Skip to main content

Official repo for Foundation Models for Quantitative Biomarker Discovery in Cancer Imaging [INSERT DOI]

Project description

Foundation Models for Quantitative Imaging Biomarker Discovery in Cancer Imaging

Overview

This is the the official repository for the paper Foundation Models for Quantitative Imaging Biomarker Discovery for Cancer Imaging
Suraj Pai, Dennis Bontempi, Vasco Prudente, Ibrahim Hadzic, Mateo Sokač, Tafadzwa L. Chaunzwa, Simon Bernatz, Ahmed Hosny, Raymond H Mak, Nicolai J Birkbak, Hugo JWL Aerts

General overview of the study. a. Foundation model pre-training. A foundation model, specifically a deep convolutional encoder model, was pre-trained by contrasting volumes with and without lesions. b. Clinical application of the foundation model. The foundation model was used to extract biomarkers and then evaluated for three classification tasks on diverse datasets. c. Foundation model implementation approaches The foundation model was adapted to specific use cases by extracting features or through fine-tuning (left). d. Evaluation against supervised models with selected performance metrics. We compared the performance of the foundation models against conventional supervised implementations, trained from scratch (left) and fine-tuned from a different task (right). The comparison was made through several criteria for the different use cases, including quantitative performance, stability, and biological analysis. Biological, clinical, and stability analyses are limited to use case 2 due to the availability of associated data.

Repository Structure

├── foundation_image_biomarker/ # Python package for the repository. You don't need to use this unless you want to contribute to the repo
├── data/ # Download and pre-processing scripts for the data
├── experiments/ # Configuration files to reproduce the final experiments of the paper
├── models/ # Notebooks to download and test each of the final models used in the paper
└── tests/

Quick Start

The easiest way to use this repository is by installing it through pip. This will load all the dependencies and allow you to load model weights dynamically.

pip install foundation-image-biomarker

Once you do this, you can look run each of our described pipelines

Data pre-processing

You can find detailed instructions for this here

Using all trained models

You can find detailed instructions for this here

Re-running all the experiments outlined in the paper

You can find detailed instructions for this here

🛡 License

License

This project is licensed under the terms of the MIT license. See LICENSE for more details.

📃 Citation

@misc{foundation-image-biomarker,
  author = {AIM-Harvard},
  title = {Official repo for "Foundation Models for Quantitative Biomarker Discovery in Cancer Imaging"},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/AIM-Harvard/foundation-cancer-image-biomarker}}
}

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

Built Distribution

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page