UNeCt: MLP-based image segmentation network

Abstract

Medical image segmentation is a necessary prerequisite for the development of healthcare systems, especially for disease diagnosis and treatment planning. UNet has become the de facto standard in various medical image segmentation tasks with great success. However, because the inherent local nature of convolutional operations makes UNet usually limited in explicitly modeling long-term dependencies, and because the huge parameters and computational complexity of UNet and its variants make UNet and its variants perform poorly for fast image segmentation in medical applications, we propose a new network structure (UNeCt) based on the UNet structure. U-sing a tokenized MLP in the latent space reduces the number of parameters and computational complexity, while being able to produce a better representation to aid segmentation. The network also includes skip connections between encoders and decoders at all levels. The results show that we achieve a good balance between the number of parameters, computational complexity and segmentation performance.