Intel and Penn Medicine (Perelman School of Medicine, University of Pennsylvania) have developed AI capable of detecting brain tumors based on federated learning.
This AI, however, places importance on patient privacy. More than 29 healthcare establishments located in 7 different countries (United States, Canada, United Kingdom, Germany, Netherlands, Switzerland and India) are participating in this collaborative project. It is based on medical data, but no case shares it.
This research is funded by the Informatics Technology for Cancer Research (ITCRI) program run by the National Institutes of Health (NIH), in the form of a grant of $ 1.2 million over three years.
Reconciling the need for analysis of medical data with respect for privacy
For this work to be effective, they must analyze a colossal amount of medical data. However, this need for mass analysis contradicts privacy laws or medical confidentiality. These two requirements therefore had to be reconciled in order to develop an effective tool while complying with the law.
” Together, we will begin to develop algorithms this year to identify brain tumors from a greatly expanded version of the international brain tumor segmentation dataset. This federation will allow medical researchers to access much greater amounts of health care data while ensuring that it is protected. Said Dr. Spyridon Bakas, a medical imaging specialist at the University of Pennsylvania and principal investigator for this study.
With this machine learning model that encrypts data, this work will now be done with Intel software and hardware to help protect privacy.
This federated learning model seems to be efficient since it can form a model with more than 99% of accuracy compared to a model with the traditional method.