Differentiation of tumor progression from pseudoprogression in patients with posttreatment glioblastoma using multiparametric histogram analysis

Abstract

Background and purpose: The multiparametric imaging can show us different aspects of tumor behavior and may help differentiation of tumor recurrence from treatment related change. Our aim was to differentiate tumor progression from pseudoprogression in patients with glioblastoma by using multiparametric histogram analysis of 2 consecutive MR imaging studies with relative cerebral blood volume and ADC values.

Materials and methods: Thirty-five consecutive patients with glioblastoma with new or increased size of enhancing lesions after concomitant chemoradiation therapy following surgical resection were included. Combined histograms were made by using the relative cerebral blood volume and ADC values of enhancing areas for initial and follow-up MR imaging, and subtracted histograms were also prepared. The histogram parameters between groups were compared. The diagnostic accuracy of tumor progression based on the histogram parameters of initial and follow-up MR imaging and subtracted histograms was compared and correlated with overall survival.

Results: Twenty-four pseudoprogressions and 11 tumor progressions were determined. Diagnosis based on the subtracted histogram mode with a multiparametric approach was more accurate than the diagnosis based on the uniparametric approach (area under the receiver operating characteristic curve of 0.877 versus 0.801), with 81.8% sensitivity and 100% specificity. A high mode of relative cerebral blood volume on the subtracted histogram by using a multiparametric approach (relative cerebral blood volume ×ADC) was the best predictor of true tumor progression (P < .001) and worse survival (P = .003).

Conclusions: Multiparametric histogram analysis of posttreatment glioblastoma was useful to predict true tumor progression and worse survival.

Fig 1.

Histogram parameters. Mode is the value with the maximum frequency on the histogram. Median is the middle value. Kurtosis is a measure of whether the data are peaked (high kurtosis) or flat (low kurtosis) relative to a normal distribution. Skewness is a measure of asymmetry. Negative skew means that the left tail is longer; positive skew means that the right tail is longer.

Fig 2.

A 54-year-old man with glioblastoma who underwent tumor removal followed by CCRT. Sixty-eight days after CCRT, the enhancing lesion developed around the surgical cavity (A). On follow-up MR imaging performed 52 days after CCRT (B), the size of enhancing lesion was increased, suggestive of tumor progression. However, a subtracted 3D histogram shows a decreased population (blue) of high rCBV voxels (C), and an increased population (red) of low rCBV/high ADC components. Craniotomy and tumor removal were performed, and pathologic findings showed radiation necrosis, suggestive of pseudoprogression.

Fig 3.

A 70-year-old man with glioblastoma who underwent tumor removal followed by CCRT. One hundred forty-two days after CCRT, the enhancing lesion developed in the genu of corpus callosum (A). On follow-up MR imaging performed 54 days after CCRT (B), the enhancing lesion was stable or somewhat decreased in size, suggestive of pseudoprogression. However, the 3D histogram shows a markedly increased population (coded as red) of high rCBV voxels with relatively low ADC values (C). Eventually, the enhancing lesion increased in size on follow-up MR imaging (D) performed after 67 days from first follow-up MR imaging (B), and tumor progression was diagnosed.

Fig 4.

The subtracted 3D histogram of each group by using a multiparametric approach. A, The tumor progression group shows an increased voxel population with high rCBV and relatively low ADC values (800∼1200 × 10⁻⁶mm²/s) and a decreased population with low rCBV with a wide range of ADC values. B, The pseudoprogression group shows a decreased voxel population with high rCBV and relatively low ADC, which shows increased population in the tumor progression group. In contrast, voxels with low rCBV with a wide range of ADCs (1000∼2000 × 10⁻⁶mm²/s) show an increased population in the pseudoprogression group.

Fig 5.

Receiver operating characteristic curve analysis for the mode of rCBV on the subtracted histogram by using uniparamteric and multiparametric approaches for differentiating tumor progression from pseudoprogression. Diagonal line = 50% of the area under the receiver operating characteristic curve analysis.

Fig 6.

Graph data indicate that overall cumulative (Cum) survival is significantly better for patients with pseudoprogression (A) and mode of rCBV < 1.8 on the subtracted histogram with a multiparametric approach (B) after CCRT.