GRASP [1] is an iterative reconstruction technique for DCE-MRI, based on a radial stack-of-stars k-space trajectory. Due to the radial sampling scheme, GRASP is inherently motion-robust and, thus, promises dynamic abdominal imaging at high spatio-temporal resolution during free breathing.

However, if the subject is breathing deeply, the motion can lead to blurring and reduces the effectiveness of the applied through-time regularization. A recent extension, XD-GRASP [2], addresses this limitation by separating R=4-5 respiratory phases for each temporal phase and introduces an additional regularization term along the respiratory dimension. While this approach mitigates the effects of inter-frame and intra-frame motion, it significantly increases the computational effort because in total an R-times higher number of image volumes has to be reconstructed. Furthermore, the achieved temporal resolution is relatively low because a sufficiently high number of k-space samples must be available for every respiratory bin. Here, we propose the combination of respiratory gating with a higher gating acceptance to overcome the aforementioned challenges.

The golden-angle radial sampling scheme utilized by GRASP MRI has three important properties. First, the temporal resolution can be selected flexibly for reconstruction by grouping the desired number of radial readout lines into each temporal frame. Second, because the k-space center is read out with every acquisition angle, a respiratory self-gating signal can be extracted from the raw measurement data. And third, retrospective gating can be applied without deviating much from a uniform angular distribution of the k-space samples, due to the relatively high angular increment between subsequently acquired radial spokes.

The effect of motion in a GRASP study can therefore be reduced by applying retrospective respiratory gating, such that only a defined fraction of the acquired data in each time-point is utilized for reconstruction. To minimize inter-frame motion, the temporal resolution should be selected that each temporal phase contains at least one respiratory cycle.

Ten clinical patients were examined with IRB approval on a 3 T MRI scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) using the 18-channel body coil and the 32-channel spine coil array. Image acquisition was performed using a prototypical T1-weighted, fat-saturated spoiled gradient echo pulse sequence with golden-angle stack-of-stars k-space sampling. Scan parameters were as follows: TR/TE = 3.5 / 1.6 ms, FA=12°, FOV = 332 x 332 x 200 mm³, base resolution = 256 samples, 1456 radial angles, number of partitions = 80, resulting in a spatial resolution of 1.3 x 1.3 x 2.5 mm³ and a scan time of 246 seconds. The contrast agent injection started approximately 20 seconds after the beginning of the scan and consisted of a 18-20 ml bolus of Dotarem (Guerbet, France) followed by a 20 ml saline flush.

The respiratory self-gating signal was extracted automatically from the k-space center [3], and a 50-second sliding mean filter was used to suppress a signal baseline drift due to the arrival of the contrast agent. The raw data was partitioned into temporal frames with a temporal duration of 42 radial spokes (7.1 seconds).

First, non-gated GRASP reconstructions were computed using FISTA-based optimization [4]. Then, gated GRASP (gGRASP) was carried out by retrospectively gating each time-frame at end-expiration with 50 % acceptance and applying the same reconstruction algorithm.

In radial imaging, motion manifests as blurring of structures, which can lead to a reduced contrast compared to the surrounding tissue. Therefore, to assess the effectiveness of the proposed gating, manually drawn line profiles perpendicular to the posterior sectoral branch of the right portal vein were extracted using Fiji (NIH, USA) and analyzed in Matlab (The MathWorks, USA). Vessel contrast was defined as the ratio of the peak intensity to the profile baseline, where the peak intensity was averaged over the highest three voxel intensities, and the baseline was averaged over the first and last three intensity values of the line profile. For each patient, the maximum vessel contrast was determined for both GRASP and gGRASP.Respiratory signal extraction and gating succeeded in all patients. An exemplary respiratory signal and the corresponding gating and temporal windows are shown in Fig. 1a. Line profiles across one vessel in non-gated and gated venous-phase images of one patient are shown in Fig. 1b-d. The vessel-to-liver contrast is improved in the gGRASP reconstruction. This was also reflected quantitatively, with 7.0-27.5 % higher maximum vessel contrast ratio (mean/std: 15.0 ± 7.3 %).Respiratory gating allows to improve the motion-robustness of abdominal GRASP DCE-MRI. The blurring effect of motion is reduced, which leads to improved conspicuity of vessels and, possibly, lesions in the liver and kidneys.