Summary
There is to date no study that attempted to integrate imaging biomarkers and tumor gene expression into a statistical model that would potentially constitute a robust predictor of patient outcome than either individual data. The purpose of this study was to explore whether such a model would allow one to reliably predict patient survival and time to tumor recurrence based on a combination of MRI imaging features and tumor gene expression.
The study is aimed at incorporating imaging features and genomic biomarkers into statistical models to reliably predict glioblastoma patients outcome. MRI images of 70 GBM patients were reviewed by six neuroradiologists using the VASARI scoring system. 620 angiogenesis genes were tested. Patient outcome was measured as duration of survival and time to recurrence. Eight MRI features were associated to survival with an unadjusted p-value < 0.05. Ependymal extension (feature F19) was correlated with the shortest survival (P= 0.0012). Expression of ANG and TGFB2 genes correlated with shorter survival. CCL5 and TNF genes correlated with longer survival. A statistical model incorporating F19 with the expression of the above genes correctly predicted survival for 82% of patients. Left hemispheric tumor location (feature F2) correlated with the longest time to recurrence (P= 0.0084). The optimal linear regression model constructed included F2 and expression of the STAT1, ARHGAP24 and SSTR2 genes. Our study demonstrated that a subset of VASARI imaging features does correlate with survival and time to recurrence. Linear regression models incorporating one or multiple imaging features and tumor gene expression can reliably predict patient outcome.
Preliminary analysis has been derived from the Round 1 of the VASARI Research Project and presented at the following conferences:
- RSNA 2011 Annual Meeting: Prediction of Glioblastoma Multiforme (GBM) Time to Recurrence Using MRI Image Features and Gene Expression
- ASNR 2011 Annual Meeting: Prediction of Glioblastoma Multiforme (GBM) Patient Survival Using MRI Image Features and Gene Expression
The study is still ongoing and is now being reviewed with the inclusion of Round 2 data as well.
Supporting Documentation and Metadata
Shared Lists
The following Shared Lists have been created to easily obtain the subset of TCGA-GBM relevant to this study.
- TCGA-GBM Outcome Prediction: Consists of only the 70 subjects from Round 1 of the VASARI Research Project which were utilized in preliminary analysis.
Note: See Section 3.7 of TCIA User Guide for help with Shared Lists.
Clinical and genetic data
The corresponding gene, survival, and recurrence data was obtained from the TCGA Data Portal. The following text file contains the full list of sample IDs from the data portal which were used in the preliminary analysis: