ECG Electrode Placements for Magnetohydrodynamic Voltage Suppression

Abstract

This study aims to investigate a set of electrocardiogram (ECG) electrode lead locations to improve the quality of four-lead ECG signals acquired during magnetic resonance imaging (MRI). This was achieved by identifying electrode placements that minimized the amount of induced magnetohydrodynamic voltages (V_MHD) in the ECG signals. Reducing V_MHD can improve the accuracy of QRS complex detection in ECG as well as heartbeat synchronization between MRI and ECG during the acquisition of cardiac cine. A vector model based on thoracic geometry was developed to predict induced V_MHD and to optimize four-lead ECG electrode placement for the purposes of improved MRI gating. Four human subjects were recruited for vector model establishment (Group 1), and five human subjects were recruited for validation of V_MHD reduction in the proposed four-lead ECG (Group 2). The vector model was established using 12-lead ECG data recorded from Group 1 of four healthy subjects at 3 Tesla, and a gradient descent optimization routine was utilized to predict optimal four-lead ECG placement based on V_MHD vector alignment. The optimized four-lead ECG was then validated in Group 2 of five healthy subjects by comparing the standard and proposed lead placements. A 43.41% reduction in V_MHD was observed in ECGs using the proposed electrode placement, and the QRS complex was preserved. A V_MHD-minimized electrode placement for four-lead ECG gating was presented and shown to reduce induced magnetohydrodynamic (MHD) signals, potentially allowing for improved cardiac MRI physiological monitoring.

1. Introduction

The success of cardiovascular magnetic resonance imaging (MRI) depends largely on heartbeat synchronization between electrocardiograms (ECG) and MRI as well as the accuracy of real-time QRS complex detection in ECGs [1,2,3]. This presents a problem for MRI solutions since ECGs acquired during MRI are largely considered to be non-diagnostic [4].

The distortion of ECG signals is caused by the magnetohydrodynamic (MHD) effect, in which an induced voltage is overlaid onto ECG recordings, due to the interaction between the static magnetic field of the MRI (B₀) and rapidly moving blood ejected into the aortic arch during early systole [2,5,6,7,8,9]. The resultant ECG signal distortion from these sources can lead to intermittent QRS complex detection and subsequent motion-based image artifacts, prolonged scan time, and may require repeat scans [10]. These issues are exacerbated at higher field strengths and with irregular or elevated heart rates.

Several approaches currently exist to achieve ECG acquisition for physiological monitoring. The predominant technique relies on the reduction of inter-electrode distances to 5–10 cm in four-lead ECGs, thereby decreasing induced noise levels by decreasing electrode distance from the synthesized Wilson Central Terminal (WCT) [11]. Additional approaches exist to reduce magnetohydrodynamic voltages (V_MHD) through software adaptive filtering and independent component analysis (ICA). However, these methodologies are computationally intensive and are difficult to realize in real time for the purposes of MRI scanner gating [10,12,13,14]. The presented work was inspired by a study conducted by Dimick et al. to optimize ECG electrode placement for cardiac MRI gating by selecting electrode placements that were less susceptible to V_MHD because the electrical path of V_MHD was obscured by the left lobe of the lung [15].

In this study, we investigate four-lead ECG electrode placements to reduce the magnitude of induced V_MHD obtained in intra-MRI ECG recordings. The derivation of the proposed electrode placements is intended to serve as a methodology to determine electrode placements that minimize V_MHD. The methodology includes three central steps: (1) development of an MHD vector model; (2) four-lead ECG electrode position optimization; and (3) evaluation of the efficacy of V_MHD reduction. Part of the work in this manuscript has been presented in the 19th Annual Society for Cardiovascular Magnetic Resonance Scientific Sessions in Los Angeles, CA, USA, 27–30th January 2016 [16].

2. Materials and Methods

Ethical approval of human trials was obtained from the University of Georgia Office of Research Institutional Review Board (IRB)—reference number MOD00003704. Informed consent to participate in the study and for publication of their individual details in this manuscript was obtained from all participants.

2.1. MHD-Vector Model

In order to accurately represent ECG reference planes in 3D space and predict measurements on the body surface, a thoracic geometric model was extracted from ECGSIM (StITPro, Nijmegan, the Netherlands), an open source simulation program designed to study the relationship between the heart’s electrical activity and resulting signals on the body surface [17,18]. An accurate model of the aortic arch was incorporated into the thoracic cavity based on simulations of the MHD effect taken from the literature [19,20]. Standardized locations for ECGs based on subject anthropometry [21,22] were used to select nodal locations on the thorax for the standard ECG electrode positions [23] (

Table 1. Standardized location of electrodes used during 12-lead electrocardiograph (ECG) recording.

Electrode	Location	Mean Electrical Vector	ECGSIM Node
RA, LA	Right/Left shoulder; avoiding thick muscle	Refer to Hexaxial Plane (Figure 1)	1, 2
RL, LL	Right/Left pelvic iliac crest	Refer to Hexaxial Plane (Figure 1)	9, 56
V1	4th intercostal space; left sternal border	Interventricular Septum	19
V2	4th intercostal space; right sternal border	Interventricular Septum	26
V3	Midway between V2 and V4	Left/Right ventricle anterior wall	33
V4	5th intercostal space; mid-clavicular line	Left/Right ventricle anterior wall	41
V5	5th intercostal space; anterior axillary line	Left ventricle lateral wall	48
V6	5th intercostal space; mid axillary line	Left ventricle lateral wall	54

).

The induced V_MHD vector intra-MRI can be predicted using the cross-product of the magnetic field ($\stackrel{\rightharpoonup }{B_0}$) and the direction of blood flow in the aortic arch ($\stackrel{\rightharpoonup }{u}$), across an electrode pair of separation distance L, assuming the primary contribution to $\stackrel{\rightharpoonup }{V_{MHD}}$ is due to arch blood flow (Equation (1)) [8,24,25]. We will commonly reference the magnitude of this vector, V_MHD, in this study.

$$\stackrel{\rightharpoonup }{V_{MHD}}={{\displaystyle \int }}_0^L\stackrel{\rightharpoonup }{u}\times \stackrel{\rightharpoonup }{B_0}·dL$$

(1)

The predicted V_MHD vector was calculated based on an applied magnetic field in the +Z direction and the direction of flow in the aortic arch.

2.1.1. Human Subjects

Nine human subjects were recruited in this study and subjects were separated into Group 1 (four subjects) and Group 2 (five subjects). ECG recordings of Group 1 were used for MHD-vector model establishment (Section 2.1.3), and ECG recordings of Group 2 were used to validate V_MHD reduction in ECG signals acquired with the proposed four-lead electrode placements (Section 2.3). All ECGs in this study were recorded at Emory University Hospital using a Nasiff CardioResting 12-lead ECG monitoring system (Central Square, NY, USA [26]) modified to be 3 Tesla (3T) MRI-compatible with a Siemens Trio TIM 3T scanner based on the method described in [10].

2.1.2. ECG Recordings for MHD-Vector Model Establishment

First, 12-lead ECG recordings were acquired for each subject in Group 1 during a 20 s breath-hold with the subject completely outside of the MRI scanner bore to register baseline physiological readings while avoiding the induction of motion and gradient artifacts into the ECG recordings. Next, the subject was positioned supine on the MRI scanner table and inserted head-first into the scanner bore. Finally, 12-lead ECG recordings were taken with the subject’s heart at the isocenter of the MRI scanner (3T). The purpose of the Group 1’s 12-lead ECG recording was to establish the MHD-vector model.

2.1.3. MHD-Vector Model Validation

The V_MHD vector predicted in Section 2.1.1 was overlaid onto an anatomical reference plane to allow observation of the relationship between the predicted V_MHD and mean electrical vectors recorded from ECG leads in Group 1.

The correlation coefficients between 12-lead ECG traces obtained inside (ECG_real + V_MHD) and outside (ECG_real) the MRI bore were determined for each subject and averaged. The correlation coefficients followed the conventional definition where 1 represents perfect correlation and 0 represents no correlation. The mean inverse correlation coefficients were then calculated by subtracting the mean correlation coefficients from one. Therefore, the scale used for the inverse correlation coefficients ranged from 0 to 1, with 0 representing perfect correlation and 1 representing no correlation. When the signals matched (inverse correlation coefficient = 0), there was no V_MHD present. When the signals did not match (inverse correlation coefficient = 1), there was maximum V_MHD present. Thus, the inverse correlation coefficients represented the quantity of V_MHD present in the ECG taken inside the MRI bore at each of the 12 leads.

The MHD-vector model was used to predict the V_MHD vector. A vector model of the mean electrical vectors from 12-lead ECGs acquired from Group 1 was constructed and overlaid on the thoracic body model in MATLAB (Mathworks, Natick, MA, USA). This updated thoracic body model was used to calculate the angular displacement between the predicted V_MHD vector and each of the 12 leads. The angular displacements represented the quantity of V_MHD predicted to be present in each of the 12 ECG leads based on the MHD-vector model.

The mean inverse correlation coefficients and angular displacements calculated for each lead were plotted in the same graph to compare the trends of the actual and predicted V_MHD. Spearman’s rank correlation coefficient was calculated between the inverse correlation coefficients and angular displacements for each of the 12 leads.

2.2. Electrode Position Optimization

To predict if unconventional electrode placements would be able to obtain ECGs with less induced V_MHD, an optimization routine was implemented. The cost function was V_MHD, and the input data were the electrode placement. Each point in the thorax geometry was taken and grouped into sets of four considering all possible combinations, and its alignment with respect to the heart and the induced MHD vector was measured. From the optimization routine, sets of electrode positions corresponding to maximum and minimum predicted induced V_MHD were deduced. A line search method based on conjugate gradient descent was used to find the electrode set corresponding to a global minimum of the predicted V_MHD [27]. This method was employed across the thorax.

2.3. Validation of V_MHD Reduction

ECGs of each of the five subjects in Group 2 were recorded both outside the MRI bore during a 20 s breath hold and with the subject’s heart at the isocenter of the same MRI scanner used on Group 1 at 3T during a 20 s breath-hold using (i) the standard four-lead electrode placement for cardiac triggering in MRI for systems with a wireless triggering module [28], and (ii) the optimized V_MHD-minimized electrode placement. QRS complexes were manually identified, time-stamped, and verified by a cardiologist to segment each cardiac cycle individually. Cardiac cycles were time-scaled to 60 beats per minute (bpm) as a baseline time scale for accurate and direct comparison [10,14,29,30].

The purpose of the Group 2’s ECG recording was to validate the MHD-vector model for isolation of MHD signals at certain positions of the ECG electrode placement in the human torso so as to enhance ECG gating during cardiac MR imaging. The ability of the proposed electrode positions to optimize induced V_MHD for the purposes of cardiac synchronization was evaluated based on the percentage of MHD suppression in the P-wave, QRS complex, and S-T segment of the ECG recordings.

3. Results

3.1. MHD Vector Model

In Figure 1b, the addition of the extremity leads’ hexaxial reference plane onto the frontal plane illustrates qualitatively two major areas of vectorial alignment:

• The superior to inferior alignment of aVF (shown as a light blue arrow pointing down) with the direction of the MRI magnetic field, B₀ (shown as a red arrow), resulting in a predicted minimization of induced V_MHD;
• The lateral alignment of aVL (shown as a light blue arrow pointing right) with the V_MHD vector (shown as a red arrow pointing left), resulting in a predicted V_MHD maxima. The remaining augmented lead (aVR, shown as a light blue arrow pointing left) and limb leads (I, II, and III, shown as dark blue arrows) are shown to be directed in an approximately 30° projection off the V_MHD and B₀ vectors, predicting a similar trend to that of lead I and aVF, respectively.

In Figure 1c, the mean electrical vectors of planar precordial lead signals are illustrated on a transverse thoracic slice. Progressing from lead V1 to V6, it can be observed that the signals approached parallel alignment with the predicted direction of induced V_MHD.

Some representative ECGs acquired from Group 1 inside the MRI bore are plotted in Figure 2a. The thoracic body model used to calculate the angular displacements between the ECG leads and the predicted V_MHD vector is displayed in Figure 2b. The inverse correlation coefficients and angular displacements for each lead are shown in Figure 2c to illustrate the relationship between trends in actual and predicted V_MHD.

Through the evaluation of the inverse correlation coefficients in relationship to the angular displacements between the predicted V_MHD vector and each electrode, a trend can be observed (Spearman’s rank correlation coefficient = 0.80). A peak in induced V_MHD was observed at aVL due to its close parallel alignment to V_MHD, and a gradual increase in V_MHD was detected as the precordial electrode placements shifted laterally around the chest (V4–V6). Among the 12 leads, limb leads I and II and precordial leads V1 and V2 yielded the least amount of induced V_MHD.

3.2. Electrode Position Optimization

As observed in the thoracic model, the V_MHD-minimized locations had limited aorta visibility compared to standard positioning due to the left superior lung lobe (Figure 3a,b). Based on mapped ECG body surface potentials using ECGSIM, it was predicted that these locations would also correlate to QRS complexes with high amplitudes (Figure 3c). From the body surface maps obtained using ECGSIM [17], a less than 3% decrease in the signal amplitude of the R_peak at the body surface was predicted for the V_MHD-minimized four-lead electrode placement compared to the standard four-lead electrode placement.

3.3. Validation of V_MHD Reduction

The ECGs of the subjects in Group 2 are displayed in Figure 4. Four-lead ECGs were acquired with the standard and V_MHD-minimized electrode placements during 20 s breath-holds outside (Figure 4a) and inside (Figure 4b) the MRI bore.

In ECGs acquired using the V_MHD-minimized electrode placements, a 43.41% reduction in V_MHD was observed, and the QRS complex was preserved.

Table 2. Evaluation of magnetohydrodynamic (MHD) reduction in each subject for QRS complex detection.

	Level of MHD Suppression in ECG Segments (Peak-to-Peak)
	P-Wave	QRS Complex	S-T Segment
Subject 1 M/20 yrs old/108 kg	0.44 mV (25.41%)	0.27 mV (11.32%)	2.55 mV (64.38%)
Subject 2 M/20 yrs old/72 kg	0.06 mV (23.19%)	0.14 mV (17.25%)	0.87 mV (55.39%)
Subject 3 M/21 yrs old/68 kg	0.13 mV (18.50%)	0.36 mV (38.91%)	0.52 mV (28.21%)
Subject 4 M/27 yrs old/82 kg	0.09 mV (33.72%)	0.18 mV (44.61%)	0.10 mV (19.27%)
Subject 5 M/26 yrs old/77 kg	0.33 mV (52.64%)	0.07 mV (10.36%)	0.89 mV (49.89%)
Mean	0.21 mV (30.69%)	0.20 mV (24.49%)	0.99 mV (43.41%)
Standard Deviation	0.16 mV (13.45%)	0.12 mV (16.11%)	0.93 mV (18.96%)

outlines the level of MHD reduction in each ECG segment as demonstrated in each subject.

4. Discussion

Twelve-lead ECGs were processed in Group 1 (four healthy subjects) to establish an MHD-vector model based on thoracic geometry obtained from ECGSIM. Leads found to align with the predicted direction of the induced V_MHD vector were shown to maintain the highest levels of V_MHD. Similarly, induced V_MHD was minimized as the angle between the electrode’s mean electrical vector and the predicted direction of induced V_MHD increased. Among the 12-lead electrode placements, limb leads I and II and precordial leads V1 and V2 exhibited the least amount of induced V_MHD, whereas the remaining leads showed increased signal magnitudes. Then an optimization routine was performed using the vector model to determine novel electrode positions for four-lead ECG that exhibited minimal induced V_MHD.

The proposed V_MHD-minimized electrode positions, illustrated in Figure 3b, were validated with ECG recordings acquired from five healthy subjects. A 43.41% reduction in V_MHD was observed, so the quality of the ECG signal was improved. Additionally, the R_peak amplitude was maintained since only a 3% decrease in amplitude was observed. These improvements may lead to improved capabilities for heartbeat synchronization between ECG and MRI as well as real-time QRS complex detection during MRI. Therefore, ECG would be able to yield improved cardiac monitoring during MRI scanning and interventional procedures without any additional hardware or software. This would be a significant development because intra-MRI ECG is currently considered to be non-diagnostic.

The method discussed in this paper focuses only on MHD suppression. The removal of gradient artifacts is not within the scope of this study. However, future studies should also consider the effects of fast switching gradient magnetic fields used during MRI on the ECG signal. Like the MHD effect, fast switching gradient magnetic fields cause interfering signals that are superimposed on the ECG signal [31,32]. Thus, many techniques have been developed to reduce the effects of these signals. For example, gradient artifacts can be reduced with algorithms based on adaptive filter techniques [33], Bayesian filters [34], and independent component analysis [35]. For a more thorough evaluation of the improvement in QRS detection provided by the presented method to reduce V_MHD, the ECG recordings may need to be preprocessed to reduce the effects of gradient artifacts.

5. Limitations

As this method was only evaluated in a limited number of healthy subjects, its accuracy and precision must be verified in a larger population and in patients diagnosed with cardiac arrhythmias. The ECG recording time should also be increased so that a larger number of QRS complexes can be evaluated from each subject. Once the population size and recording time have been increased, it may be appropriate to calculate the sensitivity and specificity of QRS complex detection to evaluate the improvement of the ECGs obtained with the V_MHD-minimized four-lead electrode placement compared to the standard four-lead electrode placement. Additionally, the effect of each subject’s breathing motion on the ECG should be taken into consideration because this movement can have a major impact on QRS detection. Similarly, the effect of the angle of the aorta on electrode positioning should be further examined as it may lead to the formation of subject-specific ECG recommendations.

6. Conclusions

A set of electrode placements to minimize V_MHD in intra-MRI four-lead ECG was presented to further improve cardiac gating standards and then validated in subjects. Future research will include applying this method in a population of patients with different cardiac arrhythmias because abnormal ECGs may contain signatures which are different from healthy ECGs and, thus, difficult to isolate from the V_MHD signal.