This study uses data from 934 medulloblastoma patients to develop MRI image-based machine learning methods for molecular subgroup classification. Each patient underwent a brain MRI scan, and a tumor segmentation deep learning model was then developed to determine the primary tumor area on the MRI image. Important features, such as tumor shape, intensity, and texture are then extracted from the images and these MR imaging signatures are ultimately used to develop a three-class classifier that places patients into either the WNT, SHH, or G3/G4 (non- WNT/non-SHH) classes. Three-fold cross-validation was used to validate the performance of the classifier; the area under the curve (AUCs) of the three-class classifier were 0.924 for the WNT sub-group, 0.819 for SHH sub-group, and 0.810 for the non-WNT/non-SHH sub-group, where each sub-group was compared against the other two classes. The researchers also developed a binary classifier for discriminating between non-WNT and non-SHH and achieved AUCs of 0.822 in the primary dataset and 0.859 in the external validation dataset. To further evaluate the performance of the developed classifier, researchers collected data from 40 more patients and obtained an AUC of 0.900 for the three-class classifier and 0.852 for the binary classifier. In addition, an analysis of feature importance was conducted and the researchers found that intra- tumoral features were the most influential in accurately predicting the molecular subgroup of a given tumor. Ultimately, this study develops a machine learning model that predicts medulloblastoma molecular subgroups thus highlighting the potential of using imaging-based AI methods for medulloblastoma diagnosis.