Synopsis

Keywords: Kidney, Perfusion, renal perfusion

We performed a series of optimization on the encoding scheme, pseudo-continuous ASL (pCASL) parameters, and post-processing of time-encoded pCASL (te-pCASL), then proposed Time-encoded Arterial Spin labeling to cover the whole Kidneys (TASK) to achieve multiple time-points renal perfusion measurement efficiently. With G_ave of 0.4-0.6 mT/m and G_max/G_ave around 10, Walsh-Hadamard encoding scheme, and retrospective registration, TASK was able to provide accurate and reproducible RBF and ATT measurement covering the whole kidneys with single 5-min scan.

Introduction

Arterial spin labeling (ASL) has become a novel MRI technique for renal perfusion measurement without contrast agent injection, whereas suboptimal post-labeling dalay (PLD) choice influences the accuracy of renal blood flow (RBF) measurement. Traditional multi-delay ASL requires an intolerant long scan time for repeated scans, though it provides more accurate and repeatable renal perfusion quantification. In this study, we performed a series of optimization on the encoding scheme, pCASL parameters, and post-processing of time-encoded pseudo-continuous ASL (te-pCASL), then proposed Time-encoded Arterial Spin labeling to cover the whole Kidneys (TASK) to achieve multiple time-points renal perfusion measurement in a single scan. The accuracy and reproducibility of TASK were evaluated in the young and elder healthy subjects.

Methods

Sequence Design: The sequence diagram of TASK is shown in Figure 1A. Pre-saturation is applied to eliminate the impact on the signal dependence of TR and the residual labeled blood. The te-pcasl in 7×8 blocks is encoded in the Walsh-Hadamard scheme¹ and followed by multi-slice EPI acquisitions.
Sequence Optimization: 1) Considering the large variety of labeling efficiency for pcasl applied in the abdomen, the net average gradient (G_ave) and rate of the amplitude of the slice-selective gradient (G_max) and G_ave (G_max/G_ave) of te-pcasl were determined by Bloch simulation according to the labeling blood velocity at the abdominal aorta measured by 2D phase contrast (PC) imaging; 2) Retrospective groupwise registration² was performed to eliminate the motion artifacts; 3) The Walsh-Hadamard encoding scheme¹ was implemented to achieve decoding with incomplete data in case of severe movements and halfway abortion during the scans.
MR Experiment: Ten young (age: 25.9 ± 3.7 years old, 5 female) and ten elder healthy subjects (age: 53.2 ± 5.8 years old, 5 female) and four patients with different kidney diseases were recruited for MRI experiments with informed consent. The experiments were conducted on the same MR scanner with a 16-channel torso coil and a 12-channel posterior coil. For all subjects, a 2D PC was firstly conducted as a scout scan to obtain the mean labeling blood velocity at the labeling plane in the abdominal aorta. The best G_ave and G_max/Gave values were immediately calculated based on the average labeling blood velocity by Bloch simulation and applied to the all following ASL sequences. Subsequently, a traditional sequential multi-delay pcasl sequences (SMD) and three versions of TASK (TASK_fix, TASK_adj and TASK_free) were performed in secession (Figure 1C). The PLD and labeling duration of SMD and all TASK techniques were optimized with the method used by Woods JG et al¹. Five subjects of the young (mean age: 26.0 ± 3.2 years old, 2 female) and five subjects of the elder groups (mean age: 51.6 ± 6.2 years old, 2 female) were randomly selected to undergo the reproducibility experiments.
Image Analysis: All analyses were performed on MATLAB (Mathworks, Natick). To minimize the impact of respiratory motion on image quality, all acquisitions of SMD, TASK_fix, TASK_adj and TASK_free were internally registered by the PCA-based groupwise registration² using Elastix³ mentioned above. For the TASK sequences, the acquisitions were firstly decoded according to the Walsh-Hadamard encoding scheme⁴ to obtain the PWI at 7 PLDs. The classic Buxton kinetic model5 was used for quantitating RBF and ATT for the four sequences. The PWI decoded from the incomplete data (TASK-2 for 1×2 encoding blocks and TASK-4 for 3×4 encoding blocks) for the three versions of TASKs were also calculated, and corresponding RBF and ATT results were fitted by the same kinetic model. The root-mean-squared error (RMSE) of fitting was also calculated. The ground truth RBF and ATT estimates were generated by fitting the combined data from all of the ASL sequences (~20 min scan time of data), similar to the method proposed by Woods JG et al⁶.

Results

Figure 2 shows the representative of the RBF and ATT ground truth and the RBF, ATT, RBF error, ATT error and RMSE maps obtained by SMD, TASK_fix, TASK_adj, and TASK_free. Figure 3 shows the representative quantitative image of the four patients. There was no significant difference in RBF between the young and elder groups (211.25±42.06 vs. 196.39±35.71 mL/100g/min, P>0.05), while ATT of the elder group was longer than that of the young group (663.05±199.91 vs. 789.68±149.89 ms). G_ave of 0.4-0.6 mT/m and G_max/G_ave around 10 was executable for TASK techniques. Compared with SMD, RBF and ATT measured by TASK techniques were in better and excellent agreement (ICC: 0.932-0.986, R² = 0.92-0.95 vs. ICC: 0.769-0.902, R² = 0.40-0.74) and had significantly smaller errors (P<0.001) with the ground truth (Figure 4). Among TASK techniques, the RMSE of TASK_free was relatively higher than the others (P<0.05) (Figure 4). All TASK techniques showed excellent reproducibility (ICC, 0.897-0.987, R² = 0.70-0.99, CV = 3.45-9.28%), which were better than SMD (ICC: 0.655-0.940, R² = 0.57-0.81, CV = 7.71-15.20%) (Figure 5).

Discussion and Conclusion

With G_ave of 0.4-0.6 mT/m and G_max/G_ave around 10, Walsh-Hadamard encoding scheme, and retrospective registration, TASK_fix and TASK_adj were able to provide accurate and reproducible RBF and ATT measurement covering the whole kidneys with single 5-min scan.

Acknowledgements

None.