Accurate brain tumor segmentation is essential for effective diagnosis and treatment planning. This study proposes DE-UNet, an enhanced U-Net architecture optimized using Differential Evolution (DE) to improve segmentation of multimodal MRI scans. The model was evaluated on two benchmark datasets: Figshare Brain Tumor Segmentation (FBTS) and BraTS 2021 datasets, focusing on whole tumor segmentation across four MRI modalities: FLAIR, T1, T1-CE, and T2. DE-UNet outperformed state-of-the-art methods, achieving Dice Similarity Coefficient (DSC) and Jaccard Index (JI) scores of 0.9160/0.8472 on FBTS and 0.9094/0.8371 on BraTS 2021. DE effectively optimized key hyperparameters—learning rate, dropout, batch size, and filter sizes—enhancing the model generalization across tumor types and imaging conditions. Visual analysis confirmed accurate tumor boundary delineation. These results highlight the potential of DE-UNet as a robust and precise tool for clinical brain tumor segmentation.