Evaluation of cardiac magnetic resonance thermometry in patients
Valery Ozenne1, Solenn Toupin1,2, Pierre Bour1, Baudouin Denis de Senneville3, Alexis Vaussy2, Matthieu Lepetit-Coiffé2, Pierre Jaïs4, Hubert Cochet4, and Bruno Quesson1

1Electrophysiology and Heart Modeling Institute, Bordeaux, France, 2Siemens Healthcare, Paris, France, 3Mathematical Institute of Bordeaux, Bordeaux, France, 4Department of Cardiac Electrophysiology, Hôpital Cardiologique de Haut-Lévêque, Bordeaux, France


Recent studies have proposed to monitor radiofrequency ablation on the heart using real-time MR-thermometry. Methods rely on ECG triggering which can fail in presence of arrhythmia. This study evaluates the precision of MR-thermometry on patients (N=15) even in presence of cardiac arrhythmia. Phase images were acquired using a single-shot multi-slice echo planar imaging and temperature maps were calculated and displayed on the fly. ECG was recorded simultaneously for further analysis of cardiac rhythm and post-processing of temperature images. Stability of temperature mapping without RF-heating was evaluated in each pixel and correlated to the prevalence of arrhythmia.


Catheter-based radiofrequency ablation (RFA) is a widely accepted method for clinical treatment of cardiac arrhythmia, with 200 000 procedures performed in Europe in 2014 [1]. However, both safety and efficacy could be improved by monitoring lesion formation in real time, thereby reducing recurrences which often require redo procedures [2]. MR-thermometry has the potential to visualize lesion formation in real-time during RFA [3]. Recent studies reported reliable cardiac thermometry [4, 5] during RFA by combining ECG-triggered imaging and online correction of respiratory motion and associated susceptibility artifacts. However, variations in cardiac cycle duration may lead to erroneous temperature estimate. The purpose of the present study is to evaluate the precision of MR-thermometry without RF-heating, including in patients with cardiac arrhythmia and to demonstrate feasibility of the method in clinical environment.


Patients: the study was approved by the Institutional Review Board and all subjects (N=15, 58.5 ± 16.2 years old and eleven male) gave informed consent to be included in the study. Five patients were in sinus rhythm and did not show substantial variations of the RR duration during scanning, whereas ten patients displayed irregular rhythm. MR Imaging: 4 to 5 temperature slices in coronal orientation were acquired sequentially under free breathing at each heartbeat during approximately 3’30 minutes on a 1.5 T MRI (Avanto, Siemens Healthcare). The sequence was a single shot gradient EPI (TE=20ms, TR=85ms, Grappa=2) with 110x110 voxels corresponding to a 1.6x1.6x3mm3 voxel size (zero filled to 0.8x0.8x3mm3). Image reconstruction, correction of residual in-plane respiratory motion and associated susceptibility variations, compensation of spatial-temporal phase drift and low pass temporal filtering were implemented in the Gadgetron framework [6], ensuring online visualization of temperature images [5]. To assess the precision of cardiac MR-thermometry, the temporal standard deviation of temperature (σT) over time was computed in each voxel from all slices during the procedure. The distribution of σT values was analyzed on a manually drawn ROI surrounding the left ventricle. Voxels where σT was higher than 5°C were removed from the statistical analysis. Categorization of Beats: ECG was recorded using standard 3-lead (ECG) acquisition. An algorithm was designed to identify, synchronize and categorize beats based on two successive RR durations, as proposed by Contijoch et al [7].


The mean ± SD heart rate across all patients was 69 ± 10 bpm, ranging from 54 to 90 bpm (Table 1). Data were first sorted in three groups: 1) patients in sinus rhythm, 2) patients with low frequency of event of arrhythmia (less than 15% prevalence) and 3) patients with high frequency of event of arrhythmia (more than 15% prevalence). In each group, a representative case study is detailed in Figures 1, 2 and 3, respectively. The statistical analysis of temperature distribution over the patients is presented in Figure 4 using a box and whisker plot. Over the patients in group 1, σT was 1.57 ± 0.34 °C with a total of 4343 ± 1817 voxels per slice. Only 3 % of the voxels in the ROI were excluded due to residual phase unwraps or σT higher than 5°C. In group 2, σT was 1.83 ± 0.70 °C (7 % of voxels excluded) and 2.17 ± 0.46 °C in group 3 (19 % of voxels excluded).


Temperature uncertainty remains below 2°C in more than 60% of the voxels of the left ventricle in 12/15 patients (Fig 4C). In patient with sinus rhythm, temperature uncertainty was consistent across subjects, with variation below 0.5°C. Although significant cardiac motion inconsistency during arrhythmia can lead to image artifacts, in most cases phase images showed good quality, as illustrated in Figures 2D, 3D, 4D, resulting in stable temperature all over the left myocardium with the exception of small area. Nevertheless, both temperature uncertainty and number of voxel with imprecise temperature increased with prevalence of arrhythmia. Additional method to automatically identify and discard artifacted images in the time series might help to improve cardiac thermometry in presence of arrhythmia.


This study presents the first evaluation of cardiac MR-thermometry on patients during free-breathing. The precision of temperature estimate was found of sufficient quality to monitor catheter-based RF ablation procedures in most cases.


This work received the financial support from the French National Founding Agency (ANR) within the context of the Investments for the Future Program: referenced ANR-10-LABX-57 and named TRAIL and referenced ANR-10-IAHU-04 and named IHU LIRYC. This study was also supported by public grants from the French ANR: program TACIT ANR-11-TecSan-003-01; Equipex MUSIC ANR-11-EQPX-0030; and program MIGAT ANR-13-PRTS-0014-01


1. Raatikainen MJ, Arnar DO, Zeppenfeld K, Merino JL, Levya F, Hindriks G, et al. Statistics on the use of cardiac electronic devices and electrophysiological procedures in the European Society of Cardiology countries: 2014 report from the European Heart Rhythm Association. Europace. 2015;17 Suppl 1:i1-75.

2. Tanner H, Hindricks G, Volkmer M, Furniss S, Kuhlkamp V, Lacroix D, et al. Catheter ablation of recurrent scar-related ventricular tachycardia using electroanatomical mapping and irrigated ablation technology: results of the prospective multicenter Euro-VT-study. J Cardiovasc Electrophysiol. 2010;21(1):47-53.

3. Kolandaivelu A, Zviman MM, Castro V, Lardo AC, Berger RD, Halperin HR. Noninvasive assessment of tissue heating during cardiac radiofrequency ablation using MRI thermography. Circ Arrhythm Electrophysiol. 2010;3(5):521-9.

4. de Senneville BD, Roujol S, Jaïs P, Moonen CTW, Herigault G, Quesson B. Feasibility of fast MR-thermometry during cardiac radiofrequency ablation. NMR in Biomedicine. 2012;25(4):556-62.

5. Ozenne V, Toupin S, Bour P, de Senneville BD, Lepetit-Coiffe M, Boissenin M, et al. Improved cardiac magnetic resonance thermometry and dosimetry for monitoring lesion formation during catheter ablation. Magn Reson Med. 2016.

6. Hansen MS, Sorensen TS. Gadgetron: an open source framework for medical image reconstruction. Magn Reson Med. 2013;69(6):1768-76.

7. Contijoch F, Rogers K, Rears H, Shahid M, Kellman P, Gorman J, 3rd, et al. Quantification of Left Ventricular Function With Premature Ventricular Complexes Reveals Variable Hemodynamics. Circ Arrhythm Electrophysiol. 2016;9(4).


Figure 1: A) Synchronously recorded ECG (red) with detected QRS peaks (blue triangles). EPI artefacts of the five slices are visible. B) Two dimensional RR duration plot with one beat cluster. The subject #5 was in sinus rhythm with RR durations intervals of 1031 ± 37 ms. Maps of temporal standard deviation of temperature (σT, computed over 210 successive repetitions) are overlaid in (E) on averaged registered magnitudes images (C). Orange arrows show phase unwraps and corresponding excluded voxels. Dotted lines show contour of the ROI surrounding the myocardium used for statistical analysis of temperature data.

Figure 2: A) Synchronously recorded ECG (red) with detected QRS peaks (blue triangles). Window acquisition was set on mid-diastole to reduce the impact of arrhythmia. B) Two dimensional RR duration plot with four beat clusters. Prevalence of arrhythmia is low with 20 beats over 251 beats. This subject #7 had RR durations intervals of 904 ± 99 ms. Temporal standard deviation temperature maps are overlaid in (E) on averaged registered magnitudes images (C). Contraction of left ventricle can be observed on each slice with smaller myocardium wall for the last images closer to the diastole.

Figure 3: A) Synchronously recorded ECG (red) with detected QRS peaks (blue triangles). B) Two dimensional RR duration plot with five beat clusters. Prevalence of arrhythmia is high, this subject #6 had RR durations intervals of 849 ± 286 ms. Blurring on magnitude image is present which denotes wrong registration and results in unstable temperature on the right part of the ventricle. Orange arrows indicate area with excluded voxels.

Figure 4: Statistics on 15 patients during free-breathing acquisition. Box-and-whisker plots of the distribution of temperature σT (A) and µT (B) over all patients. The data were retrieved from ROIs surrounding the left ventricle using all dynamic images excluding voxels where σT was higher than 5°C. Percentage of voxels below 1, 2, 5 °C temperature uncertainty are plotted in a diagram representation (C) and percentage of excluded voxels are indicated in red.

Table 1: summarizes the patient information, the scan duration, the RR durations intervals and the number of arrhythmia-related beats. The mean ± SD total time where temperature mapping was available was 3 min 13 s ± 24 s. Prevalence of arrhythmia and the corresponding number of beats is quantified using recorded ECG.

Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)