DiffMix: Diffusion Model-based Data Synthesis for Nuclei Segmentation and Classification in Imbalanced Pathology Image Datasets

By Hyun-Jic Oh and Won-Ki Jeong

This is official implementation code of DiffMix (MICCAI2023)

---

[Paper]

Updates

(Oct. 14th. 2024)

• Changed google drive link for the GLySAC dataset, as the original link is broken. * Original GLySAC data was provided by SONNET.
• Uploaded checkpoints

Abstract

• Nuclei segmentation and classification is a significant process in pathology image analysis. Deep learning-based approaches have greatly contributed to the higher accuracy of this task. However, those approaches suffer from the imbalanced nuclei data composition, which shows lower classification performance on the rare nuclei class. In this paper, we propose a realistic data synthesis method using a diffusion model. We generate two types of virtual patches to enlarge the training data distribution, which is for balancing the nuclei class variance and for enlarging the chance to look at various nuclei. After that, we use a semantic-label-conditioned diffusion model to generate realistic and high-quality image samples. We demonstrate the efficacy of our method by experiment results on two imbalanced nuclei datasets, improving the state-of-the-art networks. The experimental results suggest that the proposed method improves the classification performance of the rare type nuclei classification, while showing superior segmentation and classification performance in imbalanced pathology nuclei datasets.

Dataset Preparation

• CoNSeP
• GLySAC

Implementation

• Requirements

pip install -r requirements.txt

You can train SDM with the script "semantic-diffusion-model/scripts_ddim_sample" for two public pathology nuclei datasets "CoNSeP" and "GlySAC".

• Training code: image_train.py
• Sampling code: image_sample.py

Checkpoints

Pretrained checkpoints are available on Hugging Face.

• CoNSeP
• GLySAC

Acknowledgement

Our code is developed based on Semantic Image Synthesis via Diffusion Models (SDM) and GradMix.

Citation

Please cite us if you use our work

@inproceedings{oh2023diffmix,
  title={Diffmix: Diffusion model-based data synthesis for nuclei segmentation and classification in imbalanced pathology image datasets},
  author={Oh, Hyun-Jic and Jeong, Won-Ki},
  booktitle={International conference on medical image computing and computer-assisted intervention},
  pages={337--345},
  year={2023},
  organization={Springer}
}