F Tixier¹*, H Um¹ , D Bermudez² , A Iyer¹ , A Apte¹ , J Deasy¹ , I Mellinghoff³ , R Young² , H Veeraraghavan¹ , (1) Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, (2) Department of Radiology, Memorial Sloan Kettering Cancer Center, New York, New York, (3) Department of Neurology, Memorial Sloan Kettering Cancer Center, New York, NY

• 
  F Tixier

(Wednesday, 8/1/2018) 7:30 AM - 9:30 AM

Room: Room 202

Purpose: In glioblastoma, several molecular biomarkers, such as the methylation of the promoter of MGMT, have been shown to be useful in distinguishing patients with different prognosis. Moreover, many cancer models have shown the utility of quantitative (radiomics) features extracted from diagnostic images in predicting patient outcomes. In this context, we hypothesized that a combination model using both molecular status and MRI-radiomics may offer superior prognostication than either technique alone.

Methods: Ninety-eight glioblastoma patients were retrospectively included in a training cohort, and a separate 61 glioblastoma patients in a validation cohort. A total of 172 MRI-radiomic features (from fluid attenuated inversion recovery [FLAIR], T1 pre- and post-gadolinium contrast images) were extracted for each patient. Feature selection was performed on the training cohort using a Least Absolute Shrinkage and Selection Operator (LASSO) regression model. The prognostic value of the retained features for overall survival (OS) was evaluated by Kaplan-Meier analysis by choosing the median values from the training set and applying them in the validation set. We separately evaluated the clinical utility of MGMT and MRI-radiomic features each alone and in combination to predict OS.

Results: Five radiomics features were selected through the LASSO regression model. Of those five, a measure of the edge enhancement from T1 post gadolinium contrast images predicted OS in the discovery (p=0.004) and the validation (p=0.02) cohort. The combination of the methylation status of MGMT promoter with the selected MRI-radiomics feature further improved these results and was able to identify a subgroup of patients with a better prognosis (28 months [range, 8-43] vs. 13 months [range, 3-43], p=0.0005).

Conclusion: The combination of biomarkers and MRI-radiomics features can help to better stratify patients with different prognosis in glioblastoma.

Funding Support, Disclosures, and Conflict of Interest: This work was in part through the NIH/NCI Cancer Center Core Support Grant P30 CA008748, and the MSK Department of Radiology, Brain Tumor Center and Neuro-Oncology Research Translation in Humans.

Keywords

Image Analysis, Brain, MRI

Taxonomy

IM- MRI : Quantitative Imaging/Analysis

Contact Email