Medical image segmentation across various modalities and domains is a challenging task. This paper presents a novel model designed to address these challenges effectively. Our proposed model incorporates several key innovations. In the decoder path, we utilize Bilinear Unpooling, global attention gates, and Residual Convolution Blocks (RCBs) to up-sample and refine feature maps. The encoder path alternates between max pooling and RCBs to progressively capture higher-level features, ensuring efficient feature extraction. We incorporate skip connections between the corresponding encoder and decoder layers to preserve spatial information. The final segmented output is generated through a 1 × 1 convolution layer. We combine Dice Loss and cross entropy loss for training to optimize segmentation performance. We evaluated our method on multiple state-of-the-art datasets, achieving an average accuracy exceeding 96.5% across all modalities and data types—outperforming current state-of-the-art approaches.