The study was approved by the Institutional Review Board and was HIPAA compliant. Patients were enrolled if they had symptoms consistent with cerebral ischemia (acute, subacute, or chronic) or transient ischemic attack (TIA), and signed written prior informed consent to participate in the study. Exclusion criteria was a level of consciousness score of 2 or greater as defined by the National Institutes of Health stroke scale; symptoms likely related to psychoactive drugs or patients with symptoms related to an active inflammatory disease such as AIDS, meningitis, or cerebritis; psychiatric or substance abuse disorder or dementia that interfered with evaluation or interpretation of the neurologic and mental assessment; severe coexisting or terminal systemic disease that limited life expectancy or otherwise interfered with the conduct of the study; symptoms related to an alternative diagnosis such as seizures or migraine; or use of any thrombolytic agent or acute stroke investigational drug therapy. Patients were recruited between October 2004 and July 2008.
Eighteen patients (9 men, 9 women; mean age 47±17 yrs, range 19-87 yrs) with cerebrovascular disease (4 acute stroke, 6 subacute stroke, 2 TIA, 6 Moyamoya; of these, 6 had unilateral internal carotid (ICA) occlusion, while 2 had bilateral ICA occlusion) were enrolled in the study and underwent both xeCT and MRI CBF measurements. The mean time difference between the two examinations was 18±10 hrs with a range of -21 to +34 hrs). In 6/18 patients, the MRI study preceded the xeCT study. Diffusion positive lesions representing acute or early subacute cerebral ischemia were seen in 9 of 18 patients (50%). describes the demographics of the patients included in the study.
Computed tomography (CT) was performed using a GE Lightspeed 8 detector scanner integrated with a stable xenon enhancer system (Diversified Diagnostic Products, Houston, TX, USA). The xeCT protocol imaged 4 contiguous 10 mm slices (80 kVp, 240 mA) with the lowest slice at the level of the basal ganglia. 8 sets of images were acquired at 45 s intervals. The first 2 timepoints were acquired during room air inhalation, while the remaining 6 timepoints were acquired during 28% Xe gas inhalation. End-tidal Xe concentration was assumed equal to arterial Xe concentration, a reasonable approximation except in patients with severe respiratory disease. CBF was calculated using the Kety autoradiographic method by the manufacturer's commercial software according to reference (18
), yielding CBF maps with a nominal in-plane 1 mm spatial resolution. The true in-plane resolution is on the order of 2-3 mm, and all image calculations (see below) were performed on regions-of-interest (ROI's) measuring 10 × 10 mm in-plane.
Dynamic susceptibility contrast MRI
MRI scans were performed at 1.5T (Signa LX/i, GE Medical Systems, Waukesha, WI, USA). Anatomic imaging was performed in addition to DSC, and always included fluid-attenuated inversion recovery (FLAIR) and diffusion-weighted imaging (DWI) with an isotropic b-value of 1000 s/mm2. DSC was performed using GRE EPI with the following parameters: 12 slices, thickness 7.5 mm, FOV 24 cm, flip angle 60°, TR/TE 2000/60 ms, matrix 128×128, 40 cine time points. A power injector was used to inject 20 ml of either gadopentetate dimeglumine or gadodiamide followed by 20 ml saline at a rate of 4 ml/s.
CBF maps were created using a block-circulant (delay-insensitive) singular value decomposition (SVD) in line with the work of Wu et al. (19
). Transverse relaxivity change (ΔR2*) was calculated using:
is the mean signal intensity before contrast, excluding the first 3 time points to ensure a steady-state value. To avoid subjective selection of AIF and VOF, we employed automatic selection, based on location, peak value, peak width, and contrast arrival time (20
). 10 voxels were used for the AIF and VOF ROIs (17
). This algorithm resulted in AIF locations in middle cerebral arteries, anterior cerebral arteries, basilar artery, or ICAs. The VOF locations were typically in the superior sagittal sinus, transverse sinus, or straight sinus. We did not account for possible clipping of the AIF or VOF curves due to MR signal saturation. In our experience, it is difficult to determine the presence of saturation based purely on shape, in accordance with theoretical work (14
Four separate post-processing corrections were applied to the DSC CBF maps: no corrections, PV correction only, BB correction only, and both PV and BB corrections. PV was defined as the ratio of the area under the AIF and the VOF curves (), using trapezoidal integration. These timecurves were sampled over the entire scanning period, with only the post-contrast segment contributing to the area (as the mean tracer concentration before bolus arrival is by definition zero). Corrections for the BB quadratic relaxivity relationship (described below) were applied before deconvolution. Given the linear relationship between the AIF underestimation and CBF (22
), the DSC CBF maps were multiplied by the PV level in each patient, such that PV correction led to reduced CBF.
Figure 1 (a) Example of the automated selection of AIF (red) and VOF (blue). Only 4 of 12 slices are shown, as these were the locations of the chosen AIF and VOF. (b) Concentration versus time curves. In this particular patient, the amount of AIF partial volume (more ...)
BB correction was performed according to references (13
). For the uncorrected images, a linear relationship between relaxivity and concentration was used for both the tissue and vessels (AIF and VOF) (13
where r = 0.044 (ms mM)-1
. For the BB corrected maps, a quadratic relationship was used for the AIF and VOF (13
where a = 7.6 × 10-3 (ms mM)-1 and b = 574 × 10-6 (ms mM2)-1, while the linear relationship in Eq. 2 was applied for tissue. BB correction leads to a relative increase in the estimated AIF concentration, resulting in decreased calculated CBF.
Within patients CBF measurements
Rigid body rotation based on mutual information using SPM2 (University College of London, available at www.fil.ion.ucl.ac.uk/spm/software/spm2
) was used to co-register the MR and CT images. A 1 × 1 cm square grid were laid over each of the 4 slices, resulting in about 125 individual 1 cc ROIs per slice, such that each patient's mean CBF measurement was calculated as the mean of about 500 small cubic ROIs. Voxels belonging to the ventricles and cortical sulcal CSF were excluded by manual thresholding the diffusion-weighted images. In each patient, scatterplots of the individual DSC and xeCT CBF ROI yielded slope, intercept, and correlation coefficient.
Between patients CBF measurements
To compare between patients, we chose to examine the global CBF, which we define as the mean CBF of all voxels within the co-registered volumes, as this will be independent of intrinsic spatial resolution. It should be noted that this represents a subset of the entire MR CBF dataset, since it covered a larger volume of brain than the xeCT measurements. For each patient, this yielded a single xeCT measurement and 4 separate DSC measurements, corresponding to each of the correction methods described above. To compare the two measurements, the CBF ratio was used:
Ideally, this value should be 1, representing exact correspondence (i.e., no bias) between the two techniques.
Once the CBF ratios for each of the post-processing conditions were calculated for each patient, comparisons between patients were performed. The precision of the measurement was measured using the coefficient of variation (COV), also known as the normalized between-patients standard deviation:
where the overbar represents the mean of all patients. A low COV represents a more precise measurement. Finally, linear regression was performed between the xeCT and each of the separate DSC CBF measurements.
Bland-Altman plots were created using the xeCT CBF (gold standard) measurement as the x-axis. Mean difference and 95% limits of agreement are reported. To assess for possible bias in the measurement based upon underlying xeCT CBF, we created rank-ordered maps. To test the significance of the global xeCT and various MRI-based CBF measurements in different patients, we have calculated the simple Pearson correlation coefficient and the corresponding p-value.