Introduction
Renal perfusion is a valuable physiological parameter for assessing kidney function and identifying pathology [
1]. In recent years, arterial spin labeling (ASL) magnetic resonance imaging (MRI) has been emerging as a method for measurement of renal perfusion [
2] that does not warrant administration of contrast agent [
3‐
5] or the involvement of lengthy invasive clearance measurements [
6,
7]. An additional advantage of ASL is that it enables measurement of local perfusion, in contrast to clearance techniques which only assess renal blood flow of both kidneys combined. In ASL, images are acquired with (label) and without (control) inverting/saturating magnetization of arterial blood flowing into the tissue. The signal intensity present in the subtracted label–control images is proportional to perfusion of the tissue [
8].
Renal perfusion is a relatively new application area of ASL-MRI. In the white paper for brain ASL [
9], pseudo-continuous ASL (pCASL) is the recommended labeling approach. However, the most commonly used labeling approach for renal ASL thus far has been flow-sensitive alternating inversion-recovery pulsed ASL (FAIR) [
2]. In general, FAIR has the advantage of higher labeling efficiency and lower specific absorption rate, whereas pCASL can achieve a higher intrinsic signal-to-noise ratio (SNR) [
9,
10].
ASL-MRI in the abdomen brings new challenges with respect to the brain, like respiratory motion, complex vasculature, and increased magnetic field inhomogeneities due to the proximity of air in the lungs and in the digestive tract. Thus far, no studies have been performed directly comparing different labeling approaches for renal ASL, and therefore, the effect on the obtained perfusion signal remains unclear. The aim of this study was to compare ASL-MRI of the kidneys with the two most commonly used labeling approaches FAIR and pCASL, to obtain a better insight into the reliability and repeatability of each method and to identify the most efficient method to perform renal perfusion measurements.
Discussion
The current study focused on comparing multi-delay ASL measurements with the two labeling approaches most commonly used in the kidney, FAIR and pCASL. The goal was to obtain a better insight into the reliability and repeatability of each method and to identify the most efficient method to perform multi-delay renal perfusion quantification at 3T. The study showed that measured renal perfusion values depend on the labeling approach. Perfusion values were significantly higher for FAIR than for pCASL, and showed substantially larger variability between visits for pCASL compared to FAIR.
Previous renal ASL studies at 3T in healthy volunteers have reported mean cortical RBF in the study population ranging from 199–399 mL/min/100 g and 138–296 mL/min/100 g for FAIR and pCASL, respectively [
2]. Studies using multiple delay times reported mean values ranging from 151–309 ml/min/100 g and 117–215 mL/min/100 g for FAIR and pCASL, respectively [
27]. These values highlight a wide range between studies that used the same labeling approach, as well as indicate a tendency of FAIR to measure higher RBF values, as found in our results. The wide variation of perfusion values makes it difficult to compare results between studies, even when the same labeling approach was used. Apart from differences in the used scanner hardware and image readout [
28], choices of parameters, such as labeling efficiency, blood-tissue partition coefficient, used in the kinetic model for perfusion quantification largely influence the obtained values. In this study, RBF values obtained with pCASL may be underestimated due to a lower labeling efficiency then assumed in the quantification. For renal application, the pCASL labeling efficiency is most likely influenced by a combination of several factors which will be discussed in more detail below.
In this study at 3T, repeatability was better for FAIR compared to pCASL. For some pCASL data sets (3 out of 27), the averaged perfusion-weighted images showed very low signal corresponding to cortical RBF values < 100 mL/min/100 g. Compared with RBF values obtained at the other visit or with FAIR labeling in the same subject, it seems that these pCASL measurements failed. In the Supplementary Figures S3, S4 and Table S1, pCASL results are presented after exclusion of these data sets. Exclusion of these data sets made the repeatability of pCASL much more comparable with FAIR. After elimination, FAIR still gave RBF values that were on average 147 mL/min/100 g higher for cortex and 54 mL/min/100 g higher for medulla than when measured using pCASL.
Despite hydration instructions and planning both visits at the same time of the day, intra-subject variation of perfusion values between visits was observed for both labeling approaches. Physiological variation of blood flow in the aorta [
29,
30] might play a role here in the amount of labeled blood that is created and delivered to the kidneys. Although there was intra-subject variability between visits, the diagnostic value of perfusion imaging, showing regional differences within or differences between kidneys, might still be unaffected. Moreover, studies comparing patients with impaired kidney function and healthy controls found relatively large perfusion differences of the average cortical RBF between both groups, ranging from 66 to 202 mL/min/100 g [
31,
32].
The number of studies reporting reliability and repeatability of renal ASL is limited, especially at 3T [
2]. Previous studies using FAIR at 3T in healthy volunteers [
33,
34] have reported inter-visit ICCs of 0.80 and 0.85, and CVs of 9.2 and 9.3% for cortical RBF. To our knowledge, no studies have been published yet reporting inter-visit reliability and repeatability at 3T for pCASL. However, there is one study that has investigated intra-visit reliability and repeatability of pCASL at 3T [
35] reporting an ICC of 0.93, and CV of 14.4% for cortical RBF. Reliability and repeatability of ATT measurements have been investigated even less. One study reported intra-visit an ICC of 0.32, and CV of 33.6% for cortical ATT with pCASL at 3T [
35].
The ATT is dependent on the used measurement method, and was, therefore, not directly compared between both labeling approaches. Previous studies using multiple delay times reported average cortical ATT values of 0.11–0.30 s for FAIR and 0.96–1.23 s for pCASL [
27]. The obtained ATT is very much determined by the timing of the (first) delay times. Based on preliminary measurements in healthy volunteers, we chose four equidistantly spaced delay times covering the perfusion signal over time both before and after the expected perfusion signal peak in the cortex for most subjects, to enable proper fitting of the perfusion model for quantification. The perfusion signal curves averaged over all subjects (Fig.
4) indicate that the delay times were correctly placed to capture the signal peak, for both labeling approaches. To improve the accuracy of the ATT estimation, more measurements with short delay times could be added. In general, accuracy of perfusion estimation increases with the number of sampled delay times [
36]. However, for clinical applicability of renal ASL, total acquisition time is restricted, limiting the number of delay times (and label–control repetitions) that can be obtained.
Measuring medullary perfusion is challenging. The medulla is much less perfused compared to the cortex (only ~ 10% of blood flowing into the kidneys flows through the medulla [
37]), and the transit time of labeled blood is much longer, which results in more T
1 decay before entering the medulla and lower tSNR, as was observed in this study. It is thus not surprising that medullary perfusion showed lower repeatability than cortical perfusion. Thus far, not many studies have reported medullary RBF and ATT values, and mainly focused on measuring cortical perfusion.
In the current study, lower PWS and tSNR values were found for pCASL compared with FAIR. This finding is not in line with the notion that the inherent SNR is higher for pCASL than for PASL techniques such as FAIR. For the brain, this has been shown by theoretical modeling of the perfusion signal, and has been demonstrated experimentally [
9,
38]. The higher intrinsic SNR of pCASL in the brain builds first on the longer temporal duration of the labeled bolus, which is proportional to a larger volume of labeled blood that is delivered to the tissue and second on the closer proximity of the labeling location and the imaging slab, which reduces T
1 decay. The geometry of pCASL used in the kidney, with labeling taking place approximately 15 cm upstream of the tissue, reverses the labeling proximity argument to the advantage of FAIR. For FAIR, spatial coverage of the non-selective labeling slab is limited by the transmit RF coil, resulting in a smaller volume of labeled blood in brain. However, for kidneys, body coil transmission is used and kidneys are positioned nearly at isocenter, so this is expected to result in a very minor difference in the labeled blood volume compared to pCASL. In addition, FAIR has an essentially flow velocity independent inversion efficiency [
38], which may be beneficial for application areas with a broad distribution of flow velocities and pulsatile flow.
As previously indicated, the poorer performance of pCASL might also be a result of reduced labeling efficiency caused by (a combination of) several factors. First, labeling efficiency is sensitive to B
0 inhomogeneities present at the pCASL labeling location (due to the proximity of air in the lungs), especially for this study performed at 3T. The labeling efficiency of the balanced pCASL implementation used is more sensitive to B
0 offsets than that of the unbalanced variant. However, separate B
0 shimming at the labeling location during acquisition was performed to mitigate this effect. Still, B
0 shimming at the labeling location may fail, thus resulting in compromised labeling efficiency. Second, the presence of B
1 inhomogeneities may have resulted in lower B
1 than expected at the labeling location. Finally, in contrast to FAIR, the inversion efficiency of pCASL is dependent on flow velocity [
38]. In the current study, default implemented labeling settings were used for pCASL, which have been optimized for brain application. Blood flow characteristics such as maximum blood flow velocity and pulsatility of the descending aorta are typically different compared with those in the brain feeding arteries [
39,
40]. At higher flow velocities, the adiabatic condition will be violated resulting in less optimal inversion of blood spins [
38]. Optimization of pCASL labeling parameters for renal application has been shown to improve robustness to off-resonance effects and aortic flow pulsatility [
41]. To improve repeatability of perfusion quantification and to detect failed measurements on a subject level, we think that it is recommended to measure the labeling efficiency at the labeling location, similar as has been proposed for the brain [
42,
43]. This will permit to judge the technical validity of the ASL measurement and to correct for the labeling efficiency in the quantification.
This study has limitations. First, for the pCASL labeling approach, two variants have been proposed, balanced and unbalanced [
18]. The balanced variant, which was used in this study, has been shown to be more sensitive to B
0 offsets. Switching to unbalanced pCASL may improve robustness to off-resonance effects, as has been demonstrated for brain ASL [
44] and preliminary data is available that shows this for the kidneys as well [
45]. Nevertheless, unbalanced pCASL has its own disadvantages such as vulnerability to subtraction artifacts caused by eddy currents due to the usage of different gradient waveforms in label and control condition [
18]. A more extensive direct comparison between both pCASL variants should be made in future studies to further optimize this labeling approach and improve robustness. Second, renal perfusion lacks a gold standard technique to enable validation of RBF values obtained with both labeling approaches. Alternative techniques for measurement of renal perfusion include the para-aminohippurate (PAH) clearance method [
7] and PET imaging (using
15O-labeled water) [
46]. PAH clearance measurement involves multiple blood and urine samples over a time course of several hours and provides only information on total perfusion of both kidneys combined. PET imaging involves the infusion of a radioactive tracer, but enables localized perfusion measurements. However, even in the absence of a gold standard, the clinical value of a perfusion measurement technique is determined by its capability to measure clinically relevant perfusion differences and changes. Finally, the acquisition order of FAIR and pCASL scans was not randomized. All FAIR scans with varying delay times were always performed before the pCASL scans, resulting in a difference of ~ 10 min between acquisition of both labeling approaches. This small time difference will probably not explain the difference found between the two methods.
In conclusion, in this comparative study between multi-delay FAIR and balanced pCASL for renal perfusion measurements at 3T in healthy middle-aged volunteers, FAIR showed favorable repeatability. To improve repeatability of perfusion quantification and assess the technical validity of an ASL measurement, addition of a labeling efficiency measurement is recommended, especially for the balanced pCASL variant used in this study.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.