VM-UNet++: Advanced Nested Vision Mamba UNet for Precise Medical Image Segmentation

In the realm of medical image segmentation, the U-shaped structure comprising an encoder and a decoder has been demonstrated to be an extremely effective approach. They can be classified into models based on convolutional neural network (CNN), such as U-Net, and models based on Transformer, like Swin-UNet. Nevertheless, methods based on CNN are naturally constrained by the receptive field and have limitations in their capabilities. Although methods based on Transformer can capture global information, their computational complexity is extremely high and demands sufficient computing power support. Recently, state space model (SSM) exemplified by Mamba have emerged as a promising solution. They exhibit excellent performance in modeling long-range interactions, and maintain linear computational complexity, as well. Consequently, numerous works have employed Mamba to address computer vision tasks and achieved remarkable results. Among these, researchers combined U-Net with Vision Mamba (Vim) and proposed Vision Mamba U-Net (VM-UNet) to tackle the problem of medical image segmentation. Building on VM-UNet, we referred to the design concepts of UNet++ and implemented Vision Mamba UNet++ (VM-UNet++). Moreover, we conducted experiments on two public datasets, the ISIC17 and ISIC18 datasets, and the results indicated that our method shows a significant improvement over VM-UNet in medical image segmentation tasks.