Dynamic susceptibility contrast magnetic resonance imaging (DSC-MRI) is used for the estimation of cerebral perfusion parameters such as cerebral blood volume (CBV), cerebral blood flow (CBF), and mean transit time (MTT) using the theory of intravascular tracers (1). Considering the ability of MRI to combine morphological and functional information during a single imaging session, this method can be used in both clinical and research environments. In the clinic, DSC-MRI is widely used for visualizing perfusion deficits in patients with vascular disease, particularly acute stroke, and for evaluating angiogenesis and blood volume in brain tumors.
High magnetic field strength (B0) has a number of theoretical advantages for DSC-MRI; the larger magnetization can be used either to create images with higher signal-to-noise ratio or higher spatial resolution, while the increased susceptibility effect (which scales linearly with field strength) increases contrast. The increased susceptibility effect can be used to reduce contrast agent (CA) dose or adjust imaging parameters (eg, shorter echo time [TE]) while maintaining similar contrast. However, when performing DSC-MRI at higher field strengths, susceptibility and chemical shift artifacts, B1 inhomogeneity, and concerns about specific absorption rate are amplified. However, with appropriate sequence design, these problems can be overcome. The purpose of this study was to design a DSC-MRI protocol for 7 T that provides images with a visual appearance similar to those seen at lower field strengths. An additional aspect of this study was to investigate whether CBV and CBF estimates, in absolute terms, show any apparent field strength dependence when applying a conventional DSC-MRI quantification approach, that is, assuming equal and concentration-independent R2* relaxivities in blood and tissue. Such an apparent CBV and CBF field-strength dependence would be related to previous predictions of an increased R2* relaxivity in tissue, accompanied by a larger difference in ΔR2*-versus-concentration relationships between tissue and large vessels, when moving from 1.5 T to 3 T (2). Hence, an additional aim of this study was to assess the apparent field-strength dependence of CBV and CBF by extracting quantitative perfusion estimates in healthy volunteers and comparing the 7 T estimates with corresponding literature values obtained at 1.5 T and 3 T.
In a typical DSC-MRI experiment, the limited spatial resolution implies that the arterial input function (AIF) data are likely to include signal components from the tissue surrounding the vessel (3). This may lead to distorted AIFs, and these partial volume effects (PVEs) often contribute to an overestimation of CBV and CBF when absolute quantification is performed. In addition, this source of overestimation may interfere with the field-strength-related overestimation effects to be evaluated in this study. Inclusion of a PVE-reduction approach in the analysis of the present data was therefore warranted (4–9). We used the PVE-reduction method referred to as “tail-scaling” (9), which uses the ratio of the steady-state concentration levels of the AIF and venous output function (VOF) as a scaling factor to correct for PVEs in the AIF.
In this study, the feasibility of DSC-MRI at 7 T was investigated in 8 healthy volunteers, and the apparent field-strength dependence of CBV and CBF after applying tail-scaling PVE correction was investigated through comparison with previously published literature values obtained by a similar methodology at 1.5 T and 3 T.
Materials and Methods
Subjects and MR Imaging
In total 8 healthy volunteers (males, 5; age, 24–50 years) were examined using an actively shielded 7 T MRI scanner (‘Achieva’, Philips Healthcare LLC, Best, the Netherlands) and a 32-channel phased-array head coil (Nova Medical, Wilmington, MA). The project was approved by the local ethics committee (The Regional Ethical Review Board in Lund), and written informed consent was obtained from each volunteer. Morphological images (MPRAGE, before and after CA administration) of each volunteer were examined by an experienced neuroradiologist (PCS) to exclude any pathology.
A reduced dose (ie, half of the conventional dose) of gadoterate meglumine CA (0.05 mmol/kg body weight, Dotarem, Guerbet, Paris, France) was administered at an injection rate of 5 mL/s, and the DSC-MRI experiment was performed using single-shot gradient-echo echo-planar imaging with the following parameters: in-plane spatial resolution = 2 × 2 mm2, section thickness = 3 mm, number of sections = 14–21, flip angle = 70°, repetition time = 1500 milliseconds, TE = 30 milliseconds, and SENSE acceleration factor = 2.0 with 120 dynamics. A 20-mL saline flush was injected at a rate of 5 mL/s after the CA administration. The parameters selected were similar to our clinical and research 3 T DSC-MRI protocol parameters, although with better resolution in the slice direction and with a slightly decreased in-plane resolution, giving us a voxel volume of 12 mm3 in our 7 T experiment compared with 15 mm3 in our 3 T protocol. As part of an extended MRI protocol, an additional 0.1 mmol/kg CA dose, not used in this study, was injected ∼9 minutes before the DSC-MRI CA injection.
Concentrations C from the tissue and artery were obtained using the relationship C(t) = −k · (1/TE) · ln[S(t)/S0], where S(t) is the signal at time t, S0 is the baseline (pre-CA) signal, and k is a constant reflecting the transverse relaxivity r2* when gradient-echo imaging is used. Because tissue and blood were presumed to have equal and concentration-independent relaxivities, k was set to 1. The arterial concentration time course, that is, the arterial input function, is denoted as AIF(t) below.
Hlarge and Hsmall are the hematocrit values in large and small vessels, respectively; ρ is the mean whole-brain tissue density; and the factor (1 − Hlarge)/[ρ (1 − Hsmall)] was set to 0.705 cm3/g (12). The measured tissue concentration–time curve Ct(t) was deconvolved with the AIF to obtain the tissue impulse response function R(t) (and its peak value Rmax). A block-circulant singular value decomposition algorithm (13), with a fixed cutoff value of 10%, was used for the deconvolution. A global AIF was obtained by locating 4–8 arterial pixels in the middle cerebral artery branches in the Sylvian fissure region and averaging the corresponding concentration curves. The individual arterial curves were required to show nondistorted peaks with an early rise and high maximum concentration (14). MTT was calculated using the central volume theorem, that is, MTT = CBV/CBF.
The VOF was obtained by taking the mean value of concentration–time curves corresponding to ∼3 pixels within the sagittal sinus (from 2 to 3 different slices). Geometric distortions and/or signal saturation in the venous first-passage bolus peak were interpreted as signs of a high fraction of vascular signal, so VOFs based on venous curves with distorted first-passage shape were favored for calibration, because these voxels are likely to show less PVEs than undistorted venous concentration–time curves. Another issue to take into account is that VOFs showing a signal pixel shift that is visible throughout the steady-state phase should be avoided, because such an effect implies that the steady-state signal in a given image pixel originates from a different tissue location compared with the pre-CA baseline, which affects the concentration calculation. Hence, the criteria for selecting sagittal sinus VOFs in this study were that the individual concentration curves should display distortions (eg, flat or double-peak) during the first passage of the CA bolus, as this implies high CA content in the voxel and thereby minimal PVEs, and show either a minimal or absent pixel shift throughout the steady-state phase (15).
Subject-specific correction factors using the tail-scaling approach were obtained by calculating the mean concentration level of data points registered at the steady-state phase (40 time points) of the AIF and by dividing this mean value by the mean concentration level of the data points registered at the steady-state phase of the VOF (40 time points). No part of the recirculation phase was included in these data points.
For the perfusion evaluation, gray matter (GM) CBV and CBF estimates from the tail-scaling approach were extracted from manually selected regions of interest (ROIs) in the thalamus. In a similar fashion, frontal white matter (WM) estimates were also retrieved using manually applied ROIs. The correction factors from tail scaling were thereafter applied to the CBV and CBF GM and WM ROI mean values to correct the estimates for PVEs.
A mean GM CBF estimate of (114 ± 20) mL/min/100 g and a mean GM CBV estimate of (12.6 ± 2.8) mL/100 g were obtained (n = 8). For WM, the estimated mean CBF value was (47.3 ± 9.9) mL/min/100 g and the CBV estimate was (6.3 ± 1.0) mL/100 g. The GM-to-WM ratios were 2.4 for CBF and 2.0 for CBV. The mean MTT estimates for GM and WM were 6.6 and 8.0, respectively. Note that no pronounced field-strength dependence is expected in the MTT calculation because relaxivity changes are divided out when using the central volume theorem.
With regard to image quality, Figure 1A shows representative 7 T CBF, CBV and MTT maps from one volunteer, with similar visual appearance as the DSC-MRI maps normally obtained at lower field strengths. Note that no obvious distortions are seen in the images despite being acquired at 7 T. In addition, the CBV and CBF contrast between GM and WM is clearly visible. For comparison of image quality, a 3 T DSC-MRI data set from Knutsson et al.'s study (15) is also displayed (Figure 1B).
Figure 2 shows an example of AIF and VOF concentration–time curves in one subject. The VOF concentration–time curve shows peak distortions and a noisier steady-state concentration phase. Note that the last 40 time points were used for the tail-scaling PVE correction
To the best of our knowledge, this is the first report showing that DSC-MRI is feasible at 7 T. Although numerous challenges occur when moving from 3 T to 7 T MRI, these issues can, as shown in this study, largely be remediated by appropriate protocol designs. The question whether there is any advantage of using an ultrahigh field system for clinical perfusion research is also relevant in this context, and a reasonable answer would be to point out that one only needs half the normal amount of the CA, which may reduce the costs and also the potential risks associated with nephrogenic systemic fibrosis and the effects of long-term tissue accumulation after repeated use of gadolinium, as recently reported by the Food and Drug Administration (16).
The primary finding of this study was the elevated CBV and CBF estimates compared with those given in prior literature values at 1.5 T and 3 T, which may be related to differences in tissue and blood R2* relaxivity as a function of field strength. In the present study, mean absolute CBF estimates of 114 mL/min/100 g in GM and 47.3 mL/min/100 g in WM were obtained, whereas mean absolute CBV estimates were 12.6 mL/100 g in GM and 6.3 mL/100 g in WM. These estimates should be compared with data from previous studies using tail scaling (Table 1). In healthy volunteers, Knutsson et al. observed an absolute CBF value of 80 mL/min/100 g in GM (CBV and WM data not available) (15). Bjørnerud et al. (9) applied tail scaling to patient DSC-MRI data obtained at 1.5 T and observed CBF values of 58 mL/min/100 g in GM and 30 mL/min/100 g in WM, whereas the CBV values were 4.57 mL/100 g in GM and 2.3 mL/100 g in WM.
|1.5 T (Ref. 9)||3 T (Ref. 15)||7 T (this study)|
|CBF GM [mL/min/100 g]||58||80||114|
|CBF WM [mL/min/100 g]||30||47.3|
|CBV GM [mL/100 g]||4.57||12.6|
|CBV WM [mL/100 g]||2.3||6.3|
There are differences and similarities between these 3 studies. The study at 1.5 T included 101 patients, and the experiments were conducted using a CA dose of 0.2 mmol/kg with a spatial voxel volume of 16 mm3. Furthermore, the VOF was selected using automated cluster analysis, and, similar to the present approach, the steady-state phase of the concentration–time curve was used together with the same tracer-kinetic model and deconvolution method. However, a difference between the present study and the 1.5 T study was that Bjørnerud et al. made a gamma-variate fitting to the measured concentration and time. The 3 T study included 20 healthy volunteers and the VOF was selected manually, as in the present study, but the tail-scaling data points, for both AIFs and VOFs, also included the recirculation peak. In addition, the CA dose was 0.1 mmol/kg in the 3 T study, and the spatial resolution was slightly lower than that in the present study.
The 7 T GM-to-WM CBF and CBV ratios of the present study were 2.4 and 2.0, respectively, compared with those of the 1.5 T study (9), which showed ratios of 1.9 and 2.0, respectively. In a study by Wirestam et al. (17), at 1.5 T, the GM-to-WM CBF ratio was 2.2 using standard singular value decomposition. The comprehensive PET study by Leenders et al. (18) showed GM-to-WM ratios of 2.5 for CBF and 1.9 for CBV, which are in good agreement with the present study. From these results, no conclusions can be drawn regarding any field-strength dependence of relative DSC-MRI values in different tissue types.
With regard to the absolute CBV and CBF estimates, the present study results and those of the comparisons with previous studies are not entirely straightforward to interpret because CA dose, TE, and spatial resolution differed between the 1.5 T (9), 3 T (15) and 7 T investigations. Kjølby et al. (2) used simulations to calculate an r2* value of 44 mM−1−1 at 1.5 T and 87 mM−1−1 at 3 T, that is, indicating a substantial linear field-strength dependence of the tissue relaxivity. Wirestam et al. (19) estimated the in vivo R2* relaxivity at 3 T to be ∼32 mM−1−1, based on a comparison with a reference arterial spin labeling CBF method. In two studies investigating oxygenated whole-blood relaxivity (20, 21), a nonlinear relationship was observed between the concentration C and ΔR2*, contrary to the linear ΔR2*-versus-C relationship found typically in tissue. One plausible cause of CBV and CBF overestimations is that relaxivities of blood and tissue differ (otherwise k, defined above, would be the same for blood and tissue and cancel out in the kinetic equations), and a CBV and CBF field-strength dependence requires that this difference changes with increased B0. When CA is added to the plasma volume, the susceptibility differences between the plasma compartment and its surroundings will change, leading to altered local gradients and thereby altered R2*. The topography of the capillary–tissue interface differs substantially from the plasma–red blood cells interface in whole blood, and the R2* relaxivity of CA residing in intravascular plasma is thus dependent on the local environment, and it has been shown to be higher in tissue compartments than in whole blood (2). At higher field strengths, the susceptibility effects increase proportionally, and the change in R2* at a given CA concentration will be more prominent (ie, the R2* relaxivity is increased), and the increased susceptibility effects also imply that the difference in CA-induced ΔR2* between tissue and blood is enhanced at a higher field strength (2), which may explain our elevated estimates compared with literature values obtained at lower field strengths.
Furthermore, in whole blood, the susceptibility difference between red blood cells and plasma varies with CA concentration in a rather complex manner (22), leading to a nonlinear ΔR2*-versus-C relationship, implying that absolute estimates according to equations (1)–(2) also depend on CA dose. In theory, the blood-versus-tissue differences in CA response are reduced at higher dose because of the nonlinear ΔR2*-versus-C relationship in blood (2), and the low dose used at the 7 T would thus increase this difference. However, it is not evident that the registered AIF signal originates from whole blood only, and in such a case, the exact shape of the ΔR2*-versus-C relationship is unknown.
A drawback of this quantification approach at ultrahigh field strength is that venous steady-state pixel shifts are more pronounced in 7 T images, making it more difficult for the user to find suitable VOFs. An approach that would facilitate VOF registration would be to use a prebolus injection with a low dose, as in the study by Knutsson et al. (8). Alternatively, a sufficiently high bandwidth would remove the pixel shifts, but it would be difficult to achieve total volume coverage with reasonable temporal resolution (∼1.5 seconds) with high-bandwidth readout techniques. The problem of competing T1 effects (1) may also pose a problem because of the longer T1 in blood at 7 T, resulting in AIF and VOF underestimations.
In conclusion, a well-designed protocol can reduce susceptibility artifacts at ultrahigh field strength, and perfusion maps of clinically useful quality, at reduced CA dose, can be obtained. The study further indicated that an apparent field-strength dependence of absolute CBV and CBF estimates may be found, potentially related to an increased tissue R2* relaxivity accompanied by a larger blood-versus-tissue difference in ΔR2*-versus-concentration relationships at higher field strengths. Obviously, true CBV and CBF are not field strength-dependent and this means that the theory for quantifying CBV and CBF from DSC-MRI experiments remains incomplete.