Magnetohydrodynamic Distortion Correction for In-Bore Acquired 12-lead ECGs during MRI-guided Cardiac Interventions in Pigs

Background: Reliable and clinically interpretable electrocardiographic (ECG) monitoring inside the magnetic resonance imaging (MRI) environment is essential for advancing interventional cardiac MRI (iCMR), especially MR-guided ablations. However, ECG recording systems must overcome additional noise, such as the magnetohydrodynamic (MHD)-effect. This phenomenon, caused by blood flow in the static magnetic field, distorts the ECG, which is particularly present in the ST-segment and T-wave, complicating accurate monitoring and interpretation which can be vital during ablations. Therefore, an MHD-correction strategy is necessary to improve the quality of in-bore 12-lead ECG recordings.

Methods: Three healthy pigs underwent real-time MR-guided ventricular ablation in a 1.5-Tesla MRI system (Aera, Siemens, Erlangen, Germany). Continuous 12-lead ECGs were recorded during the procedure using a CE-marked MR-conditional ECG system (MiRTLE, MiRTLE Medical, North Andover, MA, USA). Two ECG segments were extracted: 1) a clean (~50-100 sec.) segment recorded in the MRI-room but outside the bore, and 2) MHD-affected segments (~1000 sec.) obtained while the pig was positioned in-bore without scanning. A correction was applied by constructing an average MHD template per lead and subtracting it beat-wise, creating an MHD-corrected dataset (Fig. 1). The effectiveness was quantified by Pearson correlation (J-point to end T-wave) of MHD-corrected and uncorrected beats against the clean template. Statistical analysis included paired-samples t-tests, Cohen’s kappa, and a linear mixed model (LMM).

Results: In total 53,400 beats were analyzed (153 clean vs. 53,247 MHD-affected beats). The LMM indicated that the MHD-correction was associated with a small decrease in correlation (β = −0.015, p < 0.01). The effect of the MHD-correction per subject was heterogeneous: paired t-tests showed a negligible change in correlation for subject 1 (dz = 0.11), a medium worsening for subject 2 (dz = −0.77), and a moderate improvement for subject 3 (dz = 0.45) (Table 1). Lastly, per-lead analysis showed small negative effects in six leads (aVR, aVL, V2, and V4-V6) (Table 2).

Conclusion: This study demonstrates, for the first time, continuous in-bore 12-lead ECG acquisition during MRI-guided interventions in a porcine model. The recordings supported procedural monitoring, underscoring the clinical potential of multi-lead ECG inside the MRI. However, the tested template-based subtraction method was insufficient to correct the ECG while the subject is in-bore, where non-stationary MHD morphology, due to differences in heart rate and cardiac output, and gradient/radiofrequency artefacts coexist. This further explains the difference in performance in pigs, highlighting the multifactorial nature of the MHD-effect. While limited improvements were achieved in one subject, these results provide insights into MHD waveform characteristics and establish a foundation for more robust filtering strategies.

Paired-sample t-test on corrected vs uncorrected (MHD). Per-subject paired comparison of beat-to-template correlation before and after MHD correction. N (beats) = number of paired beats. Mean r values = Pearson correlations (statistical tests were performed on Fisher-z transformed correlations). Δr = corrected – uncorrected ECGs. dz = Cohen’s dz for paired samples. T (paired) = paired t-test on Fisher-z; two-sided p-values are shown. Asterisks mark statistical significance.

Subject	n (beats)	Mean r (MHD)	Mean r (Corrected)	Δr	dz (Cohen's dz)	t (paired)	p-value
Subject 1	8844	0.148	0.152	0.004	0.112	10.515	< 0.001**
Subject 2	19644	0.191	0.117	-0.074	-0.772	-108.22	< 0.001**
Subject 3	24912	-0.013	0.016	0.030	0.450	71.016	< 0.001**

Linear mixed model for Fisher-z transformed correlation per lead. Per-lead mixed-effects estimates of the correction effect on beat–template correlation. Values are Δz = Corrected – uncorrected on the Fisher-z scale. SE = model-based standard error. 95% CI = Wald confidence interval. p-value = two-sided Wald test. Estimates come from an LMM fit. Per-lead effects were obtained as linear contrasts. Asterisks mark statistical significance.

Lead	Δz (Corrected - MHD)	SE	[95% CI]	p-value
I	0.002	0.005	[-0.008; 0.013]	0.663
II	-0.009	0.005	[-0.019; 0.002]	0.101
III	0.001	0.005	[-0.009; 0.012]	0.782
aVR	-0.016	0.005	[-0.026; -0.005]	0.004 **
aVL	-0.017	0.005	[-0.027; -0.006]	0.002 **
aVF	0.001	0.005	[-0.009; 0.012]	0.789
V1	-0.009	0.005	[-0.020; 0.001]	0.077
V2	-0.039	0.005	[-0.050; -0.029]	< 0.001 **
V3	-0.010	0.005	[-0.020; 0.001]	0.069
V4	-0.017	0.005	[-0.028; -0.007]	0.001 **
V5	-0.032	0.005	[-0.043; -0.022]	< 0.001 **
V6	-0.038	0.005	[-0.049; -0.028]	< 0.001 **

Example of the average MHD effect and clean template vs median for subject 2. (Top panel) The average MHD effect waveform before (dashed blue, “Original avg_mhd”) and after padding to align with the J-point (solid orange, “Padded avg_mhd”). An example MHD-contaminated ECG beat is shown for reference (gray trace). The vertical dashed red line indicates the J-point, marking the onset of the MHD effect and start of the evaluation window for the correlation calculations. (Bottom panel) Comparison of the clean ECG template (solid blue) with median beats from the MHD-contaminated signal (dashed orange) and the MHD-corrected ECG (green). The vertical dashed black line indicates the R-peak.