We found significant differences in uncorrected and leakage-corrected rCBV when comparing low- and high-grade pediatric brain tumors scanned prior to treatment at multiple centers. This finding is despite large differences in the dynamic susceptibility-contrast MRI protocols employed across centers. While other single-center pediatric studies have shown that rCBV can discriminate between high- and low-grade brain tumors [1, 2, 15, 24, 25], this has not been established in a large multi-center study, and while leakage correction was often used, the results were generally not presented. By analyzing data acquired at multiple centers, we included many children with a variety of tumor types. In particular, pilocytic astrocytomas tended to have low perfusion despite being contrast-enhancing, and an important message of this study is that leakage correction in this tumor group is particularly important if an erroneous rCBV is to be avoided. Overall, these results suggest that, despite differences in dynamic susceptibility-contrast MRI protocols, data can be used to aid clinicians in classifying tumors as low- or high-grade.
Uncorrected rCBV was the most significant parameter for discriminating between high- and low-grade tumors. Low-grade tumors, particularly pilocytic astrocytomas, often had negative rCBV. On applying leakage correction, rCBV became positive, reducing the significance of differences between values in the low- and high-grade tumor groups. Other studies [1, 2] found significantly higher rCBV in high- compared to low-grade tumors. Overlap between values in the two groups in both of those studies as well as ours is extensive, suggesting that dynamic susceptibility-contrast MRI results alone should not be used for tumor grading but should be viewed in conjunction with other MR imaging. Diffusion-weighted imaging in addition to dynamic susceptibility-contrast MRI was found to have high predictive diagnostic accuracy when grading pediatric brain tumors . In a clinical setting, dynamic susceptibility-contrast MRI should be part of a diagnostic pathway that is refined in a stepwise manner as more information becomes available, starting with clinical history and examination findings, being refined by conventional imaging and then advanced MRI. In this way, perfusion might provide reassurance in the putative diagnosis, or challenge it. In general, the greater the perfusion, the greater the suspicion would be that the tumor is of a high grade.
In Ho et al. , pilocytic astrocytomas had the lowest maximum rCBV. High-grade atypical teratoid rhabdoid tumors, medulloblastomas and ependymomas had the highest maximum rCBV. We found that only glioblastomas and pineoblastomas had higher rCBV than ependymomas, with medulloblastomas having the next highest rCBV. A previous study showed that of all gliomas included, glioblastomas had the highest maximum white-matter-normalized rCBV, 7.32 . Some tumor groups in our study were limited in number and so results should be treated with caution. Medulloblastomas had a large range of rCBV values, possibly because of differences in molecular subgroups . Ho et al.  also presented average histograms of rCBV for different tumor types, showing that pilocytic astrocytomas had a higher proportion of low rCBV values than high-grade tumors including ependymomas and medulloblastomas. Similarly, we found differences between rCBV histogram centiles. As in Ho’s study, skew and kurtosis did not differ significantly between the low- and high-grade groups in our study, suggesting that while histograms from different tumor types might differ in shape, when taken over a larger population, these differences are not significant.
Another study  did not employ leakage correction but classified dynamic susceptibility-contrast MRI signal–time courses as having no leakage, T1- or T2*-dominant leakage depending on whether they returned to baseline, continued above baseline or failed to return to baseline, respectively. Sensitivity tests found that a T1-dominant leakage pattern predicted lateral gene transfer in 66% of cases, rising to 91% in pilocytic astrocytomas; a T2*-dominant or baseline pattern predicted horizontal gene transfer in 100% of cases. We found median rCBVcorr had the highest sensitivity (76%) and specificity (65%) for detecting high-grade tumors using a cut-off of 1.70. A threshold of 1.60 for rCBVuncorr resulted in reduced sensitivity (53%) while specificity was improved (83%). These cut-off values lie above and below the 1.38 for maximum rCBV found in Ho et al.  and 1.07 for rCBV resulting in 100% sensitivity found by Dallery et al. , but they are close to the 1.75 cut-off presented by Law et al.  when grading adult gliomas. The low sensitivity and specificity found in our study again emphasize the importance of not using these as single tests but rather as adding information to other clinical and imaging characteristics to achieve the most likely noninvasive diagnosis, with rCBV values well above or below the cut-off having more influence. While cases that have an rCBVcorr close to the cut-off cannot be confidently assigned as low- or high-grade, those with a value below 0.785 are highly likely to be a low-grade tumor. Similarly, tumors with an rCBVcorr that is much higher than the cut-off might have an aggressive phenotype even if low-grade, as seen in the two oligodendrogliomas, known to be the more aggressive of pediatric low-grade tumors. It could be that dynamic susceptibility-contrast MRI parameters give prognostic information, as has been shown in other pediatric studies .
Only one pediatric study has presented K2 results. Provenzale et al.  found that K2 was significantly higher in high-grade than in low-grade tumors. This contradicts our results, although the leakage-correction model used  differs from ours  in not including correction for T2*-dominant effects. In Liu et al. , T1-dominant tumors had higher, positive K2 values whereas T2*-dominant tumors had lower, often negative, K2 values. In agreement with Dallery et al. , pilocytic astrocytomas demonstrated significant T1 effects, suggesting that K2 should be raised in these tumors. K2 provides a measure of the amount of leakage correction that has been applied and so will be reduced by the administration of a pre-bolus of contrast agent. It is known to depend on sequence parameters (TR, TE), pre-contrast T1 value, blood volume and permeability–surface area product. Studies have shown that K2 correlates well with Ktrans obtained from dynamic contrast-enhanced MRI , which represents a combination of permeability–surface area product and blood flow .
K2 and rCBVuncorr had lower sensitivity and specificity than rCBVcorr, likely caused by differences in injection protocols and the need to correct for negative rCBVuncorr values. Leakage correction is therefore essential to improve the accuracy of rCBV values and to account for differences in injection protocols across a pooled dataset such as this, improving the differentiation of the low- and high-grade groups. Uncorrected relative cerebral blood volume and K2 should be treated with caution unless injection protocols are consistent across the patient population. While we have presented thresholds that maximize the sensitivity and specificity at identifying high-grade tumors across this whole dataset, thresholds vary with the protocol used and so the optimal threshold should be established on a site-by-site basis.
Previous studies recommended administering a pre-bolus of contrast agent to minimize T1 effects in enhancing brain tumors that result in underestimation of uncorrected rCBV [6, 7]. While a pre-load of contrast agent reduces the effects of leakage, it does not eliminate them. This is particularly the case in the pediatric population, where administration of a single dose of gadolinium is recommended, being split between the pre-bolus and the main bolus. Consequently, if the size of the pre-bolus is increased, then leakage effects are more successfully suppressed but the size of the main bolus is reduced, leading to a reduced signal drop to noise ratio of the time-course. Use of leakage correction reduces the variability that results from the use of different pre-bolus volumes and is particularly useful if rCBV is to be compared across multiple protocols employing different extents of leakage suppression by use of a pre-bolus, as in our study. K2 and rCBVuncorr depend on the size of pre-bolus given, whereas a leakage-corrected rCBV should be more robust, and in pooled data with multiple injection protocols, leakage correction is essential to provide comparable data.
One center in our study consistently did not employ a pre-bolus of contrast agent. Leakage-uncorrected rCBV in the high- and low-grade tumor groups from this center had the best separation and highest significance using a cut-off of 0.70. K2 was always greater than 0.005 in low-grade tumors, with high-grade tumors consistently having values below this, suggesting that K2 can differentiate between low- and high-grade tumors and might hold valuable information if comparing across a dataset with consistent injection protocols. These results, while interesting, should be viewed with caution — only 12 children were scanned at this center, although there was a good split between low-grade (n=5) and high-grade (n=7) tumors. We also suggest that a pre-bolus of contrast agent is not necessary if leakage correction is applied and that administering a pre-bolus affects K2 values by compensating for leakage correction. It was recently reported  that a pre-bolus might not be necessary in adult brain tumors and that a low-flip-angle protocol with leakage correction might be preferable . In our study, children who received a pre-bolus of contrast agent had increased uncorrected rCBV (suggesting a reduction in T1 leakage effects) and reduced K2 (indicating reduced need for leakage correction). Leakage correction reduced differences in rCBV from injection protocol. It should be noted that we did not test for all differences in injection protocols, comparing only between those who received a pre-bolus and those who did not.
While too many differences exist between the dynamic susceptibility-contrast MRI protocols in this study to draw any conclusions regarding protocol optimization, certain factors (field strength and pulse sequence) did not result in significant differences between parameters obtained across the dataset. This suggests that differences in median parameters between high and low tumor grading are greater than those introduced by the variation in scan protocols.
Data in this study were acquired at multiple centers with variable protocols, creating challenges for data analysis. Thirty-two datasets were excluded because of technical issues — data corruption, incomplete data and poor quality. While the sPRESTO and gradient echo echoplanar imaging sequences produced comparable cerebral blood volumes in simulations and animal studies , poor temporal stability has been observed with the sPRESTO sequence . Signal-to-noise ratio was variable between protocols, reduced by use of a pre-bolus and low flip angle, while the use of 3 T and no pre-bolus boosted signal-to-noise ratio. Trade-offs were made between spatial resolution and whole-head coverage versus temporal resolution and signal-to-noise ratio. All centers administered a standard single dose of contrast agent in line with current recommendations . Reproducibility of parameters from regions of interest defined by two users suggests that region definition can be undertaken by multiple users across centers. We defined regions of interest encompassing the whole tumor to investigate differences in whole-tumor median parameters as well as the distributions of parameters across the tumor. Other studies have measured rCBV in hot spots, showing significant differences in maximal perfusion in the tumor. Choosing a hot spot is subject to location, being affected by both protocol and analysis method, and has a risk of being unduly affected by artifacts ; therefore, we expect a whole-tumor method to be more robust in a multicenter study. Finally, there were three versions of the World Health Organization guidance on classifying central nervous system tumors over the long accrual period in this study [19,20,21]. Tumors were classified according to the guidance available at the time. Tumor gradings were not affected by any changes.
In recent years, there have been concerns about the use of gadolinium contrast agents. People with poor renal function have been shown to be at risk of developing nephrogenic systemic fibrosis following gadolinium exposure , while recent studies have shown increased signal caused by T1 shortening on MRI scans from contrast agent deposition in areas of the brain including the dentate nucleus and globus pallidus  following earlier exposure to gadolinium. Children are at of low risk of nephrogenic systemic fibrosis ; however, there are concerns about the long-term effects of gadolinium deposition in children’s brains, particularly in those undergoing repeated MR examinations with contrast agents . Guidance mandates use of macrocylic rather than linear agents to minimize risks, the use of single dose titrated by weight, and risk-versus-benefit analysis before prescribing contrast agent, with consideration given to non-contrast methods . In patient groups such as children with brain tumors, it is still recommended that a single-dose contrast agent be administered during MRI scans at diagnosis and follow-up for acquisition of post-contrast conventional MRI  and, while this remains the case, acquiring dynamic susceptibility-contrast MRI after the contrast injection has no added risks compared to the routine imaging. Indeed, it provides an efficient use of resources. A power injector is recommended for reproducible administration of contrast agent during dynamic susceptibility-contrast MRI [17, 32]. This requires venous access via a cannula, which is invasive and can be tricky, particularly in children . The majority of our pediatric brain tumor patients have a cannula in situ at the time of their staging scan or because they are undergoing an MRI under general anesthetic.
Arterial spin-labeling measures perfusion without the need for a contrast agent. It has been shown to agree with dynamic susceptibility-contrast MRI measures of perfusion in children , with increased perfusion observed in high-grade pediatric brain tumors compared to low-grade tumors . It is gaining popularity as a method, particularly in populations at risk of nephrogenic systemic fibrosis or in those undergoing repeat MRIs. However, in comparison to dynamic susceptibility-contrast MRI, it suffers from long scan times, low signal-to-noise ratio and poor spatial resolution, and leakage information, shown to be of use in this study, is not available. Arterial spin labeling is difficult in children because of age-related variations in blood flow. While there is a recommended protocol for clinical applications in adults , this method is difficult in children, where the optimal post-labeling delay required has been shown to vary with age . Other advanced MRI methods, including MR spectroscopy and diffusion-weighted MRI, have also been shown to provide information on tumor grading in pediatric brain tumors .