Fast myocardial T1 mapping using cardiac motion correction

To improve the efficiency of native and postcontrast high‐resolution cardiac T1 mapping by utilizing cardiac motion correction.


| INTRODUCTION
Cardiovascular MR is a widely used imaging modality for the diagnosis of cardiac diseases. The quantification of T 1 relaxation is an important intrinsic marker to detect ischemic and nonischemic cardiomyopathies. [1][2][3] Postcontrast and native T 1 mapping are relevant for the characterization of pathological changes of myocardial tissue. Native myocardial T 1 has the potential to detect diffuse cardiomyopathies without the need of contrast media administration. 4,5 Furthermore, the combination of native and postcontrast T 1 maps allows for calculation of the extracellular volume fraction, permitting quantification of diffuse fibrosis. 6 Whereas respiratory motion may be largely suppressed by holding the breath during a 2D MR scan, cardiac motion is always present during data acquisition. Common myocardial T 1 mapping techniques therefore utilize electrocardiogram (ECG) triggering in which qualitative images with different TIs (TI images) are only acquired at 1 specific cardiac motion state, such as mid-diastole, after an inversion or saturation recovery preparation. However, such ECG-triggered acquisitions limit the amount of data used for image reconstruction to 10% to 20% per cardiac cycle, leading to low scan efficiencies. [7][8][9] This is especially challenging for T 1 mapping because multiple TI images need to be obtained for the calculation of 1 quantitative T 1 map. In order to ensure clinically feasible breath-hold times, cardiac T 1 mapping is commonly limited in terms of spatial resolution. In addition, ECG triggering requires reproducible detection of each cardiac cycle during data acquisition. Heart rate variations or arrhythmia can lead to TI images, which are obtained in different cardiac phases, and the mismatch between these images could lead to errors in voxel-based T 1 mapping. 9,10 Continuous data acquisition with retrospective cardiac motion correction has been proposed for 3D anatomical cardiovascular MR in order to improve scan efficiencies and make data acquisition not reliant on accurate ECG-triggering. 11,12 For quantitative cardiovascular MR, thus far only respiratory motion correction has been explored to improve parameter map quality. [13][14][15][16][17][18] These techniques correct for small residual respiratory motion in breath-held and free-breathing acquisitions. Cardiac motion leads to a more complex deformation of the heart than respiratory motion. Therefore, accurate cardiac motion estimation requires high-quality cardiac motion-resolved images with constant and high contrast between myocardium and blood.
In this study, we present a cardiac motion-corrected T 1 mapping approach with an increased T 1 mapping efficiency for high-resolution T 1 mapping. This is done by continuous acquisition, allowing reconstruction of cine MRI and TI images from the same data. Nonrigid cardiac motion estimation was integrated to correct for motion before carrying out voxel-based T 1 mapping. The approach was evaluated in 10 healthy volunteers for native T 1 mapping. Furthermore, clinical feasibility was tested in 5 patients with suspected myocardial fibrosis after contrast agent administration by comparing postcontrast T 1 maps to late gadolinium enhancement (LGE) images, which is the gold standard modality for detection of focal fibrosis.

| METHODS
The proposed approach can be separated into 3 steps: First, cine images are reconstructed from the acquired 2D golden angle radial data. In a second step, these images are registered to obtain nonrigid motion fields describing the cardiac movement of each voxel during the cardiac cycle. Lastly, cardiac phase-resolved images are reconstructed for different TIs from the same data. Each cardiac phase is transformed to a reference phase using the motion fields obtained in step 2. Systolic cardiac phases were excluded due to through-plane motion allowing for 80% of the total scan time to be used for T 1 mapping. A voxel-wise 3-parameter T 1 fit is applied to obtain the final quantitative cardiac motion-corrected T 1 maps. F I G U R E 1 Data acquisition and cardiac motion estimation. (A) Data is continuously acquired with golden angle radial readout. Multiple inversion RF-pulses were applied. (B) Cine images with 15 cardiac phases were reconstructed prior to T 1 mapping. Only part of the data (red) is used for reconstruction to guarantee a constant contrast over all cardiac phases. Negative signal associated with healthy myocardium is rejected (dashed line). (C) For each cardiac phase, a cardiac motion field is obtained 2.1 | Data acquisition 2D slices were acquired continuously by a low flip angle spoiled-gradient echo sequence with golden angle radial sampling, an angular increment of 111.25°, 19 and employing adiabatic inversion recovery preparation at constant time intervals (Figure 1).Data was acquired within a breath-hold on 3 Tesla (Verio, Siemens Healthineers, Erlangen, Germany)-flip angle: 5°,TE/TR: 2.03/4.93 ms, FOV: 320 × 320 mm 2 , 2-fold oversampling in radial direction, resolution: 1.3 × 1.3 × 8.0 mm 3 . Nonselective inversion RF-pulses were applied every 2276 ms. The ECG was recorded for retrospective cardiac gating and data selection. Total acquisition time per slice was 8 s.

| Cine reconstruction
Data was retrospectively binned into 15 cardiac motion states using the ECG signal providing cardiac motion-resolved images (I Cine [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15] ) ( Figure 1B). The contrast over the entire scan is not constant due to the application of multiple inversion pulses. Nevertheless, each image I Cine is reconstructed from data at various time points after the inversion pulses. This leads to an average contrast for I Cine [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15] , which is constant over all reconstructed cardiac motion states. In order to enhance the contrast between blood and myocardium, data was only used for cine reconstruction if the signal intensity of myocardium was positive. Data acquired immediately after all inversions was therefore excluded for reconstruction depending on the T 1 times associated with healthy myocardium. In noncontrastenhanced imaging, data obtained 1175 to 2276 ms after the first inversion and 400 to 2276 ms after all other inversions were included ( Figure 1B). Thus, only the positive signal of myocardium is used. Blood has a longer native T 1 time, so the signal from blood is lower and partly negative, resulting in dark blood images. The acceleration factor of native cine images was 5.0. In contrast-enhanced imaging, only 200 ms after each inversion were excluded due to shorter T 1 times. Here, myocardial T 1 is longer than T 1 of blood, resulting in a higher blood signal in the corresponding cine images and thus bright blood images. I Cine 1-15 were reconstructed iteratively using spatial and temporal total variation regularization (10 iterations, λ spatial : 10 −6 , λ temporal : 5 × 10 −6 ). 20 All image reconstruction was performed offline in MATLAB 2016b (MathWorks, Natick, MA).

| Signal model of longitudinal magnetization
The longitudinal magnetization M during continuous spoiled gradient echo acquisition after an inversion pulse can be described as Look-Locker concept. 23 In our model, the inversion recovery Look-Locker concept was extended to multiple inversions with a constant time between the inversion (T ic ) to characterize the longitudinal magnetization of the entire data acquisition without magnetization recovery periods 24 : where j is the number of the applied inversions at a specific time point, and M + is the magnetization right before the next inversion. M 0 is proportional to the spin density, and T eff 1 describes the effective relaxation time when a low flip angle α is applied 23 : In this model, M eff 0 is the steady state magnetization For robustness of the fitting, the inversion efficiency is assumed to be 1.

| Motion-corrected T 1 mapping
Cardiac motion-corrected T 1 mapping (T moco 1 ) contains the following steps ( Figure 2): 1. Gating: Data was retrospectively gated, yielding exactly the same 15 cardiac motion states as in cine reconstruction. In contrast to the cine reconstruction before, 15 phases for each cardiac cycle are obtained (i.e., real-time sequence) rather than retrospectively combining data from multiple cardiac cycles. 2. Data selection: Based on the reconstructed cine images of the same acquisition, 3 systolic motion states were excluded in each cardiac cycle to minimize through-plane motion artifacts in the T 1 maps, which cannot be corrected for in a 2D acquisition. 3. Image reconstruction: In each cardiac cycle (C1 to Cn), 12 highly undersampled TI images (TI 1-12 ) were reconstructed of the same cardiac motion states as selected for cine reconstruction (TI C1-Cn 1-12 ). TI images were reconstructed using non-Cartesian iterative SENSE 25 without temporal or spatial filtering of the data. The total data window for T 1 mapping was 80% per cardiac cycle, divided into 12 TI images. 4. Motion correction: Prior to T 1 mapping, corresponding inverse deformation fields iMF [1][2][3][4][5][6][7][8][9][10][11][12] were applied to all TI images TI C1-Cn Two additional T 1 maps were calculated based on the same raw data. To verify the influence of motion correction on T 1 estimation, the same TI images were selected as above; however, no motion correction step was performed prior to T 1 mapping, yielding T uncorr 1 maps. Besides this, the influence of the increased amount of k-space lines compared to standard cardiac triggered T 1 mapping approaches was evaluated. For this, data during mid-diastole was selected within a single window per cardiac cycle with a duration of 168 ms. 24 Here, the quality of each TI image is better due to the larger window length and thus F I G U R E 2 T 1 mapping. (A) T 1 mapping using 80% of the cardiac cycle. The cardiac cycle is split into the same cardiac phases as for cine reconstruction, leading to 15 qualitative images per cardiac cycle and 3 midsystolic images were excluded. Cardiac motion is corrected in TI images and T 1 is fitted, leading to T moco 1 . Furthermore, T 1 is also fitted on motion-uncorrected TI images (T uncorr 1 ). (B) For comparison purpose, T 1 mapping is performed using reconstructed TI images of only 1 cardiac phase of the same data acquired as in (A) (T 1phase 1 ). The window length was 168 ms per cardiac cycle, comparable to a standard T 1 mapping approach 442 | BECKER Et al. more available k-space data per TI image, but total amount of data used for T 1 mapping is lower. These T 1 maps are called T 1phase 1 maps throughout the study.

| Native T 1 mapping
The approach was applied without contrast administration in 10 healthy subjects (6 males/4 females, aged 31.3 ± 8.4 years) in a midventricular short axis (SAX) orientation. A stack of 9 SAX images was obtained in 1 heathy volunteer (male, 34 years) with the approach and 5(3)3 MOLLI. In accordance with the institutions' ethical committees, all participants gave written informed consent before examination. In order to study potential effects on different scan orientation, additional images were obtained in 4-chamber view in 2 of the subjects. T moco 1 , T 1phase 1 , and T uncorr 1 maps were calculated for each acquisition.

| Postcontrast T 1 mapping in patients with suspected focal fibrosis
To test feasibility of the approach for the detection of fibrosis, 5 subjects with suspected focal fibrosis were scanned (3 female/2 male, age: 60.6 ± 8.4 years, height: 168.8 ± 9.7 cm,weight: 76.0 ± 12.1, body mass index: 26.9 ± 5.2). The patients were included as part of the Berlin Longterm Observation of Vascular Events study. This study was approved by the local ethic committee (EA1/066/17), and all subjects gave written informed consent. To detect focal fibrosis, a stack of SAX LGE phase-sensitive inversion recovery images with gradient echo readout was acquired 15 minutes after contrast administration (0.15 mmol/[kg body weight] gadoteridol) with the following scan parameter: flip angle: 20°, TE/TR: 1.56/4.10 ms, acquisition time: 10 heartbeats, spatial resolution: 2.1 × 1.4 × 7.0 mm 3 , FOV: 350 × 263 mm 2 ,which was adapted to the patient. For postcontrast T 1 mapping, 1 2D slice was acquired such that fibrotic tissue was best visible, and T moco 1 , T 1phase 1 , and T uncorr 1 maps were calculated.

| Analysis
In the phantom, T 1 times were averaged over each tube. Pearson's linear correlation was tested, and differences in T 1 time between the reference and the proposed approach as well as the reference and MOLLI were calculated and averaged over all tubes.
In vivo analysis was performed in midventricular SAX images. Six midventricular myocardial segments of the left ventricle, adapted from the American Heart Association consensus statement, and the left ventricular blood pool were segmented. 27 Segmentation and evaluation were performed using in-house software in MATLAB 2016b (MathWorks).
Motion fields 33 were applied to the cine images (MF 1-15 • I Cine [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15] ) in order to visually assess motion correction. To investigate the amplitudes of estimated motion, the deformation fields of each cardiac motion state were quantified. For quantification, the norm of the 2D deformation vectors in each voxel were averaged over all 6 segments for each cardiac phase. The maximum deformation was determined in each subject and averaged over all volunteers. After exclusion of 3 systolic phases, the remaining maximum deformation was averaged over all volunteers. Additionally, the deformation was additionally obtained over the highest 10% of each segment and cardiac phase; subsequently, the highest deformation was averaged over all volunteers.
To test reproducibility of the motion estimation, an additional test was performed on the motion fields. Estimated inverse motion fields of the 12 included cardiac phases were applied to the reference cine images (iMF [1][2][3][4][5][6][7][8][9][10][11][12] . In a second step, these obtained cine images were again registered to I Cine ref , obtaining new cardiac motion fields. This was done for all 10 healthy subjects. Differences between both motion fields were calculated for each cardiac phase and each segment.
Native and postcontrast T 1 times were assessed in all myocardial and blood segments. Normal distribution of T 1 times was tested by a Shapiro-Wilk test. To verify overall effects of the T 1 mapping method on T 1 times, T 1 times of T moco 1 , T 1phase 1 , and T uncorr 1 maps were averaged over the 6 segments and compared using a Friedman test with Dunn's multiple comparison correction.
In this study, the SD of T 1 times across a segment (spatial variability) was used to quantify the precision of the measurement in healthy subjects because in healthy volunteers the T 1 times of the myocardium can be assumed to be constant.
The SD of native T 1 was determined in each segment. Normal distribution of SDs was tested by a Shapiro-Wilk test. Overall differences in SD between the 3 methods were assessed by a comparison of the SD over the whole , and T uncorr 1 maps by analysis of variance with Tukey's multiple comparison correction across the 10 healthy volunteers. For comparison to MOLLI, T1 times were assessed segment-wise in basal, midventricular, and apical slices and averaged. Spatial variation was averaged over all segments for analysis of precision.
In postcontrast T moco 1 maps with visible focal fibrosis, T 1 times of fibrotic segments were assessed, and T moco 1 values of the fibrotic segment were compared against T 1 values of all healthy segments using a Kruskal-Wallis test. The statistical analysis was performed in GraphPad Prism 6, and P < 0.05 was considered to be significant.

| Phantom experiment
Cardiac motion-compensated T 1 mapping and MOLLI showed good correlation with the reference scan (R 2 > 0.99). Relative differences between the reference and both techniques averaged over all tubes were 0.28% ± 3.70% (−5.42% to 3.88%) using our approach and 1.43% ± 1.99% (−1.42% to 4.61%) using 5(3)3 MOLLI. A comparison between the reference and our approach can be found in Supporting Information Figure S1.

| Cine reconstruction
Cine images had a consistent contrast over all cardiac phases, which can be observed in the temporal slice profiles of Figures 3 and 4. Noncontrast-enhanced images resulted in a dark-blood contrast (Figure 3), and postcontrast cine images showed a bright-blood contrast (Figure 4) due to longer myocardial postcontrast T 1 times compared to blood. The temporal resolution of each cardiac phase was between 55 and 87 ms, depending on the heart rate (46 to 73 beats per minute [bpm]), and only few streaking artefacts were present in cine images.

| Motion estimation
Motion estimation succeeded in all volunteers, with only small residual motion in cardiac motion-corrected cine images (MF 1-15 • I Cine [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15] ). The effect of motion correction can also be seen in the temporal slice profiles of the cine images in Figures 3 and 4. Animated cine images with and without F I G U R E 3 Cine images and motion correction without contrast administration. The deformation of the heart is estimated for each cardiac phase (middle), and motion is corrected (right) to match the reference mid-diastolic phase (left). The temporal slice profiles of cine images excluding 3 systolic phases are shown in the lower row (red line). Animated cine images with and without motion correction can be found in the Supporting Information Video S1 motion correction, in which 3 systolic motion states were already excluded, can be found in Supporting Information Video S1 (precontrast) and Supporting Information Video S2 (postcontrast), and motion fields of all cardiac phases are shown in Supporting Information Video S3. The highest deformation over the whole myocardium and all healthy subjects was 4.51 ± 0.98 mm in systole. Excluding 3 systolic motion states, the remaining highest deformation was still 3.30 ± 0.88 mm. This averaged deformation underestimates the maximum deformation of the myocardium due to large variations over the myocardium. The highest deformation in healthy subjects was 7.54 ± 1.89 mm. The estimated motion per cardiac phase depended strongly on the heart rate, with an averaged motion of more than a voxel length of 1.3 mm in only 5 of 15 cardiac phases for a very low heart rate (46 bpm) and 9 cardiac phases for a higher heart rate (66 bpm) ( Figure 5A).
The repeatability experiment showed small differences between motion estimations. Differences between both motion estimations were 0.25 mm ± 0.17 mm [0.04 to 1.32 mm], averaged over all 15 cardiac phases, segments, and volunteers. Only in 1 systolic cardiac phase of 1 healthy volunteer was the error in the range of the voxel size (1.32 mm).

| Native T 1 mapping
In SAX orientation as well as 4 chamber view, motion-corrected T 1 maps had less blurring and partial volume effects compared to uncorrected T 1 maps (Figures 6 and 7). The myocardium and papillary muscles can be well distinguished in the T moco  Figure S2.
In Figure 5, the influence of cardiac motion on T uncorr 1 maps is shown for different heart rates. The main F I G U R E 4 Cine images and motion correction after contrast administration. The deformation of the heart is estimated for each cardiac phase (middle), and motion is corrected (right) to match the reference mid-diastolic phase (left). The temporal slice profiles of cine images excluding 3 systolic phases are shown in the lower row (red line). Animated cine images with and without motion correction can be found in the Supporting Information Video S2 difference between the high and low heart rate was that, after leaving out 3 systolic phases (marked as blue area), only 2 cardiac phases show a deformation larger than a voxel length of 1.3 mm for a low heart rate (46 bpm). For the higher heart rate (66 bpm), 6 remaining cardiac phases show a deformation larger than 1.3 mm. Therefore, the impact of cardiac motion on the final image quality is higher for the high heart-rate case than for the low-heart rate case ( Figure 5B,C). However, for heart rates in the normal range or higher (here shown for 66 bpm), cardiac motion resulted in inaccurate estimation of myocardial T uncorr 1 , which can be observed by regional prolonged T 1 times due to a mismatch of motion state in uncorrected TI images ( Figure 5B).

F I G U R E 5
Impact of motion on uncorrected T 1 map at different heart rates. (A) Estimated cardiac motion averaged over all 6 segments per cardiac phase (mean ± SD). For normal heart rates, even after excluding systolic phases, motion amplitudes are larger than the spatial resolution of 1.3 mm for most cardiac phases (orange). In the corresponding T uncorr 1 map (B), motion led to wrong T 1 estimation. For low heart rates, only a few cardiac phases show motion amplitudes larger than the voxel size (blue), and no motion artifacts are visible in the corresponding T uncorr 1 map after exclusion of 3 systolic phases (C) F I G U R E 6 Native T 1 maps of 3 healthy subjects. T 1 maps estimated using 1 cardiac phase (T 1phase 1 , left) showed higher SD compared to T 1 maps calculated from images covering 80% of each cardiac cycle. In motion-uncorrected T 1 maps (T uncorr 1 , middle), motion led to blurring and partial volume effects (arrows), which is not present in motion-corrected T 1 maps (T moco 1 , right) T 1 times and SDs of all myocardial segments and blood are shown in Figure 8. T  (Figure 9). Basal, midventricular, apical, and blood T 1 times were 1217 ± 17 ms, 1196 ± 37 ms, 1224 ± 43 ms, and 1898 ms in MOLLI, respectively. In T moco 1 maps, T 1 times were 1272 ± 40 ms, 1267 ± 35 ms, F I G U R E 7 Native T 1 maps in 4 chamber view. Cardiac motion-corrected T 1 mapping is not restricted to short-axis orientation. Motioncorrected T 1 maps showed better T 1 mapping quality than uncorrected T 1 maps, especially in the lateral wall of the left ventricle (arrows). T 1 maps using 80% of the cardiac cycle showed visually a lower spatial variation of T 1 compared to T 1 mapping with only 1 cardiac phase F I G U R E 8 T 1 times and spatial variation of T 1 (SD). (A) T 1 times across healthy subjects in 6 myocardial segments and in the left ventricular blood pool of each segment (mean ± SD). Motion-compensated T 1 and T 1 obtained with only 1 cardiac phase did not differ significantly. (B) The T 1 SD of the motion-corrected T 1 across healthy volunteers was significantly lower in all segments compared to T 1 maps obtained with only 1 cardiac phase and motion uncorrected T 1 (P = 0.036 and P < 0.0001, respectively) | 447

| Postcontrast T 1 mapping
In postcontrast T 1 mapping, slices were oriented in basal (subject C) and midventricular short-axis orientation (subject A and B). In the LGE images, focal fibrosis was visible in 2 patients. In subject B, focal fibrosis was depicted in a lateral segment, and even small focal fibrotic spots were detected in the anterior segment of subject C (Figure 10). In the subjects without focal fibrosis, T 1 times were homogenous over the myocardium (subject A).
Postcontrast T 1 mapping was successful with the same acquisition as for native T 1 mapping. In postcontrast T uncorr 1 maps, different motion states of the TI images led to blurring of the myocardial border. In T moco 1 maps, the border between blood and myocardium was better defined than in T uncorr 1 maps. In postcontrast T moco 1 maps, focal fibrosis could be detected at the same location as in the LGE images. In these spots, T 1 times were shorter compared to surrounding healthy tissue. In the 2 subjects with focal fibrosis, the averaged T moco 1 of the fibrotic segment were −69.5 ± 11.9 ms and −51.8 ± 12.0 ms shorter compared to all healthy segments in the same subject (P < 0.002 for all segments) (Figure 9). In T uncorr 1 maps, differences were −67.3 ± 14.2 ms and −28.6 ± 17.4 ms, and small fibrotic spots were poorly visible in these T1 maps. In T 1phase 1 maps, differences were 60.5 ± 19.3 ms and 46.4 ± 14.3 ms.

| DISCUSSION
In this study, a novel cardiac motion-corrected T 1 mapping approach was presented that provides high-resolution T 1 maps in a short 8 s breath-hold. Scan efficiency was increased by continuous data acquisition and correction for cardiac motion. Hereby, it was possible to use 80% of the total scan time for T 1 mapping instead of only about 20%. Thus, the amount of data used for T 1 mapping was increased by a factor of 5 compared to a cardiac-triggered T 1 mapping approach. Cardiac motion estimation was carried out from the same raw data used for T 1 mapping. An increase in precision compared to a cardiac triggered approach using only 1 specific motion state was demonstrated, and feasibility for postcontrast T 1 mapping in patients was shown.
The spatial variation of T 1 times (SD) in healthy myocardium is related to the precision of the measurement. By integration of motion correction, the precision of T 1 mapping was improved compared to T uncorr 1 mapping, which could be explained by less blurring in T moco 1 maps. Improved sharpness could be important for the detection of small epi-or endocardial fibrosis.
The individual TI images of T moco 1 and T uncorr 1 were reconstructed from data acquired in a 55 to 87 ms window compared to the 3 times longer window used for T 1phase 1 mapping. Therefore, the TI images of T moco 1 and T uncorr 1 showed higher undersampling artifacts. Nevertheless, for the voxel-wise T 1 fit, 12 TI images per cardiac cycle were used for T moco 1 and T uncorr 1 rather than 1 for T 1phase 1 . This led on average to 5.2 (4.3 to 6.8) times more data being utilized for the T moco 1 fitting. For the proposed golden angle radial acquisition, the undersampling artifacts between different TI images were incoherent along the recovery curve; thus, a robust T 1 fit was achieved, 28 leading to more reproducible T 1 times and an increase in precision by 34% compared to the standard T 1phase 1 mapping. In vivo T 1 times obtained with T uncorr 1 , T moco 1 , and T 1phase 1 mapping varied slightly, but all were within the range of previously published native T 1 times at 3 Tesla. Compared to those studies, T 1 times were longer than MOLLI, also observed in our study, and longer than shortened MOLLI plus multitasking, [29][30][31] but shorter than SASHA and SAPPHIRE. 32 A study evaluating the precision of different T 1 mapping approaches showed a comparable precision in MOLLI and SAPPHIRE (54.0 and 58.6 ms, respectively) and a lower precision of SASHA compared to our approach (68.2 vs. 57.0 ms). 32 However, comparison to other studies is difficult due to differences in acquisition and reconstruction settings as well as postprocessing procedures such as spatial filtering and the number of fit parameters, strongly affecting the precision. 9 Radial sampling is very robust in the presence of physiological motion. Therefore, good quality T 1 maps can also be obtained for T uncorr 1 mapping, especially for subjects with low heart rates. Figure 5 shows that for a volunteer with a heart rate of 46 bpm, removing 20% of the data in systole leads to TI images in which the residual cardiac motion is below the spatial resolution of the images in more than 80% of the remaining cardiac phases. Therefore, the T 1 fit is mainly determined by these 80%, leading to a high-quality T 1 map. For a volunteer with a heart rate in the normal range (66 bpm), removing 20% of data still leaves 60% of data with residual cardiac motion larger than the voxel size. In that case, T 1 fitting is impaired by cardiac motion, and the obtained T 1 map shows severe motion artifacts due to motion state mismatches in TI images.
The separation of data into different cardiac phases both for the cine reconstruction and T 1 mapping was carried out retrospectively. The selection of the systolic phases to be removed prior to motion-corrected T 1 fitting was done based on the reconstructed cine images rather than using a predefined delay time. Having the flexibility to retrospectively optimize the data selection can be especially important for arrhythmic patients in order to minimize errors due to ECG mistriggering. 10 In the phantom, only small differences were found between validation T 1 times and our model, but T 1 estimation has limitations. The inversion efficiency is not included in the signal model. It was shown in previous studies that this leads to T 1 underestimation of up to 5%, depending on the type of inversion pulse and relaxation times. 9,33 This will affect T moco 1 as well as the IR-SE validation experiment. Because expected errors are in the range of the SD of T 1 errors in the phantom experiment and the inversion efficiency depends on T 1 as well as T 2 , the inversion efficiency is not considered in this study. Furthermore, through-plane motion occurs during the systolic phase of the cardiac cycle, which is not considered in the signal model for continuous acquisition 1 used for T 1 fitting. Also, the slice profile is not explicitly considered in the model, which could lead to underestimation of T 1 , 34 but its impact is assumed to be small for small flip angles used in our method. Both through-plane motion and slice profile F I G U R E 1 0 LGE images, postcontrast T 1 maps, and bull's-eye plots. The myocardium was more precisely defined in T moco 1 maps (right) compared to uncorrected T 1 maps (T uncorr 1 , middle). In subject A, no focal fibrosis is present and T 1 times are homogenous over the myocardium. In subject B and C, focal fibrosis can be visually depicted in T moco 1 maps and confirmed by LGE images obtained at same slice position (arrows). In subject B and C, T 1 times in the fibrotic segment were shorter compared to all healthy segments of the same subject (subject B: −69.5 ± 11.9 ms and subject C: −51.8 ± 12.0 ms, P < 0.002). In subject C, the segment intersecting the left ventricular outflow tract was excluded in the analysis | 449 BECKER Et al.
could cause an estimated effective flip angle that is smaller compared to the nominal flip angle (Supporting Information Figure S2). Further investigation has to be performed on the influence of inversion efficiency, through-plane motion, and the slice profile on T 1 accuracy. The time between inversions was fixed to 2276 ms for simplicity, which was optimized in a previous study for 1 cardiac phase. 24 A different time or varying intervals would also possibly work due to the dense sampling by using more TI images. For analysis, we have not carried out a blinded review of the image quality between uncorrected and corrected T 1 maps.
Estimated cardiac contraction during systole was 4.51 ± 0.98 mm averaged over the whole myocardium of the left ventricle, which is in the range of published deformations of the left ventricle (4.6 to 5.5 mm in a midventricular section). 35 Small differences between repeated motion estimations (0.25 mm ± 0.17 mm) suggest reproducibility of the estimation. However, accuracy of the motion estimation cannot be evaluated with this experiment because no independent reference was used. Future research should therefore focus on obtaining cardiac motion estimation using, for example, tissue phase mapping or tagging 36,37 as independent reference measures.
Motion estimation with this technique is not only restricted to cine images obtained in short axis orientation; motion-corrected T 1 mapping was also successful carried out in a 4-chamber view. Here, especially visualization of the lateral wall was strongly improved using motion correction, which can be seen by more homogeneous T 1 times in the lateral wall than in T 1 mapping without motion correction. In addition, different software or techniques could be used for motion estimation in the future. For example, feature tracking could be implemented, which seems promising for the detection of cardiac motion. 38 However, the myocardium has to be contoured in this approach, which is not the case in NiftyReg. So far, motion transformation is performed in image space requiring interpolation, which could lead to partial volume effects and residual blurring of image data. Nevertheless, cardiac motion correction could possibly be integrated in TI image reconstruction to further increase sharpness of the anatomy by iteratively optimizing image quality. 11 Cine images with and without contrast agent administration showed constant contrast over all 12 cardiac phases, although the contrast changed during magnetization recovery (Figures 3 and 4) (Supporting Information Video S1 and S2). The radial data acquisition ensures that combining raw data with different image contrasts leads to a high image quality with only small artifacts due to the inconsistencies in k-space caused by the contrast change. Compared to standard cine imaging, complementary contrast was obtained due to dependencies of T 1 instead of the inflow of blood.
In this study, motion was estimated in 2D; however, the motion estimation algorithm can also be used for a 3D acquisition. This could be beneficial for structures with a high degree of through-plane motion, such as the papillary muscles and basal segments of the heart. 35 Therefore, a 3D acquisition could improve to detect complex cardiac motion. It could also help to reduce the scan time of a 3D T 1 mapping acquisition by improving the scan efficiency compared to proposed 3D T 1 mapping techniques 39,40 and eventually allow for T 1 mapping of the entire heart in a clinically feasible scan time. Respiratory motion also would have to be accounted for because a 3D acquisition would be too long for a breath-hold. For that, a rigid motion correction could possibly be integrated for which the k-space center could serve as navigator signal. Using a 3D acquisition, throughplane motion could be detected, and systolic cardiac phases would not have to be excluded, which would further increase scan efficiency.
TI images were reconstructed with iterative SENSE, without spatial or temporal filtering of the data. These images have severe streaking artifacts. Other advanced reconstruction techniques could have also been applied, such as compressed sensing, model-based reconstruction, low-rank methods, or integration of motion-compensation in image reconstruction to further improve image quality. 11,31,[41][42][43] Postcontrast T 1 maps were in good agreement with LGE images in subjects with focal fibrosis in midventricular and basal SAX. In future research, the approach should be evaluated in a larger patient cohort, especially in patients with irregular or more complex heart motion.
In only 8 s, a 1.3 × 1.3 mm 2 image resolution was achieved. This short scan duration has the potential for the acquisition of 2 slices within a single breath-hold, which would allow for the number of breath-holds to be reduced in the clinical routine. In addition, the image resolution could possibly also be further increased by prolonging the scan to a breath-hold duration of 10 to 16 s, which is still feasible in patients.

| CONCLUSION
In this study, native and postcontrast cardiac T 1 mapping with in-plane resolution of 1.3 × 1.3 mm 2 obtained in an 8 s breath-hold was shown. Continuous data acquisition and the integration of cardiac motion correction improved the data acquisition efficiency by a factor of 5 compared to standard triggered cardiac T 1 mapping techniques. This increased efficiency resulted in more precise T 1 maps (40% lower SD in the myocardium). Accurate detection of focal fibrosis in postcontrast T 1 maps was feasible using this technique.