The paper introduces Swin-UMamba, a novel Mamba-based model designed for medical image segmentation tasks, leveraging ImageNet-based pretraining. The authors highlight the challenges in capturing long-range global information in medical images, which are often overlooked by existing methods like CNNs and ViTs. Mamba-based models, known for their ability to model long sequences efficiently, are proposed to address these challenges. The study demonstrates that ImageNet-based pretraining significantly enhances the performance of Mamba-based models, outperforming both CNNs and ViTs. Experimental results on datasets such as AbdomenMRI, Endoscopy, and Microscopy show that Swin-UMamba achieves superior segmentation accuracy with a notable margin over its closest competitors. The code and models are publicly available, providing a valuable resource for further research and practical applications in medical image analysis.The paper introduces Swin-UMamba, a novel Mamba-based model designed for medical image segmentation tasks, leveraging ImageNet-based pretraining. The authors highlight the challenges in capturing long-range global information in medical images, which are often overlooked by existing methods like CNNs and ViTs. Mamba-based models, known for their ability to model long sequences efficiently, are proposed to address these challenges. The study demonstrates that ImageNet-based pretraining significantly enhances the performance of Mamba-based models, outperforming both CNNs and ViTs. Experimental results on datasets such as AbdomenMRI, Endoscopy, and Microscopy show that Swin-UMamba achieves superior segmentation accuracy with a notable margin over its closest competitors. The code and models are publicly available, providing a valuable resource for further research and practical applications in medical image analysis.