VM-UNet++: Vision Mamba with Adaptive Feature Reweighting and Multiscale Fusion for Medical Image Segmentation

Medical image segmentation plays a vital role in contemporary clinical practice, offering clinicians quantitative analysis and visualization of lesions. This process aids in early diagnosis, disease evaluation, surgical planning, and monitoring the effects of treatment.However, how to effectively suppress background noise and accurately segment lesion boundaries poses challenges to the model. In order to solve these problems, we propose an enhanced medical image segmentation model, VM-UNet ++.Specifically, we have developed a novel multilevel feature enhancement mechanism that incorporates bilinear interpolation downsampling and large-kernel grouped attention gate (LGAG) modules, aimed at enhancing detail retention and boundary clarity. Furthermore, we introduce an SE-VSS module designed to dynamically modify the weights of feature maps, thereby improving the model's emphasis on critical features. Additionally, an asymmetric encoder-decoder architecture is utilized in our approach. We performed extensive experimental validation on several publicly available datasets, including Synapse, ISIC17, and ISIC18 datasets. The experimental results demonstrate that the VM-UNet++ model achieves outstanding performance across various evaluation metrics, validating its effectiveness and value in the domain of medical image segmentation.