Predicting MGMT and IDH1 status in gliomas using radiomic features

Brain and nervous system cancers represent a significant cancer burden in the United States with 23,210 new cases reported in 2022 and 18,203 deaths in 2023. Given this landscape, better prognostication and better understanding of tumor function is crucial. Imaging-based models to predict tumor characteristics and behavior provide a noninvasive and rapid means to evaluate brain tumors. Our research focuses specifically on using radiomic features to predict glioma characteristics, which represent up to 50 percent of malignant brain tumors. Although previous studies have investigated the predictive value of radiomic features, these studies do not evaluate both MGMT and IDH1 independently on the same dataset. Our dataset allows us to generate a model that independently predicts both factors. MRI images, MGMT status, and IDH1 status were collected for 100 patients with gliomas. Random forest classification was run on radiomic features to predict MGMT status and IDH1 status. Our analysis found that radiomic features of T1C images were the best predictor of IDH1 status (92.8% accuracy), while radiomic features of T2W images were the best predictor of MGMT status (76.6% accuracy). This difference in MRI sequence reflects distinct imaging-genomic relationships. Overall, the model performed better for prediction of IDH1 status than MGMT status. Also of note, our data and radiomics-based model allowed independent prediction of MGMT and IDH1 on the same dataset, which has not previously been reported.