Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diff-Lung: Diffusion-Based Texture Synthesis for Enhanced Pathological Tissue Segmentation in Lung CT Scans

Rezkellah Noureddine Khiati, Pierre-Yves Brillet, Radu Ispas, Catalin Fetita

TL;DR

This work targets the challenge of segmenting diverse ILD-related pathologies in lung CT where class imbalance hampers learning. It introduces DiffLung, a diffusion-based texture synthesis method conditioned on multi-class masks $m$ to generate diverse pathological textures, complemented by Class-Balanced Mask Ablated Training (CBMAT) and a mask-aware generation process. Quantitative results show DiffLung-CBMAT achieving higher $PSNR$ and $SSIM$ than GAN-based and Seg-guided diffusion baselines, and yielding Dice improvements for emphysema and fibrosis when used to augment a UNet. The approach enhances dataset diversity and class balance, improving automated ILD quantification and supporting clinical decision-making.

Abstract

Accurate quantification of the extent of lung pathological patterns (fibrosis, ground-glass opacity, emphysema, consolidation) is prerequisite for diagnosis and follow-up of interstitial lung diseases. However, segmentation is challenging due to the significant class imbalance between healthy and pathological tissues. This paper addresses this issue by leveraging a diffusion model for data augmentation applied during training an AI model. Our approach generates synthetic pathological tissue patches while preserving essential shape characteristics and intricate details specific to each tissue type. This method enhances the segmentation process by increasing the occurence of underrepresented classes in the training data. We demonstrate that our diffusion-based augmentation technique improves segmentation accuracy across all pathological tissue types, particularly for the less common patterns. This advancement contributes to more reliable automated analysis of lung CT scans, potentially improving clinical decision-making and patient outcomes

Diff-Lung: Diffusion-Based Texture Synthesis for Enhanced Pathological Tissue Segmentation in Lung CT Scans

TL;DR

This work targets the challenge of segmenting diverse ILD-related pathologies in lung CT where class imbalance hampers learning. It introduces DiffLung, a diffusion-based texture synthesis method conditioned on multi-class masks to generate diverse pathological textures, complemented by Class-Balanced Mask Ablated Training (CBMAT) and a mask-aware generation process. Quantitative results show DiffLung-CBMAT achieving higher and than GAN-based and Seg-guided diffusion baselines, and yielding Dice improvements for emphysema and fibrosis when used to augment a UNet. The approach enhances dataset diversity and class balance, improving automated ILD quantification and supporting clinical decision-making.

Abstract

Accurate quantification of the extent of lung pathological patterns (fibrosis, ground-glass opacity, emphysema, consolidation) is prerequisite for diagnosis and follow-up of interstitial lung diseases. However, segmentation is challenging due to the significant class imbalance between healthy and pathological tissues. This paper addresses this issue by leveraging a diffusion model for data augmentation applied during training an AI model. Our approach generates synthetic pathological tissue patches while preserving essential shape characteristics and intricate details specific to each tissue type. This method enhances the segmentation process by increasing the occurence of underrepresented classes in the training data. We demonstrate that our diffusion-based augmentation technique improves segmentation accuracy across all pathological tissue types, particularly for the less common patterns. This advancement contributes to more reliable automated analysis of lung CT scans, potentially improving clinical decision-making and patient outcomes
Paper Structure (13 sections, 9 equations, 1 figure, 2 tables, 1 algorithm)

This paper contains 13 sections, 9 equations, 1 figure, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Method
  3. Diffusion models
  4. Class-Balanced Mask Ablated Training (CBMAT)
  5. Mask Generation Process
  6. Experiments
  7. Dataset
  8. Generative model
  9. Quantitative Results
  10. Segmentation model
  11. Results and discussion
  12. Conclusion
  13. Acknowledgments

Figures (1)

  • Figure 1: Qualitative comparison for pathological texture synthesis in lung CT images (Green: Healthy Tissue, Blue: Emphysema, Red: Fibrosis, Pink: thorax cage and mediastinum).