Brain tumor segmentation is a critical task in medical imaging, aiding in diagnosis and treatment planning for brain tumor patients. This study presents an enhanced 3D U-Net architecture integrated with Feature Pyramid Networks (FPNs) and attention mechanisms to improve segmentation accuracy. Utilizing the BraTS 2020 dataset, which comprises multimodal MRI scans (T1, T2, T1ce, and FLAIR sequences), the data preprocessing involved intensity normalization, alignment of segmentation masks, and organization of 3D MRI scans. The architecture incorporates an encoder, bottleneck, and decoder path, leveraging FPN for multi-resolution feature extraction and attention mechanisms for tumor region focus. A combination of categorical cross-entropy and Dice loss optimized with the Adam optimizer was employed during training. The model achieved a training accuracy of 96.16%, Dice loss of 0.0086, a test accuracy of 96.14%, and F1-score of 98.03%, outperforming other baselines, including U-Net and its variants. Practical applications include automating tumor segmentation to assist in diagnosis, surgical planning, and treatment monitoring, improving radiological workflows’ speed and precision. However, challenges such as high computational demands and limited dataset diversity remain, requiring further optimization for resource-constrained environments and validation on larger, multi-institutional datasets. This study highlights the potential of combining FPN and attention mechanisms to advance clinical tumor segmentation workflows, with scope for scalability and integration into real-world medical imaging systems.