‘Federated learning’ is a machine learning technology which enables interconnectivity between various entities, without requiring them to directly reveal and share their sensitive data.
According to NVIDIA, the application of federated learning in a healthcare system could allow organizations, such as hospitals, to collaborate with each other and develop artificial intelligence (AI) models out of their data assets, while concurrently protecting patients’ sensitive clinical data.
Therefore, the collaboration will see Owkin, NVIDIA and the King’s College of London (KCL) gradually create a dataset across healthcare organizations throughout the UK, which would form a potentially valuable reference point for research, while concurrently ensuring protection of patients’ privacy.
According to Owkin, the dataset will be used to pursue improvements in several clinical research areas, including cancer, heart failure, dementia and stroke.
During the first phase of the partnership, the King’s College of London (KCL) will adopt the federated learning software developed by Owkin, known as Owkin Connect, which utilizes real-world clinical trial data to predict patient outcomes and develop biomarkers.
Adoption of Owkin Connect will be enabled through NVIDIA’s matching software platforms, EGX, DGX, and Clara, with which Owkin’s service is connected.
KCL’s Medical Imaging and AI Centre for Value Based Healthcare will initially take on the federation of data from three universities; King's College London, Imperial College, and Queen Mary University of London, as well as four hospitals: Kings College Hospital, South London & Maudsley, Guy's & St. Thomas', and Barts Health.
Subsequently, the data network will be expanded to include 12 more hospitals across the country.
Sebastien Ourselin, professor of Healthcare Engineering at KCL, commented that Owkin Connect will become “the software layer that allows models to be built, orchestrated, secured and traced as they travel between our hospital and university partners.”
The professor added that the service will ensure that “the predictive models developed from patient data are representative and unbiased,” due to the fact that they will be ‘trained’ on the ‘widest possible’ population.
For the first phase of the dataset building, this would include one third of the population of London, with the professor noting that this “will extend far beyond London in coming years.”
“We truly see this architecture as the future of healthcare informatics," Ourselin concluded.