Decentralized AI in Healthcare
Achieve Decentralized model training between the different region on real patient data.
Photo from AI SwedenDecentralized AI in Healthcare
One solution for issues with sharing health data to develop better models is decentralised AI, where a model is trained locally, and only model parameters get shared between sites. Sahlgrenska University Hospital team is currently involved in a large project on decentralised AI where we work together with AI Sweden and Region Halland (neighbouring healthcare providing area) to test this in practice.
What and why?
In the first step of the project, we are working on publicly available data and simulating the decentralised structure internally. The main goal is to share learnings via presentation in AI Sweden and Information driven healthcare community. We have two use cases that we’re actively working on (one with imaging data, one with tabular data). My main intrest is connected with Melanoma Image classification using more fair and accurate models.
My contribiution
Main tasks:
- Simulate FL setting between different datasets (internal/opensource, synthetic/real, etc) for two nodes (imitation of using VGR and Region Halland data),
- Build knowledge base on Decentralized AI.
Technologies used: Python, Pytorch
Methods used: Deep Neural Networks, Swarm Learning, Federated Learning, Classification of Skin Diseases
Learn more about the project
Project’s website: www.ai.se