Clinical experience with an MR-compatible 12-lead ECG during routine cardiac MRI and interventional MRI: beat detection performance evaluation

Background: Synchronization of cardiac MRI (CMR) and interventional CMR (iCMR) to the cardiac cycle is essential for image quality and procedural efficiency [1,2]. Arrhythmias and magnet/gradient-induced artifacts challenge beat detection (BD) [3], increasing rescans and extending procedure time. We report the use of an MR-compatible 12-lead ECG system (MiRTLE Medical, North Andover, MA, USA) and a new digital filtering approach to improve BD during iCMR and routine CMR, and benchmark performance against ≥60-year volunteers to age-match patients.

Methods: Setting & hardware. Single-site study at Amsterdam UMC under local IRB on a 1.5 T scanner (MAGNETOM Sola, Siemens Healthineers).
Cohorts. Patients n=9 (iCMR: AF ablation n=4; VT ablation n=1 [4]; routine CMR n=4). Volunteers: n=4 (subset of prior healthy cohort aged ≥60 y).
Instrumentation & algorithm. MiRTLEView MR-compatible 12-lead BD device. Using data from one AF ablation subject (AF01), we developed a Blind Source Separation (BSS) based approach [5] with automated “best-source” selection via Naïve-Bayesian-like Gaussian scoring of independent peak features to emphasize QRS-like components. The same approach was then applied across the remaining iCMR, routine CMR, and volunteer datasets.
Analysis. Online BD outputs were compared to manual “true beat” labels from two reviewers. Matches within ±25 ms were true positives (TP); missed beats were false negatives (FN); detections outside the window were false positives (FP). Per-subject Positive Predictive Value (PPV), True Positive Rate (TPR; sensitivity), and F1 were computed and summarized as mean ± SD.

Results: Thirteen datasets (9 patients + 4 ≥60 y volunteers; 47,355 labeled beats) were analyzed. With the new BSS-based approach, per-subject performance was PPV 0.91 ± 0.11, TPR 0.86 ± 0.08, and F1 0.88 ± 0.09. By subgroup, performance was highest in volunteers (PPV/TPR/F1: 0.93 ± 0.06 / 0.91 ± 0.07 / 0.92 ± 0.06) and routine CMR (0.92 ± 0.08 / 0.88 ± 0.07 / 0.89 ± 0.08). For iCMR ablations, AF cases achieved 0.98 ± 0.03 / 0.82 ± 0.06 / 0.89 ± 0.04; including the single VT case yielded an overall iCMR mean of 0.90 ± 0.17 / 0.81 ± 0.06 / 0.85 ± 0.11, with VT representing the lowest PPV (0.60) and F1 (0.66). Relative to the live run using the prior approach, mean F1 increased from 0.64 to 0.88 overall (Δ + 0.24), driven by gains during ablation (AF: 0.24 → 0.89; VT: 0.22 → 0.66), while volunteer and routine CMR performance remained high.

Conclusion: An MR-compatible 12-lead ECG with BSS-based filtering and automated source selection delivered consistently high BD accuracy across routine CMR and iCMR, with one VT outlier. These findings support technical feasibility and motivate prospective studies evaluating real-time gating, image quality, and workflow impact. Limitations include the small, single-site cohort and parameter derivation from one subject; broader multicenter studies across arrhythmia types are planned.

Table 1. Beat Detection Performance by MiRTLE.

			Live Run						New BSS-based cleaning and BD
SUBJECT		TRUE Beats	TP	FN	FP	PPV	TPR	F1	TP	FN	FP	PPV	TPR	F1
H01	healthy	1612	1183	429	302	0.80	0.73	0.76	1336	276	203	0.87	0.83	0.85
H02	healthy	1363	1265	98	47	0.96	0.93	0.95	1323	70	25	0.98	0.95	0.97
H03	healthy	1350	1230	120	102	0.92	0.91	0.92	1185	165	126	0.90	0.88	0.89
H04	healthy	1124	1107	17	54	0.95	0.98	0.97	1107	17	25	0.98	0.98	0.98
CMR01	CMRI	2885	2808	77	113	0.96	0.97	0.97	2748	137	92	0.97	0.95	0.96
CMR02	CMRI	2273	2109	164	179	0.92	0.93	0.92	1956	317	217	0.90	0.86	0.88
CMR03	CMRI	2854	2101	753	605	0.78	0.74	0.76	2227	627	530	0.81	0.78	0.79
CMR04	CMRI	1945	1698	247	30	0.98	0.87	0.92	1764	181	33	0.98	0.91	0.94
AFA01	iCMRI	4693	1498	3195	3687	0.29	0.32	0.30	4256	437	52	0.99	0.91	0.95
AFA02	iCMRI	4794	1366	3428	4844	0.22	0.28	0.25	3757	1037	40	0.99	0.78	0.87
AFA03	iCMRI	6775	1666	5109	6736	0.20	0.25	0.22	5448	1327	82	0.99	0.80	0.89
AFA04	iCMRI	10160	1810	8350	9691	0.16	0.18	0.17	8134	2026	591	0.93	0.80	0.86
VTA01	iCMRI	5527	1402	4125	5912	0.19	0.25	0.22	4092	1435	2715	0.60	0.74	0.66

Figure 1. Summary of MiRTLE BD performance improvement