AttRes-UNet: A Dual-Model Approach for Brain Tumor Segmentation

Accurate segmentation of brain tumors in MRI images is crucial for diagnosis, treatment planning, and disease monitoring. Traditional manual segmentation methods are time- consuming and prone to variability, while standard deep learning architectures such as U-Net face challenges in capturing the complex, heterogeneous structures of brain tumors. In this study, we proposed AttRes-UNet, a novel dual-model architecture combining Attention U-Net and Residual U-Net to enhance segmentation accuracy. Attention Gates allow the model to focus on critical tumor regions, while Residual Blocks improve deep feature learning, enabling the model to handle diverse tumor morphologies with greater precision. We evaluated the model using the BraTS 2021 dataset, achieving high performance with an accuracy of 0.996, a dice coefficient of 0.8340, and an Intersection over Union of 0.8210. These results demonstrate the effectiveness of AttRes-UNet in improving segmentation accuracy, making it a valuable tool for automated brain tumor delineation, ultimately reducing manual workload and improving clinical outcomes.