Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

VM-DDPM: Vision Mamba Diffusion for Medical Image Synthesis

Zhihan Ju, Wanting Zhou

TL;DR

This paper addresses data scarcity in medical imaging by introducing VM-DDPM, a diffusion-based synthesis framework that fuses CNN local perception with Vision Mamba's global modeling to achieve linear computational complexity. It presents a multi-level feature extractor (MSSBlock) and a basic encoder–decoder unit (SSLayer), and adds a Plug-and-Play Sequence Regeneration strategy for Cross-Scan Module operations to enhance spatial continuity. Through experiments on ChestXRay, BraTS2018, and ACDC, the approach shows state-of-the-art or competitive performance in FID and gains strong qualitative validation from radiologists. The findings suggest that SSM–CNN hybrids can effectively synthesize realistic, texture-consistent medical images at scale, addressing data bottlenecks while maintaining efficiency.

Abstract

In the realm of smart healthcare, researchers enhance the scale and diversity of medical datasets through medical image synthesis. However, existing methods are limited by CNN local perception and Transformer quadratic complexity, making it difficult to balance structural texture consistency. To this end, we propose the Vision Mamba DDPM (VM-DDPM) based on State Space Model (SSM), fully combining CNN local perception and SSM global modeling capabilities, while maintaining linear computational complexity. Specifically, we designed a multi-level feature extraction module called Multi-level State Space Block (MSSBlock), and a basic unit of encoder-decoder structure called State Space Layer (SSLayer) for medical pathological images. Besides, we designed a simple, Plug-and-Play, zero-parameter Sequence Regeneration strategy for the Cross-Scan Module (CSM), which enabled the S6 module to fully perceive the spatial features of the 2D image and stimulate the generalization potential of the model. To our best knowledge, this is the first medical image synthesis model based on the SSM-CNN hybrid architecture. Our experimental evaluation on three datasets of different scales, i.e., ACDC, BraTS2018, and ChestXRay, as well as qualitative evaluation by radiologists, demonstrate that VM-DDPM achieves state-of-the-art performance.

VM-DDPM: Vision Mamba Diffusion for Medical Image Synthesis

TL;DR

This paper addresses data scarcity in medical imaging by introducing VM-DDPM, a diffusion-based synthesis framework that fuses CNN local perception with Vision Mamba's global modeling to achieve linear computational complexity. It presents a multi-level feature extractor (MSSBlock) and a basic encoder–decoder unit (SSLayer), and adds a Plug-and-Play Sequence Regeneration strategy for Cross-Scan Module operations to enhance spatial continuity. Through experiments on ChestXRay, BraTS2018, and ACDC, the approach shows state-of-the-art or competitive performance in FID and gains strong qualitative validation from radiologists. The findings suggest that SSM–CNN hybrids can effectively synthesize realistic, texture-consistent medical images at scale, addressing data bottlenecks while maintaining efficiency.

Abstract

In the realm of smart healthcare, researchers enhance the scale and diversity of medical datasets through medical image synthesis. However, existing methods are limited by CNN local perception and Transformer quadratic complexity, making it difficult to balance structural texture consistency. To this end, we propose the Vision Mamba DDPM (VM-DDPM) based on State Space Model (SSM), fully combining CNN local perception and SSM global modeling capabilities, while maintaining linear computational complexity. Specifically, we designed a multi-level feature extraction module called Multi-level State Space Block (MSSBlock), and a basic unit of encoder-decoder structure called State Space Layer (SSLayer) for medical pathological images. Besides, we designed a simple, Plug-and-Play, zero-parameter Sequence Regeneration strategy for the Cross-Scan Module (CSM), which enabled the S6 module to fully perceive the spatial features of the 2D image and stimulate the generalization potential of the model. To our best knowledge, this is the first medical image synthesis model based on the SSM-CNN hybrid architecture. Our experimental evaluation on three datasets of different scales, i.e., ACDC, BraTS2018, and ChestXRay, as well as qualitative evaluation by radiologists, demonstrate that VM-DDPM achieves state-of-the-art performance.
Paper Structure (13 sections, 7 equations, 2 figures, 5 tables)

This paper contains 13 sections, 7 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Methods
  4. Preliminaries
  5. Vision Mamba DDPM (VM-DDPM)
  6. Multi-Scale State Space Block (MSSBlock)
  7. CSM with Sequence Regeneration
  8. Experiment
  9. Experimental Setup
  10. Main Results
  11. Ablation Studies
  12. Qualitative Evaluation by Radiologist
  13. Conclusion

Figures (2)

  • Figure 1: (a) The overall architecture of VM-DDPM. (b) SSLayer is the main construction layer of VM-DDPM. (c) MSSBlock is the core component of SSLayer, which includes time embedding and CSM operations.
  • Figure 2: Medical image synthesis samples and the real images.