Minimally spaced electrode positions for multi-functional chest sensors: ECG and respiratory signal estimation

Michael Klum; Tobias Minn; Timo Tigges; Alexandru-Gabriel Pielmus; Reinhold Orglmeister

doi:10.1515/cdbme-2016-0151

Article Open Access

Minimally spaced electrode positions for multi-functional chest sensors: ECG and respiratory signal estimation

Michael Klum , Tobias Minn , Timo Tigges , Alexandru-Gabriel Pielmus and Reinhold Orglmeister

Published/Copyright: September 30, 2016

Published by

Become an author with De Gruyter Brill

Author Information Explore this Subject

From the journal Current Directions in Biomedical Engineering Volume 2 Issue 1

Abstract

Unobtrusive medical instrumentation is a key in continuous patient monitoring. To increase compliance, multi-functional sensor concepts and measurement sites different from gold-standards are used. In this work, we aim to combine both approaches. We focus on minimally spaced electrode positions with high signal correlations to gold-standards. We present twofold experimental data from six and eleven healthy volunteers and provide chest positions with individual correlations up to 0.83 ± 0.06 for ECG and 0.73 ± 0.28 for the respiratory frequency. Using a performance index, we assess positions with correlations up to 0.77 ± 0.12 for ECG and 0.65 ± 0.35 for the respiratory frequency with 24 mm electrode distance.

Keywords: chest; ecg derived respiration; electrocardiogram; minimally spaced; multi-functional sensors; respiratory signal estimation; unobtrusive monitoring

1 Introduction

In recent years, a strong trend towards the development of unobtrusive medical instrumentation is evident [1]. Especially in the context of home monitoring, patient acceptance is a key aspect for acquiring continuous data.

Increasing miniaturization, integration and focus on acceptance comes at the cost of potentially using uncommon measurement sites. In the electrocardiogram (ECG) case, the electrodes in standard lead systems are spaced far apart, thus reducing comfort and needing body-spanning wiring. To increase compactness, different measurement sites and plaster approaches have been proposed [2]. Another concept is to extract parameters from other biosignals by using multi-functional sensors [3]. A prominent example is the estimation of respiratory signals from the ECG, the ECG derived respiration (EDR).

For ECG it was shown that diagonal electrode pairs around the positions V2 to V4 provide the best QRS-complex and P-wave detection [4]. Using the ePatch system, clinically usable ECG recordings at the sternum were generated and extracting a derived 12-lead ECG from three systems at three different positions was possible [5].

Different methods are available for directly and indirectly assessing the respiration pattern. Most methods use the exhaled air by attaching a sensor to the airways [6]. If, however, an ECG signal is available, an estimate of the respiratory signals can be derived, potentially making other sensors obsolete. When analyzing lead positions with respect to EDR quality, locations at the standard positions V1 and V2 and above did show best results [7], [8].

In this work we aim to contribute to the field of unobtrusive medical monitors. With focus on multi-functional sensors we provide recording sites on the chest with both high correlation to standard ECG leads and good EDR performance. We present twofold experimental results with the goal of determining optimal chest positions for minimally spaced single-lead ECG recordings and EDR extraction, thus combining the two described approaches of localization and multi-functional electrodes.

2 Methods

We first recorded the ECG at 27 differential chest positions, organized in nine groups as shown in Figure 1, along with Einthoven leads I, II and III. The horizontal electrode distance was 24 mm, the vertical distance was defined by the intercostal spaces. We measured six healthy volunteers (24.7 ± 1 years) in an upright sitting position for 300 seconds. After visual inspection, 60 s intervals were chosen. Pre-processing included band-pass filtering (5 Hz to 90 Hz) and 50 Hz notch filtering. Thereafter, the correlation coefficient (CC) was calculated between each differential position and the standard Einthoven leads. The largest absolute of the three resulting CCs for each position pos is given by |C⁢C^E⁢C⁢Gp⁢o⁢s|.

Figure 1

Electrode positions for minimally spaced recording.

In the second experimental stage we used 15 differential positions organized in five groups, chosen on the basis of the results of the first stage. Additionally, a reference respiration signal using respiratory inductance plethysmography was recorded. The data were derived from 11 healthy, normally breathing volunteers (30.7 ± 11.4 years) in an upright sitting position for 300 s. After visual inspection, 180 s intervals were chosen. Pre-processing included band-pass filtering (5 Hz to 140 Hz) and 50 Hz notch filtering. Three EDR algorithms have been implemented. The used ECG features are visualized in Figure 2.

Figure 2

ECG features used for EDR algorithms.

The EDR derived from the heart rate variability (HRV) [9] was calculated as the time difference between the R peaks in the QRS complex as given by Equation 1.
(1)E⁢D⁢RH⁢R⁢V⁢(n)=tQ⁢R⁢S⁢(n+1)−tQ⁢R⁢S⁢(n)
The QRS amplitude strongly correlates to the respiration. As non-standard QRS-complexes are expected in the recorded data and the R-to-S amplitude is more reliable than the R-to-baseline amplitude [10], the EDR value was calculated as the R-to-minimum value in 250 ms window around the R-peak as given in Equation 2.
(2)E⁢D⁢RA⁢M⁢P⁢(n)=VR⁢(n)−Vm⁢i⁢n⁢(n)
The third algorithm implemented is based on the linear PCA. Windows of 250 ms around each R-peak are used to cut out and combine QRS complexes. The eigenvector of the covariance matrix with the largest eigenvalue was calculated using PCA, which strongly corresponds to the respiratory signal [11].

For direct comparison, the magnitude squared coherence (MSC) between the EDR signals and the reference signal was calculated. The maximum value of the MSC in the frequency range of 0 Hz to 1 Hz was used as a similarity measure in the frequency domain. A mean MSC value over all three available EDR methods for each position was derived (M⁢S⁢C¯R⁢E⁢S⁢Pp⁢o⁢s).

Using the EDR as well as the reference signals, the respiratory frequency was calculated in windows of 10 s and the CC was used to assess their linear dependency in the time domain. A mean CC value over all three available EDR methods for each position was derived (C⁢C¯F⁢R⁢E⁢Qp⁢o⁢s).

Finally, the performance index given in Equation 3 was calculated for each position pos using the three available similarity measures; the largest absolute ECG correlation |C⁢C^E⁢C⁢Gp⁢o⁢s|, the mean respiration coherence M⁢S⁢C¯R⁢E⁢S⁢Pp⁢o⁢s and mean absolute respiratory frequency correlation |C⁢C¯F⁢R⁢E⁢Qp⁢o⁢s|. The latter two resulted from averaging over the three available EDR methods as described before.

(3)i⁢d⁢xp⁢o⁢s=|C⁢C^E⁢C⁢Gp⁢o⁢s|+M⁢S⁢C¯R⁢E⁢S⁢Pp⁢o⁢s+|C⁢C¯F⁢R⁢E⁢Qp⁢o⁢s|3

3 Results

The results of the first experimental stage are given in Table 1. They are organized along the 27 minimal lead positions which are further divided into nine groups with three minimally spaced leads each. For each of these positions, the correlation coefficients to the three standard Einthoven leads are given. The three highest absolute correlations for individual leads have been obtained at position C.I to Einthoven II (0.75 ± 0.19), D.II to Einthoven I (0.83 ± 0.06) and F.I to Einthoven III (−0.77 ± 0.12).

Table 1

CCs (mean ± SD) between minimally spaced and Einthoven leads with three largest CCs marked.

ECG	EH.I	EH.II	EH.III
A.I	0.21 ± 0.38	−0.64 ± 0.22**	−0.73 ± 0.15**
A.II	0.64 ± 0.21**	0.11 ± 0.29	−0.10 ± 0.26
A.III	0.62 ± 0.22**	0.33 ± 0.32	0.13 ± 0.31
B.I	0.33 ± 0.41	−0.54 ± 0.21**	−0.68 ± 0.17**
B.II	0.72 ± 0.13**	0.35 ± 0.21*	0.15 ± 0.18
B.III	0.64 ± 0.19**	0.72 ± 0.20**	0.59 ± 0.23**
C.I	0.48 ± 0.53	0.75 ± 0.19**	0.65 ± 0.29**
C.II	0.58 ± 0.44*	0.67 ± 0.31**	0.53 ± 0.42*
C.III	0.60 ± 0.38*	0.62 ± 0.35*	0.48 ± 0.46
D.I	0.80 ± 0.11**	0.42 ± 0.36*	0.22 ± 0.37
D.II	0.83 ± 0.06**	0.25 ± 0.32	0.02 ± 0.30
D.III	0.76 ± 0.10**	0.09 ± 0.23	−0.13 ± 0.20
E.I	0.32 ± 0.34	−0.58 ± 0.25**	−0.72 ± 0.19**
E.II	0.57 ± 0.32*	−0.27 ± 0.18*	−0.47 ± 0.11**
E.III	0.58 ± 0.30**	−0.03 ± 0.35	−0.23 ± 0.30
F.I	0.33 ± 0.41	−0.63 ± 0.16**	−0.77 ± 0.12**
F.II	0.72 ± 0.23**	−0.08 ± 0.40	−0.27 ± 0.38
F.III	0.68 ± 0.08**	0.33 ± 0.34	0.18 ± 0.40
G.I	0.44 ± 0.23**	−0.32 ± 0.37	−0.49 ± 0.33*
G.II	0.76 ± 0.07**	0.06 ± 0.45	−0.15 ± 0.45
G.III	0.72 ± 0.06**	0.24 ± 0.43	0.08 ± 0.48
H.I	0.64 ± 0.19**	−0.30 ± 0.27	−0.51 ± 0.24**
H.II	0.64 ± 0.27**	−0.25 ± 0.19*	−0.47 ± 0.12**
H.III	0.58 ± 0.32*	−0.20 ± 0.24	−0.41 ± 0.15**
K.I	0.52 ± 0.30*	−0.42 ± 0.28*	−0.63 ± 0.22**
K.II	0.62 ± 0.20**	−0.27 ± 0.54	−0.45 ± 0.52
K.III	0.53 ± 0.18**	−0.1 ± 0.61	−0.30 ± 0.61

Three largest CCs for each standard lead are bold. *:p < 0.05, **:p < 0.01.

For the next experimental stage we chose the three groups which contain the minimal lead with the highest correlation to a particular Einthoven lead; namely group C, D and F. In addition, we chose group B and G, as they contain potentially interesting positions for unobtrusive sensing applications.

Figure 3 shows a recorded minimally spaced ECG signal along with the reference respiratory signals, the three EDR estimations thereof and the windowed respiratory frequency derived from all four respiratory signals.

Figure 3

Sample minimally spaced ECG signal, reference respiratory signal, EDR and extracted respiratory frequency feature.

The MSC values between the reference signal and the EDRs are given in Table 2. The derived values are in the same range for all positions and show a high maximum coherence in the analyzed frequency band. The highest mean coherence was found for the positions C.III, F.II and G.III (0.85 ± 0.09 each). Regarding the EDR methods, the linear PCA outperformed the two other methods on average, closely followed by the QRS amplitude concept.

The correlation between the windowed respiratory frequency derived from the reference signal and the EDR signals show higher dependencies on the positions as given in Table 3. The three strongest correlating positions are C.II (0.73 ± 0.28), C.III (0.69 ± 0.28) and F.III (0.67 ± 0.33). Here, the linear PCA method outperformed the two other methods to a much higher degree, followed by the QRS amplitude concept.

Table 2

MSC (mean ± SD) between reference and EDR signals.

EDR	HRV	QRS Amp.	Lin. PCA	𝑴⁢𝑺⁢𝑪¯𝑹⁢𝑬⁢𝑺⁢𝑷𝒑⁢𝒐⁢𝒔
B.I	0.81 ± 0.10	0.81 ± 0.11	0.80 ± 0.13	0.81 ± 0.11
B.II	0.80 ± 0.10	0.78 ± 0.13	0.82 ± 0.11	0.80 ± 0.11
B.III	0.82 ± 0.08	0.83 ± 0.10	0.81 ± 0.11	0.82 ± 0.10
C.I	0.80 ± 0.13	0.81 ± 0.16	0.86 ± 0.10	0.82 ± 0.13
C.II	0.84 ± 0.09	0.86 ± 0.06	0.83 ± 0.09	0.84 ± 0.08
C.III	0.83 ± 0.10	0.86 ± 0.08	0.86 ± 0.10	0.85 ± 0.09
D.I	0.76 ± 0.14	0.74 ± 0.12	0.72 ± 0.15	0.74 ± 0.13
D.II	0.82 ± 0.09	0.79 ± 0.16	0.81 ± 0.14	0.81 ± 0.13
D.III	0.82 ± 0.09	0.84 ± 0.14	0.85 ± 0.12	0.84 ± 0.12
F.I	0.79 ± 0.10	0.86 ± 0.08	0.87 ± 0.08	0.84 ± 0.09
F.II	0.80 ± 0.09	0.87 ± 0.09	0.88 ± 0.08	0.85 ± 0.09
F.III	0.80 ± 0.09	0.85 ± 0.12	0.86 ± 0.09	0.84 ± 0.10
G.I	0.84 ± 0.07	0.82 ± 0.09	0.78 ± 0.14	0.81 ± 0.11
G.II	0.85 ± 0.08	0.83 ± 0.09	0.84 ± 0.10	0.84 ± 0.09
G.III	0.84 ± 0.08	0.86 ± 0.11	0.86 ± 0.09	0.85 ± 0.09
Mean	0.81 ± 0.02	0.83 ± 0.034	0.83 ± 0.04

Three positions with maximum average MSC are marked, previously found ECG positions are bold. p< 0.01 for all values.

Table 3

CCs (mean ± SD) between windowed respiratory frequency from reference signal and EDR estimations.

Freq.	HRV	QRS Amp.	Lin. PCA	\|𝑪⁢𝑪¯𝑭⁢𝑹⁢𝑬⁢𝑸𝒑⁢𝒐⁢𝒔\|
B.I	0.32 ± 0.27	0.47 ± 0.56	0.55 ± 0.46	0.45 ± 0.44
B.II	0.35 ± 0.28	0.45 ± 0.44	0.66 ± 0.32	0.49 ± 0.36
B.III	0.35 ± 0.28	0.49 ± 0.38	0.59 ± 0.32	0.47 ± 0.33
C.I	0.50 ± 0.31	0.61 ± 0.32	0.62 ± 0.46	0.58 ± 0.36
C.II	0.58 ± 0.29	0.75 ± 0.26	0.86 ± 0.23	0.73 ± 0.28
C.III	0.58 ± 0.29	0.69 ± 0.30	0.80 ± 0.21	0.69 ± 0.28
D.I	0.56 ± 0.28	0.19 ± 0.36	0.30 ± 0.44	0.35 ± 0.39
D.II	0.63 ± 0.33	0.45 ± 0.43	0.55 ± 0.43	0.54 ± 0.39
D.III	0.63 ± 0.32	0.50 ± 0.49	0.56 ± 0.42	0.56 ± 0.41
F.I	0.42 ± 0.37	0.73 ± 0.35	0.80 ± 0.23	0.65 ± 0.35
F.II	0.42 ± 0.38	0.58 ± 0.40	0.65 ± 0.33	0.55 ± 0.37
F.III	0.45 ± 0.39	0.72 ± 0.30	0.82 ± 0.16	0.67 ± 0.33
G.I	0.47 ± 0.35	0.59 ± 0.35	0.66 ± 0.31	0.57 ± 0.34
G.II	0.53 ± 0.31	0.57 ± 0.46	0.79 ± 0.21	0.63 ± 0.35
G.III	0.50 ± 0.34	0.61 ± 0.50	0.65 ± 0.35	0.59 ± 0.40
Mean	0.49 ± 0.10	0.56 ± 0.14	0.66 ± 0.14

Three positions with maximum average CC are marked, previously found ECG positions are bold. p < 0.01 for all values.

4 Discussion

Minimally spaced electrodes as defined in Figure 1 can be vital for patient compliance. Using a twofold experimental setup, we identified three chest positions for deriving estimates of standard ECG leads and respiration signals. By using 24 mm of horizontal inter-electrode distance, we effectively halved their spacing compared to other studies.

In the first experiment, we identified positions for estimating standard Einthoven (EH) leads based on the correlation coefficient (CC); D.II for EH.I (0.83 ± 0.06), C.I for EH.II (0.75 ± 0.19) and F.I for EH.III (−0.77 ± 0.12). These findings correspond to the literature in that leads in the direction of the electrical heart-axis and around the standard leads V1 to V4 provide a good signal quality.

We extracted respiratory signals using EDR and compared them with a reference with the MSC. Our results indicate that basic respiratory features such as a long-term average frequency can be obtained even from adverse positions. To assess their usability for derived features, we calculated the windowed respiratory frequency. We found higher position dependencies in the CC; C.II (0.73 ± 0.28), C.III (0.69 ± 0.28) and F.III (0.68 ± 0.33).

We evaluated the performance index given in Equation 3 which combines the ECG CC with the MSC and the CC derived from the respiratory signal and the respiratory frequency. The positions F.I (0.754), C.II (0.746) and G.II (0.743) scored best as shown in Table 4.

Table 4

Performance index for each position. The three positions with the largest index are marked.

Position	𝒊⁢𝒅⁢𝒙𝒑⁢𝒐⁢𝒔	Position	𝒊⁢𝒅⁢𝒙𝒑⁢𝒐⁢𝒔	Position	𝒊⁢𝒅⁢𝒙𝒑⁢𝒐⁢𝒔
B.I	0.644	C.III	0.719	F.II	0.704
B.II	0.671	D.I	0.630	F.III	0.727
B.III	0.671	D.II	0.728	G.I	0.624
C.I	0.716	D.III	0.720	G.II	0.743
C.II	0.746	F.I	0.754	G.III	0.720

In summary, we examined 27 positions on the chest for minimally spaced ECG recordings down to 24 mm inter-electrode distance. In addition to the correlation with standard ECG leads, we analyzed the applicability of EDR methods using a 15 position subset. Despite using a relatively small sample size, we found three distinct positions (F.I, C.II, G.II) which combine both good EDR quality and high correlations to standard Einthoven leads. Using this information will be critical in designing small-sized unobtrusive multi-purpose sensor systems. Future research may address recording additional signals using actimetry, body-sounds or reflective photoplethysmography, thus broadening the range of applications while maintaining a compact system design.

Author’s Statement

Research funding: The author state no funding involved. Conflict of interest: Authors state no conflict of interest. Informed consent has been obtained from all individuals included in this study. Ethical approval: The research related to human use has been complies with all the relevant national regulations, institutional policies and in accordance the tenets of the Helsinki Declaration.

References

[1] Y.-L. Zheng, et al. Unobtrusive sensing and wearable devices for health informatics. Biomedical Engineering, IEEE Transactions on, 2014;61:1538–54.10.1109/TBME.2014.2309951Search in Google Scholar

[2] D. R. Zhang, et al. A wireless ecg plaster for real-time cardiac health monitoring in body sensor networks. Biomedical Circuits and Systems Conference (BioCAS), 2011 IEEE, pages 205–208. IEEE, 2011.10.1109/BioCAS.2011.6107763Search in Google Scholar

[3] R. Trobec, et al. Multi-functionality of wireless body sensors. The IPSI BgD transactions on internet research, page 23, 2014.Search in Google Scholar

[4] Puurtinen M, Viik J, Hyttinen J. Best electrode locations for a small bipolar ecg device: signal strength analysis of clinical data. Ann Biomed Eng. 2009;37:331–6.10.1007/s10439-008-9604-ySearch in Google Scholar

[5] Hansen IH, Hoppe K, Gjerde A, Kanters JK, Sorensen HB. Comparing twelve-lead electrocardiography with close-to-heart patch based electrocardiography. In Conf Proc IEEE Eng Med Biol Soc. 2015;2015:330–3.10.1109/EMBC.2015.7318366Search in Google Scholar

[6] Folke M, Cernerud L, Ekström M, Hök B. Critical review of non-invasive respiratory monitoring in medical care. Med Biol Eng Comput. 2003;41:377–83.10.1007/BF02348078Search in Google Scholar

[7] A. Sobron, et al. Evaluation of methods for estimation of respiratory frequency from the ecg. Comput Cardiol. 2010;513–516.Search in Google Scholar

[8] R. Trobec, et al. Two proximal skin electrodes—a respiration rate body sensor. Sensors (Basel). 2012;12:13813–28.10.3390/s121013813Search in Google Scholar

[9] Bernardi L, Porta C, Gabutti A, Spicuzza L, Sleight P. Modulatory effects of respiration. Auton Neurosci. 2001;90:47–56.10.1016/S1566-0702(01)00267-3Search in Google Scholar

[10] Mason Cl, Tarassenko L. Quantitative assessment of respiratory derivation algorithms. Engineering in Medicine and Biology Society, 2001. Proceedings of the 23rd Annual International Conference of the IEEE 2001;2:1998–2001.Search in Google Scholar

[11] Langley P, Bowers EJ, Murray A. Principal component analysis as a tool for analyzing beat-to-beat changes in ecg features: application to ecg-derived respiration. IEEE Trans Biomed Eng. 2010;57:821–9.10.1109/TBME.2009.2018297Search in Google Scholar PubMed

Published Online: 2016-9-30

Published in Print: 2016-9-1

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Articles in the same Issue

https://doi.org/10.1515/cdbme-2016-0151

Keywords for this article

chest; ecg derived respiration; electrocardiogram; minimally spaced; multi-functional sensors; respiratory signal estimation; unobtrusive monitoring

Creative Commons

BY-NC-ND 4.0