Brain tumors must be accurately and promptly detected in order to improve patient outcomes and allow for prompt treatment. Brain tumors, which are caused by abnormal tissue development, can be benign or malignant, and, if left untreated, they can pose serious threats to neurological health. Magnetic resonance imaging (MRI) acts as a major diagnostic technique, delivering precise pictures for tumor diagnosis. However, variability in diagnostic frameworks among professionals frequently results in inconsistent conclusions, delaying treatment and reducing survival rates. This work tackles these issues by presenting ADE_DieT, a composite model for multiclass brain tumor categorization. This model synergizes a DieT transformer architecture for hierarchical feature extraction, principal component analysis (PCA) to mitigate high-dimensional data complexity, and an adaptive differential evolution (ADE) algorithm to optimize discriminative feature selection. Evaluated on a comprehensive MRI dataset, ADE_DieT achieved a classification accuracy of 95.69%, surpassing cutting-edge architectures like EfficientNet, DenseNet121, and InceptionV3. By streamlining MRI-based tumor detection, ADE_DieT minimizes reliance on manual diagnostics, expediting the diagnostic process. This advancement highlights the transformative potential of integrating evolutionary optimization with transformer-based learning in clinical oncology, offering a scalable solution to refine diagnostic precision, streamline decision-making, and ultimately improve patient care pathways. The results underscore the viability of AI-driven tools in addressing complex medical imaging challenges while fostering reproducibility in oncology diagnostics.