README.md

Synthetic Nuclei and Anntation Wizard (SNOW) Data Set

This repository contains the code for the following manuscript:

A Large-scale Synthetic Pathological Dataset for Deep Learning-enabled Segmentation of Breast Cancer

Citation

Please cite our paper if you use the SNOW dataset for any purpose.

@article{ding2023large,
  title={A large-scale synthetic pathological dataset for deep learning-enabled segmentation of breast cancer},
  author={Ding, Kexin and Zhou, Mu and Wang, He and Gevaert, Olivier and Metaxas, Dimitris and Zhang, Shaoting},
  journal={Scientific Data},
  volume={10},
  number={1},
  pages={231},
  year={2023},
  publisher={Nature Publishing Group UK London}
}

Dataset Access

SNOW dataset is uploaded to https://zenodo.org/record/6633721#.YuE33OzMJhE

Dependencies

Pytorch 1.6.0

Torchvision 0.7.0

segmentation-models-pytorch 0.2.1

Pillow 6.2.0

numpy 1.16.4

pandas 0.25.1

scikit-image 0.15.0

scikit-learn 0.21.3

Pillow 6.2.0

h5py 2.8.0

Usage

Step 1. Generating synthetic image.

https://github.com/NVlabs/stylegan2-ada-pytorch

Step 2. Annotating synthetic image automatically.

https://github.com/vqdang/hover_net

Step 3. Training teacher model on SNOW data set.

Using the code

CUDA_VISIBLE_DEVICES=1,2 python 1. self_training_teacher_train_val.py

Note: please set supervised = False for training teacher model under semi-supervised training. Otherwise, the code is used for supervised training experiments in Table 2.

Step 4. Training student model on SNOW data set.

Using the code

CUDA_VISIBLE_DEVICES=1,2 python 2. self_training_student_train_val.py

Step 5. Evaluating on TNBC data set.

Using the code

python 3. evaluation.py

Example