IMG-14. DEVELOPING A PREDICTIVE GRADING MODEL FOR CHILDREN WITH GLIOMAS BASED ON DIFFUSION KURTOSIS IMAGING METRICS: ACCURACY AND CLINICAL CORRELATIONS WITH SURVIVAL

Abstract

PURPOSETo develop a predictive grading model based on diffusion kurtosis imaging (DKI) metrics in children affected by gliomas, and to investigate the clinical impact of the model via correlations with overall survival and progression-free survival.MATERIALS AND METHODSWe retrospectively studied 59 children (33M, 26F, median age 7.2 years) affected by gliomas on a 3T magnet. Patients with tumor locations other than infratentorial midline were included. Conventional and DKI sequences were obtained. Mean kurtosis (MK), axial kurtosis (AK), radial kurtosis (RK), fractional anisotropy (FA) and apparent diffusion coefficient (ADC) maps were obtained. Whole tumor volumes (VOIs) were segmented semiautomatically. Mean DKI values were calculated for each metric. The quantitative values from DKI-derived metrics were used to develop a predictive grading model with penalized logistic regression (glmnet package, R). Elasticnet regularization was used to avoid model overfitting. Fitted model coefficients from each metric were used to develop a probability prediction of a high-grade glioma (HGG). Grading accuracy of the resulting probabilities was tested with ROC analysis. Finally, model predictions were correlated to progression-free survival (PFS) with a Kaplan-Meier analysis.RESULTSThe cohort included 46 patients with low-grade gliomas (LGG) and 13 patients with HGG. The developed model predictions yielded an AUC of 0.946 (95%CI: 0.890–1). Model predictions were significantly correlated with PFS (23.1 months for HGG vs 34.7 months for LGG, p<0.004).CONCLUSIONIn our cohort, a DKI-based predictive model was highly accurate for pediatric glioma grading. DKI-based model predictions were significantly correlated with progression-free survival.