HealthAI: TNM multi-centre dashboard

This repository contains the code for the HealthAI dashboard, which aims in demonstrating the potential of multi-centre analysis in a federated fashion. This dashboard showcases the feasibility of the federated approach using TNM data of non-small cell lung cancer patients. The data is distributed across five different Dutch hospitals: NKI, Isala, UMCG, Radboudumc and Maastro Clinic.

The project goal was to setup data nodes in each of the hospitals and investigate the feasibility of federated analysis for different data science typical tasks: simple statistics, supervised learning, and unsupervised learning. There has been numerous proof-of-concepts that shows the usefulness of federated analysis in the healthcare and other sectors. With this project, we show how to execute this approach in a real-life scenario and provide material that can be used as a template in any sector.

Federated learning infrastructures

In this project we use two different federated learning infrastructures: Vantage6 and International Data Spaces - IDS. Their documentations explain how to deploy the different infrastructures.

Input data

We assume that data nodes were deployed and that they hold TNM data of non-small cell lung cancer patients. In particular, a patient identifier (id), TNM stages (t, n, m), overall cancer stage (stage), diagnosis date (date_of_diagnosis), date of last follow-up (date_of_fu), and patient vital status (vital_status). The data should follow the common data model described in the cdm.json file in the input directory.

Run dashboard

You can run the dashboard either locally or using the docker image. Below you can see the required steps for each case.

Run with docker

You can run the dashboard with docker. Firstly, you need to create and edit a config.py file with the appropriate input for the vantage6 user client and the tasks. Please see an example located at pages/config_example.py. After that, you can simply run the following docker command to start the dashboard:

docker run --rm -d \
    --name healthai-dashboard \
    -p 5000:5000 \
    -v $(pwd)/config.py:/pages/config.py:ro \
    ghcr.io/maastrichtu-cds/healthai-dashboard:latest

Run locally

Install the dependencies

We advise you to create a Python virtual environment, using your favourite method, and install the dependencies from the requirements list:

pip install -r requirements.txt 
cp pages/config_example.py pages/config.py

Notice that you need to edit the config.py file with the appropriate input for the vantage6 user client and the tasks. This code was developed and tested with Python 3.10.

Run dashboard

You can run the dashboard by simply activating your virtual environment and then running the index.py script, an example of how to do it:

source .venv/bin/activate 
python index.py

Acknowledgments

This project was financially supported by the AiNed foundation.