Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Paired Image Generation with Diffusion-Guided Diffusion Models

Haoxuan Zhang, Wenju Cui, Yuzhu Cao, Tao Tan, Jie Liu, Yunsong Peng, Jian Zheng

TL;DR

This work tackles the shortage of annotated DBT mass lesion data caused by high concealment of lesions. It introduces Paired Image Generation (PIG), an unconditional but paired-diffusion approach that trains a diffusion guider to generate paired DBT slices and lesion masks without external conditioning, enabling their use in supervised segmentation. The authors derive that paired image generation can be realized by two mutually guiding diffusion processes and demonstrate a training/sampling framework that yields higher-quality paired data (FID around 15.9) and improved segmentation performance (Dice around 60%) when augmenting DBTMassSeg. The approach improves data efficiency for DBT mass segmentation and provides a practical, code-released pipeline for generating annotated medical image pairs to boost downstream tasks.

Abstract

The segmentation of mass lesions in digital breast tomosynthesis (DBT) images is very significant for the early screening of breast cancer. However, the high-density breast tissue often leads to high concealment of the mass lesions, which makes manual annotation difficult and time-consuming. As a result, there is a lack of annotated data for model training. Diffusion models are commonly used for data augmentation, but the existing methods face two challenges. First, due to the high concealment of lesions, it is difficult for the model to learn the features of the lesion area. This leads to the low generation quality of the lesion areas, thus limiting the quality of the generated images. Second, existing methods can only generate images and cannot generate corresponding annotations, which restricts the usability of the generated images in supervised training. In this work, we propose a paired image generation method. The method does not require external conditions and can achieve the generation of paired images by training an extra diffusion guider for the conditional diffusion model. During the experimental phase, we generated paired DBT slices and mass lesion masks. Then, we incorporated them into the supervised training process of the mass lesion segmentation task. The experimental results show that our method can improve the generation quality without external conditions. Moreover, it contributes to alleviating the shortage of annotated data, thus enhancing the performance of downstream tasks. The source code is available at https://github.com/zhanghx1320/PIG.

Paired Image Generation with Diffusion-Guided Diffusion Models

TL;DR

This work tackles the shortage of annotated DBT mass lesion data caused by high concealment of lesions. It introduces Paired Image Generation (PIG), an unconditional but paired-diffusion approach that trains a diffusion guider to generate paired DBT slices and lesion masks without external conditioning, enabling their use in supervised segmentation. The authors derive that paired image generation can be realized by two mutually guiding diffusion processes and demonstrate a training/sampling framework that yields higher-quality paired data (FID around 15.9) and improved segmentation performance (Dice around 60%) when augmenting DBTMassSeg. The approach improves data efficiency for DBT mass segmentation and provides a practical, code-released pipeline for generating annotated medical image pairs to boost downstream tasks.

Abstract

The segmentation of mass lesions in digital breast tomosynthesis (DBT) images is very significant for the early screening of breast cancer. However, the high-density breast tissue often leads to high concealment of the mass lesions, which makes manual annotation difficult and time-consuming. As a result, there is a lack of annotated data for model training. Diffusion models are commonly used for data augmentation, but the existing methods face two challenges. First, due to the high concealment of lesions, it is difficult for the model to learn the features of the lesion area. This leads to the low generation quality of the lesion areas, thus limiting the quality of the generated images. Second, existing methods can only generate images and cannot generate corresponding annotations, which restricts the usability of the generated images in supervised training. In this work, we propose a paired image generation method. The method does not require external conditions and can achieve the generation of paired images by training an extra diffusion guider for the conditional diffusion model. During the experimental phase, we generated paired DBT slices and mass lesion masks. Then, we incorporated them into the supervised training process of the mass lesion segmentation task. The experimental results show that our method can improve the generation quality without external conditions. Moreover, it contributes to alleviating the shortage of annotated data, thus enhancing the performance of downstream tasks. The source code is available at https://github.com/zhanghx1320/PIG.

Paper Structure

This paper contains 20 sections, 1 theorem, 12 equations, 3 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Methodology
  3. Revisiting Diffusion Models
  4. The Forward Process.
  5. The Reverse Process.
  6. Limitations.
  7. Paired Image Generation
  8. The Forward Process.
  9. The Reverse Process.
  10. Experiments
  11. Datasets and Experimental Setup
  12. DBTMassSeg.
  13. Experimental Setup.
  14. Results
  15. Focused Generation of Lesion Regions.
  16. ...and 5 more sections

Key Result

proposition thmcounterproposition

The joint denoising process $p_\theta$ of the paired images can be implemented with two conditional denoising processes $p_x$ and $p_y$.

Figures (3)

  • Figure 1: Hidden mass lesions in DBT slices.
  • Figure 2: Graphical model for paired image generation process.
  • Figure 3: The progressive generation process of PIG.

Theorems & Definitions (2)

  • proposition thmcounterproposition
  • proof