Feasibility of Predicting O-6-methylguanine-DNA Methyltransferase Status in Glioblastoma Using MRI-based Radiomics Features

O-6-methylguanine-DNA methyltransferase (MGMT), a notable gene promoter, is strongly related with the treatment effectiveness in glioblastoma (GBM). Analyzing tumor tissues is currently the sole dependable method to ascertain the MGMT promoter’s status. Considering the complications associated with tissue-based approaches, there is a preference for utilizing non-invasive methods. This study aimed to examine the feasibility of non-invasive prediction of MGMT status using radiomics features, with feature extraction conducted on both the original and wavelet-transformed FLAIR and T1-GD MRI images. Further, feature selection was performed based on mutual information and the Mann-Whitney U test. Then, machine learning (ML) models were trained through a pipeline for parameter optimization using 5-fold cross-validation. Finally, random forest (RF) achieved the highest prediction efficacy with 77% accuracy, 72.41% sensitivity, 81.48% specificity, and 77% AUC score. The proposed approach demonstrates the capability of radiomic features for predicting MGMT status in glioblastoma through non-invasive means.