Ginu A. Thomas MBBS, Presenter: Nothing to Disclose

Jixin Wang PhD, Abstract Co-Author: Nothing to Disclose

Pascal O. Zinn MD, Abstract Co-Author: Nothing to Disclose

Rivka Rachel Colen MD, Abstract Co-Author: Nothing to Disclose

To compare multiple predictive classification models used to predict MGMT methylation status in Glioblastoma. 

We identified 86 treatment-naïve patients from The Cancer Genome Atlas (TCGA) who had both gene and microRNA expression profiles (MGMT methylation status) and pretreatment MRI from The Cancer Imaging Archive (TCIA). Qualitative VASARI imaging features for these 86 patients were assessed by 3 independent neuroradiologists and consensus was reached. Quantitative volumetric analysis was done in the 3D Slicer software 3.6(http://www.slicer.org) using segmentation module. Fluid Attenuated Inversion Recovery (FLAIR) was used for segmentation of the edema and post-contrast T1 weighted imaging (T1W1) for segmentation of enhancement (defined as tumor) and necrosis. Each qualitative and quantitative feature was correlated to MGMT methylation status both independently and as groups and subgroups. Multiple classification models were created via regression modeling and partition analysis using various combinations of variables. JMP Pro 11 was used for modeling and statistical analysis.

Multiple classification models to predict MGMT promoter methylation status were created and compared. The logistic regression model with quantitative volumetric variables, clinical variables and the qualitative variable ‘diffusion’ could predict MGMT methylation with an AUC of 0.847 with a sensitivity of 82% and a specificity of 83.8%.

MGMT methylation status plays an important role in patient predictive and prognostic stratification of patients with GBM. The identification of a non-invasive biomarker signature as a surrogate for MGMT methylation can help stratify patients in specific therapy and predict response versus non response to therapy. An imaging genomic signature can be expected to promote a more robust personalized approach to patient care and accelerate drug development and clinical trials.

Imaging prediction of MGMT methylation status will help to specifically identify and treat those patients who respond to therapy with Temozolomide.

Thomas, G, Wang, J, Zinn, P, Colen, R, Comparative Study of Predictive Classification Models for MGMT Promoter Methylation Using Imaging Features in Glioblastoma.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14010941.html