Overall Survival Prediction Experiments with Pycox

This repository contains the code used for our submission to the SNMMI AI Challenge 2023, where we secured 2nd place. The challenge focuses on using machine learning for overall survival prediction. Our approach leverages the Pycox library to build and evaluate survival models.

For more details about the competition, visit the official SNMMI AI Challenge page.

Getting Started

Prerequisites

• Conda: Ensure you have Anaconda or Miniconda installed.
• Python: Recommended version is Python 3.10.

Repository Setup

Clone this repository with submodules to ensure you have all the dependencies:

git clone --recursive [email protected]:ipolharvard/snmmi-ai-challenge-2023.git
cd snmmi-ai-challenge-2023

Environment Setup

Create and activate a Conda environment for the project:

conda create -n snmmi-challenge python=3.10 -y
conda activate snmmi-challenge

Install the package in development mode along with the necessary dependencies:

pip install -e .[jupyter,dev]

Verify Setup

To ensure the CLI is installed correctly, run:

os run --help

This should display the available commands and options.

Data Preparation

The dataset required for the experiments is not included in the repository. You need to place the challenge datasets in the data directory as follows:

snmmi-ai-challenge-2023/
├── pyproject.toml
├── README.md
├── results/                <-- JSON results will be saved here
├── scripts/
├── ospkg/
│   └── data/               <-- Place your data files here
│       ├── SNMMI_CHALLENGE_TRAINING_V22OCT2023.xlsx
│       └── SNMMI_CHALLENGE_TESTING_V01112023.xlsx

Notebooks

The repository includes Jupyter notebooks for exploratory data analysis and result visualization. After setting up the environment, launch Jupyter:

jupyter notebook

Ensure that the required data is placed correctly in the data directory as the notebooks assume the following structure:

results/  <-- Results JSONs generated by experiments
data/     <-- Input datasets

Experiment Example

An example configuration and script to run an experiment can be found in the test folder. This provides a reference for setting up and executing your own experiments. Or available models you can find at ospkg/constants.

Acknowledgments

We thank the organizers of the SNMMI AI Challenge 2023 and the contributors of the Pycox library for their excellent work.