Unlock 1000x more data for medical research with our new research community for federated learning in medical imaging. Scroll for more information ↓
Today in the field of medical research, far more data exists than any particular researcher is able to use because that data is spread across thousands of medical institutions around the world. Acquiring a copy of the data from another institution is often time consuming, legally burdensome, and expensive for a variety of reasons (privacy, security, IP, competitive advantage, legal limits, to name but a few). As a result, a majority of research is conducted only on data within one or two institutions.
If we could allow researchers to conduct research across institutions without needing to acquire a copy of the data; the costs of data acquisition would decrease, the amount of data used in the typical medical research project would multiply by several orders of magnitude, and a wave of scientific breakthroughs would ensue.
We believe privacy enhancing technologies (PETs) are the key to unlocking cross-institutional collaboration. They have already been shown to facilitate cross-organization scientific collaboration at small scales (2-10 institutions), but at their core privacy enhancing technologies are partnership enhancing technologies, which means that they won’t actually unlock orders of magnitude more data in a scientific field until the members of that field have confidence in their efficacy, understand the positive outcomes from their own participation, and acquire the skills to actively create PET-driven data consortiums.
The scientific community needs a demonstration of how PETs can create data partnerships at scale (without sharing data) — a template that every scientific discipline can subsequently follow, with significant evidence that this end-to-end partnership procedure is safe, effective, and greatly advantageous for all participants.
We’d like to invite you to participate in OpenMined’s Medical Federated Learning Program, a new program designed to help large numbers of medical researchers meet each other, establish research projects, and demonstrate to the world that they can achieve scientific breakthroughs by pooling their training data using privacy enhancing technologies (i.e. without sharing data with each other) for key medical research tasks.
To help get things started, we are recruiting for our first project, training the most accurate medical classifier to date on the largest-scale federated network yet. The first phase of this project will be educational and will only use public data that OpenMined provides. In later phases when you are more comfortable with the technology, we will build into more advanced exercises where you and other participants can bring your own training data.
As a participant in The Medical Federated Network you will have a chance to…
Be a co-author on the largest-scale federated learning pilot ever executed, and the first federated learning pilot to train across 100+ institutions.
Be a co-author on training the most accurate medical classifier to date, with the chance to definitively solve the task using standard AI techniques applied to orders of magnitude more data.
Learn hands-on how to take these techniques into your own research, unlocking orders of magnitude more data in your own field of study.
See progress on The Medical Federated Learning Program's Phase 01
Currently in collaboration with researchers from 108 different institutions. With a total of 123 nodes launched.
Were able to train a model across 100 of the domains deployed
4 Open Houses held with events like speaker presentations, small group discussion, and hands-on PETs experimentation.