To analyze the precision of cerebral blood flow (CBF) measurements made with continuous arterial spin labeling(CASL) perfusion magnetic resonance imaging (MRI) over experimentally relevant intervals.
To analyze the precision of cerebral blood flow (CBF) measurements made with continuous arterial spin labeling(CASL) perfusion magnetic resonance imaging (MRI) over experimentally relevant intervals.
CASL perfusion MRI measurements of CBF on a 1.5-T GE Signa magnet were repeated in young healthy male and female subjects at one hour and one week. Precision of the measurement was evaluated at both time intervals.
CASL perfusion MRI measurements of CBF yielded within-subject coefficients of variation (wsCV) of 5.8% for global and 13% for individual vascular regions when measurements were repeated within one hour. Differences in these values represent the error in post-processing. Global and regional CBF measurements over one week yielded wsCVs of 13% and 14%, respectively. At one week, error secondary to physiologic variability affected global and regional measurements to the same degree and masked the software post-processing error seen at one hour. The magnitude of the difference in repeated measures correlated with the magnitude of the measurement.
CASL perfusion MRI CBF measurements are accurate and precise. Variability over longer periods of time appears attributable to physiologic factors. Repeatability of the CASL measurement is sensitive to the magnitude of the measurement. This should be taken into account when studies requiring repeated measures involve subjects with significant variability in CBF. J. Magn. Reson. Imaging 2003;18:649–655. © 2003 Wiley-Liss, Inc.
CONTINUOUS ARTERIAL SPIN labeling(CASL) perfusion magnetic resonance imaging (MRI) provides a noninvasive means of measuring global and regional cerebral blood flow(CBF). In this technique (1), protons on water molecules in flowing arterial blood are magnetically inverted using a plane proximal to the brain. Labeled arterial spins decay with T1, which is sufficiently long to allow visualization of perfusion in brain parenchyma. The effects of arterial spin labeling are determined by comparison with control conditions where arterial water is not labeled. Because the CASL effect is small, on the order of 1%, multiple control/label pairs are acquired during each study for signal averaging, with each low resolution perfusion image obtained during approximately five to six minutes at 1.5 Tesla. Using this technique, measurements of CBF may be made sequentially without contamination from tracer from previous measurements. The ability to make frequent measurements may be particularly appealing to those interested in tracking the effects of behavioral state changes, pharmacologic manipulation, or therapeutic intervention on global or regional CBF.
Ye et al (2) recently assessed the accuracy of the CASL perfusion MRI technique compared to H215O positron emission tomography (PET) and found that the results compared very favorably. Yen et al (3) recently examined the reproducibility of a flow sensitive alternating inversion recovery pulsed arterial spin labeling technique(FAIR) over a 24-hour epoch at baseline and after acetazolamide challenge. They found that test-retest differences in baseline CBF varied by 23% in gray matter and 27% in white matter over a period of 24 hours.
Our intention in this study was to analyze the precision of global CBF measurements made with the CASL perfusion MRI technique over periods of time that might typically be used in experimental studies: one hour and one week. Also, because we are interested in the application of CASL perfusion MRI to the study of cerebrovascular disorders, we examined the reproducibility of regional blood flow measurements in the three major vascular distributions in each hemisphere.
After approval from the institutional review board and after informed consent, imaging was performed on five men and seven women between the ages of 20 and 41 years (27 ± 8 years) with no known cerebrovascular disease. Two CASL image sets were obtained one hour apart. A subsequent group of 11 healthy subjects, which included six women and five men between the ages of 19 and 38 years (27 ± 7 years), underwent two CBF imaging sessions separated by a period of 6 ± 3 days. Subjects were removed from and repositioned in the scanner between image sets in the one-hour study as had been done in the one-week study. Subject retention was 100%.
All CASL perfusion MRI studies were performed in a 1.5-T GE Signa Echospeed scanner (GE Medical Systems, Milwaukee, WI) as previously described (1, 4–6). CASL control labeling was applied at the level of the cervico-medullary junction using a post-labeling delay of 1.5 seconds (7). Images were acquired using a gradient echo-planar sequence (field of view [FOV] = 24 × 15 cm, matrix 64 × 40, bandwidth 62.5 kHz, TR/TE = 4000/22 msec, slice thickness = 8 mm with a 2-mm gap). Eight slices were acquired from inferior to superior in an interleaved order, and each slice acquisition took 45 msec. The imaging volume was chosen to include supratentorial structures. Each perfusion measurement consisted of 90 acquisitions with a scan time of approximately six minutes.
CBF quantification data were reconstructed off line and corrected for geometric distortion (8) using custom software written in the Interactive Data Language (Research Systems, Boulder, CO). The raw images in each scan were separated into 45 pairs of label and control images and then pair-wise subtracted. The resulting series of 45 perfusion difference images were corrected for motion and physiologic fluctuation using an algorithm based on principal component analysis (9), followed by averaging across the image series to produce a single set of perfusion sensitive images. Absolute CBF was quantified from the mean difference of perfusion images using a modification (4) of the approach described previously (7), where the intensity of cerebrospinal fluid (CSF), corrected for partial saturation, is now used as a calibration standard to correct for sensitivity of the perfusion image to water spins. This is in place of the averaging of this effect at each pixel, which tended to impair the quality and consistency of the CASL images. The following equation (4) is thus used:
where SCASL is the difference between the control and labeled image intensities, SCSF is the average intensity of the control image in the manually defined ventricular region, T1CSF (4.2 seconds) is the longitudinal relaxation time of CSF, T1a and T2a are the longitudinal and transverse relaxation times of arterial blood, w is the post-labeling delay (1.5 seconds), α is the labeling efficiency (71%), λ is the water fraction of arterial blood (0.76), and ρ is the density of brain tissue (1.05g/mL). T1a and T2a are assumed to be constant for this study at T1a = 1100 msec and T2a = 240 msec. Use of fixed T1a rather than measuring tissue T1 assumes that the labeled spins remain primarily in the vasculature and microvasculature rather than exchanging completely with tissue water. An example of quantitative perfusion images obtained from a six-minute acquisition is shown in Fig. 1a.
Whole brain CBF (WBR) was determined by averaging perfusion values across all brain voxels. Segmentation of perfusion images into major vascular territories was achieved using an automated region of interest (ROI) analysis based on published templates (10) following manual transformation of CASL perfusion images into Talairach space using T1-weighted anatomic images obtained concurrently. Three primary vascular distributions (10) in each hemisphere were examined in this study: leptomeningeal anterior cerebral artery (RACA, LACA), leptomeningeal middle cerebral artery (RMCA, LMCA), and leptomeningeal posterior cerebral artery (RPCA, LPCA). Territories contain both gray and white matter. These territories are shown in Fig. 1b.
Criteria for conducting a reproducibility analysis on a measurement technique requires that the data be normally distributed, that the magnitude of any difference between measurements be independent of the magnitude of the measurement, and that the distribution of the data between measurements be comparable so as to exclude significant change in the physiologic status of the subject that might create error independent of the test error (11). Statistical guidelines for determining measurement error or precision have been described by Bland and Altman (12–14).
For each ROI (WBR, RACA, RMCA, RPCA, LACA, LMCA, LPCA), the distribution of differences between repeated measurements was examined for normality using box-plots and the Shapiro-Wilk W-test. Additionally, to determine whether the magnitude of any difference in repeated measurements was dependent upon the amplitude of the measurement, we performed a nonparametric analysis of the differences in measurements vs. the mean using a ranked correlation coefficient (Kendall's τb). Lastly, we compared the distributions between both examinations using Wilcoxon's signed rank test because wide fluctuations in distributions between measurements could indicate error due to physiologic changes between examinations rather than error in the measurement technique. Shapiro-Wilk, Kendall's τb, and Wilcoxon's signed rank results are presented for each ROI. P-values for the Wilcoxon's signed rank test are based on a two tailed z-test.
If a distribution appeared abnormal or the magnitude of the difference increased with the magnitude of the measurement, a log10 transformation of the data was conducted (13). Further analysis was conducted on the log10 data if the model assumptions were met after transformation. If logarithmic transformation did not normalize the distribution or eliminate a relationship between the magnitude of the difference and the magnitude of the measurement, original values are presented along with appropriate annotation.
The following calculations were made (12):
Analysis of variance (ANOVA) and students t-test were used to describe the difference in CBF between men and women. Statistical analysis was performed with JMP Professional software (SAS, Cary, NC).
|Regional Means ± SD (mL/100g/minute)|
|Baseline (N = 12)||56.6 ± 11.5||52.1 ± 12.7||64.7 ± 14.6||59.8 ± 13.6||55.1 ± 11.9||53.7 ± 12.5||53.7 ± 10.4|
|One hour (N = 12)||55.8 ± 12.7||55.1 ± 15.4||68.9 ± 14.6||64.1 ± 18.4||59.4 ± 18.9||55.3 ± 17.4||55.1 ± 16.6|
|All measurements (N = 24)||56.2 ± 11.9||53.6 ± 13.5||66.8 ± 21.2||54.6 ± 20.0||57.3 ± 15.6||54.5 ± 14.8||54.4 ± 13.6|
|Assessment of model assumptions|
|Shapiro-Wilk||P = .82||P = .002a||P = .18||P = .23||P = .41||P = .67||P = .28|
|Kendall's τ||P = .30||P = .89||P = .89||P = .04b||P = .41||P = .49||P = .13|
|Wilcoxon's signed rank test||P = .71||P = .62||P = .84||P = .02c||P = .51||P = .89||P = .93|
|Repeatability (α = .05) (mL/100g/minute)||8.38||.14||25.3||17.4||20.5||17.9||21.4|
|Regional Mean ± SD (mL/100g/minute)|
|Baseline (N = 11)||55.1 ± 13.7||50.6 ± 15.4||71.9 ± 14.8||67.4 ± 16.6||64.4 ± 14.6||57.6 ± 16.6||57.9 ± 14.4|
|One week (N = 11)||56.0 ± 16.0||50.0 ± 16.6||72.7 ± 16.9||67.5 ± 18.6||66.1 ± 16.6||58.6 ± 17.9||60.2 ± 15.0|
|All measurements. (N = 22)||55.5 ± 14.5||50.3 ± 15.6||72.3 ± 15.5||67.4 ± 17.2||65.2 ± 15.3||58.1 ± 16.7||59.0 ± 14.4|
|Assessment of model assumptions (N = 11)|
|Shapiro-Wilk||P = .19||P = .35||P = .14||P = .04a||P = .59||P = .29||P = .62|
|Kendall's τ||P = .04b||P = .35||P = .06b||P = .43||P = .04b||P = .53||P = .35|
|Wilcoxon's signed rank test||P = .95||P = .79||P = .80||P = .95||P = .95||P = .90||P = .79|
|Repeatability (α = .05)||18.8||20.5||22.3||22.8||24.0||19.9||20.6|
For the one-hour experiment, a log10 transformation was carried out for the RACA only after box plot and Shapiro-Wilk test indicated an abnormal distribution. For the RMCA, Kendall's τb confirmed that there was a significant rank correlation between the magnitude of the test differences and the magnitude of the measurement. Log10 transformation of the data for this region did not eliminate this correlation; therefore the original untransformed data are presented. For the one-week experiment, the box plot and Shapiro-Wilk test indicated an abnormal distribution for the RMCA only. Kendall's τb confirmed a correlation between the magnitude of the test differences and the magnitude of the measurement for WBR, RMCA, LACA, and LMCA. Log10 transformation of the original data in these instances did not adequately correct this correlation, and therefore the original data are presented for all regions. Deviation from the strict criteria for a reproducibility analysis are indicated in Table 1 and Table 2. Plots of the absolute test difference in CBF measurements vs. magnitude of the measurement of CBF for the right hemispheric vascular regions are shown in Fig. 2.
For the one-hour experiment, the whole brain estimate of CBF demonstrated a wsCV of 5.8%, while smaller vascular regions demonstrated a mean wsCV of 13% ± 2%, range 12–15% (Table 1). For the one-week experiment, the whole-brain estimate of CBF demonstrated a wsCV of 13%, while smaller vascular regions demonstrated a mean wsCV of 14% ± 2%, range 12–17% (Table 2).
Repeatability ranged from 8 mL/100 g/minute for WBR to 25 mL/100 g/minute for the LACA in the one-hour experiment (Table 1). In the one-week experiment, repeatability ranged from 19 mL/100 g/minute for WBR to 24 mL/100 g/minute for LMCA (Table 2).
Average CBF for all participants over all time periods in this study was 55.9 ± 13.1 mL/100 g/minute (N = 46 total measurements). Average CBF for men in this study was 45.9 ± 10.7 mL/100 g/minute (N = 20 measurements) and 63.6 ± 8.8 mL/100 g/minute for women (N = 26) (P < 0.0001).
Values obtained for global or WBR CBF using the CASL perfusion MRI technique are consistent with previous measurements validated against 15O PET (2). Average WBR CBF in this study was 45.9 ± 10.7 mL/100 g/minute for men and 63.6 ± 8.8 mL/100 g/minute for women, again consistent with previous measurements describing sex differences in CBF using 133Xenon inhalation (15). Thus, this technique is accurate when compared to other accepted modalities, including 15O PET and 133xenon inhalation.
To evaluate the precision of a measurement technique, measurements must occur over a time frame or in an environment where other influences such as physiologic variations are minimal. We have evaluated two time frames, one hour and one week. We would expect that physiologic and environmental variation would be minimized by the shorter time frame, yet variation due to varying degrees of wakefulness or head motion, even during a one-hour study, cannot be entirely eliminated. Precision in CBF measurements using CASL perfusion MRI is also dependent on noise related to the scanning technique, including scanner condition and experimental environment. The effects of scanner noise on the accuracy of MR measurements are complicated and not well understood. Drift in the MR baseline has been reported by several groups primarily based on gradient-echo echo planar imaging (EPI) techniques (16–19). This effect can be well characterized as having power that varies inversely with temporal frequency (16); in other words, the longer the time span between measures, the greater the fluctuation in baseline. However, due to the pair-wise subtraction of adjacently acquired images and the subsequent calibration process to produce an absolute measure of CBF, the slow drift effects are eliminated in arterial spin labeling (ASL) (19). Perfusion-based functional MRI studies of sensorimotor activation over a spectrum of time periods from 30 seconds to 24 hours demonstrated consistent sensitivity of ASL contrast in all tested conditions (20), suggesting ASL technique per se is unlikely to be the primary reason for the difference in repeatability between the two time frames in our experiment. Finally, precision in these measurements is potentially subject to noise introduced by the post-processing technique, for example due to errors in normalization, segmentation, and identification of CSF used in quantification. It is reasonable to assume that error due to software analysis would be relatively constant in both one-hour and one-week studies because subjects were repositioned in the scanner under both conditions. Variability in repeatability between the two time frames would therefore primarily reflect physiologic variation.
One of our goals in this study was to determine the precision of this technique over a one-hour epoch of time. This time frame might be used in physiologic experiments on CBF, such as when one desires to examine the effects of various levels of CO2, O2, or anesthetics on cerebrovascular reactivity. It is clear from our results that the CASL perfusion MRI technique is associated with a high degree of precision when determining whole brain cerebral blood flow, as the wsCV was 5.8%. This level of precision closely approximates more invasive techniques claiming differences between measurement of 2.5–3% (21, 22) and would allow one to, in a very sensitive fashion, determine meaningful physiologic changes over this period of time. The precision of the measurements in the vascular regions was predictably less accurate, ranging from 10–14%, even over the one-hour time frame. Obtaining these measurements involves the automated application of vascular templates on the brain after manual transformation of CASL perfusion MRI data into Talairach space. Additionally, the CBF values for these vascular regions represent averaging of signal from much smaller populations of pixels.
We can see that the precision of the WBR measurement declined as the wsCV increased from 5.8% (Table 1) at one hour to 13% at one week (Table 2). Interestingly, the precision of the CBF measurements obtained from the vascular regions at the one-week interval, which ranged from 12–16%, was not significantly different from that obtained in the one-hour test, and the WBR wsCV closely matched the wsCV for the smaller vascular regions. We speculate that the difference in WBR precision between the one-hour and one-week experiments may be due to greater physiologic variability over the longer time frame. We further speculate that errors in normalization and segmentation, important in the regional analysis, likely dominate physiologic variability for the vascular region data.
It is apparent from the data in some ROIs, including WBR, that the magnitude of the difference between CBF measurements was affected by the magnitude of the measurement. Figure 2 demonstrates the pervasive relationship between the magnitude of any difference in measurements and the magnitude of the measurement during both the one-hour and one-week time interval. This relationship is consistent during both time intervals. Using our lowest initial CBF from the one-week test (31.8 mL/100 g/minute) and highest initial CBF from the same data group (74.1 mL/100 g/minute), we might expect a range of wsCV from 7–21%, depending on the magnitude of the measured CBF. The relationship between the magnitude of the difference and magnitude of the measurement is not totally unexpected because CASL perfusion MRI essentially measures blood flow and is inevitably affected by flow related physiologic fluctuations, most of which are magnified by increased flow. It is well known that CBF is higher in women than in men (15) and in states of severe anemia (23), and thus one might expect to see a higher coefficient of variation in these situations.
In conclusion, CASL perfusion MRI provides accurate and precise measurements of CBF, particularly over short periods of time. Variability over longer periods of time appears to be attributable to physiologic factors rather than measurement errors. Repeatability of the CASL measurement is sensitive to the magnitude of the measurement. This should be taken into account when studies requiring repeated measures involve subjects with large variability in CBF, to include sex differences. The CASL technique, by virtue of its high level of precision, noninvasive nature, and brevity of half-life of the label, is ideally suited for physiologic studies over short periods of time.
The authors thank Joseph A. Maldjian, MD, and David Alsop, PhD, for their technical assistance on this project.