The effect of artifact correction on spectral estimates of heart
rate variability.
Citation for published version (APA):
Peters, C. H. L., Vullings, R., Bergmans, J. W. M., Oei, S. G., & Wijn, P. F. F. (2010). The effect of artifact correction on spectral estimates of heart rate variability. In Proceedings on the 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society 2008, EMBS 2008, 20-25 August 2008, Vancouver, British Columbia (pp. 2669-2672). Institute of Electrical and Electronics Engineers.
https://doi.org/10.1109/IEMBS.2008.4649751
DOI:
10.1109/IEMBS.2008.4649751 Document status and date: Published: 01/01/2010
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
providing details and we will investigate your claim.
Abstract— Spectral analysis of fetal heart rate variability
might offer additional information that can be used for assessing the fetal condition more reliably. Clinical recordings of fetal heart rate are usually contaminated by artifacts. These artifacts can be detected and corrected or removed, but this can affect the spectral estimates obtained from the heart rate data. To determine what level of artifact correction is still acceptable for reliable calculation of spectral heart rate variability parameters, artifact correction is simulated on neonatal and fetal data that did originally not contain artifacts. 2000 data segments with various levels of artifact correction are analyzed spectrally, and calculated spectral estimates are compared to the values obtained from the original, artifact free data. In the very low (< 0.04 Hz) and low (0.04 – 0.15 Hz) frequency range, powers can be calculated reliably when up to 25% of the data are missing due to artifact correction. Powers in the high frequency range (0.15 – 0.4 Hz for adults, 0.4 – 1.5 Hz for newborns) cannot be calculated reliably when data are missing due to artifact correction. This is a major limitation for application in clinical practice, which might be solved by calculating power in the high frequency range at a shorter time scale than power in the low frequency range. Short segments of heart rate data that are free of artifacts can then be used to calculate powers in the high frequency range reliably, while segments that contain artifacts are excluded.
I. INTRODUCTION
PECTRAL analysis of heart rate variability (HRV) can provide valuable insight in cardiovascular regulation by the autonomic nervous system, as sympathetic and parasympathetic nervous activity make frequency specific contributions to the power spectrum of the heart rate [1]. Currently, spectral analysis of heart rate variability is widely used in clinical research, but direct application in medical devices is scarce. Partially, this is caused by the complexity of the control mechanisms that coordinate the total pattern of activity in a subject. As changes in activity may demand alterations of the heart rate, information on this total activity is required for correct interpretation of heart rate variability [2]. In clinical practice, this information may not be available or may not be taken into account sufficiently.
This work was supported by the Dutch technology foundation STW C.H.L. Peters is with the Department of Clinical Physics, Amphia Hospital, Breda, The Netherlands; e-mail: cpeters@amphia.nl
R. Vullings is with the Faculty of Electrical Engineering, Eindhoven University of Technology, Eindhoven, The Netherlands.
J.W.M. Bergmans is with the Faculty of Electrical Engineering, Eindhoven University of Technology, Eindhoven, The Netherlands.
S.G. Oei is with the Department of Gynaecology and Obstetrics, Máxima Medical Center, Veldhoven, The Netherlands.
P.F.F. Wijn is with the Department of Clinical Physics, Máxima Medical Center, Veldhoven, The Netherlands and the Faculty of Applied Physics, Eindhoven University of Technology, Eindhoven, The Netherlands.
Additional limitations arise from the sensitivity of spectral HRV parameters to artifacts that may be present in heart rate data. Methods used for automated spectral analysis of heart rate data should be insensitive to these artifacts [3].
The worldwide used standard method for fetal monitoring, cardiotocography (CTG), is based on visual evaluation of patterns of fetal heart rate and maternal uterine activity. The specificity of cardiotocography is however poor, which is reported to result in increased rates of unnecessary operative delivery without noticeable improvement of fetal outcome [4]. Therefore, continuous need exists for additional information that can be used to assess the fetal condition more reliably. Spectral analysis of fetal heart rate variability might offer additional information that can be used for fetal monitoring [5]. To a certain extent, spectral analysis of fetal heart rate variability reflects fetal compromise during delivery, offers the potential to predict severe fetal acidose [6,7] and may be used to monitor the development of the autonomic nervous system in fetuses [8].
Fetal heart rate data obtained in clinical practice are often corrupted by artifacts, for example due to patient movement, especially when fetal data are obtained non-invasively from the maternal abdomen. Generally, in fetal heart rate data, artifacts can easily be detected and removed or corrected. Single ectopic or falsely detected heart beats can be corrected by phantom beat replacement [3]. When artifacts corrupt multiple heart beats, the corresponding part is usually removed from the dataset. Both single beat and multiple beat artifact corrections can affect the spectral estimates obtained from heart rate variability data. However, it remains unclear what level of artifact correction is still acceptable for reliable calculation of spectral HRV parameters. To provide insight in the effect of artifact correction on spectral estimates of heart rate variability, in this paper artifact correction is simulated using heart rate variability data that were originally free of artifacts. Datasets with varying levels of corrected artifacts are generated and analyzed spectrally. The resulting spectral estimates are compared to the results calculated from the original heart rate variability data.
II. METHODOLOGY
A. Datasets
1) Neonatal data: Neonatal heart rate data are obtained
from 3-leads ECG recordings in a neonatal intensive care unit (NICU). The ECG signals are recorded at a sample rate of 240 Hz using a General Electric Solar 8000M patient
The Effect of Artifact Correction on Spectral Estimates of Heart
Rate Variability
Chris Peters, Rik Vullings, Jan Bergmans, Guid Oei, Pieter Wijn
m d q a th f a s s E N d d th f a s n ti I a d th g l d d a monitor. From detected and, b quadratic fit is a dataset conta hat were detec five segments o are selected fo series for one o
2) Fetal dat
single lead sca ECG signal is Neoventa STA detected withou dataset contain
hat were detec five segments o are selected fo series for one o
B. Simulatio For each of t neonatal datase imestamps are n this way b artifacts are si desired level o he original s generated at 10 ength, steps of different data data segment, r analyzed.
Fig. 1a. R-R inte
Fig. 1b. R-R inte
these ECG si because of the used to reduce aining time stam
cted in the rec of 192 s that a or analysis. Fi of the segments
ta: Fetal heart
alp ECG recor recorded at a AN s31 fetal m ut the use of a ning time stam cted in the rec
of 192 s that a or analysis. Fi of the segments on of artifact co the 10 segment et and 5 segm e deleted at ran both single be imulated. Tim of corrected art segments, arti 0 different lev f 5%). For each segments are resulting in a to
erval series of neon
erval series of fetal
ignals, R-peak e relatively low e sampling erro mps for all R-orded signal. F are completely ig. 1a shows s of 192 s.
rate data are rding in a del sample rate o monitor. R-peak a quadratic fit. mps for all R-p orded signal. F are completely ig. 1b shows s of 192 s. orrection ts of 192 s (5 s ments from th ndom position eat artifacts an mestamps are d tifacts is achie ifact correcte vels (5 to 50% h level of corre generated fro otal of 2000 da
natal data segment
l data segment III
k occurrences w sample rate ors. This results peak occurren From this data y free of artifac the R-R inter obtained from livery room. T f 500 Hz usin k occurrences This results in peak occurren From this data y free of artifac the R-R inter egments from he fetal datas s in the segme nd multiple b deleted until eved. For each d segments % of the segm ected artifacts, om each origi ata segments to t III are e, a s in nces aset cts, rval m a The g a are n a nces aset cts, rval the et), ent. beat the h of are ment 20 inal o be C. The t calculat R-R int Each of that ov separate average variance equidist Fourier into an resampl using a with a surface with a P subtract offset, energie the app divided window After series, t Table I are calc Fig. 2 the tota each of the pow the feta power i from 0. relative Table correspo neonata Sym VLF LF HFA HFN TP LFn HFn LF/H Spectral analy timestamps in te R-R interva terval series, th f the segments verlap for 50% ely, after wh ed to obtain e. The R-R in tantly distribut transform, th equidistant se ling the R-R i a sample & ho square wave t equal to 1. Ne Parzen window tion of the me the power s s in the calcul plication of the d by the squa w, to correct for r calculating th the power in v contains an o culated. 2a to 2c show al power calcul f the levels of wers in the LF al heart rate da in the HF band 4 to 1.5 Hz. F deviation from II shows the onding standa al and fetal). OVERV mbol F power in ve power in lo power in hi adults power in hi newborns total power LF power in HF power i HF LF/HF ratio ysis each of the da al series. For s he fast Fourier of 192 s is div %. Each subs
ich the result one power ntervals in the ted in time. To he R-R interva et of data. Thi intervals at a f old technique a that has a wid ext, the resamp w, to reduce sp ean R-R interv spectrum is c lated power sp e Parzen-wind ared Fourier t r the convoluti he power spec various freque overview of th III. RESULT the powers in lated for the ne artifact correc F and HF band ata. The HF po d as defined for or all calculate m their theoret e mean relat ard deviation TABLEI VIEW OF SPECTRAL Description ery low frequency ow frequency range
igh frequency rang igh frequency rang r (outside VLF) n normalized units in normalized unit o ata segments a spectral analys transform (FF vided in 5 subs et of 64 s is ting power s spectrum with e data segmen o be able to ca als must be tr is data set is frequency of 4 and convolutin dth of 0.5 seco pled dataset is pectral leakage val value to re calculated. Fi pectrum are co dow and the sp
transform of t on prior to resa ctra of the R-ency bands is he spectral esti TS the LF and HF eonatal heart ra ction. Fig. 3a t d and the total ower that is sho r newborns, wh ed spectral esti tical value is d tive deviation for all data
L ESTIMATES Freque range < 0.04 H e 0.04 – 0 ge for 0.15 – 0 ge for 0.4 – 1. 0.04 – s ts are used to is of these FT) is used. sets of 64 s s analyzed pectra are h reduced nts are not alculate the ransformed created by 4 Hz, after ng the data onds and a multiplied e, and after emove DC inally, the orrected for pectrum is the square ampling. -R interval calculated. imates that F band and ate data for to 3c show power for own, is the hich ranges imates, the determined. n and the asets (both ency range Hz 0.15 Hz 0.4 Hz .5 Hz 1.5 Hz 2670
Fig. 3a. LF p Fig. 2a. LF po
Fig. 2b. HF po
Fig. 2c. Total p
power for fetal hear ower for neonatal h
ower for neonatal h
power for neonata
rt rate segments o_ heart rate segments
heart rate segment
al heart rate segme
_I to o_V s n_I to n_V ts n_I to n_V ents n_I to n_V artif leve 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% artif leve 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% Fig. 3c Fig. 3b RELATIVE DEV f. el VLF % 0.1 ± 0.4 % 0.6 ± 0.9 % 0.3 ± 2.8 % 2.0 ± 2.7 % 1.1 ± 4.0 % 2.1 ± 3.0 % 3.7 ± 8.7 % 1.9 ± 5.6 % 12.5 ± 18.4 % 13.0 ± 14.5 f. el TP % -1.9 ± 0.8 % -4.0 ± 1.1 % -6.5 ± 3.1 % -7.2 ± 4.1 % -8.9 ± 6.5 % -11.3 ± 10.4 % -9.8 ± 13.2 % -16.4 ± 7.3 % -12.5 ± 23.8 % -13.1 ± 27.3 . Total power for . HF power for fe
TABLEI
VIATION FROM THE
LF -0.8 ± 0.8 -1.8 ± 1.2 -3.1 ± 3.2 -3.0 ± 4.0 -3.3 ± 7.1 -5.8 ± 8.4 -2.9 ± 10.5 -7.8 ± 6.3 -2.4 ± 27.6 -2.3 ± 28.3 LFn 1.1 ± 0.6 2.4 ± 0.9 3.8 ± 1.2 4.9 ± 1.7 6.3 ± 1.8 6.7 ± 3.1 8.3 ± 4.5 10.5 ± 3.2 12.6 ± 3.6 13.9 ± 5.0 fetal heart rate seg etal heart rate segm
II EORETICAL VALUE HF adult -3.6 ± 2.9 -7.7 ± 3.2 -11.3 ± 3.9 -12.6 ± 9.0 -16.8 ± 9.9 -14.0 ± 32.4 -12.6 ± 36.7 -30.0 ± 18.0 -24.7 ± 29.7 -24.5 ± 42.2 HFn -3.0 ± 1.1 -6.3 ± 1.9 -10.8 ± 2.6 -14.9 ± 4.2 -19.3 ± 5.6 -22.8 ± 7.0 -30.1 ± 11.2 -28.9 ± 5.5 -38.7 ± 11.7 -42.7 ± 12.4 gments o_I to o_V ments o_I to o_V
ES [%] HF newborn -4.6 ± 1.1 -10.0 ± 2.2 -16.5 ± 4.5 -20.3 ± 3.1 -26.1 ± 3.7 -31.4 ± 5.5 -35.8 ± 5.8 -39.9 ± 5.2 -47.2 ± 8.0 -50.4 ± 9.2 LF/HF 4.2 ± 1.4 9.5 ± 3.1 16.9 ± 6.0 22.9 ± 3.6 31.8 ± 8.9 39.5 ± 13.2 54.7 ± 12.9 56.4 ± 14.7 89.8 ± 43.3 102.0 ± 36.0 V
IV. DISCUSSION
Fig. 2a and 3a show that even when relatively large parts (up to 25%) of heart rate data are missing due to the correction of artifacts, calculated powers in the low frequency range only slightly deviate from their theoretical values (values at 0% corrected artifacts). At higher levels of artifact correction, the standard deviation of the calculated values rapidly increases. For powers in the very low frequency range results are similar and even higher levels of artifact correction may be acceptable. Interestingly, in the very low frequency range powers are slightly overestimated, while in the low frequency range, powers are slightly underestimated.
Fig. 2b and 3b show that already at very low levels of artifact correction, calculated powers in the high frequency range for newborns significantly deviate from their theoretical values. For the averaged relative deviation in HF power as presented in Table II, a linear inverse relationship exists between the averaged relative deviation and the level of artifact correction (y = -1.024 x, R = 0.99, p < 0.001). However, a large number of datasets was used for these calculations. In general, HF power will not be constant throughout a dataset and the actual deviation will strongly depend on which part of the dataset contains the artifacts.
Powers in the high frequency range for adults also show a significant deviation from their theoretical values, but the averaged relative deviation is smaller than for the high frequency range for newborns. The total power calculated in the range between 0.04 and 1.5 Hz (Fig. 2c and 3c) shows an underestimation of the theoretical value. This underestimation is smaller than the underestimation in the high frequency range for adults and newborns, which is due to the relatively large power in the low frequency range.
V. CONCLUSION
For neonatal and fetal heart rate data, the power in the very low frequency and low frequency range can be calculated reliably for datasets in which up to 25% of the heart rate data are missing due to artifact correction. Powers in the high frequency range, for both adults and newborns, cannot be calculated reliably when heart rate data are missing due to artifact correction. For spectral analysis of fetal heart rate data in clinical practice, this is a major limitation, as these data will virtually always be corrupted by artifacts. A possible solution for this limitation is the use of alternative methods for spectral analysis. Wavelet analysis for example, enables analysis of powers in the high frequency range at a shorter time scale than the analysis of powers in the low frequency range. Short segments of heart rate data that are free of artifacts can then be used to calculate powers in the high frequency range reliably, while segments that contain artifacts are excluded. In future work, the application of a wavelet-based method of analysis will be explored.
Spectral estimates in the low frequency range of the fetal heart rate appear to be clinically most relevant for fetal
monitoring [5]. For clinically obtained fetal heart data, the power within this range can be calculated reliably after artifact correction, as long as the corrected data do not exceed 25% of the length of the segment to be analyzed. However, if normalized powers are used (LFn = LF/Total power), the total power should also be calculated reliably. Total power contains the power in the high frequency range, and therefore, for normalized powers the same limitations exist as for spectral estimates in the high frequency range.
REFERENCES
[1] S. Akselrod, D. Gordon, F.A. Ubel, D.C. Shannon, A.C. Berger, R.J. Cohen, “Power spectrum analysis of heart rate fluctuation: a quantitative probe of beat-to-beat cardiovascular control,” Science, vol. 213(4504), Jul. 1981, pp. 220-222.
[2] J.M. Karemaker, “Heart rate variability: why do spectral analysis?,”
Heart, vol. 77(2), Feb. 1997, pp. 99-101.
[3] G.D. Clifford, L. Tarassenko, “Quantifying errors in spectral estimates of HRV due to beat replacement and resampling,” IEEE Trans.
Biomed. Eng., vol. 52(4), Apr. 2005, pp. 630-638.
[4] P.V. Nielsen, B. Stigsby, C. Nickelsen, J. Nim, “Intra- and inter-observer variability in the assessment of intrapartum cardiotocograms,” Acta Obstet. Gynecol. Scand., vol. 66(5), 1987, pp. 421-424.
[5] J.O.E.H. van Laar, M.M. Porath, C.H.L. Peters, S.G. Oei, “Spectral analysis of fetal heart rate variability for fetal surveillance: review of the literature,” Acta Obstet. Gynecol. Scand., vol. 87(3), 2008, pp. 300-306.
[6] S.M. Siira, T.H. Ojala, T.J. Vahlberg, J.O. Jalonen, I.A. Välimäki, K.G. Rosén, E.M. Ekholm, “Marked fetal acidosis and specific changes in power spectrum analysis of fetal heart rate variability recorded during the last hour of labour,” BJOG, vol 112(4), Apr. 2005, pp. 418-423.
[7] T. Rantonen, E.M. Ekholm, S.M. Siira, T. Metsälä, R. Leino, U. Ekblad, I.A. Välimäki, “Periodic spectral components of fetal heart rate variability reflect the changes in cord arterial base deficit values: a preliminary report,” Early Hum. Dev., vol. 60(3), Jan. 2001, pp. 233-238.
[8] S. Cerutti, S. Civardi, A. Bianchi, M.G. Signorini, E. Ferrazzi, G. Pardi, “Spectral analysis of antepartum heart rate variability,” Clin.
Phys. Physiol. Meas., vol. 10 Suppl. B, 1989, pp. 27-31.
