Distortion Analysis of 12-lead ECG Recordings During Cardiac MRI

Background: Accurate and clinically reliable real-time 12-lead electrocardiographic (ECG) recording during cardiac magnetic resonance (CMR) is crucial for improved 1) patient monitoring, 2) cardiac triggering, 3) real-time integration of electrical (ECG) and mechanical (CMR) data, and 4) mandatory for in-bore MRI-guided cardiac interventions. However, ECGs acquired inside the MRI bore are distorted by magnetohydrodynamic (MHD) artifacts from the static magnetic field and gradient-induced voltages (GIVs) from time-varying gradients. This study systematically quantified distortions in 12-lead ECGs during CMR by separating continuous MHD artifacts from GIVs, assessing lead-dependent susceptibility for both and MRI sequence-dependent susceptibility for GIVs only.

Methods: Continuous 12-lead ECGs were recorded in patients w/o cardiac implantable electronic devices undergoing a routine clinical CMR scan (1.5T, MAGNETOM Sola, Siemens Healthineers, Forchheim, Germany), using an MRI-compatible ECG recording system (MiRLTE Medical, North Andover, MA, USA). Reference clean ECGs were recorded outside the MRI. MHD-related artefacts were quantified using maximum absolute cross-correlation and amplitude analysis (Figure 1). GIVs were assessed by the ECG signal-to-noise ratio (SNR) across various sequences, including cine imaging (bSSFP and FLASH), T1/T2 mapping (bSSFP readout), and 2D/3D LGE (FLASH readout) in both four-chamber and short-axis orientations.

Results: A total of 25 patients were included. For MHD analysis, an average of 49.2±15.8 beats were assessed per patient. Mean cross-correlation between clean and MHD-affected beats was 0.63±0.04, with high interpatient variability. MHD effects were most pronounced in limb leads, particularly lead I, showing the lowest cross-correlation (0.56±0.12) and highest amplitude (0.40 mV) (Figure 2). In contrast, precordial leads (V1–V6) showed significantly higher cross-correlations than extremity leads (0.66±0.09 vs. 0.60±0.10; 𝑝=0.012). GIVs varied by sequence and affected ECG SNR differently (Table 1). Among cine sequences, bSSFP was associated with significantly lower ECG SNR compared to FLASH (𝑝< 0.001). Precordial leads demonstrated higher SNR than limb leads (𝑝< 0.001). No differences were observed between T1 and T2 mapping. 3D LGE had higher ECG SNR than 2D LGE (𝑝< 0.001). ECG SNR did not differ significantly between short-axis and four-chamber orientations.

Conclusion: This study quantified MHD and GIVs affecting 12-lead ECG recordings in a 1.5T MRI. Precordial leads were less affected by MHD and GIVs, indicating their clinical reliability for ECG monitoring and triggering during MRI. Sequences with slower gradient slew rates, such as FLASH cine and 3D LGE, are less prone to GIVs artifacts, but may be associated with increased scan time and compromised image quality. Therefore, improving the quality of a 12-lead ECG in the MRI scanner depends on removal of both MHD and sequence-specific distortions, while preserving important clinical ECG features.

Table 1: Summary of SNR comparisons across groups. Median SNR values (in dB) with interquartile range (IQR) and Mann-Whitney U test p-values are shown. The number of subjects for which an imaging sequence was used is indicated as (n=).

Comparison	Group	Median SNR (IQR) (dB)	Significance (𝑝-value)
Precordial vs limb leads	Precordial Limb	-33.26 (9.56) -38.22 (8.89)	𝑝 < 0.0001
V1-V3 vs V4 – V6	V1 – V3 V4 – V6	-32.64 (10.57) -33.74 (9.04)	0.0039
Short-axis vs 4- chamber view	Short-axis 4-chamber	-35.96 (8.68) -35.96 (10.08)	0.7802
Cine bSSFP vs FLASH	Cine bSSFP (n=22) Cine FLASH (n=8)	-38.67 (7.27) -36.33 (9.03)	𝑝 < 0.0001
2D vs 3D LGE	2D LGE (n=19) 3D LGE (n=6)	-33.60 (8.65) -27.72 (9.22)	𝑝 < 0.0001
T1 vs T2 Mapping	T1 map (n=8) T2 map (n=6)	-35.80 (8.80) -37.74 (10.53)	0.0794

Figure 1: Flowchart of the methodology of the study. ECG and MRI data was simultaneously recorded during a routine diagnostic scan. Subsequently, the impact of the MHD effect and imaging-related noise on the ECGs was assessed. The lower left panel shows the MHD extraction consisting of R-peak detection (using independent component analysis), beat extraction and beat-per-beat subtraction of the MHD-affected beat and clean beat. This method was performed on 60-second clean and MHD ECG segments. The right lower panel shows the determination of the signal-to-noise ratio (SNR) for different imaging sequences after alignment, selection of sequences and planes, and extraction of the gradient-induced voltages (GIVs) using a threshold-based algorithm.
Figure 2: Mean MHD effect. MHD effect is visualized for individual cases and the mean MHD effect (red) over 20 patients. Highest MHD amplitudes are seen in lead I and aVL and for most cases minimal effect in the precordial leads.