Efficient non-contrast enhanced 3D Cartesian cardiovascular magnetic resonance angiography of the thoracic aorta in 3 min

Cardiovascular magnetic resonance angiography (CMRA) is an established diagnostic tool for serial monitoring of thoracic aortic disease, which encompasses a broad range of degenerative, structural (i.e. aortic coarctation, aortic valvar stenosis, bicuspid aortic valve), acquired (i.e. hypertension, chronic obstructive pulmonary disease, smoking and inflammatory conditions), genetic-based (i.e. Marfan syndrome, Ehlers-Danlos syndrome, Loeys-Dietz syndrome), and traumatic disease states and presentations [1‐3]. There are an estimated 3000–8000 new cases of chronic thoracic aortic aneurysms in the UK each year [4].

Thoracic aortic disease is usually asymptomatic and not easily diagnosed until an acute and life-threatening complication occurs, including dissection and/or rupture [5]. Serial, life-long, robust aortic imaging is crucial to allow for monitoring of the disease progress, pre-procedural planning and post-procedural follow-up [6, 7].

Several cardiovascular magnetic resonance (CMR) techniques have been proposed for the assessment of aortic disease, including contrast-enhanced (CE) CMRA [8, 9] and CE balanced steady-state free precession (bSSFP) sequences [10]. Various impediments associated with the aforementioned techniques, including concerns regarding deposition after repeated administration of gadolinium-based contrast agent and the extremely low risk of nephrogenic systemic fibrosis in patients with kidney disease, have shifted clinicians’ interest towards non-contrast CMRA sequences (NC-CMRA) [1, 11‐13]. Recent studies have shown that electrocardiogram (ECG)-triggered, free breathing diaphragmatic-navigator gated CE [14] or NC T2-prepared bSSFP sequences can be applied for aortic disease monitoring [10, 13, 15‐18]. These sequences offer inherent high contrast between blood and background tissues, thereby allowing for optimal delineation of the aortic lumen [8, 19] and accurate, reproducible pre-operative monitoring of aortic root diameters [1, 15, 20, 21]. However, several limitations are intrinsic to this approach [22, 23]. Conventional 3D bSSFP CMRA techniques use diaphragmatic respiratory gating (dNAV) to minimize the effects of respiratory motion in the images, which results in unpredictable and excessively long acquisition times [22] and the risk of incomplete or aborted scans because of respiratory drift [23, 24]. Alternatively, a variety of 1D (one-dimensional) self-navigated techniques have been proposed [18, 25‐27], which allow the extraction of a foot-head respiratory motion signal directly from the image data that can be used for translational respiratory motion correction during post-processing. However, 1D self-navigation also suffers from certain limitations that are related to the 1D translational motion model used for correction, which may not be accurate when a wide range of respiratory motion is present, such as the case among the different thoracic aortic components [28‐30]. During the respiratory cycle, the thoracic aorta experiences a consequent displacement that follows the respiratory excursions of the thoracic wall, with significantly smaller displacement for the descending aorta than that of the ascending segment [29]. 3D-rendered computed tomography angiography (CTA) images and lumen models for patients with aortic pathology have shown that significant multi-directional translation occurred secondary to respiration, shifting the thoracic aorta and arch vessels both posteriorly and superiorly [30].

Translational respiratory motion correction using low-resolution 2D (two-dimensional) or 3D (three-dimensional) image-based navigators (iNAVs) has been applied for coronary MRA [31, 32]. With these approaches, the heart can be spatially isolated from surrounding static tissues and 2D/3D translational respiratory motion can be estimated and corrected in a beat-to-beat fashion. To correct for remaining non-rigid respiratory-induced cardiac motion, and thus to further improve coronary CMRA image quality, techniques that include non-rigid motion correction based on respiratory binning techniques have been proposed [33, 34]. In this approach, the acquired data are assigned to several states of the breathing cycle (or “bins”) and are later corrected to a reference position using non-rigid motion estimated from the binned images. While this approach has been previously demonstrated for high-resolution coronary CMRA from a ~ 10 min scan, its use for aortic imaging has not been explored so far.

The aim of this study was to extend this approach to NC thoracic CMRA in order to achieve highly resolved visualization of the thoracic aorta at 1.6 mm³, in a short and predictable scan of ~ 3 min, with a rapid, fully inline non-rigid motion corrected undersampled reconstruction in 3 min for fast clinical testing and adoption. The feasibility of this approach was evaluated in patients with different aortic pathologies, and its performance in terms of image quality and overall acquisition time was compared to clinical NC 3D T2-prepared bSSFP MRA with dNAV gating.

Both imaging sequences were completed successfully in all 35 subjects (36 ± 13 years, range 18–66 years; 14 female). Additional file 1: Table S1, presents the patient cohort with the corresponding diagnoses. The scan time for the iNAV acquisition was on average 3.1 ± 0.5 min vs. 12.0 ± 3.0 min for the clinical dNAV sequence (p = 0.005). The reconstruction time for the proposed iNAV sequence was 3 ± 0.3 min in the scanner.

The respiratory motion correction strategies affect both the scan time and the conspicuity of the reconstructed images. The conventional dNAV gating leads to prolonged (usually by a factor of two or more) and unpredictable scan times since data acquired outside a pre-defined respiratory window (usually end-expiration) is rejected and need to be reacquired at a later time [24]. The proposed iNAV approach does not rely on an operator’s expertise to properly define navigator gating and achieves 100% respiratory scan efficiency, contributing to shorter and more consistent scan times. Additionally, in conventional dNAV gating, the respiratory-induced motion of the aorta is estimated indirectly from the right hemi-diaphragmatic displacement using a gating window of 7 mm, with a slice tracking of 0.6, that corresponds only to the FH cardiac displacement. It has been previously shown that the optimal estimation of aortic displacement varies across the segments of the thoracic aorta, the vertical and the horizontal axes and also for each subject with different underlying disease, implying that motion artifacts may not be resolved fully if a fixed factor for all parts of the aorta, corresponding solely to the FH motion correction is used [29, 30, 37]; introducing residual motion blurring. Studies in diverse patient groups have shown that respiratory-induced displacement of the aortic arch differs along its segments and presents greater variability post graft interposition, which could be relevant for a subgroup of this patient cohort, rendering the non-rigid motion correction an important parameter for optimised thoracic aortic imaging [37, 38].

The blood signal in the iNAV datasets proved to be more uniform, with reduced standard deviation of the signal intensity across all aortic segments. Although the visual comparison assessment did not show any significant difference for two of the reviewers, a higher interquartile range can be observed in the clinical sequence in the quantitative measurement at the aortic segments. For the total thoracic aorta, the signal intensity is statistically significant more homogenous with the proposed sequence. Additionally, inhomogeneity in the blood signal leading to non-diagnostic images was noted in more segments for the clinical sequence compared to the proposed for all three reviewers. The signal uniformity in the entire thoracic aorta is potentially attributed to the four composite refocusing pulses applied in the T2 preparation module for the proposed sequence (MLEV4) compared to the clinical sequence (MLEV2), that mitigate artefacts due to flow in the bSSFP signal [39] along with the more sophisticated motion correction proposed (translational and non-rigid).

The improved CR of the dNAV approach in the majority of aortic segments was most likely obtained because of the larger number of slices required for acquisition in the sagittal orientation. The proposed iNAV approach (coronal orientation) requires fewer (72–88) slices to cover the entire aorta than the dNAV approach which was acquired in sagittal orientation (as per guidelines in our clinical services), thus resulting in lower signal to noise for the iNAV approach. However, coronal orientation enables simpler radiographic planning and evaluation of the subclavian arteries through their entire thoracic course (Additional file 8: Fig. S7). It is also possible that the diaphragmatic navigator may result in a more effective T2-preparation, producing a higher aortic blood pool/myocardium CR. This is however hard to prove at this point, as the start-up profiles that are used with the iNAV approach to encode the motion are also applied to catalyse the magnetisation towards the steady state. Additionally, the iNAV is implemented with a high–low profile order acquiring the centre of k-space in close temporal proximity to the 3D imaging sequence, which might potentially offset the proximity of the T2 preparation pulse to the image acquisition with the diaphragmatic navigator. However, the threefold acceleration, versus the two-fold in the clinical sequence, may also contribute to the lower aortic blood pool/myocardium contrast ratio in our proposed sequence. Lastly, the reordering of the k-space is linear for the clinical sequence, whereas the iNAV approach utilises a variable-density Cartesian trajectory with spiral-like order (VD-CASPR). The clinical evaluation showed that the proposed sequence is superior or comparable to the clinical, in decision-making and diagnostic confidence, despite the difference in this particular quality metric.

Previous approaches, including stack-of-stars [40] and radial [22, 41] k-space sampling with self-navigation of respiratory motion have been proposed for aortic CMRA, albeit with longer total acquisition times (average duration of 6 min and 13 min for spatial resolution of 1.5 mm³ and 1.1 mm³, respectively), compared to our sequence. Furthermore, the Golden-angle RAdial Sparse Parallel imaging (GRASP) reconstruction framework adopted for the stack of stars sampling and the GRASP with eXtra Dimensions (XD-GRASP) applied in radial k-space sampling [22, 40], necessitate offline reconstruction with reported duration of 4.7 h and 15–30 min for the corresponding studies, which may hinder their immediate clinical adoption.

The proposed efficient aortic CMRA framework could further benefit clinical practice as it can be gauged as a potential alternative to avoid application of ionising radiation, which is important in serial follow-up of young patients, and also obviates need for intravenous contrast injection compared to both CTA and CE-CMRA, thus reducing preparation time and costs. The fast acquisition time in concert with inline reconstruction contributes further to clinical efficiency and cost reduction.

We performed a single-centre study with a limited number of patients. However, as the observed inter-method bias for maximum thoracic aorta diameter between the clinical dNAV CMRA and the proposed iNAV CMRA approach was small with narrow limits of agreement, we consider the sample size sufficient for showing non-inferiority of iNAV CMRA. Further assessment of precision, through reproducibility studies between vendors and sites, should be investigated. The proposed sequence provided high quality CMRA in an acquisition during atrial fibrillation (Additional file 3: Fig. S2). An extensive validation will need to be performed in a multi-center study recruiting a larger patient cohort with different underlying rhythm disorders.

This study did not compare the proposed approach to a CE dNAV-based CMRA because our clinical practice favours NC acquisitions. We did not perform comparison to CTA either, in view of the current institutional guidelines, suggesting CMRA for non-acute thoracic aortic disease. Both CTA and CMRA are commonly used for aortic imaging and the current American Heart Association and European Society of Cardiology guidelines do not specify a preferred imaging modality for the assessment of non-emergent aortic disease [1, 3]. CTA has substantial drawbacks, including exposure to ionising radiation and iodinated contrast [1, 42], that are obviated with CMRA. Additionally, the sensitivity and specificity of CMRA has been shown to be equivalent to or may exceed those for CTA, and therefore CMRA is frequently preferred for patients with aortopathies, who return for follow-up examinations at frequent short-term intervals [17].

A direct comparison between the application of conventional respiratory navigator with non -rigid motion correction and the same threefold acceleration factor with variable density pattern versus the proposed sequence will be explored in future work, to allow for better understanding of the sole contribution of image-based navigation to image quality, respiratory artifact, and inhomogeneity changes in the CMRA.

The homogeneous blood pool along the entire aorta, as obtained with the proposed sequence (Figs. 2, 3, Additional file 2: Fig. S1, Additional file 3: Fig. S2 and Additional file 9: Fig. S8), can be of particular interest in paediatric and adult congenital heart disease patients because it allows the examination of all intrapericardiac and aortic structures for an efficient anatomical assessment. The rapid acquisition achieved by the proposed method may limit patients’ discomfort, which is relevant especially for children. This will be investigated in future studies.

This study describes the performance of an accelerated, free breathing, NC ECG‐triggered, thoracic CMRA approach with Cartesian sampling and image-based non-rigid respiratory motion correction. The proposed approach produces good image quality at isotropic spatial resolution (1.6 mm³) in fast acquisition (3 min) and in-line reconstruction time (3 min) facilitating adoption in clinical routine and reduction in imaging costs. Blurring from respiratory motion was mitigated with the proposed approach along with improved signal homogeneity. The overall diagnostic confidence was not significantly different between the clinical and the proposed approach, and the mean vessel diameters were in good agreement with excellent intra and inter-observer reproducibility.