Premium
TU‐D‐207B‐04: Identifying High‐Risk Tumor Volume Based On Multi‐Region and Integrated Analysis of Multi‐Parametric MR Images for Prognostication of Glioblastoma
Author(s) -
Ren X,
Cui Y,
Gao H,
Li R
Publication year - 2016
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4957512
Subject(s) - fluid attenuated inversion recovery , glioblastoma , nuclear medicine , effective diffusion coefficient , magnetic resonance imaging , medicine , hazard ratio , imaging biomarker , radiology , confidence interval , cancer research
Purpose: To identify high‐risk tumor volume for predicting survival in patients with glioblastoma on the basis of multi‐region and integrated analysis of multi‐parametric MR images. Methods: In this retrospectively study, 34 patients with glioblastoma from the Cancer Imaging Archive (TCIA) were analyzed. The patients were included if all of the pre‐operative 1) T1‐weighted contrast‐enhanced (T1c), 2) T2‐weighted fluid‐attenuation inversion recovery (FLAIR), and 3) diffusion‐weighted (DW) MR images were available. The apparent diffusion coefficient (ADC) map was calculated from the DW images. Both the FLAIR and ADC images were co‐registered onto the T1c image by rigid transformation. For each patient, gross tumor volume (GTV) was first semi‐automatically delineated with a cell automation and level‐set evolution algorithm. To fully capture the intrinsic intra‐tumor heterogeneity of glioblastoma reflected on multiparametric MR images, we further segmented the delineated tumor into several spatially distinct and phenotypically consistent subregions using k‐means clustering, with T1c, FLAIR, and ADC voxel intensities as input features. The optimal number of clusters was determined based on the Calinski‐Harabasz statistic. Finally, tumor volumes associated with potentially high‐risk subregions were evaluated in terms of overall survival (OS) prediction. Results: Three tumor subregions were identified within each glioblastoma. The tumor volume associated with the subregion of the lowest mean ADC values was prognostic of OS, with a concordance index (CI) of 0.648 (log‐rank P=0.005, hazard ratio=3.82) and outperforming GTV (CI=0.627, log‐rank P=0.019, hazard ratio=3.04). Less or similar prognostic performances were achieved for tumor volumes associated with subregions of the highest mean T1c intensities (CI=0.607, log‐rank P=0.035, hazard ratio=2.57) and the highest mean FLAIR intensities (CI=0.613, log‐rank P=0.014, hazard ratio=3.08). Conclusion: Multi‐region, integrated analysis of multi‐parametric MRI identified high‐risk tumor volume in glioblastoma. Integration of functional imaging with conventional MR sequences can potentially improve prediction of survival.