Whole tumor MR Perfusion histogram analysis in the preoperative assessment of patients with gliomas: Differentiation between high- and low-grade tumors

Purpose: To compare the diagnostic accuracy of normalized Blood Volume (nBV) histogram metrics in differentiating low from high-grade gliomas. Material and Methods: Forty-four patients (22 female, 22 male) with histologically confirmed gliomas were included. Group A comprised 10 patients with low grade gliomas (all grade II) while group B comprised 34 patients (4 grade III and 30 grade IV). Three-dimensional whole tumor segmentation was based on intensity level clustering in T2 FLAIR for the non-enhancing lesions or post contrast T1 weighted images for the enhancing lesions. Dynamic Susceptibility Contrast (DSC) perfusion was applied in all patients, and leakage corrected nBV maps were created. Corresponding histograms were generated from all the pixels included in the tumor volume. Minimum, maximum, mean, standard deviation, median, skewness, kurtosis, 5%, 30%, 70% and 95% percentiles, as well as normalized peak height and maximum peak position derived from normalized blood volume histograms were calculated for both groups. ROC analysis was performed to find optimum thresholds for differentiating between low and high grade gliomas. Results: 5% percentile of nBV normalized histogram provided the highest area under the curve (AUROC: 0.93) for the differentiation of low from high grade gliomas. A threshold value of 0.07 for the 5% percentile of nBV normalized histogram resulted in 90.8% sensitivity and 90% specificity. The positive and negative predictive values were 96.7% and 75%, respectively while the accuracy reached 91%. When removing 2 IDH mutation positive HGG patients, the corresponding AUROC increased to 0.98. Conclusion: Whole tumor normalized Blood Volume histogram analysis proved to be a very accurate method to differentiate low from high grade gliomas.