Prostate Cancer

RationAI, FAIR

Michal Jakubík, Matěj Gallo, Vít Musil, Matěj Pekár

An implementation of a pipeline for training a binary classifier to detect prostate cancer using Whole Slide Images (WSIs). Built with PyTorch Lightning, Hydra and MlFlow.

Due to the unavailability of the dataset used in this project, users will need to create their own data source to utilize the pipeline.

Installation

To install the necessary dependencies, follow these steps:

1. Install PDM
2. Clone the repository.
3. Install the dependencies using PDM.

git clone [email protected]:rationai/prostate-caner.git
cd prostate-cancer
pdm install

Getting Started

Implementing Custom Tile Extraction

First, implement your custom extract_tile method in prostate_cancer/datamodule/datasets/base_wsi.py:

def extract_tile(
    slide_fp: Path, coord_x: int, coord_y: int, tile_size: int, level: int
) -> NDArray[np.uint8]:
    """Extracts a tile from a slide using the supplied coordinate values.

    Args:
        slide_fp (Path): Path to the slide.
        coord_x (int): Coordinates of a tile to be extracted at OpenSlide level 0 resolution.
        coord_y (int): Coordinates of a tile to be extracted at OpenSlide level 0 resolution.
        tile_size (int): Size of the tile to be extracted.
        level (int): Resolution level from which tile should be extracted.

    Returns:
        NDArray: RGB Tile represented as numpy array.
    """

Creating Custom Data Source

Next, create a custom data source by subclassing the BaseDataSource class in prostate_cancer/datamodule/datasources.py.

Updating Configuration Files

Finally, add your custom data sources to the configuration files (train.yaml, train_without_valid.yaml, test.yaml) located in the conf/experiment/cancer_bc directory.

Training the Model

To train the model, run the following command:

pdm train +experiment=cancer_bc/train

License

The project is licensed under the MIT license.