<p>The transformer model has shown great potential in medical image segmentation, especially due to its ability to effectively handle complex contextual information. However, existing segmentation methods face challenges, such as the difficulty in capturing the global complexity of organ morphology and the irregular shape of lesions, which can reduce segmentation accuracy. To address these challenges, we propose a novel attention mechanism based on range-variable dynamic convolution. This mechanism is specifically designed to enhance the model’s focus on local features, allowing it to better handle the complex shapes and irregularities in medical images. In addition, we integrate this attention mechanism with dual attention-guided segmentation blocks, which capture both spatial and channel relationships across feature dimensions. This combination enables efficient feature enhancement while maintaining computational efficiency. Our method, called Dynamic Convolution and Dual Attention-guided Efficient Segmentation Network (RDENet), is evaluated on several medical image segmentation datasets (Synapse, ISIC2017, and ISIC2018). Experimental results demonstrate that RDENet outperforms existing methods in terms of both segmentation accuracy and computational efficiency.</p>