Medulloblastoma is a malignant brain tumor that is most prevalent among the children and therefore in most cases, the treatment strategies that are effective usually involve early diagnosis. Surgery biopsy is a heavily involved component of traditional diagnostic that is succeeded by histopathology and genetics. Nevertheless, the practice may be intrusive, lengthy and unavailable in resource-strained environments. To resolve this issue, the present study proposes a non-invasive diagnostic approach that employs the metabolic signature analysis, which relies on artificial intelligence (AI) and magnetic resonance imaging (MRI) and can be used to classify the medulloblastoma subtypes within a short time. This method employs the multi-parametric MRI data, which comprises of the spectroscopic sequences, to determine the metabolic patterns identifiably linked to the molecular groups, including WNT, SHH, Group 3, and Group 4. The model was built by integrating convolutional neural networks (CNN) and bidirectional long short-term memory (BiLSTM) layers to form a deep learning model capable of learning spatial and contextual information. The use of Gradient-weighted Class Activation Mapping (Grad-CAM) helped to improve interpretability by underlining important tumor locations that are relevant to classification results. The suggested model demonstrated considerable potential to be used in the real-time context in the clinical setting with the average inference time of less than 10 min per scan and the general accuracy of 94.7. This framework has the potential to greatly expedite the process of diagnosis by lowering the amount of invasive procedures and specialized laboratories needed, and at the same time ensuring high levels of precision and reliability. The results demonstrate the possibility to use AI-based MRI-aided diagnostics to enhance the early diagnosis and treatment of young brain tumors. Further research can use this framework to include additional omics data and later confirm in multicentered trials in multiple populations.