Gray blood late gadolinium enhancement cardiovascular magnetic resonance for improved detection of myocardial scar

Background Low scar-to-blood contrast in late gadolinium enhanced (LGE) MRI limits the visualization of scars adjacent to the blood pool. Nulling the blood signal improves scar detection but results in lack of contrast between myocardium and blood, which makes clinical evaluation of LGE images more difficult. Methods GB-LGE contrast is achieved through partial suppression of the blood signal using T2 magnetization preparation between the inversion pulse and acquisition. The timing parameters of GB-LGE sequence are determined by optimizing a cost-function representing the desired tissue contrast. The proposed 3D GB-LGE sequence was evaluated using phantoms, human subjects (n = 45) and a swine model of myocardial infarction (n = 5). Two independent readers subjectively evaluated the image quality and ability to identify and localize scarring in GB-LGE compared to black-blood LGE (BB-LGE) (i.e., with complete blood nulling) and conventional (bright-blood) LGE. Results GB-LGE contrast was successfully generated in phantoms and all in-vivo scans. The scar-to-blood contrast was improved in GB-LGE compared to conventional LGE in humans (1.1 ± 0.5 vs. 0.6 ± 0.4, P < 0.001) and in animals (1.5 ± 0.2 vs. -0.03 ± 0.2). In patients, GB-LGE detected more tissue scarring compared to BB-LGE and conventional LGE. The subjective scores of the GB-LGE ability for localizing LV scar and detecting papillary scar were improved as compared with both BB-LGE (P < 0.024) and conventional LGE (P < 0.001). In the swine infarction model, GB-LGE scores for the ability to localize LV scar scores were consistently higher than those of both BB-LGE and conventional-LGE. Conclusion GB-LGE imaging improves the ability to identify and localize myocardial scarring compared to both BB-LGE and conventional LGE. Further studies are warranted to histologically validate GB-LGE. Electronic supplementary material The online version of this article (10.1186/s12968-018-0442-2) contains supplementary material, which is available to authorized users.


Background
Late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) is an established technique for imaging myocardial fibrosis and left ventricular (LV) scar [1]. In addition, LGE imaging has been shown to have potential for detecting scarred tissues in other cardiac structures, including the left atrium and papillary muscles [2][3][4]. In LGE, a gadolinium-based contrast agent is injected 10-15 min prior to imaging. The LGE imaging sequence uses an inversion pulse followed by an image acquisition after an inversion delay, at which point the myocardial signal is nulled. The inversion delay is determined prior to the LGE imaging using a Look-Locker sequence [5]. The fast recovery of scar signal (shorter T 1 ) lead to scarred tissues appearing bright in LGE images. However, due to the short T 1 of blood, blood also appears bright and thus leads to low scar-to-blood contrast. This makes detection of myocardial scar in the vicinity of the blood pool challenging [6].
Several approaches have been investigated to improve blood to scar contrast by suppressing the blood pool. One approach utilizes the out-of-slice blood flow to selectively suppress the blood using dual [6] or quadruple [7] inversion pulses. However, sluggish blood flow and the sensitivity of the technique to predetermined timing parameters can be a limitation of this approach [8]. To avoid this limitation, flow-independent approaches have been proposed to null the blood signal based on T 1 differences among the blood, scar, and normal myocardium. This includes using nonselective dual inversion recovery [8] and blood signal nulling using phase-sensitive LGE [9]. Magnetization preparations before inversion have been also proposed [10][11][12] for selectively suppressing the blood pool signal relative to the myocardium. We recently proposed a dark-blood LGE (DB-LGE) sequence by inserting the T 2 magnetization preparation pulse between the inversion pulse and image acquisition [13,14] to achieve simultaneous nulling of the blood pool and the myocardium. This technique was subsequently used in combination with phase-sensitive inversion recovery (PSIR) [15]. PSIR-based DB-LGE techniques [12,[15][16][17] are optimized to null the blood pool after the myocardium and thus the negative signal appears darker. A black blood contrast is then generated by means of intensity windowing where blood and myocardium appear black and gray, respectively. This results in a more conspicuous myocardium-blood border due to a high contrast to noise ratio between blood and normal myocardium. A validation study by Francis et al. [16] showed that using T 2 -preparation after the inversion recovery pulse in PSIR LGE increases the observer confidence in detecting subendocardial scar. A flow-independent PSIR-based DB-LGE technique was recently presented and validated by Kim et al. [17]. Magnetization-transfer preparation module was used prior to the inversion recovery pulse to achieve blood suppression. The technique was validated using canine model as a reference standard and human subject data [17].
Prior work of complete suppression of the blood signal [6,10,13,14] can be limited by making the clinical interpretation of LGE images more difficult. For example, simultaneous nulling of healthy myocardium and the blood pool makes it difficult to localize a scar and to assess the transmurality. This approach may also artifactually enhance noise/artifacts in patients without scarring. Therefore, LGE with improved contrast between all 3 tissues (i.e. blood pool, healthy and scarred myocardium) may be preferred to complete blood signal nulling.
In this study, we present a 3D gray-blood LGE (GB-LGE) sequence based on our previous DB-LGE sequence [13,14] by introducing a parameter optimization approach that allows flexible adjustment of tissue contrast. Numerical simulations, phantoms and in-vivo studies in both humans and a swine model of infarction are used to study the performance of the proposed method.

Methods
A block diagram of the DB-LGE sequence is shown in Fig. 1a, indicating the three timing parameters D 1 , D 2 and D 3 that are to be determined for GB-LGE [13]. At the beginning of acquisition, the acquired signal is given by [13], Fig. 1 Pulse sequence of the black-blood (BB) late gadolinium enhancement (BB-LGE), gray-blood LGE (GB-LGE) (a); and the signal evolution of the different tissues for conventional LGE (b), BB-LGE (c) and GB-LGE (d). All tissue types are assumed having the same (unity) initial signal intensity. Image acquisition time point is indicated by a black vertical line where, T r 1 and T r 2 are the T 1 -and T 2 -parameters of the myocardium, blood, or scar, and M r ss is the tissue steadystate magnetization (normalized by the fully-recovered magnetization) available immediately before the inversionrecovery pulse. The subscript r can take the symbol: 'myo' , 'blood' , or 'scar' to indicate the tissue type (myocardium, blood, or scar, respectively). The steady state magnetization, M r ss can be analytically derived given the image acquisition sequence (Appendix).
To allow adjustment of the LGE image contrast, the following cost function, Q, is formulated to represent the desired image contrast, where, M r is given by Eq. 1, and α and β are arbitrary non-negative weights less than 1. The first term in the above equation ensures that the optimal solution minimizes the normal myocardium signal. Minimization of the second term leads to a blood signal that is approximately equal to a desired fraction (β) of the scar signal. To obtain a black-blood LGE (BB-LGE) contrast, β is set to zero to suppress both the blood and the myocardium signals (Fig. 1c). To partially attenuate the blood signal, β is set to a fraction (e.g., 0.1) to obtain GB-LGE contrast (Fig. 1c).
In this work, α is set to 0.5 in BB-LGE to equally suppress both the blood and the myocardium signals. For GB-LGE contrast, α is set to a relatively large value (= 0.9) to prioritize the myocardium nulling over strictly equating the blood signal to a given signal level. The other parameters of the cost function include the T 1 and T 2 of the myocardium, the blood, and the scar. The T 1 values are determined using a T 1 mapping scan as discussed below, while the T 2 values of the myocardium, blood, and scar are fixed to previously-reported values equal to 50, 200 and 55 ms, respectively [18,19]. Changing these fixed values can be shown to have no significant effect on the optimized signal (Additional files 1, 2, 3 and 4).
Since there are only two physical phenomena controlling the optimal solution-namely, T 2 decay and T 1 recovery-only two optimization parameters are sufficient to adjust the image contrast. In this work, we use D 2 (to control T 2 decay) and D 3 (to control T 1 recovery), while the parameter D 1 is fixed to a predetermined arbitrary value. The optimization problem in Eq. 2 is numerically solved using the Levenberg-Marquadt algorithm [20] implemented in Matlab (Mathworks Inc., Natick, Massachusetts, USA). Numerical iterations were terminated when the changes in Q were below 10 − 8 .

Selecting D 1
The feasible range of D 1 has a lower limit equal to onehalf the duration of the inversion pulse plus the associated crusher gradients. This is determined by the type of inversion pulse; e.g. adiabatic and/or water-selective, and the hardware capabilities of the CMR system. On the other hand, the upper limit of D 1 can be determined by observing that the T 2 -preparation pulse needs to start before any of the blood or the myocardium longitudinal magnetizations reaches its nulling time point. Otherwise, it would not be possible to find a time point where both signals are simultaneously nulled. Given that the blood magnetization recovers faster than the myocardium magnetization (due to its shorter T 1 ), the upper limit of the feasible range of D 1 is given by, A numerical simulation was used to study the effect of fixing D 1 on the scar signal. First, 10,000 different combinations of tissues were generated by randomly selecting (using uniform probability distribution) the tissue T 1 values from different continuous ranges for the blood ( For each combination of tissue parameters, D 1 was varied from 5 to 100 ms with a step of 5 ms and substituted in Eq. 2 to estimate the optimal D 2 and D 3 values. Each value of D 1 and the corresponding optimal D 2 and D 3 were then used in Eq. 1 to estimate the scar signal. The mean and standard deviation (mean ± SD) of the scar signal over all tissue combinations were calculated for each D 1 . The simulation was repeated twice to simulate the scar signal in BB-LGE contrast (α = 0.5, β = 0) and GB-LGE contrast (α = 0.9, β = 0.1). The results of this simulation were then used to select an optimal D 1 value that yielded the maximum average scar signal.

Phantom experiments
A phantom of NiCl 2 -doped agarose vials with different T 1 and T 2 values was used to demonstrate the impact of different β parameters in Eq. 2 as measured by different vial signals. Vials in the phantom (referred to as vial-M, vial-B, and vial-S) contain materials with T 1 /T 2 parameters mimicking those for the myocardium, blood, and scar (= 598/48, 436/174, and 343/54 ms, respectively). Images were acquired using different imaging parameters based on different β values (= 0, 0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07 and 0.08). Imaging with β> 0.08 was not feasible because the computed D 2 (given the phantom T 1 /T 2 parameters) was too short. The signal ratios of vial-B and vial-M relative to vial-S were computed.

In-vivo experiments
The imaging protocol was approved by our institutional review board and the study was HIPAA-compliant. All subjects provided written informed consent to participate in this study. For the animal study, the research protocol was approved by the Institutional Animal Care and Use Committee and conformed to the Position of the American Heart Association on Research Animal Use as well as to the Declaration of Helsinki.
For all scans, ECG-gated free-breathing 3D spoiled gradient echo based imaging sequence was used to acquire the images using a 32-channel cardiac coil. The imaging parameters for the 3D BB-LGE and 3D GB-LGE techniques were: TR/TE = 5.2/2.5 ms, α = 25°, FOV = 300-400 × 300-400 × 80-120 mm 3 , sensitivity encoding rate = 1.5-2.0, acquisition voxel size = 1.5 × 1.5 × 4-6 mm 3 , acquisition window = 125 ms, low-high phase-encoding order, with 5 startup RF pulses to establish steady-state magnetization, 24 phase encoding line per cardiac cycle, and spectral pre-saturation inversion recovery based fat suppression. Malcom-Levitt (MLEV) composite refocusing pulses were used for T 2 preparation [22]. Conventional 3D LGE imaging was performed using similar imaging parameters with 18-24 phase encoding lines per cardiac cycle and acquisition window = 95-125 ms. A Look-Locker scouting sequence [5] was used to determine the proper inversion time delay of the LGE sequence. Respiratory motion was compensated using a diaphragmatic 2D pencil beam adaptive navigator gating (window = 7 mm) [23] with slice-tracking. Assuming 100% navigator efficiency, the scan time was 0.8-2.2 min with number of reconstructed (interpolated) slices = 15-42. A factor of 2 oversampling was used in slice orientation.
To estimate the T 1 values of the myocardium and the blood, T 1 mapping was performed using a single slice MOLLI sequence 5(1)3 [21] with the following parameters: mid-ventricular short-axis, bSSFP acquisition, scan duration = 9 heartbeats, TR/TE = 2.6/1.3 ms, α = 35°, FOV = 300-400 × 300-400 × 10 mm 3 , sensitivity encoding rate = 2.0, acquisition voxel size = 2 × 2 × 10 mm 3 , acquisition window = 242 ms, linear phase-encoding order, with 10 startup RF pulses, and 93 phase encoding lines per cardiac cycle. The scar T 1 in Eq. 1 is set equal to 0.7 that of the blood. The T 1 map for each patient was automatically reconstructed on the scanner main computer and the operator manually selected two regions inside the myocardium and the blood (using the standard graphical interface of the scanner) and the average T 1 values were used to estimate the optimal imaging parameters. A workstation next to the scanner (a computer with 2.8 GHz Intel Xeon processor and 16 GB RAM) was used to run the optimization algorithm.

In-vivo image analysis
The in-vivo datasets (45 human subjects and 5 swine) were quantitatively analyzed to compare the contrast resulting from the three sequences: GB-LGE, BB-LGE, and conventional LGE. Regions of interest were drawn within the myocardium, blood-pool and the hyperenhanced area in the GB-LGE images and the corresponding location in the BB-LGE and the conventional LGE images. The slice with the maximum scar extent was selected for analysis. Due to the absence of a ground truth (i.e., histology) to verify the hyper-enhancement seen with each technique, only hyper-enhancements that were visible on all three sequences were included in the analysis. Quantitative analysis was performed for the scar-to-blood, scar-to-myocardium, and blood-to-myocardium relative contrast. The relative contrast of tissue A to tissue B is defined as the difference between the average signal of A and B divided by the average signal of B.
Subjective analysis for the dataset was performed by two independent readers (CWT with 10 years CMR experience and UN, a Level-III SCMR-accredited reader). Hyperenhancements within the LV and the papillary muscles were assessed. For each dataset, readers independently assigned 'yes' , 'no' , or 'uncertain' to specify whether a scar was present or not. Also, the readers independently evaluated the image quality of each dataset and specified whether it is of acceptable or poor quality. An image with acceptable quality was defined to have proper myocardium nulling and no severe imaging artifact that might limit the diagnosis. The diagnostic value of each technique was subjectively evaluated in humans using two measures: scar detection and scar localization. The former measures the technique's ability to determine the presence or absence of hyper-enhancements in the LV or papillary muscles and was evaluated using a 4point scale from 1 = challenging to 4 = easy. A similar scale was used to measure the ability to localize the scarring and determine its pattern and extent. In animals, the models involved only myocardial infarction and thus only the ability of the different techniques to localize the scarred tissues was evaluated. Any disagreement in the presence or absence of scar was reviewed in a subsequent consensus reading by both readers.

Statistical analysis
For human subject dataset analysis, a non-parametric Friedman statistical test with Bonferroni adjustment was used to test significant differences in the relative scar contrast or qualitative assessment scores among the GB-LGE BB-LGE, and conventional LGE sequences. Additionally, the Wilcoxon signed-rank test with Bonferroni-Holm adjustment was used to perform pairwise comparisons of the different LGE sequences. In these tests, the pairwise test between GB-LGE and conventional LGE images was performed using the two groups of patients (n = 45). In the tests involving BB-LGE, data from the first group of patients (n = 27) were used to allow paired comparison. Analyses were done using Matlab (Mathworks Inc., Natick, MA). Statistical significance was set at type-I error of 0.05. Due to the small size of the animal dataset (n = 5), we did not perform any statistical analysis and only provided the actual subjective assessment and all images.

Results
The numerical simulation of the effect of D 1 on the resulting scar signal is summarized in Fig. 2. The maximum scar signal (= 0.23 ± 0.04 and 0.24 ± 0.04 in BB-LGE and GB-LGE, respectively) occurred at the lowest value of D 1 (= 5 ms), and continued to decrease with increased D 1 . Therefore, to obtain maximum scar signal, D 1 should be set to its minimum feasible value. Based on this finding, D 1 was set to the minimum value allowing the use of adiabatic and/or water-selective inversion-recovery pulses in addition to the crushing gradients (= 20 ms).
The phantom experiment demonstrated flexible adjustment of the signal from different vials (Fig. 3). Increasing β from 0 to 0.08 resulted in a gradual increase of the signal ratio between vial-B and vial-S from 0.07 to 0.35. For all values of β, the signal ratio of vial-M to vial-S was below 0.06 showing a suppression of the myocardium-mimicking vial.
For the in-vivo imaging, the estimated timing parameters were variable among different patients depending on T 1 tissue parameters. The estimated parameter D 2 for the BB-LGE and GB-LGE sequences varied from 7 to 26 ms (median = 15 ms), and from 4 to 28 ms (median = 9 ms), respectively. The parameter D 3 varied from 146 to 208 ms (median = 183 ms), and from 184 to 273 ms (median = 217 ms) for the BB-LGE and GB-LGE sequences, respectively. The computation time for determining the optimal parameters was less than 1 s.
Both GB-LGE and BB-LGE showed improved blood and scar contrast (Fig. 4), however GB-LGE images showed improved visualization of healthy myocardium. Compared to the conventional LGE images, suppression of the blood signal in GB-LGE and BB-LGE sequences allowed clear visualization of the hyper-enhanced atria (red arrow), left and right ventricular hyper-enhancement (yellow arrows), and chordae tendineae and papillary muscles (blue arrows) in both GB-LGE and BB-LGE images ( Fig. 4 and Additional files 2 and 3).
The Friedman test of the quantitative analysis of the human subjects' images showed a significant difference among the three sequences in the scar-to-blood relative contrast (P < 0.001). Pairwise comparisons indicated that the scarto-blood relative contrast in GB-LGE (1.1 ± 0.5) was significantly higher than that in the conventional LGE images (0.6 ± 0.4) with P < 0.001 (Table 1). Due to the complete nulling of the blood signal, the BB-LGE scar-to-blood relative contrast (3.61 ± 1.83) was significantly higher than that of both GB-LGE (P < 0.001) and conventional LGE (P < 0.001). The scar-to-myocardium relative contrast in GB-LGE, BB-LGE, and standard LGE were 5.1 ± 3.1, 6.1 ± 4.1, and 5.9 ± 3.5, respectively ( Table 1). The scar-tomyocardium relative contrast of the conventional LGE was comparable to that of GB-LGE (P = 0.19) and BB-LGE Fig. 2 The simulated scar signal resulting from solving the optimization problem at different fixed values of D 1 (from 5 ms to 100 ms). Each data point is averaged over 10,000 different combinations of tissue T 1 and T 2 values, with error bars representing standard deviation. The figure shows that the lower the value of D 1 is, the higher the scar signal. BB = black blood, GB = gray blood, LGE = late gadolinium enhancement  (P = 0.59). The GB-LGE showed significantly higher scarto-myocardium relative contrast than that of BB-LGE (P = 0.023) which could be due to the shorter T 2 preparation pulses in GB-LGE compared to BB-LGE. As expected, suppression of the blood signal has significantly reduced the blood-to-myocardium relative contrast in GB-LGE (1.7 ± 1.4) and BB-LGE (0.7 ± 1.2) compared to conventional LGE (3.8 ± 3.5) with P < 0.001 for all pair-wise tests (Table 1).
In the subjective assessment, both readers assigned an image quality score of 'good' to 44 GB-LGE datasets, 44 BB-LGE datasets, and 42 conventional LGE datasets. Both readers confirmed presence of LV scar in 17 datasets (out of 45): 9 in patient group 1 and 8 in patient group 2 ( Table 2). All cases were correctly identified in GB-LGE and BB-LGE images by both readers. Among these 17 datasets, reader 1 and reader 2 missed the presence of the scar in 5 and 4 cases in the conventional LGE images, respectively. None of the GB-LGE datasets was assigned an 'uncertain' score by either of the readers. In contrast, readers 1 and 2 assigned 'uncertain' assessment scores to 2 and 4 conventional LGE datasets, respectively and both assigned 'uncertain' to 1 BB-LGE dataset. Also, more papillary scar were identified by both readers in GB-LGE compared to BB-LGE and conventional LGE images ( Table 2).
There was a statistically significant difference by the Friedman test between the subjective diagnostic value scores (i.e., the ability to detect LV scar, localize LV scar and detect papillary scar) among the three different sequences (Table 3 and Additional file 4: Table S2). Comparisons between each pair of sequences indicated that all subjective scores for the diagnostic value of GB-LGE were significantly higher than those of the conventional LGE images (P < 0.001 for all score comparisons). The GB-LGE scores were also significantly higher than those of BB-LGE images to localize LV scar (P = 0.024) and papillary muscle scar (P = 0.014). The GB-LGE was similar to BB-LGE in the ability to detect the LV scar (P = 0.10).
3D GB-LGE and 3D BB-LGE were successfully acquired in all animals (Fig. 5). The average scar-to-blood relative contrast in GB-LGE (1.5 ± 0.2) was higher than that of the conventional LGE images (− 0.03 ± 0.2) but lower than that in BB-LGE (5.2 ± 1.3) due to the complete nulling of the blood signal ( Table 4). The scarto-myocardium relative contrast in GB-LGE (8.9 ± 4.9) Table 1 Mean ± standard deviation of the slice scar-to-blood and scar-to-myocardium and myocardium-to-blood relative contrast in the black-blood (BB) late gadolinium enhancement (BB-LGE), gray-blood LGE (GB-LGE), and conventional LGE sequences in the human dataset   and conventional LGE (7.1 ± 6.2) were comparable but higher than that of BB-LGE (4.7 ± 1.5) ( Table 4). The blood-to-myocardium relative contrast was reduced in GB-LGE (2.9 ± 1.7) and BB-LGE (0.1 ± 0.2) compared to conventional LGE (7.2 ± 5.9) ( Table 4). Both reviewers assessed all images of diagnostic quality in all animals. There was no difference between readers in identifying hyper-enhancements in GB-LGE and BB-LGE images ( Table 2). The subjective assessment of the ability to localize LV scar indicated that GB-LGE had consistently higher scores (4.0 ± 0.0) compared to BB-LGE (3.2 ± 0.5) and conventional LGE (2.0 ± 0.7) ( Table 5).

Discussion
We present a 3D GB-LGE sequence that yields improved scar visualization by increasing the contrast between scar, blood and normal myocardium. Phantom and in-vivo images show partial suppression of the blood signal can be achieved by choosing the appropriate sequence parameters. In our human datasets, GB-LGE contrast improved localization of the myocardium hyper-enhancements compared to BB-LGE and conventional LGE. In the animal datasets, despite the higher scar contrast in BB-LGE images, both readers consistently scored the GB-LGE images higher for the ability to localize LV scar compared to both BB-LGE and conventional LGE. While GB-LGE shares recently presented DB-LGE techniques the use of T 2 -preparation to adjust the image contrast [13,15], it allows more flexibility to adjust the blood contrast compared to the method by Basha et al. [13].
Simple imaging parameter scouting was used in GB-LGE that included single breath-hold T 1 mapping scan (9 heartbeats) and fast numerical solution of the optimization cost function (< 1 s). Compared to conventional LGE, GB-LGE requires additional steps of manual drawing of regions of interest to determine tissue T 1 and running the optimization solver. This process usually takes 10-15 s and is performed by imaging technologists. The same concept of using T 1 mapping for scouting the imaging parameters has been employed earlier in DB-LGE by Kellman et al. [15], where parameter computations were performed on the scanner main computer.
Both GB-LGE and BB-LGE data were acquired in 1 R-R (i.e. one heart-beat), while conventional PSIR-LGE images were acquired in 2 R-R. This resulted in improved signal-to-noise ratio (SNR) and fewer artifacts due to  LGE compared to the other two sequences incorrect nulling. The 3D acquisition in GB-LGE and BB-LGE inherently improves the SNR despite the effect of T 2 preparation pulses. However, 3D PSIR images will be 2× longer which will be clinically prohibitive especially for higher-spatial resolution scar imaging where scan time is 4-7 min [25][26][27]. That said, GB-LGE sequence can be easily modified to achieve PSIR GB-LGE, similar to BB-LGE [15]. Optimizing the contrast cost function in the proposed method results in shorter T 2 preparation times (4-28 ms) compared to our prior parameter selection algorithm for DB-LGE (35 ms) [13] or PSIR-based DB-LGE (10-40 ms) [15]. A shorter T 2 preparation time will result in increased SNR in the new parameters selection scheme; however we did not directly compare the SNR improvements between the two sequences.
In general, flow-independent DB-LGE imaging sequences can potentially generate GB-LGE contrast through some modifications. For example, increasing the time delay between T 2 preparation and acquisition in DB-LGE [13] or changing the parameter selection function in PSIR DB-LGE [15] can result in GB-LGE contrast. However, this may require relaxing the constraints of fixing or minimizing the T 2 preparation echo times in DB-LGE and PSIR DB-LGE, respectively. In other T 2prepared LGE imaging sequences [10][11][12], our parameter optimization methodology can be followed to generate the GB-LGE contrast. In PSIR DB-LGE (without magnetization preparation) method [9], increasing the inversion-to-acquisition time delay can generate GB-LGE contrast but will reduce the scar-to-blood relative contrast. In flow-based DB-LGE methods [6,7], changing the sequence timing parameters results in incomplete blood suppression, which may lead to inhomogeneous and patchy signal within the blood pool.
In our study, we used different contrast agents, types and doses in animals and humans. Despite these differences, we successfully achieved GB-LGE in both human and animal studies. Use of a high-relaxivity contrast agent and dose in animal experiments resulted in higher blood pool signal in conventional PSIR-LGE images, and further demonstrated the effectiveness of GB-LGE in blood suppression and scar detection. This difference also highlights the robustness of the parameter optimization for parameter selection for different contrast types and doses. The proposed contrast function was intended to provide a simplified analytical approximation of the LGE image contrast to enable tractable optimization of the imaging parameters. The observed differences between the actual and the prescribed blood to scar contrast in GB-LGE in our study can be explained by several factors. First, the optimization solver was weighted to prioritize nulling the myocardium signal than achieving specific blood to scar ratio. Also, the actual scar T 1 is controlled by many factors (including the dose and washout rate of the contrast material) while it is assumed fixed relative to the blood T 1 in the contrast model. The latter factor leads to variability of the blood signal between patients. However, this effect also presents in conventional LGE but with a lesser extent (Additional files 1, 2, 3 and 4). This observation was confirmed in the quantified image contrast, where the standard deviation of the scar-to-blood relative contrast in GB-LGE was higher compared to conventional LGE (Table 1). Table 4 Scar-to-blood, scar-to-myocardium and blood-to-myocardium relative contrast in each animal dataset (in gray blood (GB), black blood (BB), and conventional (phase-sensitive inversion recovery) late gadolinium enhancement (LGE) sequences) Pig   Our study has several limitations. While we found differences in the ability of different sequences to detect scarring, we do not have any histological evidence of the presence/extent of scarring in those patients. We have not performed histological validation of the GB-LGE sequence. Larger studies in patients are warranted to further assess the diagnostic and prognostic value of GB-LGE for scar detection.

Conclusion
GB-LGE imaging improves the ability to identify and localize scarred tissues compared to BB-LGE and conventional LGE.