In Vivo T1 of Blood Measurements in Children with Sickle Cell Disease Improve Cerebral Blood Flow Quantification from Arterial Spin-Labeling MRI

BACKGROUND AND PURPOSE: Children with sickle cell disease have low hematocrit and elevated CBF, the latter of which can be assessed with arterial spin-labeling MR imaging. Quantitative CBF values are obtained by using an estimation of the longitudinal relaxation time of blood (T1blood). Because T1blood depends on hematocrit in healthy individuals, we investigated the importance of measuring T1blood in vivo with MR imaging versus calculating it from hematocrit or assuming an adult fixed value recommended by the literature, hypothesizing that measured T1blood would be the most suited for CBF quantification in children with sickle cell disease. MATERIALS AND METHODS: Four approaches for T1blood estimation were investigated in 39 patients with sickle cell disease and subsequently used in the CBF quantification from arterial spin-labeling MR imaging. First, we used 1650 ms as recommended by the literature (T1blood-fixed); second, T1blood calculated from hematocrit measured in patients (T1blood-hematocrit); third, T1blood measured in vivo with a Look-Locker MR imaging sequence (T1blood-measured); and finally, a mean value from T1blood measured in this study in children with sickle cell disease (T1blood–sickle cell disease). Quantitative flow measurements acquired with phase-contrast MR imaging served as reference values for CBF. RESULTS: T1blood-measured (1818 ± 107 ms) was higher than the literature recommended value of 1650 ms, was significantly lower than T1blood-hematocrit (2058 ± 123 ms, P < .001), and, most interesting, did not correlate with hematocrit measurements. Use of either T1blood-measured or T1blood–sickle cell disease provided the best agreement on CBF between arterial-spin labeling and phase-contrast MR imaging reference values. CONCLUSIONS: This work advocates the use of patient-specific measured T1blood or a standardized value (1818 ms) in the quantification of CBF from arterial spin-labeling in children with SCD.

S ickle cell disease (SCD) is associated with a considerable risk of stroke, 1 which is reduced by blood transfusion therapy 2 and identified by screening blood flow velocities in intracranial arteries with transcranial Doppler. 3 Additionally, microvascular tissue perfusion, or CBF, is also increased in patients with SCD 4,5 ; which is related to low hematocrit (Hct). 6,7 CBF measurements are in-strumental in understanding the pathophysiology of impaired perfusion in the occurrence of silent cerebral infarcts in SCD. 4,8,9 Noninvasive CBF measurements can be performed with arterial spin-labeling (ASL) and a quantification model to calculate physiological CBF values. The wide range of CBF values reported in the literature in SCD 1,4,9 emphasizes the need for either more accurate estimates or direct measurements of the often-assumed parameters required for CBF quantification models.
The longitudinal relaxation time of the blood (T1 blood ) parameter accounts for the decay of the ASL signal with time, and inaccurate estimates of T1 blood could result in over-or underestimation of CBF. [10][11][12] For healthy adults, with a stable Hct, a fixed T1 blood value of 1650 ms is recommended for CBF quantification from pseudocontinuous ASL (pCASL) at 3T. 13,14 T1 blood is inversely correlated with Hct, 10,13,[15][16][17][18][19] and a linear relationship has been proposed in the literature permitting the calculation of T1 blood from measured Hct values. 12,13,16 While Hct ranges from 38% to 45% in healthy children, 20 it is as low as 18%-30% in children with SCD. 21 Hence, if measured Hct values are available, T1 blood can be derived accordingly. However, recent studies suggest that T1 blood may additionally differ in children with SCD. 12,22,23 Owing to recent developments in MR imaging, direct measurements of the inversion recovery of T1 blood are now possible by combining a global inversion pulse and a subsequent sectionselective Look-Locker readout in the sagittal sinus. 16,17 Patientspecific, in vivo T1 blood measurements are noninvasive, robust, and fast, making them preferable to calculating T1 blood from blood samples. Our first hypothesis was that in vivo-measured T1 blood would be higher in children with SCD than the adult reference value of 1650 ms due to anemia. We also considered that conformational changes inherent to sickle red blood cells may produce additional unforeseen changes in T1 blood . 12 We investigated the importance of measuring patient-specific differences in T1 blood for the accuracy of ASL quantification in patients with SCD. We hypothesized that patient-specific T1 blood values acquired in vivo would improve CBF quantification in SCD compared with CBF quantification with T1 blood calculated from Hct or T1 blood -fixed at 1650 ms.
The aim of this study was to determine which of the following 4 T1 blood derivatives would provide the best CBF quantification compared with quantitative reference CBF values measured with 2D phase-contrast MR imaging (PC-MRI): 1) literature-recommended adult T1 blood of 1650 ms, 14 2) T1 blood calculated from Hct, 3) in vivo-measured T1 blood , or 4) a fixed average SCD value from the mean T1 blood measured in vivo in this study.

MATERIALS AND METHODS
Experiments were performed according to principles of the Declaration of Helsinki, and the study was approved by the local institutional review board at the Academic Medical Center, Amsterdam, the Netherlands.

Patients
Eligible children were approached prospectively from 2 outpatient clinics as described previously. 24 Informed consent was obtained from parents or guardians and children older than 12 years of age. Inclusion criteria were HbSS or HbS␤ 0 genotypes and 8 -17 years of age. Exclusion criteria were a history of stroke, stenosis of the intracranial arteries and velocity of Ͼ155 cm/s on transcranial Doppler imaging, current chronic blood transfusion therapy, bone marrow transplant, MR imaging contraindications, and major concomitant health problems. Patients were in a steady-state of SCD, without evidence of infection or sickle cell crisis up to 1 month before participation.

Hematocrit
Venous blood samples were drawn from an antecubital vein on the day of the MR imaging assessment and processed according to standard procedures in the hospital laboratory. Hct values were used to calculate T1 blood -Hct values.

MR Imaging Acquisition
Thirty-two children underwent 3T imaging on an Intera scanner (Philips Healthcare, Best, the Netherlands) with an 8-channel head coil, and due to a scanner upgrade, the remaining 8 children were scanned at 3T on an Ingenia (Philips Healthcare) with a 15-channel head coil. The protocol included 3D-TOF MRA, 2D T2-weighted, T1 blood , 2D pseudocontinuous ASL, and 2D phase-contrast sequences.
The T1 blood acquisition section was planned perpendicular to the posterior sagittal sinus 16 and comprised a multi time-point inversion recovery experiment. This technique uses a global inversion pulse followed by a series of 95°section-selective readout pulses, which are intended to saturate the tissue surrounding the sinus. Assuming complete replenishment of inverted blood between 2 consecutive pulses, a high contrast is achieved between tissue and blood, allowing the detection of the inversion recovery of blood. A nonselective adiabatic 180°inversion pulse (hyperbolic secant pulse, B1 value/duration of the pulse ϭ 13.5 mT/13 ms) preceded a single section Look-Locker EPI readout (flip angle, 95°; voxel size, 1.5 ϫ 1.5 mm; matrix, 240 ϫ 240 mm; section thickness, 2 mm; TE/TR, 15/10,000 ms; TI 1 , 200 ms; ⌬TI, 150 ms; 60 readouts; 6 signal averages; scan duration, 1 minute 20 seconds).
where M models the T1 recovery from the data, nTI is the readout number, abs denotes the absolute values, M 0 is the net magnetization, "Offset" accounts for imperfect inversion, TI 1 is 200 ms, and ⌬TI is the sampling interval (150 ms). The sum of squared errors of the final (optimal) iteration after solving the Nealder-Mead function indicated how well the data fitted the model and served as a quality check.
Cerebral Blood Flow. Raw pCASL data were processed as described previously 25 by using a processing pipeline for the registration and quantification of the data. A 2-compartment quantification model was used, as published in detail previously 9,26 (except that the equilibrium magnetization of arterial blood was derived from the M 0 of CSF multiplied by the blood-water partition coefficient, 27 and labeling efficiency was 0.7). The T1 blood parameter was adjusted for each CBF quantification as follows: first, adult fixed T1 blood of 1650 ms taken from literature 13 ; second, patient-specific Hct-calculated T1 blood values 16 ; third, patient-specific in vivo-measured T1 blood values; and finally an average T1 blood value obtained from the mean of in vivo T1 blood measurements in our patients with SCD. T1 blood -Hct was calculated per patient according to the relationship proposed by Varela et al 16 derived from venous blood in neonates: 2) T1 blood ϭ 1 0.5 ϫ Hct ϩ 0.37 .
PC-MRI. The internal carotid and vertebral arteries were segmented manually from phase difference images by using ITK-SNAP (http://www.itksnap.org) to obtain total flow (milliliters per minute). Total flow was then divided by brain mass (gram), which was calculated from the product of the volume (estimated from segmented anatomic images in SPM8; http://www.fil.ion.ucl.ac.uk/spm/software/spm12) and an assumed brain density of 1.05 g/L, 28 to obtain PC-MRI CBF in milliliters/100 g/min, 29 which served as the reference value for CBF. 22,29

Statistical Analysis
A Pearson correlation was performed between T1 blood -measured and Hct. Repeated-measures ANOVA was used to test the statistical significance of the differences among the 5 CBF quantification methods: 1) CBF (T1 blood -fixed at 1650 ms), 2) CBF (T1 blood -Hct), 3) CBF (T1 blood -measured in vivo, 4) CBF (T1 blood -SCD fixed at the average measured value), and 5) PC-MRI reference CBF. Paired t tests were used to test the statistical significance of individual group differences post hoc. Agreement between PC-MRI and the 4 ASL methods was investigated with linear regression and Bland-Altman analyses in Matlab (MathWorks, Natick, Massachusetts). Linear regression analysis was performed to show agreement between PC-MRI and the 4 CBF quantification methods from ASL. Bland-Altman analysis was performed to indicate the bias corresponding to over-or underestimation of the ASL CBF method compared with the PC-MRI method. The limits of agreement (dotted lines) indicate the 95% confidence intervals. Table 1. One patient's T1 blood scan was discarded due to poor image quality, so the mean CBF values from pCASL are based on 39 datasets. For PC-MRI, only 33 datasets were of sufficient quality to quantify reference CBF.

Measured T1 blood
The mean Hct was 23% Ϯ 3% for 39 children. The mean T1 bloodmeasured value was 1818 Ϯ 107 ms, which was significantly lower than mean T1 blood -Hct values (2045 Ϯ 69 ms; paired t test, P Ͻ .001) but higher compared to the fixed adult value of 1650 ms. T1 bloodmeasured was not significantly different between scanners (t test, P ϭ .94). Figure 1A shows a representative inversion recovery curve from 1 patient as a function of the sum of least-squares fit. The sum of squared errors from fitting the T1 blood -measured values to the model is shown in On-line Fig 1. T1 blood -measured values did not correlate   FIG 1. A, Representative inversion recovery of the venous T1 blood signal acquired in the sagittal sinus in a child with sickle cell disease. B, In vivo-measured T1 blood values are significantly lower than Hct-derived T1 blood values. T1 blood -measured does not correlate with patient hematocrit (mean Hct, 23% Ϯ 3%) (Pearson r ϭ 0.02, P ϭ .89; n ϭ 39). with Hct values measured from blood samples (r ϭ 0.02, P ϭ .89; Fig  1B) or with age (r ϭ 0.03, P ϭ .85) and did not differ significantly between males and females (t test, P ϭ .37).

Cerebral Blood Flow
Four CBF quantification methods were compared with PC-MRI CBF, the results of which are summarized in Table 2. Linear regression analyses between PC-MRI and pCASL CBF are shown in the left panel of Fig 2 and reveal slopes significantly different from zero for all CBF quantifications except for the T1 blood -Hct CBF quantification. The Bland-Altman plots in the right panel of Fig 2 show the bias and limits of agreement for the mean and the difference between the measurements. T1 blood -fixed overestimated CBF and T1 blood -Hct underestimated CBF, while the individual in vivo T1 blood -measured values and mean T1 blood -SCD value provided the best agreement with PC-MRI values, both on an absolute level, revealed by no significant difference between PC-MRI and CBF in the repeated-measures ANOVA analysis ( Table 2), but also on a one-to-one basis, as demonstrated in the linear regression plots (Fig 2). A representative example of CBF maps quantified with T1 bloodmeasured from 2 patients is shown in

DISCUSSION
We demonstrate that in vivo-measured venous T1 blood values in children with SCD were higher than the literature-recommended 1650 ms, were not significantly correlated with measured Hct, and were lower than the Hct-derived values for T1 blood . CBF quantified with in vivo-measured T1 blood provided better agreement with PC-MRI reference measurements than CBF quantified with fixed adult T1 blood and Hct-derived T1 blood .

T1 blood and Hematocrit
Previous literature suggests that healthy children 6 -18 years of age (assuming a stable Hct of 40%-45%) have T1 blood values between 1680 and 1880 ms. 18 In this study, in patients with a much lower Hct than healthy children, we measured T1 blood values closer to the upper range of the literature-reported T1 blood values. 18 Yet, our T1 blood values were lower than expected, considering the low Hct values obtained from our patients' blood samples. It is unlikely that we underestimated T1 blood due to sequence-related limitations because the Look-Locker T1 technique has previously provided robust results in the same ROI. 16,17,27 Reports of T1 blood values ranging from 1500 to 2100 ms follow a linear relationship with Hct between 23% and 50%. 13,16,18 It is possible that we did not have sufficient precision to detect this inverse relationship in our dataset or that the range of Hct values  A and B); T1 blood calculated from hematocrit (CBF T1 blood -Hct), T1 ϭ 0.5*Hctϩ0.37 (C and D) 16 ; in vivo-measured T1 blood (CBF T1 blood -measured) (E and F); and a fixed SCD value obtained from the mean of the in vivo-measured T1 blood (CBF T1 blood -SCD) (G and H). The left panel shows linear regressions (solid line), and the right panel shows the mean on the x-axis versus the difference on the y-axis between pCASL and PC-MRI CBF with limits of agreement (dotted lines above and below) (n ϭ 33). was too narrow in our patients (17%-32%). Abnormalities in SCD blood, other than low Hct, may account for the incongruity between T1 blood and Hct measured here. While we did not measure blood rheology, abnormalities such as decreased red blood cell deformability, increased aggregation, and increased viscosity have been demonstrated consistently. 21,[30][31][32][33][34] Furthermore, red blood cells in SCD exhibit different membrane properties and viscosity, which may have reduced T1 blood due to shrinkage of cells and therefore lower water content. 35

CBF Quantification
Our CBF results fall within the large range of reported values in children with SCD (ϳ70 -150 mL/100 g/min). 1,4,9,36,37 The necessary reliance on a quantification model for obtaining physiologically meaningful CBF values means that the method is sensitive to the assumptions of the model used, which could differ between healthy adults and children with SCD. The fact that measured T1 blood ameliorates the CBF quantification but Hct-calculated T1 blood does not opposes the use of Hct-corrected CBF quantification in SCD and, instead, advocates the use of measured T1 blood . T1 blood measurements are advantageous over Hct-calculated T1 blood because they are faster (1 minute 20 seconds) and less invasive. In the absence of T1 blood measurements, we propose using a mean value of 1818 ms, as measured in this study in children with SCD, which would suffice in improving the absolute agreement with PC-MRI for CBF quantification from ASL.

Limitations
This study should be considered in light of the technical limitations of the T1 blood measurement and the potentially inaccurate reference flow measurements from PCMR.
Whereas T1 blood measurements were acquired in venous blood, the quantification model requires arterial estimates. However, because we compared venous T1 blood measurements with T1 blood values derived from venous Hct, the potential mismatch would have been similar for both methods. Moreover, we demonstrate that the measured venous T1 blood , used to quantify CBF, improved the agreement with independently acquired flow measurements in arterial vessels with PC-MRI, which shows that although the arterial measurement may be better, the venous measurement is sufficient.
PC-MRI as a surrogate for CBF could be critiqued for CBF overestimation due to partial volume effects 38 and inaccurate brain density estimates or underestimated flow due to noncardiac-triggered acquisition. Still, recent literature suggests that errors in flow values associated with nontriggered 2D PC-MRI are Ͻ3% compared with triggered acquisitions. 29,39 Despite these limitations, a recent study has shown high agreement (intraclass correlation coefficient, 0.73) between PC-MRI and pCASL, 40 emphasizing that PC-MRI is currently the best noninvasive reference for pCASL CBF.

CONCLUSIONS
Inaccurate T1 blood estimates can be a major confounder for quantitative perfusion assessment from ASL. Patient-specific, in vivomeasured T1 blood measurements provided more accurate CBF values than T1 blood derived from Hct values. To avoid overestimation of CBF in SCD, we recommend the use of a fixed value of 1818 ms (T1 blood -SCD) for CBF quantification from ASL in SCD if measured T1 blood values are not available.