INTRODUCTION

Computed Tomography is currently the clinical gold standard for assessing lung anatomy. To avoid ionizing radiation, MRI has been proposed as an alternative modality that can provide a simultaneous evaluation of pulmonary structure and function. However, lung MRI can be challenging due to short T2* relaxation times in lung parenchyma, low proton density, and substantial respiratory motion (1). This work explores lung MRI using an ultrashort echo time (UTE) technique based on a stack-of-spirals trajectory (2,3). The sequence uses variable-duration slice encoding to minimize T2* decay, Cartesian sampling along Kz, and in-plane spiral sampling to reduce the time needed for covering k-space. To achieve motion-resolved and high-resolution pulmonary MRI during free breathing, spiral arms were grouped into different respiratory states based on self‐navigator signals (4). These data were then reconstructed with an eXtra-Dimensional (XD) compressed-sensing algorithm that used self-calibrating GRAPPA operator gridding (GROG) (5) as an initial step, to avoid repetitive gridding/regridding operations during the iterative reconstruction of the non-Cartesian data. The proposed reconstruction pipeline was tested on 5 healthy volunteers, showing improved motion-compensated imaging results as compared with conventional gridding.

METHODS

All data were acquired with a work-in-progress (WIP) Spiral UTE sequence on a 3T Prisma system (Siemens Healthcare, Erlangen, Germany) using a body coil array. All human scans were performed under free-breathing conditions. Common imaging parameters included TE/TR=0.05/3.7ms, spiral duration=1600 µs, 144 slices, 2000 spiral views, no fat suppression, 1.3mm or 1.0mm isotropic resolution, and golden-angle ordering of spiral arms. The sequence uses non-selective RF excitation. Therefore, a coronal orientation was chosen to minimize the required number of through-plane phase encoding steps. Two reconstruction techniques were compared: (1) Standard gridding (NUFFT) without motion compensation, calculated directly on the scanner, and (2) the combination of GROG and XD-GRASP reconstruction (referred to as GROG-XD-GRASP and is described in the following subsection) implemented in Matlab (MathWorks, MA, USA).
GROG-XD-GRASP Reconstruction: We aimed to extend the XD-GRASP approach by incorporating GROG, specifically for spiral UTE data. Using the spiral trajectory given by the machine, GROG reconstruction was integrated as a preprocessing step shifting all spiral k- space points onto a Cartesian grid. The k-space data reconstructed by GROG was then sorted according to respiratory motion data extracted from center of k-space. Next, the XD-GRASP iterative reconstruction was performed by solving the following optimization problem:
$$\arg\min_\mathbf{d} ||\mathbf{F}\mathbf{C}\mathbf{d}- \mathbf{m}||_2^2+λ||\mathbf{T}\mathbf{d}||_1$$
where $$${F}$$$ is the FFT operator, $$${C}$$$ represents coil sensitivity maps, $$${d}$$$ is the image series to be reconstructed, $$${m}$$$ is the sorted k-space data (after GROG) and $$${T}$$$ is the sparsifying transform applied along the respiratory dimension with a regularization parameter .

RESULTS AND DISCUSSION

Spiral UTE data were binned into four respiratory states from inspiration to expiration (Fig.1a) based on the respiratory signal derived from the center of k-space. The subject was instructed to breath deep and steady during the scan, so that changes in lung volume could clearly be noticed in the resulting images. Next, the expiratory state was reconstructed by both the conventional gridding method and the proposed method (Fig.1b). Images generated by gridding show motion blurring (see yellow arrow) and provide less specific anatomical details. A closer comparison of these methods in another volunteer with identical spatial resolution show a clear advantage of the proposed motion-corrected reconstruction (Fig. 2, left), revealing more structures of the parenchyma. These anatomical structures can be further appreciated in the zoomed images (Fig. 2, right) of these regions. Finally, a high-resolution coronal dataset with 1.0 mm isotropic spatial resolution (Fig. 3) was binned into four respiratory states and motion‐resolved inspiration (left) and expiration (right) states were reconstructed by the two methods, demonstrating slightly finer anatomical details when using the proposed reconstruction.

CONCLUSION

This work shows that self-navigation and motion-sorting enabled by spiral UTE imaging, combined with sparsity-based reconstruction, can provide additional anatomical information of potential clinical value. Moreover, when implemented with initial GROG operator to avoid repetitive gridding operations, image reconstruction can be further accelerated.

Acknowledgements

We acknowledge support from NIH grant P41 EB0171813 and R01 5R01EB018308.