CMR in arrhythmia patients with AutoCMR self-gated, 3D cine

Background: Cardiovascular magnetic resonance (CMR) provides functional metrics that are key to the evaluation of cardiovascular health^1–3. To resolve cardiac function, most CMR acquisitions rely on synchronization to an ECG, which often leads to artifacts in patients with cardiac motion inconsistencies caused by arrhythmia. Self-gating acquisitions have shown promise in rejecting irregular heartbeats, leading to preserved image quality in arrhythmia patients^4–6. AutoCMR, a push-button, free-breathing exam providing self-gated 3D cine⁷, was recently proposed with the goal of reducing technical barriers to CMR. With reduced barriers, there is potential to enable CMR in sites lacking CMR-trained technologists, but arrhythmia-robust imaging will be required. In this work, we aim to demonstrate the feasibility of AutoCMR cine imaging in a small cohort of patients with arrhythmia.

Methods: Four patients with history of arrhythmia were identified from the AutoCMR cohort. With IRB approval, 3 patients were scanned at 3T by a CMR technologist, and 1 patient was scanned at 1.5T in a community health center by a technologist with no CMR training. AutoCMR images were acquired with 1.6mm isotropic resolution and 307mm³ FOV (GRE, TE/TR≤4.4/2.9ms, FA≤15°). Self-gating lines (acquired with kx=ky=0) in the region of the heart were identified using automated image segmentation and analyzed with PCA and Fourier analysis to identify the cardiac signal (Figure 1)^8–11. A Savitzky-Golay¹² filter was applied, followed by peak detection. Cine data was adaptively binned into 30 cardiac frames, automatically rejecting data acquired during peak-to-peak (RR) intervals outside 1 standard deviation from the mode. Results were compared to conventional ECG-gated 2D cine images (2DCMR).

Results: Patients 1 (DCM, sinus arrhythmia, ventricular tachycardia), 2 (atrial fibrillation), and 3 (frequent premature ventricular complexes) were observed to have arrhythmia during conventional 2DCMR and AutoCMR. In all three patients, resulting motion artifacts are observed in the 2DCMR images, but not in AutoCMR images (Figure 2). Motion plots (single LV projection over time) highlight spatio-temporal blurring and contrast abnormalities in 2DCMR compared to AutoCMR.

Patient 4 (Turner’s syndrome, PDA, accelerated junctional rhythm) was observed to be in regular rhythm during 2DCMR and the first half of the AutoCMR cine acquisition, and to have frequent arrhythmia during the remaining exam (Figure 3). Corresponding arrhythmia was observed AutoCMR self-gating signal (Figure 3a) and RR-interval distribution (Figure 3b). Despite arrhythmia beginning mid-scan, the resulting AutoCMR cine image quality is preserved compared to 2DCMR (Figure 3c-d).

Conclusion: In a small cohort of patients, AutoCMR cine is observed to be robust to cardiac arrhythmia through rejection of data collected during heartbeats with outlier RR-intervals. Arrhythmia-robust acquisitions will be important in enabling CMR in mixed-use MRI sites lacking CMR technologists.

Figure 1. Description of cardiac binning with automated outlier rejection using data from Patient 3. a, Image segmentation is used to select the self-navigating signals near the heart. b, Principal component analysis (PCA) and Fourier frequency analysis are performed to identify the cardiac signal. c, Peak finding is applied to the cardiac signal after application of a Savitzky-Golay filter. d, The RR-interval between neighboring peaks is calculated and data acquired outside of the acceptance window (mode RR interval ± 1 standard deviation, shown in red) is rejected.
Figure 2. In three example patients with arrhythmia during scanning, 2DCMR images (a, e, i) are observed to have resulting motion blurring artifacts, while AutoCMR (ACMR) images (c, g, k) are artifact free with well-defined cardiac motion. Motion plots, generated by plotting a projection through the LV (blue) over time, show spatial and temporal blurring (red arrows) in 2DCMR images (b, f, j) compared to AutoCMR (d, h, k).
Figure 3. Patient 4 was observed with ECG monitoring to be in normal rhythm during the 2DCMR acquisition and the first 5 minutes of the 10-minute AutoCMR cine acquisition (gray), and to have frequent arrhythmia during the remaining 5 minutes (blue). The beginning of frequent arrhythmia was similarly observed 5 minutes into the AutoCMR exam on both the self-gating signal (a) and the detected RR-interval (b). With automatic rejection of data acquired during irregular heartbeats, the resulting AutoCMR images (d) are comparable to 2DCMR images acquired in normal rhythm.