GOAT-NET model: Glioma-Oligodendro astrocytoma tumor network model based on the optimization-based few-shot learning...

The diagnostic challenges linked to rare neurological diseases, including Glioblastoma, remain a significant issue, especially in resource-limited healthcare systems like India. Conventional deep learning algorithms work well for traditional medical imaging tasks, but rare disease datasets have data scarcity and an imbalance. This paper presents GOAT-NET (Glioblastoma, Oligodendroglioma, and Astrocytoma Tumor Network), a meta-learning-based diagnostic framework aimed at improving the categorisation of rare Glioblastoma and Healthy samples using a limited number of annotated MRI samples. It was primarily trained and validated on Astrocytoma, Oligodendroglioma, Glioblastoma, and Healthy classes. The framework combines a ResNeXt-50 as a backbone for strong domain-specific feature extraction with a few-shot learning module that allows the system to quickly adapt to new tumor forms. Additionally, GOAT-NET uses explainable AI methods like LIME and SHAP to make visual explanations that are easy to understand. The training and validation accuracy for the base model is 97.34% and 96.85% respectively. Experimental validation indicates that GOAT-NET model can correctly classify 97.75% of all Glioblastoma instances, which is much better than standard deep learning baselines. The technique could benefit in the presurgical characterisation of brain tumours, however, neuropathological assessment remains the highest standard for glioma subtype classification. The work further emphasizes the relevance of meta-learning in healthcare settings marked by scarce resources, diverse patient demographics, and restricted imaging infrastructures. GOAT-NET solves the computational problems that come with diagnosing rare tumours by integrating high-capacity feature learning, meta-learning framework for few-shot adaptation, and clear interpretability.