- Email: [email protected]
Comparison of multislice and single-slice acquisitions for pulsed arterial spin labeling measurements of cerebral perfusion
Magnetic Resonance Imaging 24 (2006) 869 – 876
Comparison of multislice and single-slice acquisitions for pulsed arterial spin labeling measurements of cerebral perfusion Alison M. Campbell, Christian Beaulieu4 Department of Biomedical Engineering, Faculty of Medicine and Dentistry, 1098 Research Transition Facility, University of Alberta, Edmonton, Alberta, Canada T6G 2V2 Received 4 January 2006; accepted 29 March 2006
Abstract Multislice Q2TIPS is a widely used pulsed arterial spin labeling (PASL) technique for efficient and accurate quantification of cerebral blood flow (CBF). Slices are typically acquired inferior to superior from a tagging plane. Superior slices show signal loss greater than the loss expected from blood T 1 decay. In order to assess the reasons for this additional signal loss, three single-slice acquisition studies were compared to multislice acquisition (six slices) in healthy volunteers. In Study 1 (n = 8), the tagging plane was fixed in location, and the inversion time (TI 2) was 1500 ms for each slice. For Study 2 (n = 12), the tagging plane was fixed as in Study 1; however, TI 2 increased as slices were acquired further from the tagging plane. In Study 3 (n = 9), the tagging plane was kept adjacent to the imaging slice, and TI 2 was 1500 ms for every slice. Gray matter (GM) and white matter (WM) signal-to-noise ratio (SNR) and CBF were measured per slice. GM SNR from single-slice acquisitions was significantly higher at slices 4 – 6 in Study 2 and at slices 2 – 6 in Study 3 compared to multislice acquisitions. Signal loss in distal slices of multislice acquisitions can be attributed to the destruction of tagged bolus in addition to blood T 1 decay. If limited brain coverage is acceptable, perfusion images with greater SNR are achievable with limited slices and placement of the tagging region immediately adjacent to the site of interest. D 2006 Elsevier Inc. All rights reserved. Keywords: Arterial spin labeling (ASL); Q2TIPS; Cerebral blood flow (CBF); Perfusion MRI
1. Introduction Arterial spin labeling (ASL) is a noninvasive imaging technique that has been proven to yield accurate measurements of cerebral blood flow (CBF) in cortical gray matter (GM) [1]. Arterial spins are labeled by changing their state of magnetization in an artery upstream of the site of interest. As these spins flow through the tissue vasculature, water exchange occurs between labeled blood and tissue. This causes a measurable change in both apparent tissue T 1 and tissue magnetization; either change can be measured and used to quantify blood flow. Essentially, perfusion is determined by acquiring one image with spin tagging and one image without spin tagging. Subtraction of these two images results in a difference signal image (DM) that is directly proportional to blood flow. In the early 1990s, it was shown that either pseudocontinuous saturation or flow-driven adiabatic inversion could 4 Corresponding author. Tel.: +1 780 492 0908; fax: +1 780 492 8259. E-mail address: [email protected] (C. Beaulieu). 0730-725X/$ – see front matter D 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2006.03.011
be used to tag blood water spins in the neck and to measure perfusion in a single slice within the brain [2–4]. Since the first attempts using continuous arterial spin labeling (CASL), many different arterial labeling schemes have been proposed, such as the pulsed arterial spin labeling (PASL) technique called EPISTAR, whereby a single radiofrequency pulse is used to label a thick slab close to the imaging slice of interest [5]. A 1808 inversion pulse is used to apply an upstream tag; after a delay time to allow the tagged blood to perfuse into the tissue, an image is acquired within the brain. As with CASL, the resultant DM image reflects the difference in magnetization of the blood between images acquired with and without the application of a tag pulse. Other PASL variants, such as FAIR [2,6,7], PICORE [8] and Q2TIPS [9], have also been successfully applied. Although the inherent signal-to-noise ratio (SNR) is higher in CASL, PASL has a higher inversion efficiency (close to 1) and a lower natural transit delay, making SNR efficiency between the two techniques similar [10]. With both CASL and PASL, the goal is to produce a tag and a control image where the signals from static tissues in
870
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
both images are identical. This is easy to accomplish with single-slice techniques; however, it is an increasingly complicated task when a multislice data set is required for greater brain coverage. Multislice acquisition involves application of a single magnetization preparation pulse to tag arterial blood followed by sequential acquisition of multiple slices downstream. The major complication in CASL multislice imaging is magnetization transfer (MT) effect. It is necessary that the tag image receive the same MT effects as the control image, and this case can only occur if a single slice is acquired. The most effective solution to this problem is to employ a two-coil CASL system [11]; however, this requires specialized hardware. For PASL, multislice acquisitions are difficult because of variable transit delays both within and between slices, and because of static tissue subtraction errors consistent with slice profile effects of tag and control pulses [12]. Because of these problems with multislice acquisitions, single-slice CASL and PASL techniques are ideal. However, cerebral perfusion results are often desired at multiple slice locations. To avoid this problem, one could acquire multiple single-slice images; however, this is time-consuming given that lengthy signal averaging is necessary to create perfusion maps with adequate SNR. Therefore, when time constraints are present, which is usually the case, multislice ASL acquisitions are necessary for greater brain coverage. It is well known that SNR in multislice scans decreases in more distal slices due to longer inversion times and greater T 1 decay of blood [13]. However, we have observed signal loss in distal slices, which is greater than the loss expected with blood T 1 decay. Several reasons have been proposed for this effect, including the destruction of tagged blood destined for more distal slices by the acquisition of proximal slices [8,9], inversion profile imperfection [8] and uncertainty of transit time [14]. The purpose of the present study is to determine the cause of additional signal loss in distal slices, beyond that resulting from blood T 1 decay, by comparing perfusion maps from various single-slice acquisitions to those derived from multislice acquisitions in healthy volunteers. This study looks specifically at a Q2TIPS pulse sequence with PICORE tagging and highlights several potential ASL protocol improvements for the study of neurological diseases.
2. Materials and methods 2.1. MRI Images were obtained from a group of 17 healthy volunteers (12 males, 5 females; age range, 23–34 years), all of whom gave written and informed consent. Three separate studies were conducted for this report, and each volunteer participated in one or more of the studies. Four volunteers participated in all three studies. There were eight volunteers in Study 1, 12 volunteers in Study 2 and
9 volunteers in Study 3. MR scanning was performed on a 1.5-T Sonata scanner (Siemens Medical Systems, Erlangen, Germany) equipped with gradient coils capable of an amplitude of 40 mT/m and a slew rate of 200 T/m/s. Q2TIPS with PICORE tagging (work in progress; Siemens Medical Systems) was utilized for all ASL perfusion imaging [13]. PICORE tagging involved slab-selective inversion interleaved with no inversion in order to create tag and control images. Gradient-echo (GE) echo planar imaging (EPI) was used to acquire all slices after the application of the labeling pulse. One multislice ASL scan and six single-slice ASL scans were acquired in each study. The multislice scan consisted of six contiguous 8-mm slices acquired inferior to superior, with an interslice delay time of 54 ms. The gap between an inversion band and the first imaging slice was set at 20 mm to ensure minimal interaction of slice profiles (Fig. 1A). The time of the first saturation pulse (TI 1) was fixed at 800 ms, and the total inversion time (TI 2) for the first slice was 1500 ms. This value of TI 2 was chosen as a result of previous experiments, which indicated that TI 2 =1500 ms is optimal for this age range of volunteers (23–34 years) in order to maximize tissue signal and to minimize intravascular signal. Other multislice Q2TIPS acquisition parameters included the following: field of view (FOV) =2222 cm; matrix = 6464; echo time (TE) =15 ms; repetition time (TR)= 2500 ms; slice repetition time (TR slice)= 54 ms; scan time= 3:52 min; tag–control pairs =50; bandwidth =3004 Hz/ pixel. The six slice locations in both the multislice and single-slice acquisitions were exactly aligned. In single-slice data sets, slice thickness, TI 1, FOV, matrix, TE, TR, scan time, number of tag–control pairs and BW were the same as those of the multislice data set. The major differences between single-slice and multislice acquisition parameters include the width of the gap between the inversion band and the imaging slice, and the TI 2 of each slice. Basically, the location of each slice was static; however, the position of the 10-cm inversion band and the timing of image acquisition (TI 2) varied between studies. In Study 1, the placement of the inversion band was fixed with respect to the volunteer. Therefore, the gap between the inversion band and the first imaging slice was 20 mm, and this gap became larger with increasing slice number (Fig. 1B). The distances between the inversion band and the imaging slices were 20, 28, 36, 44, 52 and 60 mm for slices 1–6, respectively. The time of image acquisition (TI2) was fixed at 1500 ms for every slice, regardless of the distance between the inversion band and the slice. In Study 2, the inversion band location was fixed, and the distances between the inversion band and the imaging slices were the same as in Study 1 (Fig. 1B). As superior slices were imaged, TI2 increased in accordance with the slice repetition time of multislice acquisitions (54 ms). Therefore, inversion times (TI2) in Study 2 for slices 1– 6 were 1500, 1554, 1608, 1662, 1716 and 1770 ms respectively (i.e., exactly the same as they would
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
871
Fig. 1. Location of the tagging plane with respect to the imaging slices. (A) The multislice study consists of six 8-mm contiguous slices placed 2 cm from the tagging plane. TI 2 increases incrementally per slice, with a slice repetition time of 54 ms. (B) For Studies 1 and 2, the placement of the tagging plane is fixed with respect to the volunteer. The difference between the two studies is that, in Study 1, TI 2 = 1500 ms for every slice, whereas in Study 2, TI 2 increases by 54 ms with every increase in slice number (identical to the multislice study). (C,D) In Study 3, the tagging plane is adjusted so that it is always located 2 cm from the imaging slice. TI 2 = 1500 ms for every slice in Study 3.
be in a multislice acquisition). In Study 3, placement of the inversion band varied so that its distance from the imaging slice was constant at 20 mm. As a result, TI 2 was also kept constant at 1500 ms (Fig. 1C and D). Several anatomical images were also acquired for each volunteer. An inversion recovery GE EPI image (TI =200 ms, TE =57 ms and TR = 5500 ms) was acquired at each ASL slice location in order to provide an image with adequate GM–white matter (WM) contrast for tissue segmentation. Two additional anatomical images were obtained in order to calculate R and M owm, which are variables necessary for the determination of M ob. A protondensity-weighted image was acquired (GE; a = 108, TE = 5 ms and TR =1000 ms), so that R (the ratio of the proton density of blood in the sagittal sinus to that in the WM) could be determined. M owm, the T 2*-weighted equilibrium signal of WM, was calculated from a single-shot GE EPI sequence with TE = 15 ms and TR = 20,000 ms. These sequences were acquired with the same receiver gain settings as the perfusion images. Together, R and M owm were used to
determine M ob (the T 2*-weighted equilibrium signal of blood) Mob ¼ RMowm expðð1=T2wm 4 1=T2b 4ÞTEÞ
ð1Þ
where T 2wm* is the T 2* of WM and T 2b* is the T 2* of blood [9]. M ob is unique to each volunteer and is necessary for the calculation of CBF. The total scan time per volunteer for one study, including anatomical images, six single-slice ASL scans and one multislice ASL scan, was approximately 30 min. 2.2. Data processing Using the Q2TIPS pulse sequence, we acquired a series of tag and control images that were motion-corrected and pairwise-subtracted to obtain individual ASL difference images. Using MRVision software (MRVision, Inc., Winchester, MA), each individual difference image was qualitatively assessed and removed if pairwise subtraction had not adequately deleted static tissue. The remaining individual difference images were added together and further processed in Matlab (The MathWorks, Inc., Natick,
872
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
MA) to create an overall signal difference image (DM image). Once DM has been produced, CBF ( f) (ml blood/ 100 ml tissue/min) was calculated using [9] DM ¼ 2Mob f TI 1 expð TI 2 =T1b ÞqðT1b ; T1t ; Tex ; f ; k; TI 2 Þ ð2Þ for TI 1 bs
and
TI2 NTI 1 þ yt
ð3Þ
where q is a correction factor accounting for the difference in T 1 decay as tagged water spins move from blood to tissue. In our study, q is assumed to be unity [13]. yt is the transit time of blood from the labeling plane to the imaging slice, and s is the time width of tagged bolus. T 1b is the longitudinal relaxation time of blood, which is assumed to be 1200 ms [15]. Inversion recovery anatomical images with good GM– WM contrast were used for the segmentation of perfusion difference images into GM and WM regions. Signal from whole slice, GM and WM was calculated for each slice in both multislice and single-slice scans. In Studies 1 and 3, the TI 2 for each slice was constant at 1500 ms. Therefore, the amount of blood T 1 decay is identical for all slices in these two single-slice studies. However, in single-slice scans in Study 2, as well as in all multislice acquisitions, TI 2 changes per slice. If not corrected for, SNR would decrease per slice in part because the T 1 decay of blood increases per slice. In order to accurately compare all studies, we corrected for blood T 1 decay prior to making the final SNR calculations. SNR was calculated for whole slice, GM and WM in all slices. In our report, SNR is calculated from DM images and is defined as the signal in the brain divided by the standard deviation of background noise. For multislice experiments,
SNR values for all volunteers were averaged per slice and compared to average SNR values for all volunteers in the corresponding single-slice experiment. In each study, the SNR in the first slice in both single-slice and multislice experiments was normalized to 1, as there are no methodological reasons for a difference in SNR in the first slice. For whole-slice measurements, the actual difference in slice 1 between single-slice and multislice experiments before normalization was 4.3% for Study 1, 0.7% for Study 2 and 16.6% for Study 3. For GM, the difference was 3.3% for Study 1, 1.9% for Study 2 and 12.2% for Study 3. Paired t tests were performed for all experiments (multislice and single-slice acquisitions) to determine if significant signal loss occurred in superior slices compared to the first slice. As well, paired t tests were used to show if any differences in SNR existed between corresponding slices acquired by single-slice and multislice methods within a study. P b.05 was considered a statistically significant difference. 3. Results Qualitative assessment of DM images in all volunteers demonstrates signal loss in superior slices in multislice as well as single-slice studies, albeit to differing degrees in the three single-slice studies (Figs. 2–4). It is not apparent in the initial evaluation of Study 1 (where a tagging plane was fixed in one position and TI 2 for each slice was held constant at 1500 ms) that any signal in superior slices is recovered (Fig. 2). In Study 2, the tagging plane is fixed in position (identical to Study 1); however, TI 2 is increased for slices farther from the tagging plane. Imaging superior slices at a later time allows more blood to reach these slices and allows perfusion signal to increase, as seen in Fig. 3. DM images in Study 3 clearly indicate that the signal in
Fig. 2. Perfusion difference images normalized to M ob (DM/M ob) for multislice and single-slice (constant TI 2 = 1500 ms) experiments in Study 1 (one volunteer). There is an obvious drop in signal with increasing slice number in multislice and single-slice experiments.
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
873
Fig. 3. Perfusion difference images normalized to M ob (DM/M ob) for multislice and single-slice (incremental TI 2) experiments in Study 2 (one volunteer). There is an obvious drop in signal with increasing slice number in the multislice experiment. The images from single-slice acquisitions appear to have greater perfusion signal.
superior slices is higher than the corresponding slice in the multislice experiment when the tagging plane is moved adjacent to the single acquired slice (Fig. 4). Although perfusion signal increases in superior slices for both Studies 2 and 3 with single-slice acquisition, the increase in signal intensity appears to be greater in Study 3. All perfusion experiments yielded DM images with adequate signal to perform SNR and CBF measurements. Difference image SNR values at slice 1 for all single-slice and multislice acquisitions were: 4.4F0.4 (whole slice), 5.6F0.7 (GM) and 1.8F0.5 (WM). The general pattern of GM SNR versus slice number for single-slice and multislice experiments in each study is demonstrated in Fig. 5. SNR measurements from whole slice and WM demonstrate nearly
identical patterns (data not shown). Although multislice acquisition results in all three studies are expected to be identical, deviation occurred due to different volunteers per study and due to inexact placement of tagging regions and slice locations. In Study 1, with a fixed TI 2 of 1500 ms per slice, GM SNR decreases in more distal slices by an amount similar to that in the multislice study (Fig. 5A). For each slice, there is no significant difference in GM SNR between single-slice and multislice experiments in Study 1. As well, there was no significant difference in whole-slice and WM SNR measurements. In Study 2, the location of the tagging plane was fixed; however, TI 2 was increased incrementally with more
Fig. 4. Perfusion difference images normalized to M ob (DM/M ob) for multislice and single-slice (adjacent tagging plane and constant TI 2 = 1500 ms) experiments in Study 3 (one volunteer). There is an obvious drop in signal with increasing slice number in the multislice experiment. There is a large increase in perfusion signal in single-slice images compared to multislice images.
874
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
superior slices. In Fig. 5B, GM SNR remains relatively constant for all slices in single-slice experiments. Therefore, additional delay (over Study 1) in acquiring more distal slices increases perfusion signal. GM results for Study 2 show that there is a significant difference between SNR measurements of single-slice and multislice acquisitions in slices 4, 5 and 6 ( P = .04, .05 and .006, respectively). The t
tests on whole-slice measurements demonstrate a significant difference in slices 4, 5 and 6 in Study 2; WM results indicate that slices 5 and 6 have significantly higher SNR in single-slice experiments than in multislice experiments. The greatest increase of SNR in whole-slice, GM and WM measurements was found in Study 3, when the tagging plane was moved to stay immediately adjacent (separation, 2 cm) to the single slice. It is apparent in Fig. 5C that the signal in superior slices in single-slice experiments is much greater than that in multislice experiments (for the same slice), and exceeds the signal acquired in Studies 1 and 2. Whole slice, GM and WM measurements for all slices are significantly different between multislice and single-slice acquisitions. Paired t tests were also performed to indicate whether SNR in slices 2– 6 differed from SNR in slice 1 for single-slice experiments. Fig. 5C demonstrates an increase in GM SNR in slices 2– 6 compared to slice 1 for Study 3. This is most likely due to the fact that the tagging plane is moved up towards the brain. After comparison to MR angiographic images, it seems that this superior movement of the tagging plane leads to the arteries in the circle of Willis being tagged. Therefore, it is possible that this greater degree of tagging leads to higher SNR in slices 2– 6. CBF results for single-slice and multislice experiments were plotted against slice number, similar to the SNR results. The graphs for whole slice, GM and WM for all three studies were nearly identical to SNR graphs (data not shown). WM displayed the most dramatic drop in CBF in multislice experiments when moving from inferior to superior slices. Whereas the average drop in WM CBF between the first three slices and the last three slices in multislice experiments was 38%, it was 23% and 10% for whole slice CBF and GM CBF, respectively. CBF was averaged across all six slices to obtain whole brain CBF results, as shown in Table 1. For Study 1, the single-slice whole-brain, GM and WM CBF are not significantly different from the corresponding multislice CBF values ( P = .53, .13 and .89, respectively). However, single-slice CBF in Study 2 is 18% (whole brain; P = .001), 16% (GM; P =.003) and 32% (WM; P =.003) larger than multislice CBF. In Study 3, single-slice CBF is 22% (whole brain;
Fig. 5. Mean GM SNRFS.D. in both single-slice and multislice acquisitions for (A) Study 1 (n = 8), (B) Study 2 (n = 12) and (C) Study 3 (n = 9). When the tagging plane location is fixed and TI 2 is held constant at 1500 ms (Study 1), single-slice GM SNR decreases with slice number by an amount similar to that in multislice studies. SNR in single-slice experiments is not significantly different from the corresponding slice in multislice studies. In Study 2, most of the signal lost in multislice acquisitions is recovered by acquiring multiple single slices. GM SNRs in slices 3, 4 and 5 in single-slice experiments are significantly higher than slices 3, 4 and 5 in multislice experiments (paired t test; *P b.05). In Study 3, the tagging plane is kept adjacent (2 cm) to single slices, hence moving up the brain for superior slices. This results in a marked increase in GM SNR for singleslice acquisitions (slices 2–5) compared to corresponding slices in multislice studies.
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876 Table 1 Average CBF (ml/100 ml/min) over six slices calculated for whole brain, GM and WM for multislice versus single-slice PASL imaging in Studies 1, 2 and 3 (paired t tests) CBF (ml/100 ml/min)
Multislice (n = 29) Study 1 (n = 8) Study 2 (n = 12) Study 3 (n = 9)
Multislice Single slice Single slice Single slice
Whole brain
GM
WM
49F5 48F17 58F144 60F154
70F7 63F20 81F174 79F174
19F4 21F12 25F134 25F154
4 P b.05.
P =.0003), 13% (GM; P = .0004) and 32% (WM; P = .02) larger than multislice CBF. 4. Discussion Many PASL schemes have been extended to multislice acquisitions and have been implemented in patient studies. However, accurate CBF quantification is difficult since there are differing blood transit times from the tagging plane to each imaging slice. Acquiring slices proximal to distal from the tagging plane means that additional T 1 decay of blood will have occurred in distal slices and blood signal will have been lower. This is qualitatively apparent in multislice image sets; however, it can be corrected for prior to CBF quantification. The introduction of quantitative imaging of perfusion using a single subtraction (Q2TIPS) [9] reduces the problem of various transit time delays. In Q2TIPS, a saturation pulse is applied to the tagging region at a time TI 1 in order to clip the tail end of the tagged bolus. Therefore, the time width of the bolus is known, and the sequence becomes insensitive to transit time. Consequently, Q2TIPS appears to be a suitable choice for multislice imaging and is the sequence used in the present report. As mentioned in the Introduction, a major difficulty in PASL multislice imaging is the destruction of blood magnetization destined for more distal slices by acquiring more proximal slices; this causes underestimation of blood flow. It has been proposed that, in order to overcome this problem, one can begin imaging with the superior slice [15]. However, this is not amenable to Q2TIPS imaging because of the necessary timing parameters involved with this technique. These timing parameters also limit the number of slices that can be acquired. In order to quantify CBF in a single subtraction, the following equation must hold: TI 2 N TI 1+yt, where TI 2 is the time of image acquisition and yt is the transit time from the tagging plane to the imaging slice. In other words, an image can only be acquired if the entire tagged bolus has been allowed to perfuse the slice and to exit by outflow. This equation is unlikely to hold if imaging is carried out with superior to inferior slice acquisition, or if too many slices are acquired. For these two reasons, we implemented an inferior to superior slice acquisition scheme while acquiring six slices.
875
Another problem with imaging superior slices prior to inferior slices is the loss/decay of tagged blood signal. The inversion time in such an experiment would have to be long in order to let the tag reach superior slices before imaging occurs. Hence, the inversion time for inferior slices would have to be even longer, and the tag will likely have already passed through these slices prior to imaging. Even if some tag remains, T 1 decay is likely to be extensive. The three studies carried out in this report help to demonstrate the extent to which the acquisition of inferior slices dampens the signal in superior slices. In particular, Study 2 is identical to multislice acquisition in all ways, except for the lack of acquisition of inferior slices. The tagging plane was applied at the same location regardless of slice position, and TI 2 for each slice in single-slice acquisitions was identical to TI 2 in multislice acquisitions. Results indicate that a large portion of the signal lost in multislice acquisitions is due to this effect. When comparing the first three slices to the last three slices in multislice acquisitions, there is a 21% drop in whole-slice SNR; however, 14% of whole-slice SNR is recovered using multiple single-slice acquisitions with parameters identical to those of multislice acquisition. The difference signal images acquired as single slices in Study 3, when the tagging plane is adjacent to the imaging slice and TI 2 is fixed at 1500 ms, show that the average GM SNR of the last three slices is 10% higher than the first three slices. This indicates that placing the tagging plane close to the imaging slice prevents blood from taking circuitous routes, increasing the amount of tagged blood reaching the slice by TI 2. As well, superior movement of the tagging plane may cause tagging of additional blood vessels, also increasing the size of the tag reaching the slice by TI 2. By comparison of Study 1 to Study 3, we are certain that the recovery of signal in Study 3 is not due to the shortening of TI 2 for more superior slices. Study 1 indicates that the effect of shortening TI 2 in all slices plays little role in the recovery of signal in more distal slices. TI 2 was shortened to 1500 ms in Study 3 so as not to miss the tagged bolus, since the distance between the tagging plane and the imaging slice was small (20 mm) and because no extra time was required for the tag to enter the slice. Mean CBF per slice, measured using multislice Q2TIPS with sequential slice acquisition, is lower in slices most distal to the tagging region. This is a drawback to 2D multislice ASL techniques, since GM and WM CBF should be similar in each slice throughout the brain. The results of this study show that the mean CBF of multiple slices, acquired using a single-slice technique with optimized parameters, might be more accurate than the mean CBF of multislice acquisitions, since CBF values are consistent over many slices. However, it is difficult to validate specific CBF values since CBF in different studies is quantified using variable acquisition parameters such as resolution, scan time and tagging methodology. A recent
876
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
study compares ASL CBF measurements to the current gold standard H215O positron emission tomography (PET) and concludes that ASL GM CBF values (64F12 ml/ 100 ml/min) are not statistically different from PET GM CBF values (67F13 ml/100 ml/min) [1]. Within the ASL literature, GM CBF ranges from 50.9F7.2 [16] to 64F 12 ml/100 ml/min [1], with GM/WM CBF ratios from 1.6 [1] to 3.2 [14]. The multislice and single-slice GM CBF values (70F7 and 80F17 ml/100 ml/min, respectively) for Studies 2 and 3, and the GM/WM CBF ratios (3.7 and 3.2, respectively) in the present study (Table 1) are high considering the ranges in the current literature. Recently, ASL has been implemented using 3D readout techniques to remove detrimental signal loss in 2D multislice schemes [14,16,17]. PASL with 3D GRASE readout schemes result in perfusion images with a mean GM SNR that is 2.8-fold higher than that with 2D EPI readouts (13.0F3.5 and 4.7F1.3 respectively), with equal nominal resolution and acquisition time [14]. As well, CASL 3D GRASE imaging at 3 T has been implemented with greater brain coverage in areas of high susceptibility and SNR comparable to CASL 2D EPI [16]. Q2TIPS with 2D EPI readout is a widely accepted sequence for perfusion quantification; however, resultant multislice data sets have lower SNR and CBF in superior slices due to T 1 decay and slice interference — a problem likely to occur with other PASL sequences using inferior to superior slice acquisition. This problem is amplified with an increasing number of slices. Therefore, minimizing slice number to cover only the region of interest (e.g., just the coverage of stroke highlighted on diffusion-weighted MRI) and placing the tagging region immediately adjacent to the slices of interest ought to be beneficial for maximizing the SNR of ASL perfusion images of the brain. Acknowledgment This work was supported by operating funds from the Heart and Stroke Foundation of Alberta, Northwest Territories and Nunavut, and the Canadian Institutes for Health Research (CIHR); by salary grants from the CIHR (C.B.), the Alberta Heritage Foundation for Medical Research (AHFMR) (C.B.) and the Natural Science and Engineering Research Council of Canada (A.C.); and by infrastructure grants from the Canada Foundation for Innovation, the AHFMR, the Alberta Science and Research Authority and the University Hospital Foundation. The authors would also like to thank Siemens Medical Systems for the ASL pulse sequence.
References [1] Ye FQ, Berman KF, Ellmore T, Esposito G, van Horn JD, Yang Y, et al. H(2)(15)O PET validation of steady-state arterial spin tagging cerebral blood flow measurements in humans. Magn Reson Med 2000;44:450 – 6. [2] Kwong KK, Chesler DA, Weisskoff RM, Donahue KM, Davis TL, Ostergaard L, et al. MR perfusion studies with T 1-weighted echo planar imaging. Magn Reson Med 1995;34:878 – 87. [3] 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. [4] Detre JA, Leigh JS, Williams DS, Koretsky AP. Perfusion imaging. Magn Reson Med 1992;23:37 – 45. [5] Edelman RR, Siewert B, Darby DG, Thangaraj V, Nobre AC, Mesulam MM, et al. Qualitative mapping of cerebral blood flow and functional localization with echo-planar MR imaging and signal targeting with alternating radio frequency. Radiology 1994;192: 513 – 20. [6] Kim SG. Quantification of relative cerebral blood flow change by flowsensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med 1995;34:293 – 301. [7] Schwarzbauer C, Morrissey SP, Haase A. Quantitative magnetic resonance imaging of perfusion using magnetic labeling of water proton spins within the detection slice. Magn Reson Med 1996;35: 540 – 6. [8] Wong EC, Buxton RB, Frank LR. Implementation of quantitative perfusion imaging techniques for functional brain mapping using pulsed arterial spin labeling. NMR Biomed 1997;10:237 – 49. [9] Wong EC, Buxton RB, Frank LR. Quantitative imaging of perfusion using a single subtraction (QUIPSS and QUIPSS II). Magn Reson Med 1998;39:702 – 8. [10] Wong EC, Buxton RB, Frank LR. A theoretical and experimental comparison of continuous and pulsed arterial spin labeling techniques for quantitative perfusion imaging. Magn Reson Med 1998; 40:348 – 55. [11] Zhang W, Silva AC, Williams DS, Koretsky AP. NMR measurement of perfusion using arterial spin labeling without saturation of macromolecular spins. Magn Reson Med 1995;33:370 – 6. [12] 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 – 96. [13] 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 – 54. [14] Gunther M, Oshio K, Feinberg DA. Single-shot 3D imaging techniques improve arterial spin labeling perfusion measurements. Magn Reson Med 2005;54:491 – 8. [15] Yang Y, Frank JA, Hou L, Ye FQ, McLaughlin AC, Duyn JH. Multislice imaging of quantitative cerebral perfusion with pulsed arterial spin labeling. Magn Reson Med 1998;39:825 – 32. [16] Fernandez-Seara MA, Wang Z, Wang J, Rao HY, Guenther M, Feinberg DA, et al. Continuous arterial spin labeling perfusion measurements using single shot 3D GRASE at 3 T. Magn Reson Med 2005;54:1241 – 7. [17] Talagala SL, Ye FQ, Ledden PJ, Chesnick S. Whole-brain 3D perfusion MRI at 3.0 T using CASL with a separate labeling coil. Magn Reson Med 2004;52:131 – 40.
Comparison of multislice and single-slice acquisitions for pulsed arterial spin labeling measurements of cerebral perfusion Alison M. Campbell, Christian Beaulieu4 Department of Biomedical Engineering, Faculty of Medicine and Dentistry, 1098 Research Transition Facility, University of Alberta, Edmonton, Alberta, Canada T6G 2V2 Received 4 January 2006; accepted 29 March 2006
Abstract Multislice Q2TIPS is a widely used pulsed arterial spin labeling (PASL) technique for efficient and accurate quantification of cerebral blood flow (CBF). Slices are typically acquired inferior to superior from a tagging plane. Superior slices show signal loss greater than the loss expected from blood T 1 decay. In order to assess the reasons for this additional signal loss, three single-slice acquisition studies were compared to multislice acquisition (six slices) in healthy volunteers. In Study 1 (n = 8), the tagging plane was fixed in location, and the inversion time (TI 2) was 1500 ms for each slice. For Study 2 (n = 12), the tagging plane was fixed as in Study 1; however, TI 2 increased as slices were acquired further from the tagging plane. In Study 3 (n = 9), the tagging plane was kept adjacent to the imaging slice, and TI 2 was 1500 ms for every slice. Gray matter (GM) and white matter (WM) signal-to-noise ratio (SNR) and CBF were measured per slice. GM SNR from single-slice acquisitions was significantly higher at slices 4 – 6 in Study 2 and at slices 2 – 6 in Study 3 compared to multislice acquisitions. Signal loss in distal slices of multislice acquisitions can be attributed to the destruction of tagged bolus in addition to blood T 1 decay. If limited brain coverage is acceptable, perfusion images with greater SNR are achievable with limited slices and placement of the tagging region immediately adjacent to the site of interest. D 2006 Elsevier Inc. All rights reserved. Keywords: Arterial spin labeling (ASL); Q2TIPS; Cerebral blood flow (CBF); Perfusion MRI
1. Introduction Arterial spin labeling (ASL) is a noninvasive imaging technique that has been proven to yield accurate measurements of cerebral blood flow (CBF) in cortical gray matter (GM) [1]. Arterial spins are labeled by changing their state of magnetization in an artery upstream of the site of interest. As these spins flow through the tissue vasculature, water exchange occurs between labeled blood and tissue. This causes a measurable change in both apparent tissue T 1 and tissue magnetization; either change can be measured and used to quantify blood flow. Essentially, perfusion is determined by acquiring one image with spin tagging and one image without spin tagging. Subtraction of these two images results in a difference signal image (DM) that is directly proportional to blood flow. In the early 1990s, it was shown that either pseudocontinuous saturation or flow-driven adiabatic inversion could 4 Corresponding author. Tel.: +1 780 492 0908; fax: +1 780 492 8259. E-mail address: [email protected] (C. Beaulieu). 0730-725X/$ – see front matter D 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2006.03.011
be used to tag blood water spins in the neck and to measure perfusion in a single slice within the brain [2–4]. Since the first attempts using continuous arterial spin labeling (CASL), many different arterial labeling schemes have been proposed, such as the pulsed arterial spin labeling (PASL) technique called EPISTAR, whereby a single radiofrequency pulse is used to label a thick slab close to the imaging slice of interest [5]. A 1808 inversion pulse is used to apply an upstream tag; after a delay time to allow the tagged blood to perfuse into the tissue, an image is acquired within the brain. As with CASL, the resultant DM image reflects the difference in magnetization of the blood between images acquired with and without the application of a tag pulse. Other PASL variants, such as FAIR [2,6,7], PICORE [8] and Q2TIPS [9], have also been successfully applied. Although the inherent signal-to-noise ratio (SNR) is higher in CASL, PASL has a higher inversion efficiency (close to 1) and a lower natural transit delay, making SNR efficiency between the two techniques similar [10]. With both CASL and PASL, the goal is to produce a tag and a control image where the signals from static tissues in
870
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
both images are identical. This is easy to accomplish with single-slice techniques; however, it is an increasingly complicated task when a multislice data set is required for greater brain coverage. Multislice acquisition involves application of a single magnetization preparation pulse to tag arterial blood followed by sequential acquisition of multiple slices downstream. The major complication in CASL multislice imaging is magnetization transfer (MT) effect. It is necessary that the tag image receive the same MT effects as the control image, and this case can only occur if a single slice is acquired. The most effective solution to this problem is to employ a two-coil CASL system [11]; however, this requires specialized hardware. For PASL, multislice acquisitions are difficult because of variable transit delays both within and between slices, and because of static tissue subtraction errors consistent with slice profile effects of tag and control pulses [12]. Because of these problems with multislice acquisitions, single-slice CASL and PASL techniques are ideal. However, cerebral perfusion results are often desired at multiple slice locations. To avoid this problem, one could acquire multiple single-slice images; however, this is time-consuming given that lengthy signal averaging is necessary to create perfusion maps with adequate SNR. Therefore, when time constraints are present, which is usually the case, multislice ASL acquisitions are necessary for greater brain coverage. It is well known that SNR in multislice scans decreases in more distal slices due to longer inversion times and greater T 1 decay of blood [13]. However, we have observed signal loss in distal slices, which is greater than the loss expected with blood T 1 decay. Several reasons have been proposed for this effect, including the destruction of tagged blood destined for more distal slices by the acquisition of proximal slices [8,9], inversion profile imperfection [8] and uncertainty of transit time [14]. The purpose of the present study is to determine the cause of additional signal loss in distal slices, beyond that resulting from blood T 1 decay, by comparing perfusion maps from various single-slice acquisitions to those derived from multislice acquisitions in healthy volunteers. This study looks specifically at a Q2TIPS pulse sequence with PICORE tagging and highlights several potential ASL protocol improvements for the study of neurological diseases.
2. Materials and methods 2.1. MRI Images were obtained from a group of 17 healthy volunteers (12 males, 5 females; age range, 23–34 years), all of whom gave written and informed consent. Three separate studies were conducted for this report, and each volunteer participated in one or more of the studies. Four volunteers participated in all three studies. There were eight volunteers in Study 1, 12 volunteers in Study 2 and
9 volunteers in Study 3. MR scanning was performed on a 1.5-T Sonata scanner (Siemens Medical Systems, Erlangen, Germany) equipped with gradient coils capable of an amplitude of 40 mT/m and a slew rate of 200 T/m/s. Q2TIPS with PICORE tagging (work in progress; Siemens Medical Systems) was utilized for all ASL perfusion imaging [13]. PICORE tagging involved slab-selective inversion interleaved with no inversion in order to create tag and control images. Gradient-echo (GE) echo planar imaging (EPI) was used to acquire all slices after the application of the labeling pulse. One multislice ASL scan and six single-slice ASL scans were acquired in each study. The multislice scan consisted of six contiguous 8-mm slices acquired inferior to superior, with an interslice delay time of 54 ms. The gap between an inversion band and the first imaging slice was set at 20 mm to ensure minimal interaction of slice profiles (Fig. 1A). The time of the first saturation pulse (TI 1) was fixed at 800 ms, and the total inversion time (TI 2) for the first slice was 1500 ms. This value of TI 2 was chosen as a result of previous experiments, which indicated that TI 2 =1500 ms is optimal for this age range of volunteers (23–34 years) in order to maximize tissue signal and to minimize intravascular signal. Other multislice Q2TIPS acquisition parameters included the following: field of view (FOV) =2222 cm; matrix = 6464; echo time (TE) =15 ms; repetition time (TR)= 2500 ms; slice repetition time (TR slice)= 54 ms; scan time= 3:52 min; tag–control pairs =50; bandwidth =3004 Hz/ pixel. The six slice locations in both the multislice and single-slice acquisitions were exactly aligned. In single-slice data sets, slice thickness, TI 1, FOV, matrix, TE, TR, scan time, number of tag–control pairs and BW were the same as those of the multislice data set. The major differences between single-slice and multislice acquisition parameters include the width of the gap between the inversion band and the imaging slice, and the TI 2 of each slice. Basically, the location of each slice was static; however, the position of the 10-cm inversion band and the timing of image acquisition (TI 2) varied between studies. In Study 1, the placement of the inversion band was fixed with respect to the volunteer. Therefore, the gap between the inversion band and the first imaging slice was 20 mm, and this gap became larger with increasing slice number (Fig. 1B). The distances between the inversion band and the imaging slices were 20, 28, 36, 44, 52 and 60 mm for slices 1–6, respectively. The time of image acquisition (TI2) was fixed at 1500 ms for every slice, regardless of the distance between the inversion band and the slice. In Study 2, the inversion band location was fixed, and the distances between the inversion band and the imaging slices were the same as in Study 1 (Fig. 1B). As superior slices were imaged, TI2 increased in accordance with the slice repetition time of multislice acquisitions (54 ms). Therefore, inversion times (TI2) in Study 2 for slices 1– 6 were 1500, 1554, 1608, 1662, 1716 and 1770 ms respectively (i.e., exactly the same as they would
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
871
Fig. 1. Location of the tagging plane with respect to the imaging slices. (A) The multislice study consists of six 8-mm contiguous slices placed 2 cm from the tagging plane. TI 2 increases incrementally per slice, with a slice repetition time of 54 ms. (B) For Studies 1 and 2, the placement of the tagging plane is fixed with respect to the volunteer. The difference between the two studies is that, in Study 1, TI 2 = 1500 ms for every slice, whereas in Study 2, TI 2 increases by 54 ms with every increase in slice number (identical to the multislice study). (C,D) In Study 3, the tagging plane is adjusted so that it is always located 2 cm from the imaging slice. TI 2 = 1500 ms for every slice in Study 3.
be in a multislice acquisition). In Study 3, placement of the inversion band varied so that its distance from the imaging slice was constant at 20 mm. As a result, TI 2 was also kept constant at 1500 ms (Fig. 1C and D). Several anatomical images were also acquired for each volunteer. An inversion recovery GE EPI image (TI =200 ms, TE =57 ms and TR = 5500 ms) was acquired at each ASL slice location in order to provide an image with adequate GM–white matter (WM) contrast for tissue segmentation. Two additional anatomical images were obtained in order to calculate R and M owm, which are variables necessary for the determination of M ob. A protondensity-weighted image was acquired (GE; a = 108, TE = 5 ms and TR =1000 ms), so that R (the ratio of the proton density of blood in the sagittal sinus to that in the WM) could be determined. M owm, the T 2*-weighted equilibrium signal of WM, was calculated from a single-shot GE EPI sequence with TE = 15 ms and TR = 20,000 ms. These sequences were acquired with the same receiver gain settings as the perfusion images. Together, R and M owm were used to
determine M ob (the T 2*-weighted equilibrium signal of blood) Mob ¼ RMowm expðð1=T2wm 4 1=T2b 4ÞTEÞ
ð1Þ
where T 2wm* is the T 2* of WM and T 2b* is the T 2* of blood [9]. M ob is unique to each volunteer and is necessary for the calculation of CBF. The total scan time per volunteer for one study, including anatomical images, six single-slice ASL scans and one multislice ASL scan, was approximately 30 min. 2.2. Data processing Using the Q2TIPS pulse sequence, we acquired a series of tag and control images that were motion-corrected and pairwise-subtracted to obtain individual ASL difference images. Using MRVision software (MRVision, Inc., Winchester, MA), each individual difference image was qualitatively assessed and removed if pairwise subtraction had not adequately deleted static tissue. The remaining individual difference images were added together and further processed in Matlab (The MathWorks, Inc., Natick,
872
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
MA) to create an overall signal difference image (DM image). Once DM has been produced, CBF ( f) (ml blood/ 100 ml tissue/min) was calculated using [9] DM ¼ 2Mob f TI 1 expð TI 2 =T1b ÞqðT1b ; T1t ; Tex ; f ; k; TI 2 Þ ð2Þ for TI 1 bs
and
TI2 NTI 1 þ yt
ð3Þ
where q is a correction factor accounting for the difference in T 1 decay as tagged water spins move from blood to tissue. In our study, q is assumed to be unity [13]. yt is the transit time of blood from the labeling plane to the imaging slice, and s is the time width of tagged bolus. T 1b is the longitudinal relaxation time of blood, which is assumed to be 1200 ms [15]. Inversion recovery anatomical images with good GM– WM contrast were used for the segmentation of perfusion difference images into GM and WM regions. Signal from whole slice, GM and WM was calculated for each slice in both multislice and single-slice scans. In Studies 1 and 3, the TI 2 for each slice was constant at 1500 ms. Therefore, the amount of blood T 1 decay is identical for all slices in these two single-slice studies. However, in single-slice scans in Study 2, as well as in all multislice acquisitions, TI 2 changes per slice. If not corrected for, SNR would decrease per slice in part because the T 1 decay of blood increases per slice. In order to accurately compare all studies, we corrected for blood T 1 decay prior to making the final SNR calculations. SNR was calculated for whole slice, GM and WM in all slices. In our report, SNR is calculated from DM images and is defined as the signal in the brain divided by the standard deviation of background noise. For multislice experiments,
SNR values for all volunteers were averaged per slice and compared to average SNR values for all volunteers in the corresponding single-slice experiment. In each study, the SNR in the first slice in both single-slice and multislice experiments was normalized to 1, as there are no methodological reasons for a difference in SNR in the first slice. For whole-slice measurements, the actual difference in slice 1 between single-slice and multislice experiments before normalization was 4.3% for Study 1, 0.7% for Study 2 and 16.6% for Study 3. For GM, the difference was 3.3% for Study 1, 1.9% for Study 2 and 12.2% for Study 3. Paired t tests were performed for all experiments (multislice and single-slice acquisitions) to determine if significant signal loss occurred in superior slices compared to the first slice. As well, paired t tests were used to show if any differences in SNR existed between corresponding slices acquired by single-slice and multislice methods within a study. P b.05 was considered a statistically significant difference. 3. Results Qualitative assessment of DM images in all volunteers demonstrates signal loss in superior slices in multislice as well as single-slice studies, albeit to differing degrees in the three single-slice studies (Figs. 2–4). It is not apparent in the initial evaluation of Study 1 (where a tagging plane was fixed in one position and TI 2 for each slice was held constant at 1500 ms) that any signal in superior slices is recovered (Fig. 2). In Study 2, the tagging plane is fixed in position (identical to Study 1); however, TI 2 is increased for slices farther from the tagging plane. Imaging superior slices at a later time allows more blood to reach these slices and allows perfusion signal to increase, as seen in Fig. 3. DM images in Study 3 clearly indicate that the signal in
Fig. 2. Perfusion difference images normalized to M ob (DM/M ob) for multislice and single-slice (constant TI 2 = 1500 ms) experiments in Study 1 (one volunteer). There is an obvious drop in signal with increasing slice number in multislice and single-slice experiments.
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
873
Fig. 3. Perfusion difference images normalized to M ob (DM/M ob) for multislice and single-slice (incremental TI 2) experiments in Study 2 (one volunteer). There is an obvious drop in signal with increasing slice number in the multislice experiment. The images from single-slice acquisitions appear to have greater perfusion signal.
superior slices is higher than the corresponding slice in the multislice experiment when the tagging plane is moved adjacent to the single acquired slice (Fig. 4). Although perfusion signal increases in superior slices for both Studies 2 and 3 with single-slice acquisition, the increase in signal intensity appears to be greater in Study 3. All perfusion experiments yielded DM images with adequate signal to perform SNR and CBF measurements. Difference image SNR values at slice 1 for all single-slice and multislice acquisitions were: 4.4F0.4 (whole slice), 5.6F0.7 (GM) and 1.8F0.5 (WM). The general pattern of GM SNR versus slice number for single-slice and multislice experiments in each study is demonstrated in Fig. 5. SNR measurements from whole slice and WM demonstrate nearly
identical patterns (data not shown). Although multislice acquisition results in all three studies are expected to be identical, deviation occurred due to different volunteers per study and due to inexact placement of tagging regions and slice locations. In Study 1, with a fixed TI 2 of 1500 ms per slice, GM SNR decreases in more distal slices by an amount similar to that in the multislice study (Fig. 5A). For each slice, there is no significant difference in GM SNR between single-slice and multislice experiments in Study 1. As well, there was no significant difference in whole-slice and WM SNR measurements. In Study 2, the location of the tagging plane was fixed; however, TI 2 was increased incrementally with more
Fig. 4. Perfusion difference images normalized to M ob (DM/M ob) for multislice and single-slice (adjacent tagging plane and constant TI 2 = 1500 ms) experiments in Study 3 (one volunteer). There is an obvious drop in signal with increasing slice number in the multislice experiment. There is a large increase in perfusion signal in single-slice images compared to multislice images.
874
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
superior slices. In Fig. 5B, GM SNR remains relatively constant for all slices in single-slice experiments. Therefore, additional delay (over Study 1) in acquiring more distal slices increases perfusion signal. GM results for Study 2 show that there is a significant difference between SNR measurements of single-slice and multislice acquisitions in slices 4, 5 and 6 ( P = .04, .05 and .006, respectively). The t
tests on whole-slice measurements demonstrate a significant difference in slices 4, 5 and 6 in Study 2; WM results indicate that slices 5 and 6 have significantly higher SNR in single-slice experiments than in multislice experiments. The greatest increase of SNR in whole-slice, GM and WM measurements was found in Study 3, when the tagging plane was moved to stay immediately adjacent (separation, 2 cm) to the single slice. It is apparent in Fig. 5C that the signal in superior slices in single-slice experiments is much greater than that in multislice experiments (for the same slice), and exceeds the signal acquired in Studies 1 and 2. Whole slice, GM and WM measurements for all slices are significantly different between multislice and single-slice acquisitions. Paired t tests were also performed to indicate whether SNR in slices 2– 6 differed from SNR in slice 1 for single-slice experiments. Fig. 5C demonstrates an increase in GM SNR in slices 2– 6 compared to slice 1 for Study 3. This is most likely due to the fact that the tagging plane is moved up towards the brain. After comparison to MR angiographic images, it seems that this superior movement of the tagging plane leads to the arteries in the circle of Willis being tagged. Therefore, it is possible that this greater degree of tagging leads to higher SNR in slices 2– 6. CBF results for single-slice and multislice experiments were plotted against slice number, similar to the SNR results. The graphs for whole slice, GM and WM for all three studies were nearly identical to SNR graphs (data not shown). WM displayed the most dramatic drop in CBF in multislice experiments when moving from inferior to superior slices. Whereas the average drop in WM CBF between the first three slices and the last three slices in multislice experiments was 38%, it was 23% and 10% for whole slice CBF and GM CBF, respectively. CBF was averaged across all six slices to obtain whole brain CBF results, as shown in Table 1. For Study 1, the single-slice whole-brain, GM and WM CBF are not significantly different from the corresponding multislice CBF values ( P = .53, .13 and .89, respectively). However, single-slice CBF in Study 2 is 18% (whole brain; P = .001), 16% (GM; P =.003) and 32% (WM; P =.003) larger than multislice CBF. In Study 3, single-slice CBF is 22% (whole brain;
Fig. 5. Mean GM SNRFS.D. in both single-slice and multislice acquisitions for (A) Study 1 (n = 8), (B) Study 2 (n = 12) and (C) Study 3 (n = 9). When the tagging plane location is fixed and TI 2 is held constant at 1500 ms (Study 1), single-slice GM SNR decreases with slice number by an amount similar to that in multislice studies. SNR in single-slice experiments is not significantly different from the corresponding slice in multislice studies. In Study 2, most of the signal lost in multislice acquisitions is recovered by acquiring multiple single slices. GM SNRs in slices 3, 4 and 5 in single-slice experiments are significantly higher than slices 3, 4 and 5 in multislice experiments (paired t test; *P b.05). In Study 3, the tagging plane is kept adjacent (2 cm) to single slices, hence moving up the brain for superior slices. This results in a marked increase in GM SNR for singleslice acquisitions (slices 2–5) compared to corresponding slices in multislice studies.
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876 Table 1 Average CBF (ml/100 ml/min) over six slices calculated for whole brain, GM and WM for multislice versus single-slice PASL imaging in Studies 1, 2 and 3 (paired t tests) CBF (ml/100 ml/min)
Multislice (n = 29) Study 1 (n = 8) Study 2 (n = 12) Study 3 (n = 9)
Multislice Single slice Single slice Single slice
Whole brain
GM
WM
49F5 48F17 58F144 60F154
70F7 63F20 81F174 79F174
19F4 21F12 25F134 25F154
4 P b.05.
P =.0003), 13% (GM; P = .0004) and 32% (WM; P = .02) larger than multislice CBF. 4. Discussion Many PASL schemes have been extended to multislice acquisitions and have been implemented in patient studies. However, accurate CBF quantification is difficult since there are differing blood transit times from the tagging plane to each imaging slice. Acquiring slices proximal to distal from the tagging plane means that additional T 1 decay of blood will have occurred in distal slices and blood signal will have been lower. This is qualitatively apparent in multislice image sets; however, it can be corrected for prior to CBF quantification. The introduction of quantitative imaging of perfusion using a single subtraction (Q2TIPS) [9] reduces the problem of various transit time delays. In Q2TIPS, a saturation pulse is applied to the tagging region at a time TI 1 in order to clip the tail end of the tagged bolus. Therefore, the time width of the bolus is known, and the sequence becomes insensitive to transit time. Consequently, Q2TIPS appears to be a suitable choice for multislice imaging and is the sequence used in the present report. As mentioned in the Introduction, a major difficulty in PASL multislice imaging is the destruction of blood magnetization destined for more distal slices by acquiring more proximal slices; this causes underestimation of blood flow. It has been proposed that, in order to overcome this problem, one can begin imaging with the superior slice [15]. However, this is not amenable to Q2TIPS imaging because of the necessary timing parameters involved with this technique. These timing parameters also limit the number of slices that can be acquired. In order to quantify CBF in a single subtraction, the following equation must hold: TI 2 N TI 1+yt, where TI 2 is the time of image acquisition and yt is the transit time from the tagging plane to the imaging slice. In other words, an image can only be acquired if the entire tagged bolus has been allowed to perfuse the slice and to exit by outflow. This equation is unlikely to hold if imaging is carried out with superior to inferior slice acquisition, or if too many slices are acquired. For these two reasons, we implemented an inferior to superior slice acquisition scheme while acquiring six slices.
875
Another problem with imaging superior slices prior to inferior slices is the loss/decay of tagged blood signal. The inversion time in such an experiment would have to be long in order to let the tag reach superior slices before imaging occurs. Hence, the inversion time for inferior slices would have to be even longer, and the tag will likely have already passed through these slices prior to imaging. Even if some tag remains, T 1 decay is likely to be extensive. The three studies carried out in this report help to demonstrate the extent to which the acquisition of inferior slices dampens the signal in superior slices. In particular, Study 2 is identical to multislice acquisition in all ways, except for the lack of acquisition of inferior slices. The tagging plane was applied at the same location regardless of slice position, and TI 2 for each slice in single-slice acquisitions was identical to TI 2 in multislice acquisitions. Results indicate that a large portion of the signal lost in multislice acquisitions is due to this effect. When comparing the first three slices to the last three slices in multislice acquisitions, there is a 21% drop in whole-slice SNR; however, 14% of whole-slice SNR is recovered using multiple single-slice acquisitions with parameters identical to those of multislice acquisition. The difference signal images acquired as single slices in Study 3, when the tagging plane is adjacent to the imaging slice and TI 2 is fixed at 1500 ms, show that the average GM SNR of the last three slices is 10% higher than the first three slices. This indicates that placing the tagging plane close to the imaging slice prevents blood from taking circuitous routes, increasing the amount of tagged blood reaching the slice by TI 2. As well, superior movement of the tagging plane may cause tagging of additional blood vessels, also increasing the size of the tag reaching the slice by TI 2. By comparison of Study 1 to Study 3, we are certain that the recovery of signal in Study 3 is not due to the shortening of TI 2 for more superior slices. Study 1 indicates that the effect of shortening TI 2 in all slices plays little role in the recovery of signal in more distal slices. TI 2 was shortened to 1500 ms in Study 3 so as not to miss the tagged bolus, since the distance between the tagging plane and the imaging slice was small (20 mm) and because no extra time was required for the tag to enter the slice. Mean CBF per slice, measured using multislice Q2TIPS with sequential slice acquisition, is lower in slices most distal to the tagging region. This is a drawback to 2D multislice ASL techniques, since GM and WM CBF should be similar in each slice throughout the brain. The results of this study show that the mean CBF of multiple slices, acquired using a single-slice technique with optimized parameters, might be more accurate than the mean CBF of multislice acquisitions, since CBF values are consistent over many slices. However, it is difficult to validate specific CBF values since CBF in different studies is quantified using variable acquisition parameters such as resolution, scan time and tagging methodology. A recent
876
A.M. Campbell, C. Beaulieu / Magnetic Resonance Imaging 24 (2006) 869 – 876
study compares ASL CBF measurements to the current gold standard H215O positron emission tomography (PET) and concludes that ASL GM CBF values (64F12 ml/ 100 ml/min) are not statistically different from PET GM CBF values (67F13 ml/100 ml/min) [1]. Within the ASL literature, GM CBF ranges from 50.9F7.2 [16] to 64F 12 ml/100 ml/min [1], with GM/WM CBF ratios from 1.6 [1] to 3.2 [14]. The multislice and single-slice GM CBF values (70F7 and 80F17 ml/100 ml/min, respectively) for Studies 2 and 3, and the GM/WM CBF ratios (3.7 and 3.2, respectively) in the present study (Table 1) are high considering the ranges in the current literature. Recently, ASL has been implemented using 3D readout techniques to remove detrimental signal loss in 2D multislice schemes [14,16,17]. PASL with 3D GRASE readout schemes result in perfusion images with a mean GM SNR that is 2.8-fold higher than that with 2D EPI readouts (13.0F3.5 and 4.7F1.3 respectively), with equal nominal resolution and acquisition time [14]. As well, CASL 3D GRASE imaging at 3 T has been implemented with greater brain coverage in areas of high susceptibility and SNR comparable to CASL 2D EPI [16]. Q2TIPS with 2D EPI readout is a widely accepted sequence for perfusion quantification; however, resultant multislice data sets have lower SNR and CBF in superior slices due to T 1 decay and slice interference — a problem likely to occur with other PASL sequences using inferior to superior slice acquisition. This problem is amplified with an increasing number of slices. Therefore, minimizing slice number to cover only the region of interest (e.g., just the coverage of stroke highlighted on diffusion-weighted MRI) and placing the tagging region immediately adjacent to the slices of interest ought to be beneficial for maximizing the SNR of ASL perfusion images of the brain. Acknowledgment This work was supported by operating funds from the Heart and Stroke Foundation of Alberta, Northwest Territories and Nunavut, and the Canadian Institutes for Health Research (CIHR); by salary grants from the CIHR (C.B.), the Alberta Heritage Foundation for Medical Research (AHFMR) (C.B.) and the Natural Science and Engineering Research Council of Canada (A.C.); and by infrastructure grants from the Canada Foundation for Innovation, the AHFMR, the Alberta Science and Research Authority and the University Hospital Foundation. The authors would also like to thank Siemens Medical Systems for the ASL pulse sequence.
References [1] Ye FQ, Berman KF, Ellmore T, Esposito G, van Horn JD, Yang Y, et al. H(2)(15)O PET validation of steady-state arterial spin tagging cerebral blood flow measurements in humans. Magn Reson Med 2000;44:450 – 6. [2] Kwong KK, Chesler DA, Weisskoff RM, Donahue KM, Davis TL, Ostergaard L, et al. MR perfusion studies with T 1-weighted echo planar imaging. Magn Reson Med 1995;34:878 – 87. [3] 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. [4] Detre JA, Leigh JS, Williams DS, Koretsky AP. Perfusion imaging. Magn Reson Med 1992;23:37 – 45. [5] Edelman RR, Siewert B, Darby DG, Thangaraj V, Nobre AC, Mesulam MM, et al. Qualitative mapping of cerebral blood flow and functional localization with echo-planar MR imaging and signal targeting with alternating radio frequency. Radiology 1994;192: 513 – 20. [6] Kim SG. Quantification of relative cerebral blood flow change by flowsensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med 1995;34:293 – 301. [7] Schwarzbauer C, Morrissey SP, Haase A. Quantitative magnetic resonance imaging of perfusion using magnetic labeling of water proton spins within the detection slice. Magn Reson Med 1996;35: 540 – 6. [8] Wong EC, Buxton RB, Frank LR. Implementation of quantitative perfusion imaging techniques for functional brain mapping using pulsed arterial spin labeling. NMR Biomed 1997;10:237 – 49. [9] Wong EC, Buxton RB, Frank LR. Quantitative imaging of perfusion using a single subtraction (QUIPSS and QUIPSS II). Magn Reson Med 1998;39:702 – 8. [10] Wong EC, Buxton RB, Frank LR. A theoretical and experimental comparison of continuous and pulsed arterial spin labeling techniques for quantitative perfusion imaging. Magn Reson Med 1998; 40:348 – 55. [11] Zhang W, Silva AC, Williams DS, Koretsky AP. NMR measurement of perfusion using arterial spin labeling without saturation of macromolecular spins. Magn Reson Med 1995;33:370 – 6. [12] 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 – 96. [13] 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 – 54. [14] Gunther M, Oshio K, Feinberg DA. Single-shot 3D imaging techniques improve arterial spin labeling perfusion measurements. Magn Reson Med 2005;54:491 – 8. [15] Yang Y, Frank JA, Hou L, Ye FQ, McLaughlin AC, Duyn JH. Multislice imaging of quantitative cerebral perfusion with pulsed arterial spin labeling. Magn Reson Med 1998;39:825 – 32. [16] Fernandez-Seara MA, Wang Z, Wang J, Rao HY, Guenther M, Feinberg DA, et al. Continuous arterial spin labeling perfusion measurements using single shot 3D GRASE at 3 T. Magn Reson Med 2005;54:1241 – 7. [17] Talagala SL, Ye FQ, Ledden PJ, Chesnick S. Whole-brain 3D perfusion MRI at 3.0 T using CASL with a separate labeling coil. Magn Reson Med 2004;52:131 – 40.