DCGAN in PyTorch for Histopathological Image Generation

This repository implements a DCGAN (Deep Convolutional Generative Adversarial Network) for generating histopathological images, specifically glomerulus pathologies of the kidney represented by 12 classes. The code is adapted from Chapter 4 & 5 of Hands-on Generative Adversarial Networks with PyTorch 1.0 by Hany, J. & Walters, G. (2019).

Prerequisites

• Linux OS
• Python 3
• CPU or NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

• Clone this repository:

  git clone https://github.com/m4ln/pytorch_dcgan.git
  cd pytorch_dcgan

• Install dependencies via pip

  pip install -r requirements.txt
  

  Note: It might be necessary to install PyTorch manually from https://pytorch.org/get-started/locally/

Training

• If no input arguments are provided, the model is trained on the MNIST dataset

  python train.py

• To train on your own data, provide the arguments via argparse (check inside train.py)
• The directory to your input data should contain a subfolder of images for each class

Generating Fake Samples/Testing

• To generate new samples, run (by default using the MNIST trained model as in train.py):

  python test.py

• To test on your own data, provide the arguments via argparse (check inside test.py)

Citation

If you use this project for your research, please cite our paper.

@article{weis2022assessment,
  title={Assessment of glomerular morphological patterns by deep learning algorithms},
  author={Weis, Cleo-Aron and Bindzus, Jan Niklas and Voigt, Jonas and Runz, Marlen and Hertjens, Svetlana and Gaida, Matthias M and Popovic, Zoran V and Porubsky, Stefan},
  journal={Journal of Nephrology},
  volume={35},
  number={2},
  pages={417--427},
  year={2022},
  doi = {10.11588/data/8LKEZF},
  publisher={Springer}
}