Computation of the gradient-induced electric field noise in 12-lead ECG traces during rapid MRI sequences

Abstract

Background: Successful physiological monitoring using a 12-lead ECG during MR imaging is essential for safe conduction of cardiovascular interventions within a MR scanner. However, ECG artifacts induced by magnetic field gradients severely affect the signal quality and fidelity. Previously, the gradient-induced artifacts were reduced by blocking ECG transmissions during all gradient ramps [1], which has been shown feasible while the method is not suitable for short-TR sequences. Theoretical and experimental studies have shown a linear relationship between electric fields and the temporal derivatives of the magnetic field gradients [2,3]. We propose an algorithm to restore the true ECG signal by subtracting system response functions, based on the MR gradient signals, from ECG signals distorted by gradient interference. Methods: Data Acquisition: An MRI-conditional 12-lead ECG system [1] was used to acquire data on two healthy volunteers inside a 3T MRI. Outside the MRI room, high-fidelity ECG traces, along with the x, y and z gradient waveforms were digitally recorded simultaneously at 62kHz. Balanced SSFP sequences with various slice orientations (axial, coronal, sagittal and oblique) were acquired. Data Analysis: The gradient-induced ECG noise was computed as the difference between aligned ECG traces with and without MR sequence running. The noise voltage (Vni) at each electrode (i) was modeled as a linear combination of gradient derivatives and system factors, Vni = ai*dGx/dt+bi*dGy/dt+gi*dGz/dt+Ci, where ai, bi, gi and Ci are positiondependent. These parameters were then used to reconstruct the noise, for comparison with the measured ECG noise, and to further derive the restored ECG. Results: The recorded ECG traces and low-pass filtered gradient derivatives are displayed in Figure 1a. The computed noise vector (Vni) and the measured noise (Figure 1b) had differences of 21% +/- 20% in normalized Euclidean distance. The restored ECG signal was comparable to the clean ECG segments (Figure 2), providing higher signal quality and fidelity relative to low-frequency filtering of the ECG signal. Vectorial display of the fitted parameters (Figure 3) demonstrated systematic changes across the precordial leads, and varied in magnitude between subjects. Conclusions: The gradient-derivative model closely fit the measured ECG noise, possibly allowing for efficient gradient-noise removal utilizing rapid calibration scans, combined with hardware blocking of extremely high noise intervals. (Table Presented).