Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diffusion-based Data Augmentation for Nuclei Image Segmentation

Xinyi Yu, Guanbin Li, Wei Lou, Siqi Liu, Xiang Wan, Yan Chen, Haofeng Li

TL;DR

The study tackles the data scarcity challenge in nuclei segmentation by formulating a diffusion-based data augmentation pipeline that synthesizes paired nuclei structures and histopathology images. It introduces an unconditional diffusion model to generate nuclei structures from real instance maps and a conditional diffusion model, guided by classifier-free guidance and SPADE, to produce realistic images aligned with those structures. The approach yields synthetic paired samples that, when added to a small set of labeled data, achieve comparable or superior segmentation performance to fully-supervised baselines across datasets and models. This demonstrates the practical impact of diffusion-based augmentation for efficient histopathology analysis and supports broader adoption in data-limited biomedical imaging tasks.

Abstract

Nuclei segmentation is a fundamental but challenging task in the quantitative analysis of histopathology images. Although fully-supervised deep learning-based methods have made significant progress, a large number of labeled images are required to achieve great segmentation performance. Considering that manually labeling all nuclei instances for a dataset is inefficient, obtaining a large-scale human-annotated dataset is time-consuming and labor-intensive. Therefore, augmenting a dataset with only a few labeled images to improve the segmentation performance is of significant research and application value. In this paper, we introduce the first diffusion-based augmentation method for nuclei segmentation. The idea is to synthesize a large number of labeled images to facilitate training the segmentation model. To achieve this, we propose a two-step strategy. In the first step, we train an unconditional diffusion model to synthesize the Nuclei Structure that is defined as the representation of pixel-level semantic and distance transform. Each synthetic nuclei structure will serve as a constraint on histopathology image synthesis and is further post-processed to be an instance map. In the second step, we train a conditioned diffusion model to synthesize histopathology images based on nuclei structures. The synthetic histopathology images paired with synthetic instance maps will be added to the real dataset for training the segmentation model. The experimental results show that by augmenting 10% labeled real dataset with synthetic samples, one can achieve comparable segmentation results with the fully-supervised baseline. The code is released in: https://github.com/lhaof/Nudiff

Diffusion-based Data Augmentation for Nuclei Image Segmentation

TL;DR

The study tackles the data scarcity challenge in nuclei segmentation by formulating a diffusion-based data augmentation pipeline that synthesizes paired nuclei structures and histopathology images. It introduces an unconditional diffusion model to generate nuclei structures from real instance maps and a conditional diffusion model, guided by classifier-free guidance and SPADE, to produce realistic images aligned with those structures. The approach yields synthetic paired samples that, when added to a small set of labeled data, achieve comparable or superior segmentation performance to fully-supervised baselines across datasets and models. This demonstrates the practical impact of diffusion-based augmentation for efficient histopathology analysis and supports broader adoption in data-limited biomedical imaging tasks.

Abstract

Nuclei segmentation is a fundamental but challenging task in the quantitative analysis of histopathology images. Although fully-supervised deep learning-based methods have made significant progress, a large number of labeled images are required to achieve great segmentation performance. Considering that manually labeling all nuclei instances for a dataset is inefficient, obtaining a large-scale human-annotated dataset is time-consuming and labor-intensive. Therefore, augmenting a dataset with only a few labeled images to improve the segmentation performance is of significant research and application value. In this paper, we introduce the first diffusion-based augmentation method for nuclei segmentation. The idea is to synthesize a large number of labeled images to facilitate training the segmentation model. To achieve this, we propose a two-step strategy. In the first step, we train an unconditional diffusion model to synthesize the Nuclei Structure that is defined as the representation of pixel-level semantic and distance transform. Each synthetic nuclei structure will serve as a constraint on histopathology image synthesis and is further post-processed to be an instance map. In the second step, we train a conditioned diffusion model to synthesize histopathology images based on nuclei structures. The synthetic histopathology images paired with synthetic instance maps will be added to the real dataset for training the segmentation model. The experimental results show that by augmenting 10% labeled real dataset with synthetic samples, one can achieve comparable segmentation results with the fully-supervised baseline. The code is released in: https://github.com/lhaof/Nudiff
Paper Structure (9 sections, 6 equations, 3 figures, 2 tables)

This paper contains 9 sections, 6 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Unconditional Nuclei Structure Synthesis
  4. Conditional Histopathology Image Synthesis
  5. Experiments and Results
  6. Implementation details
  7. Effectiveness of the proposed data augmentation method
  8. Generalization of the proposed data augmentation.
  9. Conclusion

Figures (3)

  • Figure 1: The illustration of diffusion model in the context of nuclei structure.
  • Figure 2: The proposed diffusion-based data augmentation framework. We first generate a nuclei structure with an unconditional diffusion model, and then generate images conditioned on the nuclei structure. The instance map from the nuclei structure is paired with the synthetic image to forms a new sample.
  • Figure 3: Visualization of synthetic samples. The first and second row show selected patches and corresponding nuclei structures from the 10% labeled subset of MoNuSeg dataset. The third and fourth row show selected synthetic images and corresponding nuclei with similar style as the real one in the same column.