Prediction of Pseudoprogression in Patients with Glioblastomas Using the Initial and Final Area Under the Curves Ratio Derived from Dynamic Contrast-Enhanced T1-Weighted Perfusion MR Imaging

Discussion

In this study, we chose 2 variables—IAUC₃₀ and FAUC₃₀—to describe the shape of the dynamic time–signal intensity curve. This choice was dictated by our desire to minimize the dependence on pharmacokinetic modeling.¹⁴ We found that all of the AUCR histogram parameters could distinguish retrospectively whether a progressively enhancing lesion was caused by ETP or by pseudoprogression. Our results might be explained by the fact that the IAUC₃₀ depends mostly on the blood flow as well as the total vascular surface area exposed to the contrast agent; therefore, ETP has a prominent IAUC₃₀ on the basis of hypervascularity and neoangiogenesis. In addition, FAUC₃₀ in the pseudoprogression group could be increased by greater retention of contrast agent in the EES because of less tumor cellularity and more tissue damage. Our results are also consistent with what has been described previously by Narang et al,¹⁵ who used the IAUC at 60 seconds (IAUC₆₀) in the initial vascular phase to successfully differentiate tumor progression from radiation necrosis. Contrary to this previous study,¹⁵ we chose IAUC₃₀ instead of IAUC_60. Determination of the initial 30 seconds for IAUC was based on our observation that initial contrast agent wash-in was usually achieved during the first 30-second postcontrast agent arrival.

Compared with a DCE MR study, a DSC study is more sensitive to susceptibility artifacts mainly associated with surgery- or radiation-induced hemorrhage, and its low spatial-resolution limits offer accurate characterization of the spatial distribution of brain tumor vessels. In addition, the T2* contrast agent leakage effects, which depend on the attenuation and spatial distribution of tumor cells within the EES, can lead to an additional susceptibility calibration factor. Even without contrast agent extravasation, the estimation of T2* parameters is confounded by the complicated relationship between the signal intensity and contrast agent concentration, as the vascular susceptibility calibration factor is known to vary across tissues with differing vascular geometries and architecture.¹⁶ Recently, Larsen et al¹⁷ reported that cerebral blood volume measurements by DCE MR imaging gave results very similar to FDG-PET regarding differentiation between tumor recurrence and radiation necrosis.

Although pharmacokinetic modeling (eg, Tofts) is the ideal approach, its “quantitative” potential is often undermined when a few conditions and assumptions are unmet. For example, an arterial input function is required, but its rapid and accurate measurement is challenging, and a visible vessel without inflow artifacts or partial-volume effects may not be available in the anatomic region of interest.⁷ IAUC during a given time is a model-free parameter that describes the initial uptake of contrast agent in a tissue of interest. It has advantages in that it does not require arterial input function measurement, is unlikely to be influenced significantly by variations in scanner and sequence type, and does not rely on complex postprocessing pharmacokinetic modeling techniques.¹¹ However, model-free parameters lack a clear biologic association because they are “mixed” measures of tissue blood flow and vascular permeability, as well as an indirect measure of the EES. Although Evelhoch¹⁴ suggested that it was related to blood flow, vessel permeability, and interstitial space, its physiologic meaning was investigated by Walker-Samuel et al,¹⁸ who showed that the IAUC correlated intractably with K^trans, EES volume, and plasma volume. Their results revealed that the IAUC could be made to track K^trans or EES volume, but only under specific conditions and within a narrow range of physiologic values. We proposed a modified area under the time signal-intensity curve-derived DCE parameter (AUCR) dedicated to posttreatment glioblastoma, which retains the advantages of model-free approaches. We tried to validate the AUCR by comparison with clinicopathologic results of posttreatment glioblastomas and found that all of the AUCR histogram parameters were significantly different between the ETP group and the pseudoprogression group.

However, we did not directly compare the diagnostic accuracy of AUCR with model-based DCE parameters, such as K^trans, in the same patients with posttreatment glioblastomas because of incomplete arterial input function sampling. However, compared with a recently published report,¹⁹ the sensitivity and specificity of AUCR were comparable to those of model-based DCE parameters and were higher than those of conventional IAUC in the differentiation of tumor from radiation necrosis. Moreover, the correlation of AUCR with K^trans was significantly high in 20 patients with treatment-naïve glioblastomas, who were eligible for relatively rapid arterial input function sampling.

Because AUCR histograms in posttreatment glioblastomas were often skewed, broad, and occasionally dual peaked, a 2-component mixture normal distribution was selected to model their shape, as proposed by Pope et al⁹ by use of ADC histograms in recurrent glioblastoma. One potential explanation for this observation is that some areas of contrast-enhancing lesion may have predominant tumor and limited treatment-related change resulting in a prominent higher AUCR histogram curve, whereas other areas of contrast-enhancing lesion are composed of predominant treatment-related change and a limited tumor-generating prominent lower AUCR curve. However, a great deal of information seems to be lost on the presence of both pseudoprogression and tumor in the same patient if only the histograms are evaluated. In our present study, we found the same overlap of mAUCR_H range between the ETP group and the pseudoprogression group, which might be attributable to the coexistence of viable tumor cells and treatment effects. As reported by Hu et al²¹, they can diagnose small isolated enhancing foci and identify histologically distinct subregions within large enhancing lesions by applying accurate threshold values determined by direct correlation between histopathologic features and DSC measurements. Future studies directly correlating AUCR with histopathologic patterns in ETP and pseudoprogression samples may clarify underlying pathophysiologic mechanisms.

Our study had several limitations. First, because the model's assumptions with IAUC₃₀ and FAUC₃₀ ignored multiple other effects, such as edema, back-pressure, and transport, all of which varied with tumor grade, the equations could only measure flow rates and not true permeability.²⁰ Second, our AUCR histogram parameters showed relatively lower specificity for differentiating ETP from treatment-related change compared with a previous DSC study, which was a well-designed prospective study that used image-guided neuronavigation during surgical resection to directly correlate specimen histopathologic patterns with localized DSC measurements.²¹ Third, using the histogram analysis of AUCR values to evaluate for ETP vs pseudoprogression in a binary fashion does not provide for the clinical reality that pseudoprogression and tumor regrowth often occur together. Hu et al²¹ proposed that diagnosing small isolated enhancing foci and identifying histologically distinct subregions within large enhancing lesions could be achieved by applying accurate threshold values derived from direct correlation between tissue-specimen histopathologic patterns and DSC measurements. Further studies using direct correlation between histopathologic features and AUCR is needed to make our results more convincing. Finally, our method required multiple postprocessing steps. A simpler measurement, perhaps with region-of-interest analysis, would be better suited to clinical practice. In addition, the quantitative analysis required thresholding the image in a semiautomated fashion to ensure that the entire enhancing tumor volume is included, which suggests the need for physician oversight and limiting the ability of technologists to apply this technique independently in clinical practice.