Accelerated high-resolution free-breathing 3D whole-heart T2-prepared black-blood and bright-blood cardiovascular magnetic resonance

Background The free-breathing 3D whole-heart T2-prepared Bright-blood and black-blOOd phase SensiTive inversion recovery (BOOST) cardiovascular magnetic resonance (CMR) sequence was recently proposed for simultaneous bright-blood coronary CMR angiography and black-blood late gadolinium enhancement (LGE) imaging. This sequence enables simultaneous visualization of cardiac anatomy, coronary arteries and fibrosis. However, high-resolution (< 1.4 × 1.4 × 1.4 mm3) fully-sampled BOOST requires long acquisition times of ~ 20 min. Methods In this work, we propose to extend a highly efficient respiratory-resolved motion-corrected reconstruction framework (XD-ORCCA) to T2-prepared BOOST to enable high-resolution 3D whole-heart coronary CMR angiography and black-blood LGE in a clinically feasible scan time. Twelve healthy subjects were imaged without contrast injection (pre-contrast BOOST) and 10 patients with suspected cardiovascular disease were imaged after contrast injection (post-contrast BOOST). A quantitative analysis software was used to compare accelerated pre-contrast BOOST against the fully-sampled counterpart (vessel sharpness and length of the left and right coronary arteries). Moreover, three cardiologists performed diagnostic image quality scoring for clinical 2D LGE and both bright- and black-blood 3D BOOST imaging using a 4-point scale (1–4, non-diagnostic–fully diagnostic). A two one-sided test of equivalence (TOST) was performed to compare the pre-contrast BOOST images. Nonparametric TOST was performed to compare post-contrast BOOST image quality scores. Results The proposed method produces images from 3.8 × accelerated non-contrast-enhanced BOOST acquisitions with comparable vessel length and sharpness to those obtained from fully- sampled scans in healthy subjects. Moreover, in terms of visual grading, the 3D BOOST LGE datasets (median 4) and the clinical 2D counterpart (median 3.5) were found to be statistically equivalent (p < 0.05). In addition, bright-blood BOOST images allowed for visualization of the proximal and middle left anterior descending and right coronary sections with high diagnostic quality (mean score > 3.5). Conclusions The proposed framework provides high‐resolution 3D whole-heart BOOST images from a single free-breathing acquisition in ~ 7 min.


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][4][5][6][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][9][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][14][15][16]. Most existing PSIR LGE sequences are, however, limited to 2D acquisitions that are performed under a breath-hold. Recently, free-breathing 3D wholeheart 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 Brightblood 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, blackblood 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 brightblood 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 blackblood 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 respiratoryresolved 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 postcontrast 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.

T2-prepared BOOST sequence
The T 2 -prepared BOOST sequence alternates the acquisition of a T 2 -prepared IR module in odd heartbeats (T 2 Prep-IR, magnitude image) and a T 2 -prepared module in even heartbeats (T 2 Prep, reference image), as shown in Fig. 1a and b. Both undersampled datasets are acquired using a 3D variable-density Cartesian trajectory with spiral profile order (VD-CASPR) [23][24][25]. A short tau inversion recovery (STIR)-like fat suppression approach [26] is used in odd heartbeats, whereas a spectral presaturation (SPIR) [27] is used to suppress epicardial fat signal in even heartbeats. Low-resolution 2D iNAVs are acquired at every heartbeat to estimate 2D beat-to-beat translational motion (left-right: LR and superior-inferior: SI) and enable 100% respiratory scan efficiency and a predictable scan time. Motion estimation is performed independently for both BOOST datasets.
In post-contrast BOOST a subject-specific TI is set to null the signal from the healthy viable myocardium, whereas in non-contrast-enhanced BOOST a short TI = 110 ms is set to null the fat signal.

XD-ORCCA reconstruction
The iNAVs are registered to a common respiratory position (end-expiration) and the 2D translational motion parameters are obtained via intensity-based image registration, using mutual information as a similarity measure and steepest gradient descent optimization. Then, the estimated SI information is used to obtain the respiratory signal, which is used to distribute each BOOST dataset into five equally populated respiratory bins (Fig. 2a). Subsequently, 2D translational motion correction within each bin is performed in k-space before respiratory-resolved image reconstruction to minimize intra-bin motion (Fig. 2b). This motion information is also used during the reconstruction to align the intra-bin motion-corrected images to end-expiration (Fig. 2c), in order to increase the sparsity in the respiratory domain [22]. Therefore, respiratory-resolved T 2 Prep-IR BOOST and bright-blood T 2 Prep BOOST images (Fig. 2d) are obtained by solving a modified version of XD-ORCCA, which explores the temporal sparsity in both forward and Fig. 1 Proposed 3D whole-heart high-resolution motion-resolved accelerated T 2 -prepared Bright-blood and black-blOOd phase SensiTive inversion recovery (BOOST) for simultaneous black-blood late gadolinium enhancement (LGE) and bright-blood coronary cardiovascular magnetic resonance angiography (CCMRA). a A T 2 -prepared (T 2 Prep) inversion recovery (IR) module is applied at odd heartbeats (T 2 Prep-IR BOOST, magnitude image), whereas a b T 2 -prepared module is acquired at even heartbeats (T 2 Prep BOOST, reference image). In the latter, a SPIR pulse is used for fat saturation. In the former, the inversion time (TI) is set to null the signal from healthy viable myocardium and the fat signal is suppressed using a STIR-like fat approach. Both acquisitions are performed using a variable-density 3D Cartesian trajectory with spiral profile order and a high flip-angle of 90°. A low-resolution 2D image navigator (iNAV) is acquired at every heartbeat, before each spiral interleave of the 3D data acquisition, to estimate translational respiratory motion along the superior-inferior and left-right directions. c This motion information is used in the XD-ORCCA reconstruction to generate high-quality respiratory-resolved motion-compensated T 2 Prep-IR BOOST and T 2 Prep BOOST images from undersampled acquisitions. Then, the respiratory bin images are combined in a PSIR reconstruction to d generate a black-blood LGE image (PSIR BOOST) for scar visualization. The bright-blood T 2 Prep BOOST image (reference image, b) allows for coronary lumen visualization backward directions, instead of a unique direction, to further increase the sparsity [22]: where d are the 2D intra-bin translational motion-corrected k-space data, s is the 3D spatial total variation (TV) function, α, β and γ are regularization parameters, Rx = T b x b is the motion-corrected domain, where T b is the translation transform that maps the respiratory bin image x b to the reference image x 1 (end-expiration), temporal TV functions in the forward and backward directions, respectively. The operator E = A b FS incorporates the sampling matrix A for each bin b, Fourier transform F and coil sensitivities S . The regularization parameters were selected empirically. Finally, the T 2 Prep-IR BOOST and brightblood T 2 Prep BOOST images from the bin with the smallest respiratory SI displacement (usually end-expiration, Fig. 1c) are combined in a PSIR reconstruction to generate a 3D black-blood PSIR BOOST image (Fig. 1d).

In-vivo experiments
All acquisitions reported in this study were performed on a 1.5T CMR system (MAGNETOM Aera, Siemens Healthineers, Erlangen, Germany) using an 18-channel chest-coil and a 32-channel spine coil. The study was approved by the National Research Ethics Service (1/11/12 and 15/NS/0030) and written informed consent was obtained from each participant according to institutional guidelines.
The 2D iNAVs were acquired using 14 ramp-up preparation pulses of the 3D whole-heart balanced steady state free precession (bSSFP) sequence with the same field-of-view (FOV) and orientation as the 3D BOOST acquisition.
The bright-blood T 2 Prep BOOST images (interpolated to 0.6 × 0.6 × 0.6 mm 3 ) were reformatted using SoapBubble [28] to allow simultaneous visualization of the right Low-resolution 2D iNAVs are acquired at every heartbeat and used to estimate beat-to-beat 2D translational respiratory motion along the superior-inferior (SI) and left-right (LR) directions. a The estimated SI motion information is used to separate the T 2 Prep-IR BOOST and T 2 Prep BOOST datasets into five different respiratory phases (or bins) each. b Moreover, 2D translational motion correction within each bin is performed in k-space before reconstruction to minimize intra-bin motion. In this step, data within each bin are corrected to the corresponding average bin position. c Finally, during image reconstruction, intra-bin motion-corrected images (x b ) are aligned to a reference respiratory position (end-expiration) using the 2D translational motion information (T b ), in order to increase the sparsity in the respiratory dimension and d generate high-quality motion-compensated respiratory-resolved images coronary artery (RCA), left anterior descending coronary artery (LAD) and whenever possible the left circumflex coronary artery (LCX). Using the same software, coronary percentage vessel sharpness (first 4 cm and full length) and visible vessel length of both RCA and LAD were measured for all acquisitions. A Two One-Sided Test of Equivalence (TOST), as described in [29], was used to compare the quantitative results obtained from undersampled acquisitions to those obtained from the fully-sampled reference. If p-value < 0.05, the test rejects the null hypothesis of non-equivalence and accepts the alternative hypothesis of equivalence.
Blinded qualitative grading of anonymized clinical 2D PSIR and 3D PSIR BOOST images was performed on consensus basis by three cardiologists (T.F.I, I.R., SCMR III accredited, and R.H., SCMR II accredited). Images were stored as grayscale digital imaging and communications in medicine (DICOM) images and were windowed so that the black-blood PSIR BOOST showed a black blood pool, which was darker than the healthy myocardium, whereas in the clinical 2D PSIR the normal myocardium was black (with noise still visible). LGE uptake was bright in both 3D PSIR BOOST and 2D PSIR. The diagnostic quality of the clinical 2D PSIR and 3D PSIR BOOST images was graded using a 4-point scale: (1) non-diagnostic, (2) diagnostic with major artifacts, (3) diagnostic with minor artifacts and (4) fully diagnostic with no artifacts. LGE visualization scores were compared using a nonparametric TOST based on the Wilcoxon signed-rank test [31]. The level of significance of the p-value was 0.05. The diagnostic quality of the proximal, mid and distal sections of the LAD, LCX, and RCA course, was graded using the same 4-point scale. BOOST images whose acquisition started later than 40 min after contrast injection were not considered in the analysis, because the relative contrast between normal and diseased myocardium significantly decreases due to contrast agent washout.

Non-contrast-enhanced BOOST acquisitions: selection of undersampling factor
Free-breathing BOOST acquisitions and motion-resolved reconstructions with intra-bin motion compensation were successfully obtained for all 12 subjects. The reformatted bright-blood T 2 Prep BOOST and black-blood PSIR BOOST images (respiratory phase with the smallest SI respiratory displacement) obtained using the proposed method from fully-sampled, 2.6 × and 3.8 × undersampled acquisitions for a representative healthy subject are shown in Fig. 3. In addition, the reformatted bright-blood T 2 Prep BOOST images obtained from fully-sampled and undersampled data for another two representative healthy subjects are shown in Fig. 4. For subject 3 (Fig. 3,  right), the tortuous anatomy of the RCA prevented an appropriate reformatting of the mid segment of the vessel. However, the mid-RCA was successfully reconstructed from accelerated BOOST acquisitions using the proposed method (Additional file 1).
The proposed method produces images from 2.6 × and 3.8 × accelerated BOOST acquisitions with comparable visual quality to those obtained from fully-sampled BOOST acquisitions (Additional file 2).

Contrast-enhanced BOOST acquisitions
Clinical and demographic characteristics are summarized in Table 1. Patient 6 and patient 9 were not included in the qualitative analysis because BOOST data acquisition started more than 40 min after contrast agent injection. In addition, for patient 6, residual motion artifacts were visible in the BOOST images, which resulted from the rigid translation of the chest walls and arms. Moreover, patient 9 exhibited very irregular heartbeats, very short diastolic window and was not able to breath-hold, making the acquisition of CMR exam images very challenging, including the clinical 2D PSIR images, which further delayed the BOOST acquisition. Therefore, image quality analysis was performed for 8 of 10 patients. The average time after injection for BOOST acquisitions was 32 ± 7 min.
All eight 3D PSIR BOOST datasets were considered diagnostic, with a median grade of 4 (range 3-4); specifically, 5/8 cases were graded 4 and 3 cases were graded 3. Clinical 2D PSIR images were graded 4 for 4 patients, 3 cases were graded 3 and one case was graded 2 (median score 3.5, range 2-4). In terms of visual grading, the 3D PSIR BOOST datasets and the clinical 2D PSIR counterpart were found to be statistically equivalent (p < 0.05).
The clinical 2D PSIR and 3D BOOST images (T 2 Prep, PSIR and fusion of T 2 Prep and PSIR images) are shown in Fig. 6 for a patient with myocardial infarction. LGE uptake can be seen in the mid and basal anterolateral walls in both 3D PSIR BOOST and 2D clinical images. Some LGE uptake can also be seen in the septal wall in Fig. 3 Reformatted T 2 Prep-IR BOOST and PSIR BOOST images obtained for one representative healthy subject using the proposed framework. Images were reconstructed from fully-sampled, 2.6 × and 3.8 × undersampled BOOST datasets and display the right (RCA), left anterior descending (LAD) and left circumflex artery (LCX) coronary arteries. Images obtained from accelerated BOOST acquisitions have similar quality to those obtained from the fully-sampled BOOST acquisition, allowing for clear visualization of the coronary arteries. Total acquisition times are indicated for each acquisition Fig. 4 Reformatted T 2 Prep-IR BOOST images obtained for two representative healthy subjects, displaying the right (RCA), left anterior descending (LAD) and left circumflex (LCX) coronary arteries. Images were reconstructed from fully-sampled, 2.6 × and 3.8 undersampled BOOST datasets. Images obtained from accelerated acquisitions have comparable quality to those obtained from fully-sampled BOOST acquisitions the 3D PSIR BOOST. Moreover, bright-blood T 2 Prep and black-blood PSIR BOOST images are shown in Fig. 7 for four representative patients, together with the clinical 2D PSIR acquisitions.
The LGE findings present in the 2D clinical PSIR images of patient 7 were reported as a non-ischemic cardiomyopathy (myocarditis). However, in the 3D BOOST PSIR images, LGE uptake was identified as a transmural myocardial infarction. Therefore, Fig. 5 a Vessel length of the right (RCA) and left anterior descending (LAD) coronary arteries for all 12 healthy subjects obtained from the T 2 Prep-IR BOOST reconstructions of fully-sampled (FS), 2.6 × and 3.8 × accelerated (Acc) BOOST datasets. Significant equivalence (p < 0.05) in visible vessel length was observed between the reconstructions obtained from fully-sampled and accelerated acquisitions, for both coronary arteries. b Vessel sharpness obtained from the fully-sampled and accelerated T 2 Prep-IR BOOST images for all 12 healthy subjects. Image quality was assessed by measuring the first 4 cm and full length of the RCA and LAD. Measurements of vessel sharpness obtained for both coronary arteries were statistically equivalent (p < 0.05) Table 1 Summary of patients' data. In addition, the presence of LGE findings in the clinical 2D PSIR images is stated. Patients 6 and 9 were not considered for the quantitative analysis since BOOST acquisitions started later than 40 min after contrast injection BMI body mass index, HR heart rate, bpm beats per minute, LV left ventricle

Patient Gender Age (years) Weight (kg) Height (m) BMI (kg/m 2 ) HR(bpm) Clinical condition
Positive LGE correspondence between clinical 2D PSIR and 3D PSIR BOOST images was found in all the cases except for one patient with a non-ischemic cardiomyopathy. In addition, the consensus grading scores indicate that the bright-blood T 2 Prep BOOST images allowed for visualization of the proximal and middle LAD, LCX and RCA sections with high diagnostic quality (Fig. 8) and visualization of the distal sections of the RCA and LCX was fair. Nevertheless, all segments ultimately had diagnostic value.

Discussion
In this work, the respiratory-resolved motion-corrected XD-ORCCA reconstruction has been extended to accelerate the T 2 -prepared BOOST sequence and achieve high-resolution motion-corrected 3D whole-heart blackblood LGE and CCMRA, within clinically feasible acquisition times. Respiratory-resolved images are obtained for each BOOST dataset (T 2 Prep-IR and T 2 Prep) 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 T 2 Prep-IR and T 2 Prep 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  (d, h) clinical 2D PSIR. c, g Fusion between the bright-blood T 2 Prep BOOST and the black-blood PSIR BOOST images allows the assessment of scar location and transmurality. The signal from both blood and viable myocardium is nulled in black-blood PSIR BOOST images, thus providing images with good contrast between the blood pool and scar tissue. LGE uptake can be seen (arrows) in the short-axis views b, c, d, f, g, h of the mid (top) and basal (bottom) anterolateral walls. Clinical 2D PSIR and BOOST were acquired ~ 10 min and ~ 35 min post-contrast, respectively Depiction of coronary artery course was graded using a 4-point scale system (1: non-diagnostic, 2: diagnostic with major artifacts, 3: diagnostic with minor artifacts and 4: fully diagnostic with no artifacts). Bright-blood T 2 Prep BOOST images allowed for visualization of the proximal and middle LAD, LCX and RCA segments with high diagnostic quality. All segments had diagnostic value images from 2.6-fold and 3.8-fold accelerated freebreathing 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, brightblood T 2 Prep 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 ) blackblood 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 intrabin 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 2 Prep-IR BOOST and T 2 Prep 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 2 Prep-IR BOOST and bright-blood T 2 Prep BOOST bin images was tested in a preliminary study in healthy subjects, showing that the visual quality of the final 3D blackblood 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 anteriorposterior 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 2 Prep-IR BOOST and T 2 Prep 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 2 Prep-IR and T 2 Prep 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 2 Prep-IR and T 2 Prep BOOST images. This approach could potentially provide higher quality images than reconstructing the T 2 Prep-IR and T 2 Prep BOOST images separately [33][34][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 subendocardial 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 2 Prep 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-contrastenhanced 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].

Conclusions
An accelerated 3D whole-heart T 2 -prepared BOOST framework was proposed to provide high-resolution 3D whole-heart bright-blood and black-blood PSIR BOOST images within a clinically feasible scan time of ~ 7 min. This was achieved by combining the XD-ORCCA respiratory-resolved reconstruction method with free-breathing 3D whole-heart Cartesian BOOST undersampled acquisitions. The proposed method was tested in a group of cardiac patients and showed a good agreement with the clinical 2D PSIR LGE.