Accurate brain tumor segmentation in magnetic resonance imaging (MRI) scans is crucial for diagnosis and treatment planning. This paper presents an enhanced approach to brain tumor segmentation by combining unsharp masking, normalization, and histogram equalization with a 3D U-Net architecture. When compared to several state-of-the-art techniques, our strategy considerably improves the Dice Score for Whole Tumor (WT), Enhancing Tumor (ET), and Tumor Core (TC) regions. In particular, we obtain Dice Scores of 91.84% for WT, 84.58% for ET, and 85.00% for TC on the BraTS2020 dataset. The MRI images quality is enhanced by the suggested preprocessing techniques, which helps the model train and make more accurate predictions. Our approach shows significant gains in segmentation accuracy, especially in the difficult Enhancing Tumor region. These findings support the efficacy of our method and offer significant improvements over the state-of-the-art methods for brain tumor segmentation. Better contrast and feature improvement in the images are also a result of the preprocessing stages of normalization, unsharp masking, and histogram equalization, which boost model performance. All things considered, our research validates the usefulness of the suggested approach in clinical settings, providing a solid brain tumor segmentation solution that can greatly facilitate patient diagnosis and treatment planning.