PIK3CA Mutation Detection in Breast Cancer

Project : Weakly supervised learning for the detection of PIK3CA mutations in breast cancer.

Description : This project focuses on detecting the PIK3CA mutation in breast cancer using histopathology slide images. These high-resolution images offer a detailed view of tissue samples, essential for identifying cellular abnormalities. Despite limited examples and slide-level labels, we explore the potential of deep neural architectures to overcome these challenges and provide robust and reasonable predictions. We employ the CHOWDER model proposed by Owkin, a deep learning approach tailored for weakly supervised learning in medical image analysis. Our goal is to enhance the model's performance in classifying whole-slide images as PIK3CA mutant or wild-type.

Author

• [Sébastien Mandela] - Initial work - SebX-7879
• Challenge by Owkin

Project structure

mutation_detection/
├── data/
│   ├── supplementary_data/
│   │   ├── test_metadata.csv
│   │   └── train_metadata.csv
│   ├── test_input/
│   │   ├── images/
│   │   └── moco_features/
│   ├── train_input/
│   │   ├── images/
│   │   └── moco_features/
│   └── train_output.csv
├── datasets/
│   ├── __init__.py
│   ├── core.py
├── figures/
├── logs/
├── models/
│   ├── utils/
│   ├── __init__.py
│   └── chowder.py
├── test_output/
├── trainer/
├── utils/
├── .gitignore
├── baseline.ipynb
├── download_data.py
├── LICENSE
├── main.py
├── working_notebook.ipynb
├── README.md
└── requirements.txt

Libraries

Install the required libraries using :

pip install -r requirements.txt

Data

To download the data, run :

python download_data.py

The data is expected to have the same structure as above. Due to storage restrictions (max number of authorized files), we did not include the images in the loading process, but only the moco_features and metadata files.

Training the model

The model is trained using the main.py script. The script uses the Trainer class from the trainer module to train the model. The Trainer class is responsible for loading the data, training the model, and evaluating the model on the test set. The Trainer class uses the CHOWDER model from the models module to train the model. Hyperparameters such as the model's parameters, learning rate, batch size, and number of epochs can be directly set in the main.py script.

To train the model, run :

python main.py

Additionally, a working notebook is available in the working_notebook.ipynb file. It contains the entire pipeline, from data preprocessing, model training, to evaluation. It makes use of Kfold cross-validation, and thus benefits from multiple instance learners, providing a more robust training and evaluation process than the main.py script.

License

This project is derived from another under-license project. See the LICENSE file for details.