Histogram Analysis of Intravoxel Incoherent Motion for Differentiating Recurrent Tumor from Treatment Effect in Patients with Glioblastoma: Initial Clinical Experience

Abstract

Intravoxel incoherent motion can simultaneously measure diffusion and perfusion characteristics. Our aim was to determine whether the perfusion and diffusion parameters derived from intravoxel incoherent motion could act as imaging biomarkers for distinguishing recurrent tumor from treatment effect in patients with glioblastoma. Fifty-one patients with pathologically confirmed recurrent tumor (n = 31) or treatment effect (n = 20) were assessed by means of intravoxel incoherent motion MR imaging. The histogram cutoffs of the 90th percentiles for perfusion and normalized CBV and the 10th percentiles for diffusion and ADC were calculated and correlated with the final pathology results. A leave-one-out cross-validation was used to evaluate the diagnostic performance of our classifiers. The mean 90th percentile for perfusion was significantly higher in the recurrent tumor group (0.084 ± 0.020) than in the treatment effect group (0.040 ± 0.010) (P < .001). The 90th percentile for perfusion provided a smaller number of patients within an overlap zone in which misclassifications can occur, compared with the 90th percentile for normalized CBV. The mean 10th percentile for diffusion was significantly lower in the recurrent tumor group than in the treatment effect group (P = .006). Receiver operating characteristic curve analyses showed the 90th percentile for perfusion to be a significant predictor for differentiation, with a sensitivity of 87.1% and a specificity of 95.0%. There was a significant positive correlation between the 90th percentiles for perfusion and normalized CBV (r = 0.674; P < .001). A histogram analysis of intravoxel incoherent motion parameters can be used as a noninvasive imaging biomarker for differentiating recurrent tumor from treatment effect in patients with glioblastoma.