Diagnostics, Artificial Intelligence

UPenn, Intel partner to use federated learning AI for early brain tumor detection

The project will bring in 29 institutions from North America, Europe and India and will use privacy-preserved data to train AI models. Federated learning has been described as being born at the intersection of AI, blockchain, edge computing and the Internet of Things.

The University of Pennsylvania and chipmaker Intel are forming a partnership to enable 29 heatlhcare and medical research institutions around the world to train artificial intelligence models to detect brain tumors early.

The program will rely on a technique known as federated learning, which enables institutions to collaborate on deep learning projects without sharing patient data. The partnership will bring in institutions in the U.S., Canada, U.K., Germany, Switzerland and India. The centers – which include Washington University of St. Louis; Queen’s University in Kingston, Ontario; University of Munich; Tata Memorial Hospital in Mumbai and others – will use Intel’s federated learning hardware and software.

“It is widely accepted by our scientific community that machine learning requires ample and diverse data that no single institution can hold,” said University of Pennsylvania instructor Spyridon Bakas, of the university’s Center for Biomedical Image Computing and Analytics, in a statement. “We are coordinating a federation of 29 collaborating international healthcare and research institutions, which will be able to train state-of-the-art AI models for healthcare, using privacy-preserving machine learning technologies, including federated learning.”

The idea behind federated learning is that training AI models to detect brain tumors early requires researchers to have access to large amounts of data, but it is also essential that the data remain private. Intel and the university published a paper last January in the journal Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries for a study that they said marked the first use of federated learning for multi-institutional collaboration that enabled deep learning modeling without sharing patient data. They compared the results they got from federated learning with two alternative collaboration methods, finding that they failed to match federated learning’s performance.

Federated learning has been described as a decentralized AI model, which Lenovo’s Jed Record described in an article online as “born at the intersection of on-device AI, blockchain, and edge computing/[Internet of Things].” Its techniques are further described as being privacy-preserved by design.

Photo: MF3d, Getty Images