To assess the diagnostic performance of normalized and non-normalized diffusion kurtosis imaging (DKI) metrics extracted from different tumor volume data for grading glioma according to the integrated approach of the revised 2016 WHO classification.

Sixty patients with histopathologically confirmed glioma, who provided written informed consent, were retrospectively assessed between 01/2013 and 08/2016 from a prospective trial approved by the local institutional review board. Mean kurtosis (MK) and mean diffusivity (MD) metrics from DKI were assessed by two blinded physicians from four different volumes of interest (VOI): whole solid tumor including (VOI_tu-ed) and excluding perifocal edema (VOI_tu), infiltrative zone (VOI_ed), and single slice of solid tumor core (VOI_slice). Intra-class correlation coefficient (ICC) was calculated to assess inter-rater agreement. One-way ANOVA was used to compare MK between 2016 CNS WHO tumor grades. Friedman's test compared MK and MD of each VOI. Spearman's correlation coefficient was used to correlate MK with 2016 CNS WHO tumor grades. ROC analysis was performed on MK for significant results.

The MK assessment showed excellent inter-rater agreement for each VOI (ICC, 0.906–0.955). MK was significantly lower in IDH_mutant astrocytoma (0.40±0.07), than in 1p/19q-confirmed oligodendroglioma (0.54±0.10, P=0.001) or IDH_wild-type glioblastoma (0.68±0.13, P<0.001). MK and 2016 WHO tumor grades were strongly and positively correlated (VOI_tu-ed, r=0.684; VOI_tu, r=0.734; VOI_ed, r=0.625; VOI_slice, r=0.698; P<0.001).

Non-normalized MK values obtained from VOI_tu and VOI_slice showed the best reproducibility and highest diagnostic performance for stratifying glioma according to the integrated approach of the recent 2016 WHO classification.