<p>Brain tumor segmentation from Magnetic Resonance Imaging (MRI) remains a significant clinical challenge due to diverse tumor morphologies, subtle boundary definitions, and considerable variations across imaging modalities. Precise segmentation is crucial for effective diagnosis, monitoring, and treatment planning. However, current deep learning methods often suffer from hyperparameter sensitivity, inefficient tuning processes, and limited generalization capabilities across diverse datasets. This study addresses these challenges by proposing a novel segmentation framework, DE-DeepLabV3+Res-UNet50, optimized by Differential Evolution (DE), a powerful evolutionary algorithm for robust hyperparameter tuning. Our approach combines the strengths of DeepLabV3+’s multi-scale atrous spatial pyramid pooling with ResUNet50’s efficient residual learning, significantly enhancing feature extraction and boundary detection accuracy. Extensive experiments and ablation studies were conducted on three publicly available brain tumor datasets (Figshare Brain Tumor Segmentation (FBTS), BraTS 2018, and BraTS 2021), incorporating rigorous evaluation metrics, statistical analysis and Receiver Operating Characteristic (ROC) analysis. The results demonstrate that our proposed method achieves state-of-the-art performance, with notable Dice Similarity Coefficient (DSC) and Jaccard Index (JI) scores reaching 0.9805 and 0.9620, respectively. Furthermore, ROC analysis reveals exceptional Area Under Curve (AUC) values ranging from 0.9998 to 1.0000, confirming superior segmentation reliability. Despite increased computational demands due to DE optimization, the significant gains in segmentation accuracy and generalization justify its clinical applicability. To address the remaining limitations, future work will explore extending the approach to 3D volumetric MRI segmentation, integrating self-supervised learning for improved representation, and developing lightweight, computationally efficient models for clinical deployment.</p>