CME JOURNAL OF MAGNETIC RESONANCE IMAGING 41:1591–1600 (2015)

Original Research

Cerebral Blood Flow Measurements in Infants Using Look–Locker Arterial Spin Labeling Marta Varela, PhD,1,2* Esben T. Petersen, PhD,3 Xavier Golay, PhD,4 and Joseph V. Hajnal, PhD1,2 Purpose: To measure cerebral blood flow (CBF) using Look–Locker arterial spin labeling (ASL) in children under 1 year of age and to investigate the advantages of using subject-specific estimates of ASL model parameters in this population. Materials and Methods: Of 12 scanned infants, we successfully acquired CBF maps in 7 (postmenstrual age: 32 to 78 weeks) using a Look–Locker ASL scheme and both adult literature-derived and subject-specific model parameters. ASL global CBF measurements were compared with independent global CBF measurements obtained in the same scanning session using phasecontrast angiography. Results: Measured global CBF values ranged from 24 to 56 mL/100g/min in the scanned infants, increasing significantly with postmenstrual age (rSpearman ¼ 0.89, P-value ¼ 0.01). Using subject-specific model parameters yielded CBF estimates in significantly better agreement with phasecontrast angiography values (P-value: 0.80) than when standard adult parameters were used (P-value: 0.04). Conclusion: Look–Locker ASL can be used to measure CBF in infants and its accuracy is improved with the use of infant-specific auxiliary parameters, particularly blood and tissue T1, which were much more variable in the imaged infants in than adults. Key Words: Cerebral perfusion measurements; blood flow quantification; pediatric neurology and hemodynamics; phase contrast angiography; neonatal imaging; magnetic resonance imaging J. Magn. Reson. Imaging 2015;41:1591–1600. C 2014 Wiley Periodicals, Inc. V

1 Department of Biomedical Engineering, Division of Imaging Sciences, King’s College London, London, UK. 2 Imaging Sciences Department, MRC Clinical Sciences Centre, Imperial College London, London, UK. 3 Department of Radiology, University Medical Center, Utrecht, The Netherlands. 4 UCL Institute of Neurology, University College London, London, UK. Contract grant sponsor: Studentship funding from the Portuguese Foundation for Science and Technology (FCT). *Address reprint requests to: M.V., Department of Biomedical Engineering, Division of Imaging Sciences, King’s College London, Rayne Institute, 3rd Floor, Lambeth Wing, St Thomas’ Hospital, London SE1 7EH, UK. E-mail: [email protected] Received March 31, 2014; Accepted July 15, 2014. DOI 10.1002/jmri.24716 View this article online at wileyonlinelibrary.com. C 2014 Wiley Periodicals, Inc. V

PERINATAL ASPHYXIA and hypoxic-ischemic episodes are believed to be linked with many of the brain lesions that often accompany problematic births in term infants (1). Preterm infants are also at risk of developing neurological damage, as they often suffer from cardiovascular impairments that may compromise blood flow to the brain (1). Observation of alterations in cerebral perfusion, or cerebral blood flow (CBF), could therefore shed light on the evolution of neurological pathology and help monitor patient treatment. CBF measurements are nonetheless uncommon in neonatal clinical practice, as there are few noninvasive techniques able to provide accurate CBF values in this population. These can be particularly challenging in very small children, such as preterm-born infants, whose CBF is known to be very low (2–4). Arterial spin labeling (ASL) (5) is a magnetic resonance imaging (MRI) method capable of producing quantitative perfusion maps without requiring the administration of an external contrast agent. ASL uses a bolus of magnetically labeled arterial blood water as a freely diffusible tracer. The bolus passage through the brain can be detected as a small change in signal on brain images compared to control acquisitions obtained in the absence of a labeled bolus. The signal in the subtraction of these two images, the perfusion-weighted images (PWI), depends strongly on regional cerebral perfusion values. The perfusion-related signal is very low in comparison to static brain signal. Therefore, several repeats of the experiment are usually performed in the same conditions and subsequently averaged to improve the PWI’s signal-to-noise ratio (SNR). CBF and other hemodynamic parameters, such as arterial transit time (ATT), can be estimated from the signal in the PWI, DS, through the use of a hemodynamic model, such as the multiparameter singlecompartment model proposed by Buxton et al (6). The application of the full Buxton model requires the repetition of the ASL experiment at several delays and the use of several parameters, namely: longitudinal relaxation time constant of blood and brain tissue, respectively: T1b and T1t; fully relaxed signal from blood, S0b; arterial transit time, ATT; bolus duration, t; inversion efficiency of the labeling pulse, a; and

1591

1592

Varela et al.

blood–brain partition coefficient for water, l. The ana-

lytical form of the model for pulsed ASL (modified from eq. (3) in (6)) is shown in Eq. (1):

0  DSðTIÞ ¼ 2 S0b a CBF exp TI=T1app expðk ATT Þ–expðk TIÞ=k 2 S a CBF exp TI=T expðk ATTÞ1–expðktÞ=k 0b

1app

where 1/T1tapp ¼ 1/T1t þ CBF/l; k ¼ 1/T1b – 1/T1tapp and TI is the time at which the inversion pulse was applied. To minimize the scan time of the ASL experiment, perfusion-weighted images can be acquired at several TIs following an initial labeling/control pulse. By considering the effect of the repeated excitation pulses on tissue signal only, the Buxton model can be modified by replacing T1tapp in Eq. (1) by T1tappeff, where: 1=T1tappeff ¼ 1=T1tapp –loge ðcos FA Þ=DTI

[2]

as previously shown (7). This Look–Locker Buxton model uses the flip angle (FA) of the excitation pulse as an additional parameter. CBF measurements using ASL in children are fraught with a number of difficulties. A major challenge concerns the accurate estimation of the kinetic model’s parameters in neonates. It is not clear whether values estimated in adults apply to the neonatal and pediatric population and whether the variability within the infants allows population-specific rather than subject-specific values to be used. Another challenge concerns subject motion, as infants are unlikely to stay still for the duration of the experiment and ASL, as a subtraction technique, is particularly sensitive to subject motion. The inability to acquire ASL data contiguously for long time periods also impacts negatively the SNR of the averaged PWIs. This is of particular concern in very young infants whose very low cerebral perfusion levels (2) already strongly limit the SNR of perfusion-weighted images. Although ASL experiments have been performed in neonates and young children (8–13), these issues were not explicitly dealt with and it has generally been assumed that adult model parameters can be used in the neonatal population without incurring a significant error. In this study we used a Look–Locker ASL protocol to measure CBF in infants in the first year of life. We also test whether using subject-specific estimates of model parameters (ATT, T1t, T1b, S0b, and l) improves CBF estimates, when compared to phase-contrast angiography mean CBF measurements.

for TI < ATT for ATT  TI < ATT þ t

[1]

for ATT þ t  TI

assumed: T1b ¼ 1700 msec; T1t ¼ 1200 msec; l ¼ 0.98;  FA ¼ 35 ; CBF ¼ 40 mL/100 g/min; ATT ¼ 500 msec, and t ¼ 700 msec. Sampling was performed at TI1 ¼ DTI ¼ 300 msec, and nTI ¼ 25. The contributions of l and T1t to estimates of S0b, as in Eq. (3) below, were not taken into account in this simulation. Subjects Twelve infants (eight male) with a range of gestational ages at birth (GA: 30–42 weeks, and postmenstrual ages at the time of scan, PMA: 32–78 weeks) were imaged. All infants were scanned under ethical approval, requiring written informed parental consent for each subject. All subjects were scanned under medical supervision and were spontaneously breathing air at the time of the examination. Infants were either sleeping naturally following a feed or sedated through the oral administration of chloral hydrate (30–50 mg/kg). All infants were positioned using inflatable cushions to minimize motion artifacts. The perfusion scan was typically performed at the end of a 45–60-minute long MRI scan, during which MR images were acquired for clinical reasons or for other ethically approved research studies. Perfusion-weighted images were obtained in each of the subjects by subtracting contiguous control and label image pairs for all TIs. The PWIs at each repeat and TI were inspected for inconsistencies between contiguous control and tag images and among repeats, caused by subject motion. Some subtracted images showed evidence of motion upon visual inspection, namely by displaying white-dark bands along the edges of the brain in control-tag subtraction images or when subtracting PWIs from different repeats. In these, the longest contiguous series of repeats that were not affected by motion was taken. The selected PWIs were then averaged to give a perfusion-weighted time series. Some of the infants scanned presented some degree of pathology, which is detailed in Table 1, alongside relevant clinical parameters for the subjects. Subjects are numbered in increasing PMA at the time of scan. Image Acquisition Protocol

MATERIALS AND METHODS Parameter Sensitivity Analysis A sensitivity analysis of the Look–Locker Buxton model was carried out to determine the relative impact of each auxiliary parameter on CBF estimates. The following baseline parameter values were

All infants were imaged using a Philips 3T Achieva scanner and a SENSE 8-element head coil (Philips Healthcare, Best, Netherlands). T1- and T2-weighted images were acquired in each subject using neonatalspecific acquisition parameters (2). Each examination included an ASL protocol and an auxiliary acquisition similar to the one used previously (16) for blood T1

CBF in Infants Using Look–Locker ASL

1593

Table 1 Characteristics of the Imaged Infants Infant # 1 2 3 4 5 6 7

Pathology

GA birth (weeks)

PMA scan (weeks)

Yes Yes No Yes Yes

29.7 33.0 41.7 39.3 40.0

31.5 34.1 45.1 49.0 49.6

M M M M M

No

38.0

58.9

F

Yes

40.9

77.5

M

Sedated?

Periventricular leukomalacia None None None Atrophy of basal ganglia and white matter, mild ventriculomegaly Agenesis of corpus callosum, brain atrophy, mild ventriculomegaly Mild ventriculomegaly

measurement, which provided estimates of the subject-specific parameters used in the ASL kinetic model (Eqs. (1 and 2)). The scan parameters used are shown in Table 2. ASL was performed using the QUASAR sequence (17) applied in a single transverse plane, which was manually positioned in the mid-ventricular region in all imaged subjects (Fig. 2). A single-slice implementation was chosen, because it was found in pilot tests that multiple slice implementations provided unstable CBF measurements in subsequent slices. In brief, image excitation and acquisition was performed at a time TI1 following the labeling/control pulse and then repeated at DTI intervals thereafter. To increase the SNR of the perfusion-weighted images, the entire sequence was repeated at time intervals TR for a number of repeats, NAV. NAV varied from subject to subject, depending on the ability of the subjects to remain still and the amount of time available for the ASL scan. Suppression of intravascular signal using dephasing gradients was not performed, as it was found in pilot studies in neonates using a maximum velocity encoding (vENC) of 3 cm/s, as is commonly done in adults (17), that this led to an unacceptably low SNR in the acquired PWIs. The auxiliary scan sequence used to estimate T1b and other model parameters starts with a thick-slab

inversion, applied to a wide slab centered on the imaging slice (16). This is followed by a series of excitation pulses and 2-shot interleaved EPI readouts, performed at every TI time interval. The positioning of the imaging slice was the same in the ASL and auxiliary scans (Fig. 2). In the auxiliary scan, the center of the inversion slab matched that of the imaging slice. The auxiliary scan’s bandwidth was set manually to ensure that the water-fat shift was the same in the two images. The auxiliary scan was also used to acquire a gray matter (GM) and white matter (WM) mask in the same geometry as the ASL scan. This was achieved through simple thresholding of the voxels inside the brain in the auxiliary image that showed the highest white/ gray matter contrast (the inversion time for this image being strongly age-dependent). Phase-contrast angiography (PCA) images of the internal carotid arteries (ICAs) and basilar artery (BA) were also acquired in the same scanning session to allow mean cerebral perfusion to be estimated using an independent method. The method, which has been described in detail before (2), uses a PCA scan with a resolution of 0.6  0.6  4.0 mm3, TR/TE: 7.0/4.3 msec, vENC: 60 cm/s, for infants with PMA < 40 weeks, or 140 cm/s otherwise. T1- and T2-weighted anatomical images, which are acquired in every infant for clinical reasons, were segmented to provide estimates of brain volume (2).

Table 2 Imaging Parameters for the ASL and Auxiliary Scans

Auxiliary Parameter Estimates

Parameter

Blood T1 (T1b)

Field of view (mm2) Acquisition resolution (mm3) TR/TE (msec) TI1/DTI (msec) nTI  Nominal FA ( ) TIsat (msec) NAV Nominal inversion slab width (mm) Nominal inversion slab gap (mm) # EPI shots SENSE factor Scan time

ASL scan

Auxiliary scan

200200 3.043.045.50

200200 1.841.842.00

4000/11 300/300 12 40 900 30–90 100

8000/18 300/300 12 40 — 1 500

30



1 2.5 4–12 min

2 3 24 sec

Gender

Using the auxiliary scan, blood T1 was estimated by fitting the signal in the voxels in the superior sagittal sinus to an inversion recovery model, as described previously (16). Although a measurement of arterial blood T1 would be more desirable for ASL studies, this is technically challenging and the T1 of blood is the sagittal sinus was used as a surrogate measure of arterial T1b. In the current study, the auxiliary scan was performed using multishot echo-planar imaging,   at a lower flip angle (nominally 40 , compared to 90 in the original study (16)) and in a less angulated geometry than in the original study, causing concern that partial volume effects could affect blood T1 values. To this end, the T1b scan of (16) was additionally performed in four infants and a comparison between

1594

Varela et al.

Table 3 Look-up Table for the Blood–Brain Partition Coefficient (l, in mL/g) Hematocrit (%) PMA < 37 weeks 37  PMA < 45 weeks PMA  45 weeks

65

55

45

35

25

1.18 1.08 1.02

1.15 1.05 0.99

1.12 1.02 0.99

1.09 1.00 0.94

1.06 0.97 0.92

blood T1 values acquired using the current and previous methods was carried out. Brain Tissue T1 (T1t) Brain tissue T1 was estimated by fitting the signal time course of each of the WM and GM voxels in the segmented auxiliary scan to a Look–Locker inversion recovery, using the model proposed previously (18) and assuming an inversion efficiency, a, of 1, as in previous studies (16). The mean and standard deviation of the T1 values found in all voxels of the same tissue type were then used respectively as the best estimate and the error of the T1 of the tissue. Blood–Brain Partition Coefficient for Water (l) The blood–brain partition coefficient for water, which is the ratio of the average water content in brain and capillary blood when in equilibrium, was not measured directly, but instead corrected for subject age and hematocrit (volume fraction of red blood cells in the blood). Estimates of water content in the brain as a function of age were taken from the literature (19) and the hematocrit was estimated from T1b using the relationship found in (16), given that no direct hematocrit estimates were available for the imaged infants. This allowed a look-up table (Table 3) for l as a function of the hematocrit and PMA for the whole brain to be created. Fully Relaxed Signal From Blood (S0b) To estimate S0b, we divided the signal from fully relaxed brain tissue, S0t, by l, as initially proposed (6). S0t is obtained by correcting the signal from ROIs in white and gray matter in the control images at the last TI, St(TIend), by the saturation effect caused by the Look–Locker readout (20), to give the following expression for S0b:

S0b

    St ðTIend Þ 1  exp  DTI T1t cosFA    ¼ l 1  exp  DTI T1t

[3]

For comparison purposes, the same expression was used to compute S0b using the adult parameter values listed in Table 4. Other Parameters Flip angle estimates were carried out using the method proposed by Yarnykh (21). Due to scan time constraints, FA was only measured in two neonates with PMA of 59 and 79 weeks. It was found that the FA was 83 6 5% of the nominal value in both scanned   infants. A value of 33 (instead of the nominal 40 ) was therefore used in all calculations throughout this study. A value of 0.91 was used for the inversion efficiency of the labeling pulse, as measured by for an inversion pulse with similar characteristics (17). This value is lower than the inversion efficiency of the auxiliary scan’s inversion pulse (assumed to be 1 (16)), which was applied closer to the isocenter. The duration of the bolus, t, was determined by the delay between the inversion and the application of the QUIPSS-II pulse, TIsat¼900 msec. A summary of all of the parameters in the Look– Locker Buxton model, including the method used to estimate them in neonates and literature values for adults when available, can be found in Table 4. CBF Measurements Using ASL The Look–Locker Buxton model (Eqs. (1 and 2)) was used to fit the signal from each voxel using a leastsquares method implemented in MatLab (MathWorks, Natick, MA). Maps of ATT and CBF (the only two free parameters) were computed for each subject, with additional parameters estimated as described in Table 4. To allow CBF in different parts of the brain to be compared across different subjects, regions of interest (ROIs) in the basal ganglia, occipital lobe, and frontal white matter were drawn manually. These regions were chosen over other parts of the brain, because it was possible to manually draw ROIs of one tissue type in them, with minimal partial volume contamination from other types of tissue. Mean CBF and ATT in each of these regions were computed by averaging

Table 4 List of All Model Parameters and Methods Used to Estimate Them Parameter CBF ATT T1b T1t S0b t l FA a

Method used

Adult values

Free parameter in fit Free parameter in fit Auxiliary scan Auxiliary scan Using T1t and l (see Equation [3]) QUIPSS-II saturation pulse Literature, optimized for infants (see Table 3) Measured separately in two infants Literature

approx. 20 (WM), 45 (GM) mL/100g/min (32) — 1700 msec (16) 830 (WM), 1330 (GM) msec (26) — — 0.82 (WM), 0.98 (GM) mL/g (33) — 0.91 (17)

CBF in Infants Using Look–Locker ASL

1595

computed using the standard adult values in Table 4. Global CBF values obtained using these nonoptimized model parameters were compared to the values obtained using PCA and the infant-specific parameters. RESULTS Parameter Sensitivity Analysis

Figure 1. Sensitivity analysis of the parameters affecting CBF estimation using the Look–Locker Buxton model (Eqs. (1 and 2)).

CBF and ATT values obtained in each of the voxels pertaining to the ROI. Given that no intravascular signal suppression was performed, voxels with very high CBF (more than 200 mL blood/100 g tissue/min) were assumed to be highly contaminated with intravascular signal and were discarded. Similarly, voxels with very low CBF (less than 5 mL blood/100 g tissue/min) were also discarded as being contaminated by CSF or other nonperfused tissue, as suggested by the relatively poor quality of the fits to the model in these voxels. Global CBF was then computed by performing an average of white matter CBF (measured in the frontal WM ROI) and gray matter CBF (measured in the occipital GM ROI), weighted by the volume of gray and white matter computed from segmented anatomical MR images in the same subjects, as detailed previously (2). In infants with PMA >50 weeks, brain volume was measured using T1-weighted anatomical scans, segmented using FAST (22). In younger infants, T2weighted scans, segmented using a neonatal-specific method (23), were used. BET (24) was used for brain extraction prior to segmentation in all infants.

The results of the parameter sensitivity analysis are shown in Fig. 1. It can be seen that the accuracy of CBF estimates depends strongly on the imaging parameters: an inaccuracy of 15% in a given imaging parameter (with the exception of l) leads to errors in CBF estimates of more than 4%. The direct scaling factor, S0b*a, is the most important parameter, followed by FA and the subject-specific parameters ATT, T1t, t, and T1b. Although l appears to have a modest impact on CBF estimates, it is often used to estimate S0b, as shown in Eq. (3), in which case its impact increases substantially. Auxiliary Parameter Estimates Of the initial cohort of 12 infants, some woke up during the ASL scan and the data from others showed intrashot motion artifacts and were clearly too motion-corrupted to be analyzed. Only data from seven infants were therefore used in this study. Of these, the data from three subjects showed motion artifacts in some of the PWIs, which led to the rejection of some of the repeats. The number of repeats used ranged from 30 to 60 across all subjects. Differences in the number of repeats were not directly taken into account in the data analysis. Blood T1 (T1b) Blood T1 estimates using the auxiliary scan ranged between 1861 and 2094 msec, in good agreement with the literature (16) and, as expected, spanning a considerably higher range than in healthy adults. A comparison of the current protocol with the literature scan (16) in four infants gave nonsignificant differences in T1b of 58 6 84 msec (3.2 6 4.6%, paired t-test P-value ¼ 0.26). Brain Tissue T1 (T1t)

Comparison With PCA The methods used for CBF measurements in young infants with PCA are explained in detail elsewhere (2). Briefly, global CBF estimated using ASL was compared to global CBF values that were computed by dividing flow to the brain (measured in the basilar and internal carotid arteries with PCA) by the brain volume (WMþGM volume in the cerebrum, cerebellum, and brainstem) computed using the segmented anatomical images, as described above. The global CBF values thus obtained were divided by the average neonatal brain density (1.05 g/mL (25)) to convert them into mL blood/100 g tissue/ min, the units used to express CBF in the performed ASL measurements. To assess the impact of the subject-dependent parameter estimation, voxelwise CBF values were also

Figure 3 shows the values obtained for white and gray matter T1 as a function of postmenstrual age in the scanned infants. The measured values ranged from 1312 6 100 to 3163 6 271 msec in WM and 1512 6 160 to 2162 6 115 msec in GM and, as expected, are much higher and more variable than corresponding adult values (26). The obtained values decrease with PMA, are in good agreement with the literature for neonates when available (27), and show the well-known neonatal reverse T1-weighted contrast between white and gray matter in all but the oldest infant. Fully Relaxed Signal from Blood (S0b) The relative difference in measured S0b values computed using Eq. (3) with optimized neonatal parameters and the adult literature values ranged from 32% to þ19% with an average value of 5%.

1596

Varela et al.

Figure 2. Pilot T2-weighted scans of Infant 3, showing the positioning of the imaging slice in the ASL scan (transverse image and orange rectangle in coronal and sagittal images). The labeling slab for the ASL scan (blue rectangle) is also shown. For the auxiliary scan, the inversion slab is centered in the imaging slice, over the entire head and neck regions.

CBF Measurements Figure 4 shows time series of perfusion-weighted images from Infant 1 (PMA: 32 weeks) and Infant 7 (PMA: 78 weeks). Differences in the PWIs and the auxiliary scans clearly reflect the different degree of maturation of the two subjects. As no dephasing gradients were used, some of the early images (TI < 1200 msec) clearly show intravascular signal, particularly in the oldest infant. At subsequent timepoints, the measured signal mostly comes from brain tissue, but shows a low SNR due to the low number of repeats used. On average, 15% of voxels in the brain were discarded as being contaminated by either intravascular signal or by nonbrain tissue. Figure 5 shows the computed CBF and ATT maps for the same two infants, with example fits from voxels in the cortex. As expected, white matter shows, in general, a lower CBF and a higher ATT than gray matter. Some intravascular contamination can also be observed as voxels with CBF >100 mL blood/100 g tissue/min, particularly in the older infant (Fig. 5b).

The model fitted the data well (Fig. 5), with small residuals compared to the data variance, especially in gray matter, given its higher SNR. The measured frontal WM CBF was 22 6 14 mL blood/100 g tissue/min (range: 16 to 30 mL/100g/min). Measured CBF in the basal ganglia was 40 6 18 mL blood/100 g tissue/min, ranging from 25 to 60 mL/100g/min. Occipital GM showed even higher perfusion values: 53 6 29 mL/ 100g/min, ranging from 31 to 80 mL/100g/min. Mean whole brain CBF and ATT values are shown, as a function of postmenstrual age at scan, in Fig. 6. Global CBF is in good agreement with the literature from different modalities (3,4,28) and increases significantly with PMA (rSpearman ¼ 0.89, P-value ¼ 0.01), from 24 mL/100g/min for Infant 1 to 56 mL/100g/ min for Infant 7. The ATT demonstrated a nonsignificant trend that decreased with increasing postmenstrual age, from a mean value of 492 msec in Infant 4 to 1081 msec in Infant 2 (rSpearman ¼ 0.64, Pvalue ¼ 0.14). Comparison With PCA As shown in Fig. 6, using subject-specific model parameters was found to give CBF values in significantly better agreement with those estimated using PCA (relative difference in CBF across subjects: 4 6 27%; paired t-test P-value: 0.80), than when adult model parameters (see Table 4) were used (relative difference in CBF across subjects: 43 6 49%; paired t-test P-value: 0.04). DISCUSSION

Figure 3. Measured brain tissue T1 values for white matter (WM) and gray matter (GM) in each of the scanned neonates, as a function of postmenstrual age at the time of scan (PMA).

In this study, a Look–Locker ASL protocol was used to successfully acquire cerebral perfusion maps in seven infants with a range of postmenstrual ages and degrees of pathology. A rapid (less than 30 sec) auxiliary scan was performed in the same scanning session to allow some of the ASL model parameters (ATT, T1b, T1t, S0b, l) to be estimated on a subject-by-subject basis. Of these, blood T1 was estimated using the auxiliary scan, as described previously (16), as it was found that changes in acquisition flip angle and slice

CBF in Infants Using Look–Locker ASL

1597

Figure 4. Perfusion-weighted image time series for: (a) the youngest infant, Infant 1 (PMA: 32 weeks) and (b) the oldest infant, Infant 7 (PMA: 78 weeks) at each inversion time (shown above each image). To provide an anatomical reference, an auxiliary scan image of the same infant is also shown on the right.

angulation did not significantly affect T1b estimates. The infants’ hematocrit was also estimated from the T1b values, in order to allow subject-specific estimates of l to be computed. Tissue T1 was estimated by fitting the signal in gray and white matter in the auxiliary scan to a Look– Locker recovery curve. In principle, tissue T1 could have been estimated directly from a similar fit to the ASL control images. In practice, however, it was found, by looking at the time course of the tissue signal in control images, that the applied presaturation pulses did not fully saturate the brain tissue, leaving a variable amount of magnetization that confounded T1t estimates using this method. The obtained values agree well with the literature (27), with the youngest infants showing T1 values more than twice as high as literature adult T1 values (26), highlighting the need to use age-specific model parameters for T1t in the neonatal population. In the current study, T1t was also used to estimate S0b, increasing its impact on CBF estimates even further.

S0b was computed using the fully relaxed signal in white and gray matter and l. It could have instead been estimated from the signal in the auxiliary scan, namely, from estimates of the fully relaxed signal in the voxels where T1b is estimated. This approach can, however, lead to errors due to automatic scanner signal calibration and was not explored in the current study. Scan time constraints precluded the optimization of all scan parameters, such as the bolus width, t, which was set to 900 msec, a slightly higher value than is usually used in adults for a 10 cm slab (15), to account for the slower transit time in the neck arteries in the youngest infants (2). This may mean that the bolus width is slightly overestimated in the older infants, who have a higher blood velocity in the arteries that feed the brain (2). Using these parameters and the Look–Locker Buxton model, we were able to obtain CBF maps and values in seven infants, in good agreement with the literature from other modalities (3,4,28). The

1598

Varela et al.

Figure 5. (a) Cerebral blood flow (CBF) and (b) arterial transit time (ATT) maps for Infant 1 (PMA: 32 weeks, top panel) and Infant 7 (PMA: 78 weeks, bottom panel), together with example fits from voxels in (c) the occipital cortex and (d) basal ganglia. The best line fit is shown in black with the measured signal in the perfusionweighted images (PWI) in blue.

assumptions underlying the Buxton model (namely, the negligible bolus dispersion across the arterial tree and the existence of an immediate equilibrium between water and tissue) were not tested specifically in the neonatal population. However, the good fit of the data to the model (Fig. 5) provides some assurance about the validity of the model used, although the sensitivity of the fit to the details of the model was not explored. More complex perfusion kinetic models (for example, (29,30)) would require the use of additional parameters, such as capillary permeability or blood–tissue exchange time, which would also need to be estimated in the neonatal population.

Dephasing gradients (crushers) were not used in the current study, as it was found that dephasing gradients with the same settings as for adults (vENC ¼ 3 cm/s (17)) led to an unacceptably low SNR and it is also uncertain if adult-inspired vascular crushers are appropriate for the neonatal flow regime. Instead, to minimize intravascular contamination, voxels whose CBF estimates were higher than 200 mL/100g/min were discarded from the CBF calculations. This is a nonideal approach, as partial volume effects and the fact that the point-spread function in MRI does not have an infinitesimal area imply that signal in neighboring voxels can be

CBF in Infants Using Look–Locker ASL

1599

Figure 6. (a) Global CBF estimates computed using PCA, ASL with optimized parameters (ASL opt), and ASL with adult literature values (ASL non opt) as a function of postmenstrual age at the time of scan (PMA). For clarity, for the ASL estimates only the upper or lower branch of the error bars is displayed. (b) Mean arterial transit time (ATT) as a function of PMA, computed using optimized model parameters.

increased by the presence of very high intravascular signal. Alternatively, a data analysis procedure that allows intravascular contributions to be separated from tissue signal, such as the Bayesian approach proposed by Chappell et al (31), could have been used, presumably after some modification of the priors for the infant population. The degree of intravascular contamination is likely to be age-dependent and could therefore affect CBF estimates differently in infants of different ages. The obtained CBF estimates show a large amount of uncertainty, given the low SNR of the PWIs used to estimate CBF. This is particularly severe for WM voxels, where, as in adults, SNR is particularly low. Subject motion is one of the causes of the observed low SNR and remains one of the greatest difficulties in neonatal imaging, particularly in protocols that require averaging of several repeats and/or image subtraction, such as ASL. In the current study, no image registration was performed to correct for interrepeat subject motion, as it was unlikely that this motion would be purely in-plane. In this study, the comparison between ASL and PCA CBF estimates was only performed in seven infants and should be investigated in a larger cohort in future studies. Data were acquired in a single plane, as preliminary tests in adults showed that a multislice protocol did not produce reliable data. This implies that whole-brain CBF measures rely on CBF estimates in small WM and GM ROIs, weighted by the GM and WM across the brain. This treats CBF as homogeneous in GM and WM across the brain, which is only true as a first approximation. Despite its limitations, the current study is the first to use subject-specific measures of several model parameters and to compare ASL-derived CBF estimates with those obtained with a concomitant independent measurement. Previous reports (8–12) have used adult literature values, with the exception of T1b in one recent study (13), and do not directly compare ASL estimates with other means of measuring CBF.

The multiple inversion timepoint technique used in the current study also contrasts with previous neonatal ASL reports, which relied on estimates from a single TI and used the QUIPSS-II (14) / Q2TIPS (15) formalism. This formalism depends on strong assumptions based on adult hemodynamics, particularly about expected values of ATT and t. The Look– Locker approach taken in the current study bypasses these assumptions, at the expense of the simplicity of the model used to estimate CBF. Similar to the current study, previous neonatal ASL reports have found an increase in CBF with PMA, with a range of mean CBF values (all in mL blood/100 g tissue/min) quoted for infants born preterm: 5 (11), 7 (13); term: 12 (13), 17 (8), 20 (9), and at 3 months of age: 30 (13), 40 (12). These values compare well with CBF values measured this work. In this study, some of the infants presented with some degree of pathology, which may affect the measured CBF values. However, the fact that no infants showed focal lesions suggests that this should have little impact in the comparison with PCA-derived CBF estimates. Different levels of sedation were also used, in agreement with local clinical practice. This may have a small impact on the absolute CBF estimates in each subject, but should not affect the comparison of CBF estimates performed in the same session using ASL and PCA. In comparison with PCA measurements, it was found that using infant-specific model parameters significantly improved the agreement of mean CBF measurements using both modalities. This comparison assumed that CBF values measured in WM and GM ROIs in each of the infants were representative of whole brain GM and WM values. The proportion of GM and WM in each subject’s brain was estimated from a segmented anatomical MRI, to avoid introducing a bias caused by the distribution of white and gray matter in the imaging slice. The high uncertainty of the CBF values measured using ASL, particularly in white matter, and the

1600

Varela et al.

longer scan and processing time used to obtain CBF estimates using this method suggest that PCA combined with segmented anatomical MR images (2) provide a better method to estimate global CBF values. However, PCA does not provide any spatial information about CBF, not even of the CBF ratio in white and gray matter, and is therefore of little utility when studying regional perfusion. In conclusion, in this study we successfully used an ASL technique to acquire CBF and arterial transit time maps in children less than 1 year of age. Using customized model parameters, the mean CBF values measured show good agreement with whole-brain CBF measurements performed in the same scanning session with a different MRI technique. ABBREVIATIONS a l t ASL ATT CBF FA GA GM NAV PCA PMA PWI S0b T1b T1t TI vENC WM

Inversion Efficiency Blood–Brain Partition Coefficient for Water Bolus Duration Arterial Spin Labeling Arterial Transit Time Cerebral Blood Flow Flip Angle Gestational Age Gray Matter Number of Repeats Phase Contrast Angiography Postmenstrual Age Perfusion-Weighted Images Fully Relaxed Signal from Blood Blood Longitudinal Relaxation Time Constant Tissue Longitudinal Relaxation Time Constant Inversion Time Maximum Velocity Encoding White Matter

REFERENCES 1. Volpe JJ. Neurology of the newborn, 4th ed. Philadelphia: WB Saunders; 2001;22:251–264. 2. Varela M, Groves AM, Arichi T, Hajnal JV. Mean cerebral blood flow measurements using phase contrast MRI in the first year of life. NMR Biomed 2012;25:1063–1072. 3. Edwards AD, Wyatt JS, Richardson C, Delpy DT, Cope M, Reynolds EO. Cotside measurement of cerebral blood flow in ill newborn infants by near infrared spectroscopy. Lancet 1988;2: 770–771. 4. Greisen G. Cerebral blood flow in preterm infants during the first week of life. Acta Paediatr Scand 1986;75:43–51. 5. Williams DS, Detre JA, Leigh JS, Koretsky AP. Magnetic resonance imaging of perfusion using spin inversion of arterial water. Proc Natl Acad Sci U S A 1992;89:212. 6. Buxton RB, Frank LR, Wong EC, Siewert B, Warach S, Edelman RR. A general kinetic model for quantitative perfusion imaging with arterial spin labeling. Magn Reson Med 1998;40: 383–396. 7. Gunther M, Bock M, Schad LR. Arterial spin labeling in combination with a look–Locker sampling strategy: inflow turbo-sampling EPI-FAIR (ITS-FAIR). Magn Reson Med 2001;46:974–984. 8. Miranda MJ, Olofsson K, Sidaros K. Noninvasive measurements of regional cerebral perfusion in preterm and term neonates by magnetic resonance arterial spin labeling. Pediatr Res 2006;60: 359–363. 9. Licht DJ, Wang J, Silvestre DW, et al. Preoperative cerebral blood flow is diminished in neonates with severe congenital heart defects. J Thorac Cardiovasc Surg 2004;128:841–849.

10. Wang J, Licht DJ. Pediatric perfusion MR imaging using arterial spin labeling. Neuroimaging Clin N Am 2006;16:149–167. 11. Wintermark P, Hansen A, Gregas MC, et al. Brain perfusion in asphyxiated newborns treated with therapeutic hypothermia. AJNR Am J Neuroradiol 2011;32:2023–2029. 12. Duncan AF, Caprihan A, Montague EQ, Lowe J, Schrader R, Phillips JP. Regional cerebral blood flow in children from 3 to 5 months of age. AJNR Am J Neuroradiol 2014;35:593–598. 13. De Vis JB, Petersen ET, de Vries LS, et al. Regional changes in brain perfusion during brain maturation measured noninvasively with Arterial Spin Labeling MRI in neonates. Eur J Radiol 2013;82:538–543. 14. Wong EC, Buxton RB, Frank LR. Quantitative imaging of perfusion using a single subtraction (QUIPSS and QUIPSS II). Magn Reson Med 1998;39:702–708. 15. Luh WM, Wong EC, Bandettini PA, Hyde JS. QUIPSS II with thinslice TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn Reson Med 1999;41:1246–1254. 16. Varela M, Hajnal JV, Petersen ET, Golay X, Merchant N, Larkman DJ. A method for rapid in vivo measurement of blood T1. NMR Biomed 2011;24:80–88. 17. Petersen ET, Lim T, Golay X. Model-free arterial spin labeling quantification approach for perfusion MRI. Magn Reson Med 2006;55:219–232. 18. Gowland P, Mansfield P. Accurate measurement of T1 in vivo in less than 3 seconds using echo-planar imaging. Magn Reson Med 1993;30:351. 19. Dobbing JJ, Sands JJ. Quantitative growth and development of human brain. Arch Dis Child 1973;48:757–767. 20. Look DC, Locker DR. Time saving in measurement of NMR and EPR relaxation times. Rev Sci Instrum 1970;41:250–251. 21. Yarnykh VL. Actual flip-angle imaging in the pulsed steady state: a method for rapid three-dimensional mapping of the transmitted radiofrequency field. Magn Reson Med 2007;57:192–200. 22. Zhang Y, Brady M, Smith S. Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Trans Med Imaging 2001;20:45–57. 23. Xue H, Srinivasan L, Jiang S, et al. Automatic segmentation and reconstruction of the cortex from neonatal MRI. Neuroimage 2007;38:461–477. 24. Smith SM. Fast robust automated brain extraction. Hum Brain Mapp 2002;17:143–155. 25. Delpy DT, Cope MC, Cady EB, et al. Cerebral monitoring in newborn infants by magnetic resonance and near infrared spectroscopy. Scand J Clin Lab Investig 1987;47:9–17. 26. Norris DG. High field human imaging. J Magn Reson Imaging 2003;18:519–529. 27. Williams L-A, Gelman N, Picot PA, et al. Neonatal brain: regional variability of in vivo MR imaging relaxation rates at 3.0 T—initial experience. Radiology 2005;235:595–603. 28. Takahashi T, Shirane R, Sato S, Yoshimoto T. Developmental changes of cerebral blood flow and oxygen metabolism. AJNR Am J Neuroradiol 1999;20:917–922. 29. Parkes LM, Tofts PS. Improved accuracy of human cerebral blood perfusion measurements using arterial spin labeling: accounting for capillary water permeability. Magn Reson Med 2002;48:27– 41. 30. Zhou J, Wilson DA, Ulatowski JA, Traystman RJ, van Zijl PCM. Two-compartment exchange model for perfusion quantification using arterial spin tagging. J Cereb Blood Flow Metab 2001;21: 440–455. 31. Chappell MA, MacIntosh BJ, Donahue MJ, G€ unther M, Jezzard P, Woolrich MW. Separation of macrovascular signal in multiinversion time arterial spin labeling MRI. Magn Reson Med 2010; 63:1357–1365. 32. Leenders KL, Perani D, Lammertsma AA, et al. Cerebral blood flow, blood volume and oxygen utilization: normal values and effect of age. Brain 1990;113:27–47. 33. Herscovitch P, Raichle ME. What is the correct value for the brain-blood partition coefficient for water? J Cereb Blood Flow Metab 1985;5:65–69.

Cerebral blood flow measurements in infants using look-locker arterial spin labeling.

To measure cerebral blood flow (CBF) using Look-Locker arterial spin labeling (ASL) in children under 1 year of age and to investigate the advantages ...
4MB Sizes 0 Downloads 6 Views