CSWin-UNet is a novel U-shaped segmentation method that integrates the CSWin self-attention mechanism into the UNet to enable horizontal and vertical stripe self-attention. This method significantly enhances computational efficiency and receptive field interactions. The innovative decoder uses a content-aware reassembly operator to strategically reassemble features, guided by predicted kernels, for precise image resolution restoration. Extensive empirical evaluations on diverse datasets, including synapse multi-organ CT, cardiac MRI, and skin lesions, demonstrate that CSWin-UNet maintains low model complexity while delivering high segmentation accuracy. The CSWin Transformer block, based on the CSWin self-attention mechanism, allows for efficient feature extraction and reduces computational complexity. The encoder and decoder are symmetrical and each consists of four stages. The encoder reduces resolution and increases channel count, while the decoder increases resolution and decreases channel count using the CARAFE layer. The CARAFE layer enables precise feature reassembly and enhances the precision of organ edge segmentation. Experimental results show that CSWin-UNet outperforms other methods in segmentation accuracy and computational efficiency. The method is lightweight, with a balanced architecture that improves segmentation accuracy and robustness. The CSWin-UNet achieves superior performance in various medical image segmentation tasks, including CT, MRI, and skin lesion segmentation. The method is effective in handling complex segmentation environments and provides higher segmentation accuracy. The results indicate that CSWin-UNet is more advanced and suitable for medical images of various modalities than other state-of-the-art methods. The method shows some deficiencies in challenging cases, such as significant differences in segmentation accuracy for different samples of the gallbladder and kidney regions. The pre-training of the model significantly impacts its performance, and further research is needed to explore end-to-end medical image segmentation methods.CSWin-UNet is a novel U-shaped segmentation method that integrates the CSWin self-attention mechanism into the UNet to enable horizontal and vertical stripe self-attention. This method significantly enhances computational efficiency and receptive field interactions. The innovative decoder uses a content-aware reassembly operator to strategically reassemble features, guided by predicted kernels, for precise image resolution restoration. Extensive empirical evaluations on diverse datasets, including synapse multi-organ CT, cardiac MRI, and skin lesions, demonstrate that CSWin-UNet maintains low model complexity while delivering high segmentation accuracy. The CSWin Transformer block, based on the CSWin self-attention mechanism, allows for efficient feature extraction and reduces computational complexity. The encoder and decoder are symmetrical and each consists of four stages. The encoder reduces resolution and increases channel count, while the decoder increases resolution and decreases channel count using the CARAFE layer. The CARAFE layer enables precise feature reassembly and enhances the precision of organ edge segmentation. Experimental results show that CSWin-UNet outperforms other methods in segmentation accuracy and computational efficiency. The method is lightweight, with a balanced architecture that improves segmentation accuracy and robustness. The CSWin-UNet achieves superior performance in various medical image segmentation tasks, including CT, MRI, and skin lesion segmentation. The method is effective in handling complex segmentation environments and provides higher segmentation accuracy. The results indicate that CSWin-UNet is more advanced and suitable for medical images of various modalities than other state-of-the-art methods. The method shows some deficiencies in challenging cases, such as significant differences in segmentation accuracy for different samples of the gallbladder and kidney regions. The pre-training of the model significantly impacts its performance, and further research is needed to explore end-to-end medical image segmentation methods.