Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DSVM-UNet : Enhancing VM-UNet with Dual Self-distillation for Medical Image Segmentation

Renrong Shao, Dongyang Li, Dong Xia, Lin Shao, Jiangdong Lu, Fen Zheng, Lulu Zhang

TL;DR

DSVM-UNet addresses the limitations of conventional UNet-based medical image segmentation by integrating dual self-distillation into a Vision Mamba–UNet framework. It introduces Projection Self-Distillation to align global semantic features across all layers and Progressive Self-Distillation to harmonize local features between adjacent layers, all without changing the core architecture. The approach yields state-of-the-art results on ISIC2017, ISIC2018, and Synapse benchmarks, with modest increases in mIoU and DSC and a favorable efficiency profile. The findings demonstrate that targeted feature-level knowledge transfer can substantially boost segmentation performance while maintaining computational efficiency, and the authors provide code at the project site for reproducibility.

Abstract

Vision Mamba models have been extensively researched in various fields, which address the limitations of previous models by effectively managing long-range dependencies with a linear-time overhead. Several prospective studies have further designed Vision Mamba based on UNet(VM-UNet) for medical image segmentation. These approaches primarily focus on optimizing architectural designs by creating more complex structures to enhance the model's ability to perceive semantic features. In this paper, we propose a simple yet effective approach to improve the model by Dual Self-distillation for VM-UNet (DSVM-UNet) without any complex architectural designs. To achieve this goal, we develop double self-distillation methods to align the features at both the global and local levels. Extensive experiments conducted on the ISIC2017, ISIC2018, and Synapse benchmarks demonstrate that our approach achieves state-of-the-art performance while maintaining computational efficiency. Code is available at https://github.com/RoryShao/DSVM-UNet.git.

DSVM-UNet : Enhancing VM-UNet with Dual Self-distillation for Medical Image Segmentation

TL;DR

DSVM-UNet addresses the limitations of conventional UNet-based medical image segmentation by integrating dual self-distillation into a Vision Mamba–UNet framework. It introduces Projection Self-Distillation to align global semantic features across all layers and Progressive Self-Distillation to harmonize local features between adjacent layers, all without changing the core architecture. The approach yields state-of-the-art results on ISIC2017, ISIC2018, and Synapse benchmarks, with modest increases in mIoU and DSC and a favorable efficiency profile. The findings demonstrate that targeted feature-level knowledge transfer can substantially boost segmentation performance while maintaining computational efficiency, and the authors provide code at the project site for reproducibility.

Abstract

Vision Mamba models have been extensively researched in various fields, which address the limitations of previous models by effectively managing long-range dependencies with a linear-time overhead. Several prospective studies have further designed Vision Mamba based on UNet(VM-UNet) for medical image segmentation. These approaches primarily focus on optimizing architectural designs by creating more complex structures to enhance the model's ability to perceive semantic features. In this paper, we propose a simple yet effective approach to improve the model by Dual Self-distillation for VM-UNet (DSVM-UNet) without any complex architectural designs. To achieve this goal, we develop double self-distillation methods to align the features at both the global and local levels. Extensive experiments conducted on the ISIC2017, ISIC2018, and Synapse benchmarks demonstrate that our approach achieves state-of-the-art performance while maintaining computational efficiency. Code is available at https://github.com/RoryShao/DSVM-UNet.git.
Paper Structure (13 sections, 5 equations, 1 figure, 4 tables)

This paper contains 13 sections, 5 equations, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Preliminaries
  4. Dual Self-distillation Framework
  5. Projection Self-Distillation
  6. Progressive Self-Distillation
  7. Total Loss
  8. experiments
  9. Implementation Details
  10. Comparison to the SOTA Models
  11. Ablation Studies
  12. Model Complexity
  13. Conclusion

Figures (1)

  • Figure 1: The entire framework of our DSVM-UNET methods is illustrated. On the right is the dual self-distillation framework, which comprises projection self-distillation and progressive self-distillation. VSS blocks are derived from VM-UNet ruan2024vm.