Differentiation among Glioblastoma Multiforme, Solitary Metastatic Tumor, and Lymphoma Using Whole-Tumor Histogram Analysis of the Normalized Cerebral Blood Volume in Enhancing and Perienhancing Lesions

Discussion

Our study has shown that semiquantitative histogram analysis of the normalized CBV was able to distinguish accurately among GBMs, SMTs, and lymphomas. The results are in concordance with the findings of a number of previous studies regarding the localized maximum rCBV.^9–12 In this study, we used 6 histogram parameters derived from normalized CBV maps of dynamic susceptibility contrast-enhanced MR perfusion imaging to differentiate GBMs, SMTs, and lymphomas. We found that the MV for perienhancing lesions was the most significant parameter in the differential diagnosis of GBMs from SMTs. This result was comparable with the findings of Law et al¹² and Cha et al.² We also found that HW and PHP for perienhancing lesions could differentiate GBMs and SMTs with reasonable sensitivity and specificity. Observed peritumoral hyperperfusion in GBMs is consistent with the known infiltrative nature of gliomas, with the tumors extending beyond the margin of contrast enhancement to the perienhancing region of T2 signal intensity abnormality.^12,19–21 A perienhancing lesion of a brain metastasis represents pure vasogenic edema that results from uncontrolled leakage of blood plasma into the interstitial compartment due to leaky capillaries.^8,22 There have been very few studies regarding the utility of dynamic susceptibility contrast-enhanced histogram analysis for cerebral gliomas. We found 3 histogram parameters (HW, PHP, and MV) to be valuable metrics for hyperperfusion in perienhancing lesions, in keeping with increased angiogenesis that occurs beyond the contrast-enhancing portions of a tumor.¹²

For the analyses of contrast-enhancing lesions to differentiate GBMs from SMTs, previous studies have demonstrated that perfusion MR imaging could not be used to differentiate GBMs and SMTs reliably because both lesions are highly vascular tumors that demonstrate an increased rCBV.^6,12 In the present study, only the PHP for contrast-enhancing lesions could differentiate the 2 tumor types. However, the pathologic background for this finding is not clear because unlike gliomas, histopathologic features of metastatic tumor vessels have not been established. Furthermore, the P value, sensitivity, and specificity of the PHP for contrast-enhancing lesions were not high; therefore, further studies including larger variances for individual tumor types are needed to improve the integrity of the cutoff values with higher confidence levels.

To differentiate lymphomas from GBMs or SMTs, all 3 histogram parameters were significant for contrast-enhancing lesion regions. Although there is moderate overlap of the PHP parameter for the contrast-enhancing lesion, particularly for differentiating SMT and lymphoma, PHP was the only histogram parameter that can significantly distinguish the 3 tumor types. In multiple comparison tests, MV and HW are the more significant parameters for differentiating lymphomas from GBMs or SMTs. However, in view of the differential diagnosis of solitary enhancing brain lesion, PHP can be a useful parameter. Although there are few published reports on dynamic susceptibility contrast-enhanced imaging characteristics of primary central nervous system lymphomas, our results are in accord with the findings of a previous pathologic report in which neovascularization was not a prominent feature for lymphomas, though tumor invasion of endothelial cells and even into the vessel lumen can often be seen.²³ One of the most striking histopathologic features of a primary central nervous system lymphoma is the angiocentric growth pattern, in which tumor cells form multiple thick layers around the host vessels and widen the perivascular space.⁶ Our results for lymphoma are in concordance with the findings of previous MR perfusion studies for primary central nervous lymphoma.^24,25 These studies showed that primary central nervous lymphomas tend to have low maximum CBV ratios. On the other hand, the percentage of signal-intensity recovery indicating contrast leakage showed discrepancy between the 2 previous studies.^24,25 In our small population of lymphomas, this parameter was variable. However, it was not the main issue in this study, and more large-population studies should be performed to validate the above findings.

The HW for contrast-enhancing lesions probably reflects the heterogeneity of tumor angiogenesis and microvasculature. The higher value of the HW for GBMs in this study is similar to results of previous studies in which high-grade gliomas showed a trend toward having a higher SD compared with low-grade gliomas.^13,14 Law et al¹⁴ have suggested that the histogram metric having the highest correlation with glioma grade was SD. Because SD is a measure of the dispersion of data around the arithmetic mean, increased SDs for high-grade gliomas could reflect heterogeneous tumor angiogenesis. In our study, the HW for contrast-enhancing lesions was also high for SMTs; however, the pathologic background should be established with larger population studies. The higher MV seen for contrast-enhancing lesions of GBMs is similar to results of previous studies for glioma grading using the maximum localized region-of-interest method, which have shown that the maximum value of the rCBV for gliomas correlates with the pathologic tumor grade.^9–11 As shown in previous studies^6,12 and our study, SMTs can have an equally high MV for contrast-enhancing lesions compared with GBMs. On the basis of our experience of perienhancing lesion histogram analysis, lymphomas showed variable values of the 3 histogram parameters; however, the histogram values were significantly lower than the corresponding values for GBMs.

In view of statistics, the PHP that indicates “mode” and the MV that indicates “maximum” are location statistics. The maximum is known to be a less accurate reflection of the location than the mode, so the finding that PHP might be a better discriminator than the MV is expected. However, the median is more accurate than the mode, and the mean is even more accurate than the median. The HW that indicates range is a measure of dispersion, as are the SD and interquartile range, which would be less dependent on the sample size. Nevertheless, we did not analyze mean, median, SD, or interquartile range in this study because we focused on quick semiquantitative histogram analysis based on the histogram shape rather than the detailed statistical value.

There are several limitations to this study. First, it included a small number of study patients. It would have been preferable to include more patients to strengthen the statistical power. Second, an obvious challenge with the use of the histogram method is identifying the appropriate perienhancing tumor portion region. The optimal definition of the entire tumor volume is complicated because gliomas are infiltrating tumors with indistinct borders beyond the radiologic margins.^26,27 However, a previous study has shown that variations between observers caused by imperfect tumor delineation are relatively unimportant, given the large number of data points included in the histogram.¹³ Law et al¹⁴ have suggested that the confounding effect of lesion size can be a limitation for the use of SD in predicting the glioma grade. For total tumor measurements, it is conceivable that larger tumors will have a more prominent impact on the total tumor perfusion metrics compared with smaller tumors.

The issue of tumor size is much more complex because primary glial tumors are known to infiltrate the adjacent normal-appearing brain parenchyma.¹⁴ However, in the present study, the HW, which is a histogram parameter similar to SD, weakly correlated with lesion size. Third, when confined to the intravascular space, paramagnetic contrast agent produces signal-intensity loss in the extravascular space on T2-weighted scans. However, because this contrast agent is also an effective T1 relaxation enhancer, the susceptibility contrast signal-intensity loss can be masked by signal-intensity increase in regions where T1 effects are significant. This occurs in enhancing tumors, in which contrast agent extravasates into the interstitial space of lesions with significant blood-brain barrier breakdown. Although both nonlinear γ-fitting and numeric integration to the peak of the first-pass ΔR2 curve can reduce this effect by focusing on the early first pass, the entire first-pass curve is in principle contaminated by the T1 effect.¹⁸

In this study, we did not use an administration of a preload of contrast agent, which plays an important role in diminishing the T1 leakage effects that confound dynamic susceptibility-weighted contrast-enhanced data. As described in previous reports,⁶ our dynamic susceptibility contrast-enhanced imaging protocol for brain tumor used small-flip-angle methods to reduce sensitivity to T1 leakage effects. However, this method can result in poor tumor-brain relative CBV contrast. Finally, a limited variety of metastases was analyzed in this study. Most of the metastatic tumor pathologies in the present study were lung cancer. The possibility remains that metastases of different histopathologic origins can exhibit different values of histogram parameters. Further studies with a larger population and even distribution of tumor pathologies are necessary to validate the utility of histogram parameters to differentiate each intra-axial tumor.