Apparent Diffusion Coefficient as a Predictive Biomarker for Survival in Patients with Treatment-Naive Glioblastoma Using Quantitative Multiparametric Magnetic Resonance Profiling.

BACKGROUND

The purpose of the present study was to investigate whether quantitative radiomic profiles extracted from multiparametric magnetic resonance (MR) profiles can predict the clinical outcomes for patients with newly diagnosed glioblastoma (GBM) before therapy.

METHODS

MR images from 93 treatment-naive patients with newly diagnosed GBM were analyzed. Through tumor segmentation, we selected 36 radiomic features. Using the unsupervised clustering method, we classified our patients into 2 groups and investigated their overall survival (OS) using Kaplan-Meier analyses.

RESULTS

Among the 36 radiomic features, the apparent diffusion coefficient (ADC) histogram parameters demonstrated a significant association with OS (P < 0.05). To validate this finding, unsupervised clustering analysis revealed 3 clusters with similar radiomic expression patterns. Clusters 1 and 2 showed a significant correlation with the radiomic features representing the tumor volume, and cluster 2 also showed a significant correlation with relative cerebral blood volume values. In contrast, cluster 3 showed an inverse relationship with cluster 2, mainly representing the radiomic features indicating the ADC and mean transit time. Although no statistically significant difference was found in OS between cluster 1 plus 2 and cluster 3, cluster 3 showed a trend toward longer OS compared with cluster 1 plus 2 (P = 0.067). After stratification by methylation status and radiomic feature clustering, patients with methylated O-methylguanine DNA methyltransferase and those included in cluster 3 had significantly longer OS (P = 0.029).

CONCLUSIONS

ADC histogram parameters are feasible prognostic biomarkers to predict the survival of patients with treatment-naive GBM. Quantitative MR profiles can predict the clinical outcomes of patients with GBM before therapy.