Federated learning improves brain tumour detection by 33%

Intel Labs and the Perelman School of Medicine at the University of Pennsylvania (Penn Medicine) have completed a joint research study using federated learning — a distributed machine learning (ML) artificial intelligence (AI) approach — to help international healthcare and research institutions identify malignant brain tumours.

The medical federated learning study, published in Nature Communications, featured a global dataset from 71 institutions across six continents and demonstrated the ability to improve brain tumour detection by 33%.

In 2020, Intel and Penn Medicine announced the agreement to cooperate and use federated learning to improve tumour detection and improve treatment outcomes of a rare form of cancer called glioblastoma (GBM) — the most common and fatal adult brain tumour with a median survival of just 14 months after standard treatment. While treatment options have expanded over the past 20 years, there has not been an improvement in overall survival rates.

A new AI software platform called Federated Tumor Segmentation (FeTS) was used by radiologists to determine the boundary of a tumour and improve the identification of the operable region of tumours or ‘tumour core’. Radiologists annotated their data and used open federated learning (OpenFL), an open source framework for training machine learning algorithms, to run the federated training. The platform was trained on 3.7 million images from 6314 GBM patients across six continents — the largest brain tumour dataset to date.

Jason Martin, principal engineer at Intel Labs, said, “Federated learning has tremendous potential across numerous domains, particularly within health care, as shown by our research with Penn Medicine. Its ability to protect sensitive information and data opens the door for future studies and collaboration, especially in cases where datasets would otherwise be inaccessible.”

Data accessibility has long been an issue in health care due to state and national data privacy laws in the US, including the Health Insurance Portability and Accountability Act (HIPAA). This has made medical research and data sharing at scale almost impossible to achieve without compromising patient health information. Intel’s federated learning hardware and software comply with data privacy concerns and preserve data integrity, privacy and security through confidential computing.

The Penn Medicine–Intel result was accomplished by processing high volumes of data in a decentralised system which only allowed models from the raw data to be sent to the central server, rather than the data itself.

“All of the computing power in the world can’t do much without enough data to analyse,” said Rob Enderle, principal analyst, Enderle Group.

“This inability to analyse data that has already been captured has significantly delayed the massive medical breakthroughs AI has promised. This federated learning study showcases a viable path for AI to advance and achieve its potential as the most powerful tool to fight our most difficult ailments.”

“In this study, federated learning shows its potential as a paradigm shift in securing multi-institutional collaborations by enabling access to the largest and most diverse dataset of glioblastoma patients ever considered in the literature, while all data are retained within each institution at all times,” said senior author Spyridon Bakas, PhD, assistant professor of Pathology & Laboratory Medicine and Radiology, at the Perelman School of Medicine at the University of Pennsylvania.

“The more data we can feed into machine learning models, the more accurate they become, which in turn can improve our ability to understand and treat even rare diseases, such as glioblastoma.”

Through this project, Intel Labs and Penn Medicine have created a proof of concept for using federated learning to gain knowledge from data. The solution can significantly affect health care and other study areas, particularly among other types of cancer research.

Image credit: iStock.com/Sutthaburawonk