Generalization of vision pre-trained models for histopathology

This is the repository for generalization of vision pre-trained models for histopathology:

Citing this work

If you use this code or the results for your research, please cite:

@article{sikaroudi2023generalization,
  title={Generalization of vision pre-trained models for histopathology},
  author={Sikaroudi, Milad and Hosseini, Maryam and Gonzalez, Ricardo and Rahnamayan, Shahryar and Tizhoosh, HR},
  journal={Scientific reports},
  volume={13},
  number={1},
  pages={6065},
  year={2023},
  publisher={Nature Publishing Group UK London}
}

Getting Started

These instructions will guide you on how to execute the code on your local machine for development and testing purposes.

Prerequisites

You need to have the following packages installed:

• Python 3.7+
• PyTorch 1.9.0+
• numpy
• pandas
• sklearn
• wandb
• torchvision
• PIL

You can install these packages using pip:

pip install torch torchvision numpy pandas sklearn wandb pillow

Configuration

To run the program, you need to have a configuration file (config.json). Here is a sample configuration file that you can use:

{
    "learning_rate": 0.0001,
    "momentum": 0.9,
    "epochs": 50,
    "batch_size": 64,
    "augmentation_in_training": false,
    "model": "kimianet",
    "pretrained": false,
    "kimianet_weight_path": "../kimianet_weights/KimiaNetPyTorchWeights.pth",
    "dataframe_root": "/isilon/datasets/camelyon17/",
    "trail_sites": ["center_0", "center_1", "center_2", "center_3", "center_4"],
    "holdout_trial_site": "center_0",
    "font_path": "/usr/share/fonts/type1/gsfonts/c059016l.pfb",
    "train_val_portions": [70,10]
}

Note: Please adjust the parameters according to your needs and availability of computational resources.

Running the code

To run the code, simply execute the main Python script:

python main.py

The script will then start training the model according to the parameters specified in the config.json file.

The model weights will be saved in the current directory with a name specified by the parameters in the config.json file.

Contributions

Your contributions are always welcome. If you find a bug or want to propose a new feature, feel free to open an issue or send a pull request.

Contact

If you need to get in touch with the maintainer of this project, please contact me at [email protected].

License

This project is licensed under MIT License.