Accurate Radiogenomic classification of brain tumors play an essential role in improving prognosis, diagnosis, and therapy strategy in glioblastoma patients. In this paper, we provide a novel MGMT Promoter Methylation Prediction (MGMT-PMP) approach that extracts latent features fused with radiomic data to predict the genetic subtype of glioblastoma. The novice rejection method has been shown to be highly effective at picking and isolating negative training cases from the original dataset. In this study, the fused feature vectors are used to train and test CNN classifiers. We examined classification performance for the first time in published form, using metrics such as accuracy, F1-score, and Matthews’s correlation coefficient. Jackknife tenfold cross-validation was used to train and evaluate the BraTS-2021 dataset validation. The maximum classification accuracy for detecting MGMT methylation status in glioblastoma patients is 99.13 percent. Deep learning feature extraction with Radiogenomic characteristics, merging imaging phenotypes and molecular structure, and using a rejection method were found to outperform in diagnosing the MGMT methylation status of glioblastoma patients. The approach connects genetic variance to radiomic characteristics, bridging two areas of research that may be valuable for clinical treatment planning and leading to better outcomes.