Light-UNet: An Efficient Segmentation Network for Medical Image

In recent years, the Convolutional Neural Networks (CNN) models have been used in many medical image segmentation tasks. However, these models lack consideration for the actual application scenarios. In reality, some medical image segmentation tasks lack powerful GPU devices in the actual application scenarios, and can only be deployed in convenient devices with low computing power. Many existing models cannot be applied to mobile devices and point-of-care applications for rapid medical image segmentation. To cope with this issue, we developed Light-UNet, a segmentation model that reduces parameters and computational complexity while achieving superior performance, making it ideal for point-of-care and mobile devices medical imaging applications. Specifically, we proposed Small and Robust U-block (SRU) and Positional-Encoded MLP (PEMLP) blocks, they can extract and fuse local and global information, maintaining efficiency without computational complexity and parameter surge. It has been validated on multiple datasets, demonstrating its effectiveness in improving segmentation outcomes compared to state-of-the-art medical image segmentation architectures.