Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DiffSeg: A Segmentation Model for Skin Lesions Based on Diffusion Difference

Zhihao Shuai, Yinan Chen, Shunqiang Mao, Yihan Zho, Xiaohong Zhang

TL;DR

DiffSeg tackles weakly supervised medical image segmentation for skin lesions by leveraging diffusion difference from denoising diffusion probabilistic models to extract noise-based semantic cues. The method generates multiple segmentation outputs across diffusion steps to visualize consistency and ambiguity, and it quantifies uncertainty with Generalized Energy Distance ($D_{GED}^2$), followed by boundary refinement with DenseCRF. Evaluated on the ISIC 2018 dataset, DiffSeg outperforms state-of-the-art U-Net-based methods in Dice, Precision, and Recall. The work provides interpretable uncertainty metrics and a practical post-processing pipeline that improves segmentation accuracy and supports clinical decision-making.

Abstract

Weakly supervised medical image segmentation (MIS) using generative models is crucial for clinical diagnosis. However, the accuracy of the segmentation results is often limited by insufficient supervision and the complex nature of medical imaging. Existing models also only provide a single outcome, which does not allow for the measurement of uncertainty. In this paper, we introduce DiffSeg, a segmentation model for skin lesions based on diffusion difference which exploits diffusion model principles to ex-tract noise-based features from images with diverse semantic information. By discerning difference between these noise features, the model identifies diseased areas. Moreover, its multi-output capability mimics doctors' annotation behavior, facilitating the visualization of segmentation result consistency and ambiguity. Additionally, it quantifies output uncertainty using Generalized Energy Distance (GED), aiding interpretability and decision-making for physicians. Finally, the model integrates outputs through the Dense Conditional Random Field (DenseCRF) algorithm to refine the segmentation boundaries by considering inter-pixel correlations, which improves the accuracy and optimizes the segmentation results. We demonstrate the effectiveness of DiffSeg on the ISIC 2018 Challenge dataset, outperforming state-of-the-art U-Net-based methods.

DiffSeg: A Segmentation Model for Skin Lesions Based on Diffusion Difference

TL;DR

DiffSeg tackles weakly supervised medical image segmentation for skin lesions by leveraging diffusion difference from denoising diffusion probabilistic models to extract noise-based semantic cues. The method generates multiple segmentation outputs across diffusion steps to visualize consistency and ambiguity, and it quantifies uncertainty with Generalized Energy Distance (), followed by boundary refinement with DenseCRF. Evaluated on the ISIC 2018 dataset, DiffSeg outperforms state-of-the-art U-Net-based methods in Dice, Precision, and Recall. The work provides interpretable uncertainty metrics and a practical post-processing pipeline that improves segmentation accuracy and supports clinical decision-making.

Abstract

Weakly supervised medical image segmentation (MIS) using generative models is crucial for clinical diagnosis. However, the accuracy of the segmentation results is often limited by insufficient supervision and the complex nature of medical imaging. Existing models also only provide a single outcome, which does not allow for the measurement of uncertainty. In this paper, we introduce DiffSeg, a segmentation model for skin lesions based on diffusion difference which exploits diffusion model principles to ex-tract noise-based features from images with diverse semantic information. By discerning difference between these noise features, the model identifies diseased areas. Moreover, its multi-output capability mimics doctors' annotation behavior, facilitating the visualization of segmentation result consistency and ambiguity. Additionally, it quantifies output uncertainty using Generalized Energy Distance (GED), aiding interpretability and decision-making for physicians. Finally, the model integrates outputs through the Dense Conditional Random Field (DenseCRF) algorithm to refine the segmentation boundaries by considering inter-pixel correlations, which improves the accuracy and optimizes the segmentation results. We demonstrate the effectiveness of DiffSeg on the ISIC 2018 Challenge dataset, outperforming state-of-the-art U-Net-based methods.
Paper Structure (15 sections, 13 equations, 5 figures, 1 table)

This paper contains 15 sections, 13 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. RELATED WORKS
  3. METHOD
  4. Differentiation metrics
  5. Aleatoric Uncertainty Metrics
  6. Segmentation results optimization
  7. EXPERIMENTS
  8. Model Training
  9. Index Evaluation
  10. RESULTS & DISCUSSION
  11. Medical Image Segmentation
  12. Aleatoric Uncertainty Measurement
  13. Optimization for Multi-output
  14. CONCLUSION
  15. ACKNOWLEDGMENTS

Figures (5)

  • Figure 1: Strategies for measuring ambiguity and uncertainty in multi-output results and optimizing outcomes.
  • Figure 2: The experimental process includes the original images, noise under different semantics, noise difference grayscale images, and binarized segmentations.
  • Figure 3: Segmentation results obtained under different noise addition iterations (from 60 to 150).
  • Figure 4: After calculation, the consistent and ambiguous regions are obtained.
  • Figure 5: After calculation, the consistent and ambiguous regions are obtained.