Quantitative Cerebral Perfusion Imaging in Children and Young Adults with Moyamoya Disease: Comparison of Arterial Spin-Labeling–MRI and H2[15O]-PET

BACKGROUND AND PURPOSE: Cerebral perfusion assessment is important in the preoperative evaluation and postoperative follow-up of patients with Moyamoya disease. The objective of this study was to evaluate the correlation of quantitative CBF measurements performed with arterial spin-labeling–MR imaging and H2[15O]-PET in children and young adults with Moyamoya disease. MATERIALS AND METHODS: Thirteen children and young adults (8 female patients; age, 9.7 ± 7.1 years; range, 1–23 years) with Moyamoya disease underwent cerebral perfusion imaging with H2[15O]-PET (Discovery STE PET/CT, 3D Fourier rebinning filtered back-projection, 128 × 128 × 47 matrix, 2.34 × 2.34 × 3.27 mm3 voxel spacing) and arterial spin-labeling (3T scanner, 3D pulsed continuous arterial spin-labeling sequence, 32 axial sections, TR = 5.5 seconds, TE = 25 ms, FOV = 24 cm, 128 × 128 matrix, 1.875 × 1.875 × 5 mm3 voxel spacing) within less than 2 weeks of each other. Perfusion of left and right anterior cerebral artery, MCA, and posterior cerebral artery territories was qualitatively assessed for arterial spin-labeling–MR imaging and H2[15O]-PET by 2 independent readers by use of a 3-point-Likert scale. Quantitative correlation of relative CBF with cerebellar normalization between arterial spin-labeling–MR imaging and H2[15O]-PET was evaluated in a volume-based approach for each vascular territory after 3D image coregistration. RESULTS: Interreader agreement was good (κ = 0.67–0.69), and strong and significant correlations were found between arterial spin-labeling–MR imaging and H2[15O]-PET for both qualitative perfusion scoring (ρ = 0.77; P < .001) and quantitative perfusion assessment of relative CBF with cerebellar normalization (r = 0.67, P < .001). CONCLUSIONS: In children and young adults with Moyamoya disease, quantitative evaluation of CBF is possible with the use of arterial spin-labeling–MR imaging without ionizing radiation or contrast injection with a good correlation to H2[15O]-PET after cerebellar normalization.

C hildren and young adults with Moyamoya disease have progressive bilateral stenosis of the supraclinoid ICA, anterior cerebral artery, MCA, and to a much lesser extent, the posterior cerebral artery, with formation of netlike collateral vessel networks, termed "Moyamoya" vessels. 1 "Moyamoya" is the Japanese term for "puff of smoke" and describes the characteristic angiographic appearance of the collateral vessel networks typically formed at the skull base and around the basal ganglia in Moyamoya disease. The condition is rare; however, it is implicated in approximately 6% of childhood strokes. 2 The main goal of treatment of children with Moyamoya is the prevention of future strokes by means of surgical revascularization.
To determine the optimal targets for revascularization procedures as well as for postoperative follow-up, CBF studies with the use of Xe-CT, SPECT, or PET have been shown to be of high clinical value. [3][4][5][6][7] However, these methods are associated with ionizing radiation, which should be minimized or avoided whenever possible and especially in young patients. Recently, interest in MR imaging techniques for CBF assessment has grown, and initial evaluation of these techniques has been undertaken in patients with Moyamoya disease. MR perfusion imaging by use of dynamic susceptibility contrast imaging, on the basis of the decrease of T2/T2* times of tissue during the first pass of a gadoliniumcontaining contrast agent through the cerebral capillary bed, has been shown to correlate well with the reference standard of H 2 [ 15 O]-PET 8 in the assessment of cerebrovascular occlusive disease. Arterial spin-labeling (ASL) is an alternative MR imaging technique for CBF imaging that is based on the T1 magnetization state of electromagnetically tagged (ie, spin-labeled), freely diffusible arterial water. Blood flow measurements with ASL have been shown to correlate well with those from DSC imaging in children with Moyamoya disease. 9 As a method requiring neither ionizing radiation nor exogenous contrast material injection, ASL could represent an ideal technique for the assessment of cerebral perfusion in children. The purpose of the present study was thus to assess the comparability of ASL and the reference standard of H 2 [ 15 O]-PET imaging in children and young adults with Moyamoya disease.

Patients
Children and young adults with angiographically proven Moyamoya disease who underwent both MR imaging with ASL imaging as well as H 2 [ 15 O]-PET for preoperative assessment or postoperative follow-up after surgical revascularization as a routine work-up at our institutions between January 2011 and December 2012 were included in this retrospective study. The informed consent requirement was waived by the institutional review board because of the retrospective nature of the study.
Exclusion criteria were patient age Ͼ25 years (n ϭ 0), H 2 [ 15 O]-PET not performed within 2 weeks of ASL (n ϭ 11), interventions or cerebrovascular incidents (ie, transient ischemic attacks or strokes) between the 2 examinations (n ϭ 0), postoperative status Ͻ6 months (n ϭ 0), contraindications to sedation in children Ͻ6 years old as evaluated by a board-certified anesthesiologist (n ϭ 0), contraindications to contrast-enhanced MR imaging, such as noncompatible metallic implants or known previous allergic reactions to gadolinium-based contrast material (n ϭ 0), and nondiagnostic image quality of either ASL or H 2 [ 15 O]-PET studies (n ϭ 1: nondiagnostic image quality of ASL caused by motion artifacts).
Thirteen children and young adults (8 female patients and 5 male patients; age, 9.7 Ϯ 7.1 years; range, 1-23 years) met all criteria and were thus included in the study. Suzuki stages of the angiographically confirmed Moyamoya disease in these patients were II (n ϭ 4), III (n ϭ 4), IV (n ϭ 3), and V (n ϭ 2). Anterior watershed infarctions were present in 4 patients unilaterally and in 1 patient bilaterally. The mean time interval between the 2 examinations was 2 Ϯ 3 days (range, 1-13 days). H 2 [ 15 O]-PET was performed first in 3 patients, whereas ASL was performed first in 10 patients. Because of the retrospective nature of the study, the order of acquisition was not randomized.

MR Imaging Protocol and Image Reconstruction
MR imaging was performed with the use of a 3T scanner (HD.xt TwinSpeed; GE Healthcare, Milwaukee, Wisconsin) equipped with an 8-channel receive-only head coil for signal reception in all patients. Patients under the age of 6 years (n ϭ 6), were sedated with propofol (Propofol-Lipuro; Braun Medical, Melsungen, Germany) and monitored by a board-certified anesthesiologist during the examination. A standardized MR imaging protocol for patients with Moyamoya disease consisting of multiplanar T2WI, T1WI, FLAIR, DWI, 3D-TOF, ASL, and contrast-enhanced DSC as well as multiplanar postcontrast T1WI was performed in all patients without complications.
ASL images were acquired as 32 axial sections by use of a background-suppressed, pulsed continuous arterial spin-labeling sequence, with a 3D stack of spiral fast spin-echo readout. 10 The labeling duration was 1.5 seconds, and a postlabeling delay of 1.5 seconds was used to reduce errors from transit time effects. 11 Further imaging parameters of the ASL sequence were TR, 5500 ms; TE, 25 ms; FOV, 24 cm; matrix, 128 ϫ 128; section thickness, 5 mm; and number of excitations, 3. The reconstructed voxel spacing was 1.875 ϫ 1.875 ϫ 5 mm 3 . The total scan time was 5 minutes, 34 seconds.
Quantitative CBF maps were calculated from the ASL data by use of the standard on-line perfusion image reconstruction provided by the scanner manufacturer. This reconstruction calculates the perfusion images according to the model described by Wang et al, 12 with an additional factor included to correct for incomplete recovery of the magnetization in the reference image caused by a saturation pulse applied 2 seconds before imaging. 13 No further preprocessing was performed. CBF was calculated according to the following equation. 11,13 f ϭ In this equation, f is blood flow (in mL/min per 100 mg), S ctrl -S lbl is the difference image (control-label), and S ref is a proton-attenuation-weighted reference image; is the blood-brain partition coefficient (0.9), ␣ is the inversion efficiency, T 1b is the T1 of blood (1600 ms), T 1g is the T1 of gray matter (1200 ms), w is the postlabeling delay (1.5 seconds), and is the labeling duration (1.5 seconds). 9,12,13 To investigate the effects of variations in blood T1 across the patient group, the blood T1 was calculated for n ϭ 11 patients from venous hematocrit levels according to the equation T1 ϭ 1/(0.83 Hct ϩ 0.28). ASL perfusion values were calculated by use of an assumed blood T1 of 1600 ms and by use of the individual blood T1 for the patients in whom hematocrit data were available.

PET Protocol and Image Reconstruction
The PET data were acquired in 3D mode on a full-ring PET/CT scanner (Discovery STE; GE Healthcare) and corrected for attenuation, scatter, random photons, and dead time by use of the corresponding low-dose CT (120 kV/80 mA). The scanner's axial FOV covered 15.3 cm, and transaxial images were reconstructed by use of a 3D Fourier rebinning filtered back-projection algorithm resulting in a 128 ϫ 128 ϫ 47 matrix with 2.34 ϫ 2.34 ϫ 3.27 mm 3 voxel spacing; 400 -800 MBq H 2 [ 15 O] was administered intravenously by use of an automatic injection device that delivered the predefined tracer dose over a period of 20 seconds. Patients under the age of 6 years (n ϭ 6, the same as in ASL) were sedated with the use of Propofol-Lipuro and monitored by a boardcertified anesthesiologist during the examination. The emission data were acquired as a series of eighteen, 10-second frames.
Parametric images of CBF were generated by means of a reported method 14 in which a standardized arterial input function and image scaling on the basis of the washout rate (k2) of H 2 [ 15 O] are used to derive CBF. This technique, which is based on the Alpert method, exploits the fact that k2 is related to the shape and not the scale of the arterial input function and proportional to relative CBF (rCBF): (k2 ϭ K1/p, where p is the partition coefficient). 15

MR Imaging and PET Image Analysis
Qualitative Analysis. ASL and H 2 [ 15 O]-PET CBF maps were assessed separately by 2 blinded and independent radiologists with 5 years and 7 years of experience. CBF was visually categorized on a 3-point Likert scale (1: normal, 2: reduced, 3: severely reduced) in the color-coded CBF maps (Fig 1) of each technique for anterior cerebral artery, MCA, and posterior cerebral artery territories on each side by comparison to cerebellar perfusion, which was assumed to be normal.
Quantitative Image Analysis. Semi-automated image analysis was performed in PMOD (version 3.4; PMOD Technologies, Zurich, Switzerland). With the use of a PET perfusion template image (SPM5, average H 2 [ 15 O] image from 12 subjects), ASL and PET perfusion maps were normalized to the stereotaxic spatial array of the Montreal Neurological Institute brain template by applying a mutual information-based registration including 12parameter affine and elastic transformations and transitional image smoothing with a 6-mm full width at half maximum Gaussian kernel. Successful normalization was verified visually with the aid of image-outlining contours. After normalization, a volume-ofinterest template defining the vascular territories (left/right, anterior cerebral artery, MCA, and posterior cerebral artery) and cerebellum was used to extract average perfusion by territory.
This volume-of-interest template was generated by attributing the predefined brain regions outlined by the Hammer N30R83 atlas 16 to the respective vascular territories and was identical in size and shape for both H 2 [ 15 O]-PET and ASL data (Fig 2). Relative CBF values (rCBF), that is, cerebellar normalized values, were calculated by dividing the CBF value of the respective supratentorial territory by cerebellar CBF.
The correlation between H 2 [ 15 O]-PET and ASL was assessed by use of Spearman rank correlation coefficient for qualitative perfusion scores and by Pearson correlation coefficient for quantitative CBF values and quantitative relative CBF values after cerebellar normalization. Bland-Altman analysis was used to compare mean differences in CBF and rCBF between H 2 [ 15 O]-PET and ASL.
Paired t tests were used to compare CBF and rCBF values between H 2 [ 15 O]-PET and ASL, and unpaired t tests were used to compare CBF and rCBF between territories with normal (qualitative perfusion score of 1) and reduced perfusion (score of 2 or 3) for both H 2 [ 15 O]-PET and ASL. Differences in CBF and rCBF between territories grouped by perfusion scores were assessed by means of 1-way ANOVA with Tukey honestly significant differences post hoc analysis. All statistical tests were performed with the use of the statistical package R 17 (http:// www.r-project.org/) and the R frontend RKWard Version 0.6.1 (http://rkward.sourceforge.net/). Statistical significance was inferred at P Ͻ .05.

Qualitative Image Analysis
Interreader agreement for qualitative perfusion assessment was good for both H 2 [ 15 O]-PET ( ϭ 0.67) and ASL ( ϭ 0.69). A strong and significant correlation between perfusion scores in ASL and H 2 [ 15 O]-PET was found ( ϭ 0.77; P Ͻ .001). No posterior circulation involvement that could have influenced cerebellar normalization was found in our patient cohort.
Reduced perfusion in the visual qualitative assessment was found in 33 of 78 (42%) territories in 11

Quantitative Image Analysis
Quantitative image analysis revealed a strong and significant correlation between H 2 [ 15 O]-PET and ASL (r ϭ 0.67, P Ͻ .001) after cerebellar normalization (rCBF), with a regression slope of 1.0061 and intercept of Ϫ0.0057 (Fig 3). For absolute CBF values, the correlation was much weaker (r ϭ 0.26, P ϭ .015) using a literature value (1600 ms) for the blood T1 in the ASL perfusion quantification. The correlation was not improved with the use of an individual blood T1 on the basis of the hematocrit (r ϭ 0.14, P ϭ .025).   Note:-ACA indicates anterior cerebral artery; PCA, posterior cerebral artery; rCBF, relative CBF (values are given as ratios to cerebellar CBF). a CBF values are given as mL/min per 100 g. P ϭ .016) and ASL (CBF: P ϭ .010, rCBF: P ϭ .004). Significant differences between territories with reduced perfusion (score 2) and severely reduced perfusion (score 3) were found for ASL (P ϭ .03) but not for H 2 [ 15 O]-PET (P ϭ .17).

DISCUSSION
In the present study, we compared MR imaging-based and PETbased methods for the evaluation of cerebral perfusion in children and young adults with Moyamoya disease, through the use of qualitative and quantitative approaches. Our results demonstrate that CBF assessment by use of ASL-MR imaging with cerebellar normalization correlates well with the reference standard H 2 [ 15 O]-PET method. This suggests that ASL-MR imaging could be a valuable alternative to H 2 [ 15 O]-PET, while avoiding the use of ionizing radiation and injection of an exogenous contrast agent. However, absolute values of CBF may differ between the modalities.
The most widely used MR imaging technique is gadoliniumenhanced DSC, which is based on the decrease of T2/T2* signal intensity of cerebral tissue during the first pass of gadolinium through the capillary bed. It has been shown to reveal alterations in perfusion even in apparently (structurally) normal cerebral parenchyma in patients with Moyamoya disease and may be used in the targeting of revascularization approaches. 18 8 Arterial spin-labeling is a well-established method for the quantitative assessment of cerebral perfusion. It does not require the injection of an exogenous contrast agent but instead uses the water in arterial blood as an endogenous tracer through electromagnetic tagging with radiofrequency pulses. For the calculation of blood flow, the amount of labeled blood arriving in the cerebral tissue after a predefined transit time is measured by subtraction from a previously acquired control image. 19 A high correlation between ASL and DSC has been reported, with correlation coefficients between ASL and DSC of 0.62-0.79 in patients with acute ischemic stroke. 20 In a recent study with the use of DSC as the standard of reference in children with Moyamoya disease, it could be demonstrated that ASL-based measurements of CBF showed a similarly strong and significant correlation with DSC-based perfusion (r ϭ 0.79, P Ͻ .001), enabling the detection of impaired perfusion per cerebral vascular territory with a high diagnostic accuracy. 9 In a direct comparison of ASL and H 2 [ 15 O]-PET, one study found similar cortical perfusion values with no significant  Bland-Altman plots of CBF (mL/100 g per minute) and relative CBF after cerebellar normalization (rCBF).