Background
Late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) imaging has emerged as the clinical reference standard for assessing viability following myocardial infarction [
1,
2]. Moreover, LGE imaging is gaining importance in the diagnostic and prognostic evaluation of a range of heart muscle diseases where fibrosis can develop [
3‐
7]. LGE imaging is performed after the administration of a gadolinium (Gd)-based contrast agent using inversion recovery (IR) prepared T
1-weighted gradient echo sequences [
1,
8‐
10]. The inversion time (TI) is set to null the signal of healthy myocardium in order to maximize the contrast between infarcted/fibrotic and normal myocardium. This method is highly dependent on the correct selection of TI for healthy myocardial signal nulling.
Phase-sensitive IR (PSIR) LGE is an alternative approach that is less sensitive to the TI selection [
11,
12]. Conventional PSIR sequences require an IR-prepared image (or “magnitude image”) and a proton density weighted image (or “reference image”) to provide a reference for background phase. However, conventional PSIR sequences provide suboptimal contrast between scar tissue and the blood pool, meaning that sub-endocardial infarctions could be difficult to detect or delineate.
Black-blood PSIR LGE schemes have been proposed to improve scar-blood contrast by suppressing both the healthy myocardium and blood pool signal [
13]. This can be achieved by combining an inversion pulse with a T
2 preparation or magnetization transfer preparation [
13‐
16]. Most existing PSIR LGE sequences are, however, limited to 2D acquisitions that are performed under a breath-hold. Recently, free-breathing 3D whole-heart PSIR acquisitions based on diaphragmatic navigator gating have been proposed [
17,
18]. However, these approaches have low scan efficiency that lead to long and unpredictable acquisition times, and thus, may result in reduced image quality due to the change of the TI over time as well as disrupted workflows. In addition, PSIR sequences require the acquisition of a reference image, which doubles the scan time without providing additional diagnostic information. Recently, Ginami et al
. proposed a free-breathing 3D whole-heart T
2-prepared Bright-blood and black-blOOd phase SensiTive inversion recovery sequence (BOOST) for simultaneous bright-blood coronary angiography and black-blood LGE imaging [
19]. With BOOST, magnitude and (bright-blood) reference whole heart datasets are acquired and combined in a PSIR reconstruction to obtain a third, complementary, black-blood volume. Additionally, BOOST uses image-based navigators (iNAVs) [
20] to achieve 100% respiratory scan efficiency and predictable scan times. The T
2-prepared BOOST sequence allows for visualization of myocardial infarction with excellent scar-blood contrast. In addition, the reference image provides a complementary bright-blood volume for cardiac anatomy and coronary lumen visualization (coronary CMR angiography, CCMRA). However, high spatial resolution T
2-prepared BOOST (< 1.4 × 1.4 × 1.4 mm
3) requires relatively long acquisition times of ~ 20 min. Yet, high-resolution 3D BOOST acquisitions are required to ensure accurate detection of scar location, size and transmurality, and identify coronary stenosis. This information is important to risk-stratify patients and guide revascularization decisions.
A non-contrast-enhanced version of BOOST has also shown potential for simultaneous bright- and black-blood whole-heart CMR for coronary lumen and thrombus visualization [
21]. However, like post-contrast BOOST, high-resolution non-contrast-enhanced BOOST also requires long acquisition times.
Here, we propose to accelerate the T
2-prepared BOOST sequence to enable high-resolution whole-heart 3D CCMRA and black-blood PSIR imaging in a clinically feasible scan time (< 10 min). This is achieved by extending XD-ORCCA (Optimized Respiratory resolved Cartesian CCMRA) [
22], a highly efficient respiratory-resolved motion-corrected reconstruction framework, to BOOST imaging. XD-ORCCA reconstruction, which was originally introduced for fully-sampled 3D Cartesian CCMRA, exploits 2D translational motion information (extracted from iNAVs) to increase the sparsity in the respiratory dimension and to compensate for residual respiratory motion [
22]. In this study, the feasibility of accelerating the T
2-prepared 3D BOOST sequence [
19,
21] using XD-ORCCA was tested in 10 patients with suspected cardiovascular disease, using an accelerated post-contrast BOOST sequence, at the end of a clinical CMR examination. Due to technical and ethical limitations, it is difficult to test new contrast-enhanced CMR sequences in patients and thus, the optimal acceleration factor employed was identified by first testing different acceleration factors using non-contrast-enhanced 3D BOOST acquisitions in 12 healthy subjects.
Discussion
In this work, the respiratory-resolved motion-corrected XD-ORCCA reconstruction has been extended to accelerate the T2-prepared BOOST sequence and achieve high-resolution motion-corrected 3D whole-heart black-blood LGE and CCMRA, within clinically feasible acquisition times. Respiratory-resolved images are obtained for each BOOST dataset (T2Prep-IR and T2Prep) using XD-ORCCA, which uses 2D translational motion information extracted from iNAVs to increase the sparsity in the respiratory dimension and to compensate for residual motion within each respiratory bin. Then, the T2Prep-IR and T2Prep BOOST images are combined in a PSIR reconstruction to generate a dark-blood PSIR BOOST image.
Initially, the feasibility of using XD-ORCCA to accelerate the T
2-prepared BOOST acquisitions was tested using non-contrast-enhanced 3D BOOST acquisitions in healthy subjects. Fully-sampled BOOST acquisitions were compared against 2.6-fold and 3.8-fold accelerated BOOST scans. This study showed that the proposed framework produces high quality motion-compensated images from 2.6-fold and 3.8-fold accelerated free-breathing BOOST acquisitions. Equivalence tests with TOST revealed statistically significant equivalence in visible vessel length and sharpness when comparing measurements from fully-sampled bright-blood BOOST image and undersampled BOOST images, for both coronaries. Therefore, acquisition times can be reduced from ~ 17 to ~ 6 min by using 3.8-fold accelerated BOOST acquisitions, without significantly sacrificing image quality. However, the fixed TI value used in this study is not the optimal for all subjects since TI is sensitive to the heart rate. Therefore, image quality could be improved by using of a subject-specific TI to suppress the fat signal and reduce chemical shift artefacts, such as those observed in Fig.
3.
The first study, using non-contrast-enhanced 3D BOOST acquisitions, was essential to select the optimal acceleration factor to be used in the contrast-enhanced BOOST study. It showed that XD-ORCCA can reconstruct images from 3.8 × accelerated BOOST acquisitions with comparable quality to those obtained from fully-sampled acquisitions. Thus, in the second study, 3.8-fold accelerated post-contrast BOOST acquisitions were compared against clinical 2D PSIR in patients with suspected coronary artery disease. All 3D BOOST PSIR images included in the qualitative study had diagnostic quality. Moreover, scar tissue could be identified in 3D PSIR BOOST images despite image acquisition only starting on average 32 min after contrast administration. This was facilitated by the high contrast between scar tissue and blood pool in 3D PSIR BOOST images compared to clinical 2D PSIR. Healthy viable myocardium and scar tissue (when present) were visible in matching anatomical locations for both the clinical 2D PSIR and the 3D whole-heart PSIR BOOST images. In addition, bright-blood T2Prep BOOST images allowed visualization of the origin, proximal and mid sections of the coronary arteries with high diagnostic quality. Hence, the proposed framework successfully generates high-resolution 3D whole-heart black-blood PSIR LGE and bright-blood coronary angiography images with sufficient diagnostic quality from 3.8-fold accelerated post-contrast BOOST acquisitions, which can be obtained in a clinically feasible scan time (~ 7 min).
For patient 2, septal LGE uptake was identified in the 3D BOOST PSIR, but was not detected in the 2D clinical PSIR (Fig.
6). This finding could be because the proposed method provides high-resolution (1 × 1 × 2 mm
3) black-blood PSIR BOOST images with whole-heart volumetric coverage, whereas the clinical 2D PSIR images have lower in-plane resolution (1.4 × 1.4 mm
2) and large slice thickness (8 mm). Moreover, clinical 2D PSIR requires multiple breath-holds, and hence, 2D PSIR images may not be in the same respiratory phase as the displayed 3D PSIR BOOST images. However, to draw conclusions, further studies in a large cohort of patients are necessary to compare the location and extent of LGE in accelerated 3D PSIR BOOST and clinical 2D PSIR.
The number of respiratory bins used in both studies provides a good compromise between remaining intra-bin motion and undersampling artifacts [
32]. However, better results could potentially be achieved by adapting the number of respiratory bins according to the irregularity and amplitude of the respiratory motion. One solution would be to add/remove respiratory bins until a fixed bin width is achieved (e.g. 3 mm). Additionally, arrhythmia rejection techniques can be integrated with the proposed framework to further improve imaging quality. At present, the sequence requires relatively regular R-R intervals and therefore cannot be used in patients with atrial fibrillation, although this is a limitation for many techniques employed in clinical CMR.
Currently, it is assumed that the T
2Prep-IR BOOST and T
2Prep BOOST images generated with XD-ORCCA are registered, and hence, in the same respiratory position. However, there could potentially be registration errors, which can cause phase errors in the PSIR BOOST images. The use of non-rigid registration to correct for residual non-rigid motion between the highest-quality T
2Prep-IR BOOST and bright-blood T
2Prep BOOST bin images was tested in a preliminary study in healthy subjects, showing that the visual quality of the final 3D black-blood PSIR BOOST image could be slightly improved (Additional file
3).
The proposed framework only considers translation motion along the SI and LR direction. However, respiratory-induced heart motion involves more complex motion patterns, including motion along the anterior–posterior direction and even non-rigid deformation. Therefore, residual motion could be addressed by estimating/correcting for 3D translational or non-rigid motion at each XD-ORCCA iteration. Alternatively, the high-quality respiratory XD-ORCCA images could be used to estimate 3D bin-to-bin non-rigid motion, which can subsequently be used to generate 3D T
2Prep-IR BOOST and T
2Prep BOOST non-rigid motion-corrected images [
24]. Moreover, non-rigid respiratory motion correction may help reduce ghosting artefacts that may originate from rigid translation of static tissues such as the chest wall and arms. However, all these methods are computationally more expensive.
The proposed 3D PSIR BOOST requires separate reconstructions for each T
2Prep-IR and T
2Prep BOOST dataset. However, since this produces images of the same anatomical region with different contrasts, a joint sparsity regularization term could be added to Eq. (
1) to exploit the structural similarity (sparsity) across contrasts and jointly reconstruct T
2Prep-IR and T
2Prep BOOST images. This approach could potentially provide higher quality images than reconstructing the T
2Prep-IR and T
2Prep BOOST images separately [
33‐
35].
In this study, post-contrast BOOST acquisitions were performed at the end of the clinical CMR examination so that the research protocol did not interfere with the clinical CMR exam indicated to diagnose suspected cardiovascular disease. Therefore, injection of contrast agent was optimized to provide optimal contrast for the clinical 2D PSIR sequences and the order of the 2D PSIR and 3D BOOST acquisitions was not randomized as per ethical requirements. Thus, post-contrast BOOST acquisitions had potentially sub-optimal contrast conditions. This was particularly evident for two specific cases that were excluded of the qualitative analysis (patient 6 and patient 9). Moreover, additional challenges can occur by performing BOOST acquisitions at the end of the clinical protocol, for example, patients’ tolerance to being inside the scanner reduces with time and there can be more respiratory and heart rate variability that can have a detrimental effect on image quality. Therefore, accelerated high-resolution post-contrast 3D BOOST and clinical 2D PSIR acquisitions will be compared in a future study by randomizing the order of the two acquisitions and by performing separate contrast injections to ensure similar contrast conditions. Future studies will also aim to investigate the accuracy of black-blood PSIR BOOST for the detection and quantification of scar transmurality in patients with angiographically confirmed coronary artery disease.
Patient 7 had a non-ischemic cardiomyopathy (myocarditis), which was identified as a myocardial infarct in the BOOST PSIR images. This could be due the lack of contrast between the healthy myocardium and blood pool, which gives the impression that the fibrosis is sub-endocardial when, in fact, it is located in the mid myocardial wall. Moreover, BOOST imaging started more than 30 min post-contrast injection, and hence, contrast agent washout could have prevented adequate scar depiction. However, a more accurate location and delineation of scar tissue can be obtained by fusing the black-blood PSIR image with the bright-blood T
2Prep BOOST image, which then allows both scar and myocardial anatomy visualization as shown in Fig.
6. Alternatively, a grey-blood PSIR technique can be used, which is optimized to null the blood pool, leading to a gray appearance of the blood in the PSIR image, and hence, offering better contrast between myocardium and blood pool [
36,
37]. Nevertheless, the ability of the proposed method to discriminate between ischemic and non-ischemic cardiomyopathies needs to be evaluated and will also be the subject of future studies.
The accelerated high-resolution non-contrast-enhanced BOOST showed promising results for both bright blood and black‐blood coronary artery imaging. However, the suitability of the proposed method for detecting coronary thrombus still needs to be investigated. Nevertheless, this indicates that the proposed framework is quite versatile and could potentially be used with other free-breathing LGE or PSIR techniques [
38,
39].
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.