Brain tumors are a major cause of death among adults, presenting significant challenges in neuro-oncology owing to their varied forms, sizes, and invasive nature. Accurate segmentation of these tumors from medical images is crucial for effective treatment planning and follow-up. This research presents an improved U-Net model for segmenting brain tumors in MRI images, tested on the BraTS 2020 dataset. By incorporating residual blocks and batch normalization, our model improves the segmentation accuracy and robustness. It achieves high Dice scores of 0.9295, 0.9079, and 0.9184 and Mean Intersection over Union (IoU) scores of 0.9593, 0.9758, and 0.9721 for the Whole Tumor, Tumor Core, and Enhancing Tumor regions, respectively. Further evaluation on the BraTS 2018 dataset confirmed the model’s exceptional performance, with Dice scores of 0.9074, 0.9213, and 0.9247 and mean IoU scores of 0.9346, 0.9683, and 0.9732, respectively. This demonstrates the adaptability of the model and its potential for broad clinical application.