Intel Labs and the Perelman School of Medicine at the University of Pennsylvania (Penn Medicine) have completed a joint study using federated learning – a distributed machine learning (ML) artificial intelligence (AI) approach – to help healthcare institutions and research efforts to identify malignant brain tumors. It is the largest medical federated learning study to date, examining an unprecedented global dataset from 71 institutions on six continents. The Dutch Erasmus MC from Rotterdam is also involved. The project has shown that it is possible to improve the detection of brain tumors by 33%.
“Federated learning has enormous potential in many industries, particularly in healthcare, as our research with Penn Medicine demonstrates. The ability to protect sensitive information and data opens the door for future studies and collaborations, especially in cases where datasets would otherwise be inaccessible. Our work with Penn Medicine has the potential to positively impact patients around the world, and we look forward to further exploring the promise of federated learning,” said Jason Martin, chief engineer, Intel Labs
Make data accessible
Data accessibility has long been an issue in healthcare due to national data protection laws, including the General Data Protection Regulation (GDPR). This has made it nearly impossible to carry out medical research and data sharing at scale without compromising patient health information. Intel’s federated learning hardware and software meet data privacy requirements and protect data integrity, privacy, and security through confidential processing.
The Penn Medicine-Intel achievement was achieved by processing large amounts of data in a decentralized system. This was done using Intel Federated Learning Technology in conjunction with Intel® Software Guard Extensions (SGX). This technology removes barriers to data sharing that previously hindered collaboration in similar cancer and disease research. The system solves many data privacy issues by keeping the raw data within its hospital network and only allowing model updates calculated from that data to be sent to a central server or aggregator, not the raw data.
Radiologist Prof. Dr. Smits and biomedical researcher Dr. Van der Voort from Erasmus MC: “Through this federated learning study, we at Erasmus MC have been able to help improve automatic tumor detection, without having to send patient data. Automated tumor detection is an important step in tailoring and monitoring a treatment, and to develop this methodology it is essential to use data from many different institutions. With this partnership, we have been able to do this easily, while maintaining control over our data.”
“Federated learning offers a breakthrough in ensuring secure multi-agency collaborations. It allows access to the largest and most diverse data set ever seen in the literature, while all data is kept within each institution at all times,” he said senior author Spyridon Bakas, PhD, assistant professor of pathology and laboratory medicine and radiology at the University of Pennsylvania Perelman School of Medicine. “The more data we can feed into machine learning models, the more accurate they become. This, in turn, will improve our ability to understand and treat even rare diseases, such as glioblastoma.”
To improve disease treatment, researchers need access to vast amounts of medical data, in most cases datasets that exceed the threshold an institution can produce. Research demonstrates the effectiveness of federated learning at scale and the potential benefits healthcare can realize when multi-site data silos are opened. Benefits include early diagnosis of the disease, which can improve the quality of life or extend a patient’s lifespan.
The results of the Penn Medicine-Intel Labs study have been published in the peer-reviewed journal, Nature communications.
