The classification of rare diseases, namely grade IV tumour De Novo Glioblastoma, poses significant difficulties due to the scarcity of data, which frequently leads to issues related to overfitting in traditional deep learning systems. This paper presents a novel meta-learning approach that uses neuroimaging data and advanced deep learning techniques to detect glioblastoma at an early stage. Our innovative technique intends to detect glioblastoma in its early stages, even when there is limited data available, by utilizing the specific weights of a base model that has been trained on the commonly used astrocytoma dataset. By utilizing meta-learning techniques, the system adapts in real-time to the unique characteristics of glioblastoma, outperforming conventional deep learning algorithms. This approach shows great potential in overcoming the limitations caused by limited data in the detection of rare diseases. This paper enhances the area of medical diagnostics by offering a strong framework for efficiently diagnosing rare diseases, especially in situations with limited resources. Our approach, which combines neuroimaging, deep learning, and meta-learning methodologies, has the potential to significantly transform early disease detection paradigms which can lead to prompt interventions and improved patient outcomes.