Introduction

Tissue characterization including identification and quantification of fibrosis and oedema plays an important role in many myocardial diseases. Conventionally, 2D T₁ and T₂ maps are acquired sequentially under several breath-holds^1,2. However these approaches achieve limited spatial resolution and coverage of the heart. In this work, we sought to develop a free-breathing motion compensated 3D whole-heart sequence for joint T₁/T₂ mapping with isotropic resolution and co-registered anatomical water and fat imaging for visualisation of cardiac and coronary anatomy and epicardial and pericardial fat.

Methods

The proposed framework is shown in Fig.1. Four interleaved 3D volumes are acquired with ECG-triggered two-point bipolar Dixon, spoiled gradient echo (GRE) readout and variable-density Cartesian trajectory with spiral profile order and an undersampling factor of 4x^3,4. The four datasets are acquired respectively with 1) Inversion Recovery (IR) preparation, 2 and 3) no preparation and 4) T₂ preparation pulse (T2prep). Low resolution 2D image navigators (iNAVs)⁵ are acquired prior to imaging to estimate and correct for translational respiratory motion, enabling 100% scan efficiency and predictable scan time. Translational motion correction at end-expiration is performed for each echo of each dataset separately. The motion compensated undersampled in- and out-of-phase volumes are jointly reconstructed with a 3D multi-contrast patch-based low-rank reconstruction (HD-PROST)⁶. A water-fat separation algorithm⁷ is used to generate fat and water images for each dataset, and the four water images are normalized on a voxel-by-voxel basis in time to obtain the signal evolution across the four acquired volumes. Extended phase graph (EPG) simulations⁸, matching the acquisition parameters, are carried out to generate a subject-specific dictionary. Quantitative water T₁ and T₂ maps are generated by matching each voxel measured signal evolution to the closest dictionary entry, corresponding to a T₁/T₂ pair, whereas the T₂-prepared dataset (4^th dataset) is used for water/fat anatomical and coronary visualization.

Acquisition: A standardized T₁/T₂ phantom and six healthy subjects were scanned on a 1.5T scanner (Siemens Magnetom Aera) to validate the proposed free-breathing sequence. Acquisition parameters included FA=8deg, isotropic resolution of 2mm³, FOV=320x320x88-124mm³, coronal orientation, 14 echoes for iNAV acquisition, subject specific mid-diastolic trigger-delay and acquisition window of 96-118ms, TR/TE₁/TE₂=6.71/2.38/4.76ms, bandwidth=485Hz/pixel, T2prep=50ms, TI=120ms and total scan time=9±1min48sec. Accuracy and precision of T₁ and T₂ quantification was investigated in phantom scans with respect to phantom reference values⁹, and in-vivo with respect to 2D MOLLI and T2prep-based 2D bSSFP T₂ mapping. Additionally, a standard T₂-prepared CMRA with Dixon encoding, was acquired to qualitatively asses the fat/water separation for each acquired dataset and anatomical visualization of cardiac structures and coronary anatomy. 2D clinical T₁ and T₂ maps were acquired in 3 short axis views with a spatial resolution of 1.8x1.8x8mm³ in a ~10sec breath-hold for each slice, whereas the acquisition parameters of the conventional 3D T₂-prepared (T2prep=40ms) CMRA matched the proposed sequence.

Results

T₁/T₂ phantom: Phantom results are shown in Fig.2. Excellent linear correlations of y = 1.07x-11.7 (R²=0.98) and y = 0.85x + 5.7 (R²=0.99) were found between the proposed sequence and spin echo phantom values for both T₁ and T₂ quantification respectively. An overestimation of long T₁ (T₁=1489ms corresponding to blood) and an underestimation of long T₂s (vials corresponding to blood) were observed with the proposed sequence. The resulting biases were T₁ = 40.93ms and T₂ = -8.75ms respectively. However such over and underestimations were observed also with the 2D MOLLI and 2D T₂-prepared bSSFP sequences.

Healthy subjects: The translational motion corrected water and fat images for each acquired volume are shown in Fig.3 and compared with conventional T₂-prepared CMRA with Dixon encoding. Good fat/water separation is achieved with no observable water/fat swaps in the cardiac region for each reconstructed dataset. The T₂-prepared water volume (4^th dataset of the proposed approach) showed good myocardium-blood contrast for visualization of anatomical structures. The 3D isotropic nature of the acquisition allowed to reformat the acquired anatomical water and fat images and T₁ and T₂ maps in different orientations. Good depiction of anatomical structures and homogeneous T₁ and T₂ quantification can be observed for one representative healthy subject in Fig.4. T₁ and T₂ maps obtained with the proposed approach show comparable image quality to standard 2D techniques (Fig.5). A T₁ overestimation leading to a bias of T₁ = 111.2ms was observed in the septum compared to standard 2D MOLLI. In contrast a small bias of only -0.2ms was observed for T₂ quantification compared to 2D bSSFP T₂ mapping. A lower precision, quantified by the spatial variability in the septal region, was observed with the proposed approach (joint T₁/T₂ spatial variability: T₁ = 54.0±4.9, T₂ = 3.9±0.9) compared to 2D acquisitions (2D MOLLI spatial variability: T₁= 43.3±8.1 and bSSFP T₂ map spatial variability: T2 = 3.1±0.6). However, a comparable precision of T₁ = 43.7±7.2ms and T₂ = 2.7±1.0ms were observed by reconstructing the images matching the resolution of the 2D acquisitions (slice thickness = 8mm).

Conclusion

The proposed motion compensated joint T₁/T₂ sequence showed good agreement with reference T₁ and T₂ values in the T₁/T₂ phantom and good agreement with 2D MOLLI T₁ mapping and 2D bSSFP T₂ mapping in healthy subjects. Additionally, a 3D CMRA water/fat volume for anatomical and coronary visualization is obtained. Future work will include validation in patients.

Acknowledgements

This work was supported by EPSRC (EP/L015226/1, EP/P001009/1, EP/P007619 and EP/P032311/1) and the Wellcome EPSRC Centre for Medical Engineering (NS/A000049/1).