Long-Delay Arterial Spin Labeling Provides More Accurate Cerebral Blood Flow Measurements in Moyamoya Patients: A Simultaneous Positron Emission Tomography/MRI Study.

Figure 1 shows perfusion images from a 26-year-old woman with bilateral Moyamoya disease. In areas of prolonged Tmax, standard-delay ASL showed areas of underestimation and overestimation relative to the PET scan. Improved visual consistency with PET rCBF maps was observed with multiple-delay and LLLD ASL scans. Multidelay ASL also revealed elevated ATT that corresponded to areas of long Tmax. Figure 2 shows a 31-year-old woman with severe bilateral Moyamoya disease and previous stroke in the left hemisphere. Although true perfusion deficits were identified on PET imaging in this patient, standard-delay ASL CBF maps had additional inaccuracies of reduced signal because of long ATTs. This signal loss was mitigated by multidelay ASL and was further improved visually by the LLLD ASL acquisition. In 2 such patients with regions of severely long delays (average Tmax > 8 seconds), all ASL sequences underestimated CBF in affected regions relative to PET. Across patients, the largest discrepancies between ASL and PET were observed in areas commonly affected by Moyamoya pathology, such as the anterior cerebral artery and middle cerebral artery territories. The disease status of each region was classified as mild (n=181), moderate (n=56), or severe (n=33) based on Tmax delay. Across the regions with mild delays, there was no difference in rCBF between any of the 3 ASL scans compared with PET ( Figure 3 ). Standard-delay ASL measured lower rCBF that was 73% of the PET values in regions with moderate delays (corrected P <0.001); and 66% of the PET values in regions of severe delays (corrected P <0.001). Multidelay ASL measured ATTs of 1180±371 ms in regions with mild delays, 1550±370 ms in regions with moderate delays, and 1770±340 ms in regions with severe delays ( P <0.02; 1-way ANOVA). Through correcting for ATT, multidelay acquisition improved rCBF estimation up to 83% of the PET values in areas of moderate delays but still underestimated rCBF in areas of severe delays (corrected P =0.02). No difference was detected between LLLD ASL and PET rCBF values, even in the presence of moderate or severe delays. The rCBF values from all 3 ASL acquisitions significantly correlated with PET in regions with moderate Tmax ( Figure 4A ). Standard-delay and multidelay ASL scans provided 16% and 8% underestimation bias, respectively, relative to PET while LLLD measurements provided minimal bias on the Bland–Altman plots. Similar observations were made in regions with severe Tmax delays ( Figure 4B ). Although all ASL scans correlated with PET, standard-delay and multidelay ASL scans provided 16% and 14% underestimation bias. LLLD scans had the strongest correlation with PET, as well as minimal bias (1%) relative to PET.

Fifteen patients with Moyamoya disease undergoing evaluation for bypass surgery were recruited for PET/MRI scans ( Table 1 ). Inclusion criteria included age of ≥15 years and a confirmed diagnosis of Moyamoya disease based on either 1 or more of the following diagnostic tests: cerebral angiogram, MRI/magnetic resonance angiography, and computed tomographic angiogram. Exclusion criteria included known allergies to sulfa drugs or contrast agent, impaired renal function that would preclude administration of contrast agent, or any contraindication to MRI. The study was performed in compliance with the Stanford University Institutional Review Board, and all patients provided written informed consent before the study. Participants were requested to refrain from caffeine intake for 6 hours before the study. All images were acquired on a simultaneous time-of-flight enabled 3.0 Tesla PET/MRI system (Signa, GE Healthcare, Waukesha, WI). For PET-based perfusion scans, each patient received manual, intravenous injection of [ 15 O]-water (490–960 MBq) through an antecubital vein. PET acquisition commenced with the tracer injection, and the first 2 minutes of PET counts after tracer administration were reconstructed to create a map of relative CBF. PET reconstruction was performed with a TOF-OSEM algorithm (time-of-flight ordered subset expectation maximization; 4 iterations and 28 subsets) on a 192×192 matrix, 30 cm field-of view, 2.78 mm slice thickness, and included correction for scatter, random counts, dead time, and point-spread function. During each PET scan, liver-accelerated volume acquisition-Flex 2-point Dixon MRI scans were acquired for attenuation correction with an atlas-based algorithm. 15 Three ASL MRI scans were performed simultaneously with the [ 15 O]-water PET scan ( Table 2 ). All ASL scans used pseudocontinuous labeling, had a 3-dimensional fast spin echo readout with stack-of-spirals readout trajectory, and were designed with similar scan time of ≈5 minutes. The labeling duration (LD) and PLD times of the standard-delay ASL scan were chosen based on recent consensus values for clinical patients, with LD/PLD of 1.5/2.0 seconds. 11 The multidelay ASL scan consisted of a slightly longer LD duration (2 seconds) and included 5 PLD times between 0.7 and 3.0 seconds, enabling measurement of and correction for ATT. Finally, the long-label long-delay (LLLD) ASL scan had a single, longer PLD of 4 seconds to help ensure arrival of tagged blood, with longer LD (3 seconds) to improve the signal-to-noise ratio. No vascular crushing was applied. All patients received 3-dimensional T 1 -weighted scans with an inversion-recovery prepared, fast spoiled gradient recalled sequence (repetition time=9.5 ms; echo time=3.8 ms; spatial resolution=0.93×0.93×1 mm 3 ). DSC MRI was performed using 0.1 mmol/kg of gadolinium contrast injected at 4 mL/s at the end of the study. DSC scan parameters included 60 time points, 19 slices, spatial resolution of 1.875×1.875×5.0 mm 3 , repetition time=1800 ms, and echo time=40 ms; and images were processed with RAPID software 16 to create Tmax maps. Perfusion maps from the standard-delay and LLLD ASL scans were generated from a 1-compartment model: (1) C B F = 6000 ⋅ e P L D / T 1 a 2 ε T 1 a ( 1 − e − L D / T 1 a ) ⋅ λ ⋅ M M 0 where ΔM represents the perfusion difference signal, and M 0 represents the fully relaxed, equilibrium tissue magnetization based on a saturation recovery image acquired with saturation time=2000 ms. T 1a =1650 ms is the assumed relaxation time for arterial blood, and λ=0.9 is the average partition coefficient in the brain. The constant ε=0.85×0.75 accounts for the combined effects of labeling efficiency and background suppression. For multidelay ASL scans, a 2-compartment ASL model was applied for the i th delay time PLD i : (2) C B F ( i ) = 6000 ⋅ e A T T / T 1 a 2 ε T 1 a ( e − max ( P L D i − A T T , 0 ) T 1 t − e − max ( P L D i + L D − A T T , 0 ) T 1 t ) ⋅ λ ⋅ M i M 0 where T 1t =1200 ms is the assumed relaxation for tissue. The ATT were measured using a weighted delay method and look-up table as described in Dai et al. 17 Based on the theoretical ASL signal at each PLD i , the CBF maps from each delay time were combined through a weighted average. Linear registration was performed to align the ASL and PET scans to each subject’s T 1 -weighted MRI, followed by nonlinear registration to the Montreal Neurological Institute brain template with Advanced Normalization Tools software. Regional CBF values from cortical tissue were determined in 10 perfusion territories (2 anterior, 6 middle, and 2 posterior) per hemisphere, corresponding to 2 slice locations of the Alberta Stroke Programme Early Computed Tomography Score (ASPECTS). 18 Relative CBF (rCBF) values were computed for each region through normalization to the mean CBF of the posterior territory, assuming that the posterior circulation in Moyamoya patients has sufficiently short ATTs to allow patient-specific corrections of CBF. 19 For 2 individuals with posterior disease involvement, only the unaffected posterior regions were used for normalization. The rCBF values were considered for regions mild Tmax delays (5 seconds) categorized by DSC. Differences in mean rCBF between each ASL sequence and PET were assessed by paired Wilcoxon signed-rank tests, followed by conservative Bonferroni correction for multiple comparisons. For regions with moderate and severe Tmax delays, scatter and Bland–Altman plots were generated to compare ASL and PET.

This study evaluated the accuracy of several ASL MRI acquisitions to measure cerebral perfusion in patients with Moyamoya disease compared with reference [ 15 O]-water PET scans acquired simultaneously on a hybrid system. Major findings include: Standard-delay ASL acquisitions with PLD of 2 seconds led to dramatic underestimation of rCBF in anterior and middle cerebral artery perfusion territories with moderate and long transit delays in patients; Multidelay ASL scans improved the correlation with PET through direct consideration of ATT and a 2-compartment model but still resulted in noticeable rCBF underestimation for long transit delays; and LLLD ASL acquisitions with LD of 3 seconds and PLD of 4 seconds, which was comparable to Tmax delays measured by DSC, led to the strongest CBF correlation with PET scans in cases of long transit delays. Changes in CBF and CVR are ubiquitous in clinical Moyamoya disease, requiring monitoring throughout the patient’s life. For example, accurate measurement of perfusion is critical to assess CVR, which is used clinically for initial surgical intervention and to assess the success of surgery. Although in some patient populations (eg, acute stroke), quantification may be less critical because of the focal nature of the disease; in Moyamoya disease, often both hemispheres are affected, CBF and CVR changes are less easily appreciated without quantification, and global changes are also relevant. Our results suggest that while standard-delay ASL scans are not suitable for this purpose, multidelay and LLLD scans (which are relatively untested) are appropriate imaging approaches to assess CBF in these patient cohorts. Because of its noninvasive nature and reliability compared with PET, baseline ASL scans with multiple PLDs and a long PLD can be incorporated into the MRI evaluation both before and after surgery to understand the effects of intervention. These ASL scans can also be repeated before and after a cerebrovascular challenge to assess CVR. The novel use of simultaneous PET/MRI here allowed us to image the same brain perfusion state by both modalities, removing physiological variability that can occur between different scan sessions. A recent review of studies that compared [ 15 O]-water PET and ASL in different sessions showed stronger correlations between PET and MRI if the scans were spaced closely in time, 20 suggesting that careful control of physiology is critical for CBF validation. Because PET and ASL scans were performed simultaneously, this study achieved rCBF correlations as high as R 2 =0.78 even in areas of severe disease, which are among the strongest correlations that have been reported. 20 Previous comparisons of standard ASL sequences (PLD of ≤2 seconds) with nuclear imaging by [ 123 I]-IMP SPECT 21 ([ 123 I]-iodoamphetamine single-photon emission computed tomography) and [ 15 O]-PET 22 demonstrated clear underestimation of CBF by ASL in areas of Moyamoya disease. These regions have long transit delays relative to the PLD, leading to artificially low ASL signal that is a known confounder in highly collateralized brain regions. 13 More recently, Hara et al 23 observed that ASL with a single PLD of 2.5 seconds, which is similar to our standard PLD, outperformed ASL with a shorter PLD of 1.5 seconds in areas of Moyamoya disease (Tmax > 6 seconds). However, only a modest relationship ( R 2 =0.25) between PET and ASL CBF was observed, 23 with persistent quantification bias by ASL compared with PET. ASL scans with PLD of 2 seconds have also shown local CBF hyper-intensities that likely correspond to collateral flow with long transit times, 24 and we observed similar hyperintense, intravascular signal in our standard-delay ASL scans ( Figure 1 ). ASL imaging with multiple PLDs and a 2-compartment model has been proposed to mitigate the effect of long ATTs. Previous work found that multidelay ASL more strongly correlated with computed tomographic perfusion than single-delay ASL (PLD of 2 seconds) in Moyamoya patients. 25 Two-compartment modeling of multidelay ASL 26 has shown 27% longer ATT’s in the affected side compared with the contralateral side, similar to the 31% ATT elevation we observed in regions with moderate Tmax. Tsujikawa et al 26 also demonstrated a strong correlation between PET and multidelay ASL CBF ( R 2 =0.67), which was similar to our correlations in Figure 4A . However, the longest PLD used in past multidelay ASL implementations was 3 seconds 25 , 26 or as short as 1.5 seconds. 27 Although ATT measured by ASL and Tmax measured by DSC are not necessarily the same, the PLD range that we used may be suboptimal because the PLDs are shorter than the Tmax values in many patients. 12 Our findings suggest that an optimal ASL protocol with longer PLD is necessary for accurate cortical CBF assessment in Moyamoya patients. Based on catheter angiography, these patients have arterial circulation times as long as 3.25 seconds, 28 motivating the use of longer PLDs to allow tagged blood to reach the capillaries. In this study, LLLD scans with PLD of 4 seconds showed the strongest correlation and the least bias of rCBF values compared with PET in areas of severe transit delays. The main disadvantage of LLLD scans, however, is the decay of ASL signal during the longer PLD before image acquisition, leading to challenges with signal-to-noise ratio. For instance, using a PLD of 4 seconds and assuming a T 1a of 1.65 seconds, ≈90% of labeled signal will be lost because of T1 recovery by the time of imaging. To achieve adequate signal-to-noise ratio within 5 minutes, we increased the LD and reduced the in-plane spatial resolution of the images. Even so, the LLLD ASL images had lower signal-to-noise ratio than the other ASL scans, as evidenced by higher background noise in Figures 1 and 2 . Although this study focused on cortical tissues, CBF values in the deep gray matter were noticeably reduced in LLLD ASL scans. This observation may reflect earlier blood arrival combined with faster T 1 relaxation in the tissue compared with the blood, which leads to underestimation of both ATT and CBF in early-arrival regions at longer PLDs. 17 It may also help explain why LLLD ASL did not perform as well as multidelay ASL in cases of moderate Tmax. The large range of expected ATT variations across brain regions and because of disease make it challenging to identify a single, optimal delay time for ASL in patients. Thus, comprehensive CBF assessment in Moyamoya disease may require multiple PLDs with an extended PLD range (between 0.3 and 4 seconds or longer) and use of time-efficient encoding strategies 29 to achieve clinical scan times. As a limitation, this work did not directly measure rCBF by DSC, which is the most commonly used MRI perfusion method in Moyamoya disease. Goetti et al 13 performed 2 studies to compare single-delay ASL to DSC and to [ 15 O]-water PET, 22 respectively, on a similar cohort of young Moyamoya patients. For qualitative perfusion assessment, single-delay ASL showed similar relationships to DSC (ρ=0.77) and to PET (ρ=0.77). More recently, Hara et al 23 studied ASL, PET, and DSC in the same cohort and showed that differences between ASL scans acquired at different PLD times were predictive of DSC timing parameters, such as Tmax. Unlike DSC, however, ASL uses the same diffusible tracer (water) as PET, and we, therefore, expect optimized, multidelay ASL scans to outperform DSC when compared with PET in future studies. Regional analysis in this study used regions of interest (ROIs) based on the ASPECTS scoring system because it is well established for evaluation of cerebral ischemia and delineates cortical perfusion territories that are relevant to assessment of cerebrovascular disease. In addition, one important clinical goal of CBF imaging in Moyamoya disease at our institution is the measurement of CVR, and our ROIs with a mean volume of 45 cm 3 are typically the smallest size in which such assessments can be reliably performed. However, this size of ROI may not be appropriate for regional comparisons of focal CBF lesions. Alternative analyses could use smaller ROIs or even voxel-wise analysis to improve the scale of the correlations. In this work, our Moyamoya patients had relatively preserved baseline perfusion because of Moyamoya collateral networks, and regional deficits typically occurred in areas of previous infarct. These more local PET hypointensities commonly corresponded to low ASL signal within that ROI, driving the CBF dynamic range of Figure 4 ; we would anticipate even stronger correlations with finer ROI scales. The correlations presented here, despite the large ROI volume, are nonetheless relevant to other cerebrovascular conditions, including ischemic stroke, which have large lesion volumes (median of ≈102 cm 3 across gray matter). 30 Another limitation of this study is the lack of absolute CBF quantification, which would allow serial follow-up of Moyamoya disease and assessment of cerebrovascular reactivity. To minimize patient discomfort, we did not collect arterial blood samples, which prevents kinetic modeling of absolute CBF from the PET data. Instead, we created rCBF maps from the first 2 minutes of PET counts after the [ 15 O]-water injection; similar static PET images have been shown to accurately represent CBF if the first 40 seconds after the injection are included. 31 Future CBF comparisons will focus on quantitative evaluations between ASL and PET through use of image-based input functions for CBF quantification. 32 In addition, because atlas-based magnetic resonance attenuation correction slightly underestimates PET tracer uptake, 15 alternative attenuation correction methods based on zero-echo time MRI may improve the accuracy of absolute PET CBF values. 33 Once these quantification issues are addressed, simultaneous PET/MRI provides an ideal strategy to validate advanced ASL techniques, such as velocity-selective ASL, 34 that hold promise in challenging cerebrovascular cases. Furthermore, oxygen extraction fraction and the cerebral metabolic rate of oxygen are also critical biomarkers to characterize hemodynamic impairment in Moyamoya disease 6 and evaluate its treatment. As novel MRI methods to assess oxygen extraction fraction and cerebral metabolic rate of oxygen gain traction, 35 the PET/MRI hybrid modality offers careful control of cerebral physiology to validate these new measurements against the triple [ 15 O]-PET reference standard. 36

We compared CBF measurements with multiple ASL techniques against the PET reference standard in patients with Moyamoya disease under simultaneous conditions. LLLD ASL showed the least bias in rCBF values compared with PET, enabling reliable CBF assessment even in areas of severe Tmax delay. We demonstrated that extremely long PLD’s are required for ASL studies in cerebrovascular disease, and any multiple-delay time acquisitions should include a wide range of post-label delays (eg, between 0.3 and 4 seconds).