H-vmunet: High-order Vision Mamba UNet for medical image segmentation

Abstract

In the field of medical image segmentation, variant models based on Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have been extensively developed and applied. However, CNNs often struggle with processing long-sequence information, while Vision Transformers exhibit low sensitivity to local feature extraction and face challenges with computational complexity. Recently, the emergence of state-space models (SSMs), particularly 2D-selective-scan (SS2D), has challenged the longtime dominance of traditional CNNs and ViTs as foundational modules in visual neural networks. In this paper, we extend the adaptation of SS2D by proposing a High-order Vision Mamba UNet (H-vmunet) model for medical image segmentation. Among them, the H-vmunet model includes the proposed novel High-order 2D-selective-scan (H-SS2D) and Local-SS2D module. The H-SS2D is used to reduce the introduction of redundant information when SS2D features are feature-learning in the global receptive field. The Local-SS2D module is used to improve the learning ability of local features in SS2D. We conducted comprehensive comparison and ablation experiments on three publicly available medical image datasets (ISIC2017, Spleen, and CVC-ClinicDB), and the results consistently demonstrate the strong performance of H-vmunet in medical image segmentation tasks. The code is available from https://github.com/wurenkai/H-vmunet.