Accurate brain tumor segmentation in MRI images is crucial for effective diagnosis and treatment planning. Conventional methods, like the U-Net architecture, may struggle with complex tumor geometries and boundary delineation. To address this, we integrate the residual grouped convolution module (RGCM) and the convolutional block attention module (CBAM) into the U-Net framework, enhancing its design and performance. By making the network capable of learning feature representations across numerous filter groups, the RGCM enhances important Spatial pattern recognition in MRI images. In addition, attention mechanisms across channel dimensions of the CBAM are used to enable us to have accurate segmentation results by focusing on important locations inside networks. Here is the process of our method: Shrink MRI image data AugmentationDivision dataset in training and test. The model was trained with entropy loss and Adam optimizer as proposed. Results demonstrate that our model achieves an accuracy of 97.581%, which outperforms the traditional U-Net. Our model does better than the existing models in comparison. For example, Isensee et al.’s modification of U-Net was the best, achieving an accuracy of 96.78%. Baid et al.’s SegNet reached 96.67%, Oktay et al.’s Attention U-Net achieved 96.89%, and Zhang et al.’s Residual Attention U-Net recorded 96.94%. These results demonstrate the improved performance of our model in brain tumor segmentation. Together, these results confirm the utility of our model in achieving consistent detection performance irrespective of maintaining a balance between malignant and benign brain lesions. In this work, the authors introduced a new approach for better tumor segmentation in MRI images and opportunities still remain for further image processing.