Accurate classification of grade 2 infiltrating gliomas and evaluation of tumor burden have important implications for diagnosis, prognosis, and selection of the most appropriate therapy. As was seen from the evaluation of anatomic images, most of these lesions are nonenhancing on T1-weighted postcontrast images and the lesions observed on T2-weighted images can be large, with relatively uniform intensity. This means that selecting the most appropriate region for tissue sampling to make an accurate diagnosis is problematic and would therefore benefit from having imaging information that distinguishes regions of the tumor with different physiological or metabolic characteristics. In the present study, we were interested in determining whether data from MRSI, PWI, and DWI could provide parameters that could be used for targeting tissue sampling and for providing information that would contribute to the characterization and therapeutic planning for patients with grade 2 glioma.
MRSI is a method that measures chemical markers of neoplastic activity. Spectra from brain tumors have increased Cho, which correlates with membrane biosynthesis by proliferating cells, and reduced NAA, which indicates loss of neuronal integrity due to tumor cell infiltration [8
]. In this study, reduced NAA and increased Cho were found in all grade 2 glioma subtypes with a trend toward higher Cho in oligodendroglioma than in astrocytoma and oligoastrocytoma. This trend may be explained by the fact that oligodendrogliomas have been reported to be more proliferative than astrocytomas [27,28
]. The varying levels of NAA that were observed in these lesions may be due to the limited spatial resolution of the MRSI data, which allows voxels on the edge of the tumor to include some normal brain tissue or to the heterogeneity within the tumor itself, which may include a mixture of infiltrative tumor and residual normal tissue. No significant differences in individual metabolite values were found between subtypes, which is consistent with previous studies [29–31
The CNI is a quantitative parameter that describes the differences in Cho and NAA relative to normal tissue and that has been shown to be sensitive to the presence of tumor [32
]. From our studies, it seems that obtaining a tissue sample from the region with highest CNI is likely to reveal the area with the highest density of tumor cells and lowest amount of normal brain. Thus, whereas there is an overlap between the CNI levels for different histologic subtypes, the highest CNI levels are observed in oligodendrogliomas and these in vivo
measurements may be useful in directing a surgeon to the location that gives the most definitive histologic diagnosis. Previous studies that examined the correlation between metabolite levels and tissue biopsies have suggested that the ratio of Cho to NAA and the level of Cho relative to Cre in normal brain may be more effective in defining the borders of the tumor than the region of hyperintensity on T2-weighted images [33
]. These values may also be important in postsurgical MR examinations for evaluating residual tumor or for assessing response to therapy. Although the echo time used in this study precluded the quantification of levels of glutamate, previous reports indicated that these may also be relevant in distinguishing grade 2 oligodendrogliomas from astrocytomas [30
PWI provides measurements of rCBV that reflect tumor angiogenesis and vascularity. In our study, the 75th percentile rCBV of oligodendrogliomas was significantly higher than the value for astrocytomas or oligoastrocytomas. This difference may be explained by the fact that oligodendrogliomas are more likely to exhibit microvascular proliferation than the astrocytomas and oligoastrocytomas [34,35
]. Statistical analysis showed that the 75th percentile rCBV was a significant variable in the differentiation of oligodendrogliomas from each of the other two subtypes. These results confirm observations from a previous study in a smaller population of patients, which found that the maximum rCBV within the tumor was significantly higher for oligodendrogliomas than for astrocytomas [18
ADC values derived from diffusion MR imaging provide a measurement of the movement of water molecules within tissue microstructures. Although ADC is thought to correlate to cell density in high-grade tumors [16
], both the results of this study and the observations made in previous studies have indicated that there is no clear correlation between ADC and levels of Cho for grade 2 gliomas [17,36
]. One explanation of this is that the levels of ADC observed in these lesions are also influenced by other biologic factors, such as the presence of edema and calcification. Edema represents an increase of free extracellular water content of tissue and should therefore result in a higher ADC [37,38
]. It is reasonable to expect that there would be more edema in astrocytomas than oligodendrogliomas because they are more infiltrative. Calcification is more common in oligodendrogliomas [39
] and results in a relatively lower ADC because it limits water content and restricts water movement. The values for oligodendrogliomas that were observed in this study are in agreement with previous data, which use the analysis of wholetumor ADC histograms [19
The classification accuracy of the logistic regression model based on median ADC was higher than that based on the 75th percentile rCBV. When they were both included in the regression, the accuracy showed a slight improvement over median ADC alone, with 3 of 40 additional patients for the analysis of oligodendrogliomas versus astrocytomas and 3 of 36 additional patients for the analysis of oligodendrogliomas versus oligoastrocytomas being correctly classified. The similarity in classification between the two variables was most likely due to the 75th percentile rCBV and the median ADC being inversely correlated, which indicates that increased vascularity and restricted diffusion were present in a coordinated fashion. It should be noted that there are limitations on these assessments based on the logistic models. First, the classification accuracy of the model is being evaluated using the same data set used to create the model. This will tend to overestimate the ability of the model to predict for an independent set of cases. In addition, by setting the cutoff probability to be 0.5, we assume that the proportion of cases with the specified subtypes would accurately represent the proportions seen in the general population of cases. Even with these caveats, the results provide strong evidence that both the 75th percentile rCBV and the median ADC provide useful information for distinguishing oligodendrogliomas from astrocytomas and oligoastrocytomas.
Oligoastrocytomas contain variable proportions of cell populations that display both astrocytic and oligodendroglial phenotypes [1
]. The MR findings obtained in our study reflected the mixed cellular phenotypes of these tumors in that they had 75th percentile rCBV and median ADC values that fell in between those of oligodendrogliomas and astrocytomas. It should be noted that although our results are consistent with several other studies that have reported on differences in imaging characteristics for grade 2 gliomas, the histologic criteria for distinguishing oligoastrocytomas from oligodendrogliomas and astrocytomas may vary between institutions. Low-grade gliomas display a broad range of cellular phenotypes, with the classic oligodendrogliomas and astrocytomas being at different ends of a discontinuous spectrum [40
]. It is therefore possible that some of the patients whom we defined as having oligoastrocytoma would have been diagnosed as having oligodendroglioma or astrocytoma in other institutions. For the purposes of our study, we focused on identifying only those tumors with hallmark histopathologic features as oligodendrogliomas, whereas the criteria used to separate oligoastrocytomas and astrocytomas were less distinctive. It is therefore not surprising that these two subpopulations had similar features and could not be distinguished based on the MR parameters considered.
One of the limitations of the current study is that we were unable to make a direct link between the MR imaging parameters and the histologic examinations because the sample used for diagnosis was not targeted based on metabolic and physiological imaging data. Thus, although the MR findings from this study were attributed to histopathologic factors such as neoangiogenesis, cell density, presence of edema, and calcification, we cannot be certain as to the contribution that each individual factor makes to each measurement. Further studies that directly correlate imaging with histologic observations will be important for enhancing the interpretation of MR-derived parameters. We recommend targeting locations within the region of T2 hyperintensity with the highest CNI value, elevated rCBV, or ADC values in the ranges that were found to be indicative of oligodendrogliomas or astrocytomas. Note that there was a direct correlation between ADC values at the location with maximum CNI with the median ADC values. This indicates that targeting the region with the most abnormal metabolism should provide representative data for the entire tumor.
Another factor that may be important for future studies is to use recent advances in genetics to develop more accurate and prognostically relevant tumor classification systems. Reports based on molecular biology of gliomas have suggested that loss of heterozygosity chromosomes 1p and 19q is typical of oligodendrogliomas, whereas mutation of TP53
gene is far more likely to associate with astrocytomas [5,41,42
]. These findings have increased the interest in investigating the relationship among MR imaging findings, glioma genotype, and histologic type [31,43–45
]. By performing these correlation studies, it will be possible to develop more definitive reference standards for glioma subtypes and to validate the use of MR imaging parameters as noninvasive biomarkers for directing surgical sampling, evaluating tumor burden, and assessing response to therapy.