Navigator artifact reduction in 3D late gadolinium enhancement imaging of the atria
Jennifer Keegan,1 Peter Drivas,1David N Firmin.1,2
1Cardiovascular Biomedical Research Unit, Royal Brompton and Harefield NHS Trust, London, and 2Imperial College of Science Technology and Medicine, London.
Running title:Navigator artifact reduction in 3D LGE imaging
Word count: 3221
Key words: late gadolinium enhancement imaging, navigator, artifact, pulmonary veins
Address for correspondence:
Dr Jennifer Keegan
Cardiovascular Magnetic Resonance
Royal Brompton Hospital
Sydney Street, London SW3 6NP
Email:
Tel: 020 7351 8800
Fax: 020 7351 8816
Abstract
Purpose: Navigator-gated 3D late gadolinium enhancement (LGE) imaging demonstrates scarring following ablation of atrial fibrillation. An artifact originating from the slice-selective navigator-restore pulse is frequently present in the right pulmonary veins (PVs), obscuring the walls and making quantification of enhancement difficult. We describe a simple sequence modification to greatly reduce or remove this artifact.
Methods: A navigator-gated inversion-prepared gradient echo sequence was modified so that the slice-selective navigator-restore pulse wasdelayed in timefrom the non-selective preparation (NAV-restore-delayed). BothNAV-restore-delayed and conventional3D LGE acquisitions were performed in 11 patients and the results compared.
Results: One patient was excluded due to severe respiratory motion artifact in bothNAV-restore-delayed and conventional acquisitions. Moderate – severe artifact was present in 9of the remaining 10 patients using the conventionalsequence and was considerably reduced when using theNAV-restore-delayed sequence (ostial PV to blood pool ratio: 1.7+/-0.5 vs 1.1+/-0.2respectively (p<.0001); qualitative artifact scores:2.8+/-1.1vs 1.2+/-0.4respectively (p<.001)). While navigator signal-to-noise ratio was reduced with theNAV-restore-delayed sequence, respiratory motioncompensation was unaffected.
Conclusions: Shifting the navigator-restore pulse significantly reduces or eliminates navigator artifact. This simple modification improves the quality of 3D LGE imaging and potentiallyaids late enhancement quantification in the atria.
Introduction
The most prevalent form of cardiac rhythm disturbance is atrial fibrillation (AF)1 which frequently originates from ectopic beats arising from the pulmonary veins.2Complete electrical isolation of the pulmonary veins using radio frequency ablation under X-ray fluoroscopy is the most common treatment,3 but repeat procedures are often necessary due to incomplete isolation.4Scarring following radio frequency ablation of AFhas been demonstrated5,6 and quantified7-9using 3D late gadolinium enhancement (LGE) imaging and the spatial distribution of scarring hasbeen shown to be related to the likelihood of recurrence.10LGE imaging therefore has a promising role in the assessment of RF ablation in the AF patient11 and also in directing the electrophysiologist to regions of incomplete scarring in repeat studies.12
LGE imaging is generally performed using a non-selective inversion-recovery (IR) prepared segmented gradient echo sequence with the inversion time (TI) set to null the signal from normal myocardium.13While conventional 2D LGE imaging is performed during breath-holding,14 high resolution 3D coverage of the atria requires that imaging is performed during free-breathing using diaphragmatic navigators to restrict the respiratory motion to a narrow acceptance window around the end-expiratory pause position.15,16For both pencil-beam and crossed-pairs navigators,17 a selective navigator-restore pulse18is applied immediately after the non-selective inversion preparation (Figure 1(a)) to avoid the substantial degradation of the navigator signal-to-noise ratio which would otherwise compromise the accuracy with which the diaphragm position can be determined. This navigator-restore pulse also re-inverts blood which flows into the right pulmonary veins and atria during the inversion time and can result in a characteristic artifact of high signal intensity in the region of the PVs. The intensity and extent of this artifact depends on the type of navigator implemented (crossed-pairs or 2D RF pulse), itsexact positioning, the slice thickness of the selective navigator-restore and on the amount pulmonary vein blood flow between the navigator-restore and the data acquisition. It frequently obscures the ostia of the veins and the nearby atrial wall, making both qualitative and quantitative assessment of enhancement difficult. While it has been reduced by removing the navigator-restore pulse and using a following navigator for respiratory gating,7 this precludes the use of prospective tracking or phase ordering techniques which have recently been used to reduce acquisition durations.19Respiratory bellows have also been suggested as an alternative to navigator-gating.20Alternatively, a 5cm thick diaphragmatic slab projection navigator has been proposed21 which is implemented without the artifact-forming navigator-restore pulse. However, the sharpness of the navigator edge is reduced by receiving signal from such a large slab (which may compromise the accuracy of the navigator edge detection) and the effectiveness of the respiratory gating is further reduced by the delay of 100ms which is required following the slab navigator.22 We propose a simple modification to the original navigator-gated sequence which reduces or eliminates the inflow artifact and which allows prospectively gated 3D LGE imaging without these drawbacks.
Methods
A standard crossed-pairs navigator-gated inversion-prepared segmented gradient echo sequence (TR = 2.9ms) (Fig 1(a)) was modified so that the slice-selective navigator-restore pulse is shifted in time from the non-selective IR preparation: instead of being immediately following the non-selective inversion pulse, it is delayed by 100 – 200 ms (Fig 1(b)). The benefit of this is two-fold: (i) because the blood longitudinal magnetisation has been decaying towards zero, it results in a re-inverted blood magnetisation that is lower(so that the intensity of any artifact is reduced) and (ii) any blood which is re-inverted has less time to flow into the atria resulting in any artifact being shifted away from the PV ostia.
Both conventionalandNAV-restore-delayednavigator gated 3D acquisitions were performed in 11 patients in a random orderfollowing standard clinical 2D LGE imaging on a Siemens Skyra 3Tesla scanner (Siemens Medical Systems, Erlangen, Germany). All subjects provided written informed consent according to the local ethics committee. The 3D imaging was started approximately 15-20minutes after gadolinium administration (Gadovist - gadobutrol, 0.1mmol/kg body weight). The crossed-pair navigator parameters are as follows: TE = 20 ms, TR = 30 ms, thickness 10 mm, navigator feed-back time 10 ms. The navigator restore thickness was 10 mm. 3D LGE data (TE = 1.3 ms, TR = 2.9 ms) were acquired in the transverse plane (field of view = 400 mm) as follows: 12 slices at 2mm x 2mm x 4mm, reconstructed to 24 slices at 1mm x 1mm x 2mm, generalised autocalibrating partially parallel acquisition (GRAPPA) x2, acquisition window 140ms, alternate R-wave gating, chemical shift fat suppression, flip angle 20o, left-right phase encoding, crossed-pairs navigator positioned over the dome of the right hemi-diaphragm with nominal navigator acceptance window size of 6mm.Forty eight ky lines were acquired per cardiac cycle with centric coverage. Kz coverage was also centric, with all ky phase encodings for a given kz phase encoding step being acquired before kz was changed. The nominal acquisition duration (assuming 100% respiratory efficiency) was 114 cardiac cycles. While the spatial resolution and acquisition window in the acquired datasets are poorer than those required for detailed atrial LGE imaging,5-10they are sufficient to clearly demonstrate the degree and extent of the navigator artifact in a reasonably shortimaging time so that acquisitions both with and without the sequence modification could be obtained in each subject before excessive gadolinium wash-out. The inversion time was determined by a scout 2D acquisition prior to each 3D scan. Longitudinal magnetisation recovery during the navigator-restore delay implemented in theNAV-restore-delayed sequence reduces the amount of longitudinal magnetisation re-inverted by the navigator-restore pulse and hence, the signal-to-noise ratio (SNR) in the navigator trace is reduced. The SNR reduction for a given navigator-restore delay is dependent on the inversion time implemented. In this study, the navigator-restore delay was100 – 200msand was the maximum possible that still enabled the detected diaphragm edge position to smoothly follow the respiratory motion.Raw data were stored with each acquisition to allow subsequent analysis of the effects of SNR reduction on the accuracy of diaphragm edge detection in the navigator trace.In addition, sequence simulations were also performed (MATLAB version 7 (Mathworks, Natick, MA)) to study the effect of increasing navigator-restore-delay on the longitudinal magnetisation, Mz, of liver which determined the SNR of the navigator trace. These simulations were performed at four heart rates (ranging from50 – 100 beats per minute) and assumed that the T1 of liver post-gadolinium was the same as that of normal myocardium (423 ms at ~20 minutes following .1mmol/kg gadobutroladministration.)19For each heart rate, the optimal inversion time required to null normal myocardium was determined by solving the Bloch equations. For the sequence used in vivo (as described above), the navigator TR is 30 ms, navigator feedback time is 10 ms and the duration of chemical shift fat suppression preparation is 20 ms. Consequently, the 900 excitation of the crossed-pairs navigator is output approximately 60 ms prior to the time required to null normal tissue. For each heart rate and each navigator-restore-delay (ranging from 0 – 150 ms in increments of 25 ms), the Mz of liver was consequently calculated 60 ms prior to the optimal inversion time. The ratio of this Mz with navigator-restore-delay to that without was plotted as a function of navigator-restore-delay to indicate the relative change in navigator SNR as the navigator-restore-delay increased.
Image Analysis
In images acquired with the conventional sequence, regions of interest were drawn around any ostial right superior or inferior pulmonary vein artifact and also - as a reference - in the blood pool in the descending aorta in the same image. The regions of interest were copied to the corresponding images obtained with theNAV-restore-delayed acquisition. The ratio of pulmonary vein signal to reference blood signal (PVratio) was determined for images acquired with bothNAV-restore-delayed and conventional sequences and compared using a paired t-test. In addition, consensus subjective PV image artifact scores (AS: 1=none, 2 = mild, 3 = moderate, 4 = severe) were determined by two blinded observers and compared using paired Wilcoxon analysis. The effect of reduced SNR in the navigator trace on the appearance of respiratory motion artifacts was assessed on the same 4-point scale by the same two observers and the consensus scores compared with paired Wilcoxon testing.
Navigator Analysis
In addition, offline reconstructions of the navigator raw data were performed in MATLAB version 7 (Mathworks, Natick, MA). The traces were smoothed (using a running average of 7 navigator points) and scaled to a maximum of 100. For each subject, in an end-expiratory reference navigator trace, the mean and standard deviation (SD) of the signal intensity in a region of the lung was determined, together with that in a region in the liver, just adjacent to the diaphragm edge. This lung region of interest was positioned as best possible to avoid vessels and other structures and, in the absence of a true background region of interest in the navigator data, approximated to noise. The navigator SNRsin theNAV-restore-delayed and conventional acquisitions were determined as the ratio of the mean liver signal to the SD of the lung signal in the navigator traces of the NAV-restore-delayed and conventional acquisitions respectively and compared with paired t-testing. For data acquired with theconventional sequence in each patient, the first 25 navigator traces were analysed and the diaphragm edge position for each trace was determined by a least squares fit algorithm (also over 7 navigator points), as used by the online navigator reconstruction software. This method of edge detection23 has been shown to be highly resistant to noise.Random Gaussian noise was then added to the navigator trace to reduce the SNR to the same level as that in the data acquired with the correspondingNAV-restore-delayedsequenceand the diaphragm edge detection repeated. The diaphragm edge positions with and without the addition of noise were compared using the Pearson correlation coefficient and the mean and standard deviation of the signed differences calculated.
Results
All 11 patients completed the study although one subject had severe respiratorymotion ghostingextending across the PVs (in both sequences) and was excluded from further analysis. In two further subjects, the imaging slab failed to cover the right superior PV (RSPV) so that only the right inferior PV (RIPV) could be analysed. In the NAV-restore-delayed sequence, the navigator-restore delays implemented were 100 ms (in 5 cases), 150ms (in 4 cases) and 200 ms (in 1 case). The inversion time for nulling of normal left ventricular myocardium with single R-wave gating was 278 +/- 44 ms. The implementation of the navigator-restore delay did not impact upon the acquisition durations (127 +/- 33 s and 130 +/- 47sfor the conventional and NAV-restore-delayed sequences respectively, p = ns.)
Image Analysis
Moderate – severe navigator artifact was present in either the superior or inferior (or both) PVs in 9 of the remaining 10conventionalacquisitionsand was considerably reduced, or eliminated, when using the modified sequence with a navigator-restore delay. The PVratio wassignificantly reduced when using theNAV-restore-delayed sequence (1.1 +/- 0.2 vs 1.7 +/- 0.5,p < .0001) as were theconsensus artifact scores (1.2 +/- 0.4 vs 2.8 +/- 1.1, p < .001). The degree ofnavigator artifact seen was similar for both the RIPV and RSPV when using the conventional sequence (PVratio: 1.8 +/- 0.6 vs 1.7 +/- 0.4 respectively, p = ns; consensus artifact score: 2.8 +/- 1.2 vs 2.8 +/- 1.0 respectively, p = ns) and when using theNAV-restore-delayed sequence (PVratio: 1.1 +/- 0.1 vs 1.2+/- 0.2 respectively, p = ns; consensus artifact score: 1.2 +/- 0.4 vs 1.1 +/- 0.4 respectively, p = ns). In no patient was the image quality obtained with theNAV-restore-delayed sequence worse than that with the conventional sequence. The consensus artifact score and the PVratiodid not depend upon the navigator-restore delay value (artifact score: 1.1 +/- 0.4 vs 1.2 +/- 0.4 for patients with navigator-restore delay values of 100 ms and >= 150 ms respectively (p = ns); PVratio: 1.05 +/- 0.16 vs 1.13 +/- 0.29 for patients with navigator-restore delay values of 100 ms and >= 150 ms respectively (p = ns)). Examples ofNAV-restore-delayed and conventional acquisitions in three patients are shown in Figure 2, together with the PVratio and consensus artifact scores.
The percentage navigator SNR reduction did not depend upon the navigator-restore delay value (41% +/- 16% vs 41% +/- 5%for patients with navigator-restore delay values of 100 ms and >= 150 ms respectively (p = ns).Qualitative analysis of the subjective respiratory motion artifact scores showed no significant differences between the techniques (1.7+/- 0.7vs 1.7 +/- 0.7, p = ns).
Navigator Analysis
Off-line analysis showed that, as expected, navigator SNRissignificantly lowerwhen using theNAV-restore-delayed sequence (9.7 +/- 4.0 vs 24.9 +/- 11.6, p < 0.001) although the navigator edge remains well preserved. Figures 3(a)and (e) show an end-expiratory navigator trace from the conventionalandNAV-restore-delayedacquisitions respectively in an example subject. The SNR in these navigator traces is 25.3 and 11.9 respectively. Figure 3(b) - (d) show the navigator trace in 3(a) (conventional sequence) with increasing amounts of random Gaussian noisereducing the SNR to 10.2, 7.1 and 5.3 respectively. Figure 3(f) – (g) show plots of the diaphragm positions detected in the first 25 cycles of the original trace (Figure 3(a)) against those measured with increased noise levels (Figure 3(b), 3(c) and 3(d) respectively). They show that the least squares fit algorithm used for edge detection is relatively insensitive to added noise, with high correlation coefficients for all noise levels, including levels greater than those observed in theNAV-restore-delayed acquisitions (Figure 3(e))). Figure 4 shows a plot of diaphragm edge position in the first 25 cycles of the conventional acquisitions plotted against the diaphragm edge positions obtained in noisier versions of those navigator traces for all patients, in each case the amount of noise added resulting in similar navigator SNRs as in the NAV-restore-delayed acquisitions. The addition of noise to the level of that seen in theNAV-restore-delayed sequence resulted in very little difference to the detected navigator edge positions with the mean difference between the diaphragm positions measured from the two datasets being -0.08 +/- 0.66 mm with the Pearson correlation coefficient being 0.99.
Discussion
We have presented a simple modification to a navigator-gated inversion-prepared 3D LGE sequence which significantly reduces or eliminates pulmonary vein artifact. The technique is applicable to both crossed-pairs and 2D pencil beam navigator techniques(both of which require a navigator-restore pulse) and while the SNR in the navigator traces is reduced, we have shown that this does not impact on the accuracy of the navigator edge detection nor on the subjective assessment of the respiratory motion suppression. Unlike other techniques,21 the time between the navigator and the imaging segment is unchanged so that the navigator position continues to accurately reflect the diaphragm position at the start of the data segment acquisition.
The extent and severity of the navigator artifact in standard 3D LGE imaging depends upon the amount of pulmonary artery blood flow occurring between the navigator-restore pulse and the data acquisition,24 on the pulmonary vein anatomy and on the positioning of the navigator-restore pulse (which determines the amount of pulmonary blood which is re-inverted). As such, it is highly subject-specific but in this study, it was seen in either the RSPV or the RIPV or both in 9 out of 10 subjects with moderate – severe severity. Using theNAV-restore-delayed sequence, this artifact was completely eliminated in six subjects and reduced to mild (artifact score = 2) in the remaining three. While elimination of the navigator artifactcould be achieved by removing the navigator-restore pulse and using a following navigator,7 this would not allow the use of real-time phase ordering and windowing techniques, including the continuously adaptive windowing strategy (CLAWS) which has recently been shown to be beneficial in reducing the acquisition durations of long 3D LGE whole-heart acquisitions by 26%.19 Similarly, real-time slice following techniques would not be possible with a following navigator.For centric k-space ordering, as implemented here, a following navigator would also have the additional disadavantage of being acquired further from the centre of k-space which would lead to reduced respiratory motion compensation.22