal-on, B., Brown, I. and Nunn, A., 2005, Monitoring and assess-
`ment
`of
`cardiovascular
`regulation
`in
`spinal
`cord
`injured
`patients. Proceedings of
`the IEEE Engineering in Medicine and
`Biology 27th Annual Conference, Shanghai, China, Sept 1 – 4, 6859 –
`6862.
`[9] Teng, X.F. and Zhang, Y.T., 2003, Study on the peak interval
`variability of photoplethysmographic signals. IEEE EMBS Asian –
`Pacific Conference on Biomedical Engineering, Japan, Oct 20 – 22,
`140 – 141.
`[10] Johnston, W. and Mendelson, Y., 2005, Extracting heart rate
`variability from a wearable reflectance pulse oximeter. Proceedings
`of
`the IEEE 31st Annual Northeast Bioengineering Conference,
`Hoboken, NJ, USA, Apr 2 – 3, 157 – 158.
`[11] Selvaraj, N., Santhosh, J. and Anand, S., 2007, Feasibility of
`photoplethysmographic signal for assessment of autonomic response
`using heart rate variability analysis. IFMBE Proceedings (Berlin,
`Heidelberg: Springer), 15, 391 – 395.
`[12] Bolanos, M., Nazeran, H. and Haltiwanger, 2006, Comparison of
`heart rate variability signal features derived from electrocardiography
`and photoplethysmography in healthy individuals. Proceedings of the
`28th IEEE EMBS Annual International Conference, New York, USA,
`Aug 30 – Sept 3, pp. 4289 – 4294.
`
`[13] Drinnan, M.J., Allen, J. and Murray, A., 2001, Relation between heart
`rate and pulse transit time during paced respiration. Physiological
`Measurement, 22, 425 – 432.
`[14] Johansson, A., Ahlstrom, C., Lanne, T. and Ask, P., 2006, Pulse wave
`transit time for monitoring respiration rate. Medical & Biological
`Engineering & Computing, 44, 471 – 478.
`[15] Porta, A., Gasperi, C., Nollo, G., Lucini, D., Pizzinelli, P., Antolini,
`R. and Pagani, M., 2006, Global versus local
`linear beat-to-beat
`analysis of the relationship between arterial pressure and pulse transit
`time during dynamic exercise. Medical & Biological Engineering &
`Computing, 44, 331 – 337.
`[16] Piskorski, J. and Guzik, P., 2005, Filtering Poincare´ plots. Computa-
`tional Methods in Science and Technology, 11, 39 – 48.
`[17] McKinley, P.S., Shapiro, P.A., Bagiella, E., Myers, M.M., Meersman,
`R.E., Grant, I. and Sloan, R.P., 2003, Deriving heart period variability
`from blood pressure waveforms. Journal of Applied Physiology, 95,
`1431 – 1438.
`[18] Giardino, N.D., Lehrer, P.M. and Edelberg, R., 2002, Comparison of
`finger plethysmograph to ECG in the measurement of heart rate
`variability. Psychophysiology, 39, 246 – 253.
`[19] Allen, J., 2007, Photoplethysmography and its application in clinical
`physiological measurement. Physiological Measurement, 28, R1 – R39.
`
`J Med Eng Technol Downloaded from informahealthcare.com by University of Guelph on 08/26/13
`
`For personal use only.
`
`6
`
`