The Erasmus Medical Center in Rotterdam has found a way to detect malignant brain tumors at an earlier stage during their collaborative research with Intel Labs and the Perelman School of Medicine at the University of Pennsylvania (Penn Medicine). To do this, the researchers made use of a new technique: federated learning. This is a way of combining machine learning (ML) with artificial intelligence (AI). The project demonstrated that it is possible to improve the detection of brain tumors by a third, as the organizations stated in a press release.
It is the largest medical federated learning study to date, which examined an unprecedented global dataset involving 71 institutions spanning six continents.
“Federated learning has tremendous potential in a host of domains, most notably in healthcare,” according to Jason Martin, chief engineer at Intel Labs. “The ability to protect sensitive information and data opens the door to future research and collaborations, especially in cases where datasets would otherwise be inaccessible.”
Making data accessible
Data accessibility has long been problematic in healthcare due to the national data protection laws in place, along with the General Data Protection Regulation (GDPR). This has made it practically impossible to conduct large-scale medical research and data sharing without compromising the privacy of patients. Intel’s federated learning hardware and software comply with data privacy requirements and also protect data integrity, privacy and security through the use of trusted computers.
Data privacy
The results from the Penn Medicine-Intel were achieved by processing large amounts of data in a decentralized system. This was done by making use of Intel federated learning technology combined with Intel® Software Guard Extensions (SGX). This technology removes barriers to data sharing that in the past prevented collaboration in comparable cancer and disease research. The system addresses a host of data privacy issues by keeping the raw data contained inside the hospital’s own network and only allowing model updates that are calculated based on that data to be sent to a central server (or aggregator), not the raw data itself.
Personalized treatments
“At Erasmus MC, this federated learning study made it possible for us to help improve automatic tumor detection without having to send any patient data,” radiologist Prof. Dr. Smits and biomedical researcher Dr. Van der Voort of Erasmus MC explains. “Automatic tumor detection is an important step for personalizing and monitoring treatment, and in order to develop this methodology, it was essential to use data from many different institutions. Thanks to this collaboration, we were able to do that easily, while still keeping control of our data in our own networks.”
Breakthrough in safe collaborations
“Federated learning marks a breakthrough in ensuring that multi-institutional collaborations are secure. It facilitates access to the largest and most diverse dataset ever seen so far in medical literature. All the while ensuring that all such data is kept confined to each institution at all times,” said senior author Spyridon Bakas, PhD, assistant professor of Pathology & Laboratory Medicine and of 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 will become. That, in turn, will improve our ability to understand and treat even very rarer diseases, such as glioblastoma.”
Unlocking data silos
In order to improve the treatment of disease, researchers need to be able to access large amounts of medical data – in most cases datasets that exceed the threshold that a single institution is capable of producing. The research demonstrates the effectiveness of federated learning at scale and the potential advantages that healthcare stands to benefit from when multisite data silos are unlocked. Among these is the early detection of disease, which can improve quality of life of a patient and/or extend their life span
The results of the research were published in the journal Nature Communications.
Selected for you!
Innovation Origins is the European platform for innovation news. In addition to the many reports from our own editors in 15 European countries, we select the most important press releases from reliable sources. This way you can stay up to date on what is happening in the world of innovation. Are you or do you know an organization that should not be missing from our list of selected sources? Then report to our editorial team.
Related Posts
