MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Prediction of MGMT Status for Newly Diagnosed Glioblastoma Patients Using Radiomics Feature Extraction From 18F-DOPA PET Imaging

J Qian*, M Herman , D Brinkmann , N Laack , P Korfiatis , B Kemp , C Hunt , V Lowe , D Pafundi , Mayo Clinic, Rochester, MN

Presentations

(Thursday, 7/18/2019) 7:30 AM - 9:30 AM

Room: Stars at Night Ballroom 2-3

Purpose: Methylation of the O6-methylguanine methyltransferase (MGMT) gene promoter is associated with improved treatment response and survival in patients with Glioblastoma. Although pathologic evaluation is currently the gold standard, tissue sample is inadequate or MGMT analysis is inconclusive in 20-50% of specimens. In this study, we assessed whether radiomics features from pre-treatment ¹�F-DOPA PET imaging could be used to predict pathologic MGMT status.

Methods: 68 patients with newly diagnosed Glioblastoma were randomly split into two cohorts: 59 for training and 9 for testing. Radiomics features were extracted from two ¹�F-DOPA PET contours: a “Gold� contour of the entire uptake region per expert Nuclear Medicine physician blinded to MGMT status, and a “HGG� contour based on a T/N > 2.1 representing the most aggressive tumor components within the Gold region. Feature selection was performed by comparing the weighted feature importance in multiple models. Optimization of model parameters was explored using a grid search with selected features.

Results: In addition to demographic patient information, the initial feature space included shape, intensity and texture features. Given the small patient cohort, the number of features was highly constrained to reduce overfitting. Applying the model to the randomly chosen test data, 78% accuracy in predicting MGMT status was achieved using Random Forest model based on features from the HGG contour alone. The prediction was not improved with the addition of the Gold contour.

Conclusion: Radiomics feature extraction from ¹�F-DOPA PET imaging for newly diagnosed Glioblastoma patients suggests prediction of pathologic MGMT status with 78% accuracy. ¹�F-DOPA PET radiomics may provide valuable therapeutic guidance for patients where MGMT testing is untestable or inconclusive. Future work will correlate radiomics modeling using both PET and MRI with clinical outcomes to determine utility for prognositic accuracy in patients with heterogeneous MGMT methylation who currently have clinically variable course.

Keywords

PET, Brain, Image Analysis

Taxonomy

IM/TH- Image Analysis (Single modality or Multi-modality): Imaging biomarkers and radiomics

Contact Email