New AI Data Analysis Preserves Patient Privacy
This novel method of data analysis allows for collaboration between research facilities, while preserving anonymity of patient data. Photo courtesy of Penn Medicine News.

Artificial intelligence models are becoming more widely adopted for medical purposes. One major area of growth and potential is their use in analyzing MRI scans. Machine learning models have the potential to surpass human detection rates and become more accurate and reliable. Artificial intelligence can be a preferable method of data analysis as it is not susceptible to problems such as human error. If developed further, it may be able to recognize minute differences that are beyond the identification capacity of humans. This can result in earlier detection of illnesses, leading to more effective treatments.

However, to become more accurate, artificial intelligence systems rely on large datasets, from thousands of patients. Healthcare facilities often refrain from sharing medical data due to legal concerns of patient privacy, or cultural views on sharing medical information.

Researchers at the University of Pennsylvania School of Medicine have been developing a system to provide large data sets to support machine learning models while simultaneously refraining from infringing upon patient privacy. This new technique, federated learning, allows for learning models to be shared between multiple facilities on a centralized server. With this system, the decentralized devices using the models do not store the original datasets used to train the models. This allows multiple models to be compared, and more accurate “consensus models” to be created as a result.

A collaborative study was recently conducted on federated learning for brain tumor identification models, with 10 hospitals each contributing AI models trained with their respective data. The federative learning model created was compared to other collaborative data analysis techniques that did not preserve patient privacy. The consensus model created through federated learning performed 99% identically to the other methods in this study.

This novel technique must be approved by the FDA before it can become commercially available. Recent results have been promising, and as a result, larger studies are being planned. In an upcoming study, 30 institutions spanning nine countries will jointly research federated learning, after receiving a $1.2 million grant from the National Cancer Institute, of the National Institutes of Health. This study will focus on training a consensus AI model to analyze brain tumor data in brain scans. However, this technology has the potential to be used for other medical purposes outside of brain tumor identification.

Spyridon Bakas, PhD., from the University of Pennsylvania, said:

“The more data the computational model sees, the better it learns the problem, and the better it can address the question that it was designed to answer. Traditionally, machine learning has used data from a single institution, and then it became apparent that those models do not perform or generalize well on data from other institutions.”

Via: Penn Medicine News

For Further Information

This recent development presents interesting implications for the future of the medical imaging device market. iData Research has several medical imaging market reports that contain detailed market analyses, including in-depth interviews and procedural volumes.