The development of Intel and Penn Medicine uses an artificial intelligence (AI) technique to protect privacy, called federated learning . The work of the Perleman School of Medicine is supported by the Information Technology for Cancer Research program of the National Cancer Institute and some of the National Institutes of Health through a three-year grant valued at $ 1.2 million allocated to the investigator. Principal, Spyridon Bakas, of the Center for Biomedical Imaging Computing and Analytics at the University of Pennsylvania.
They say at Intel that “artificial intelligence has great potential in the early detection of brain tumors, but to realize all that potential, more data is required than any medical center can collect on its own. Employing our software and hardware, as well as the support of some of the brightest minds at Intel Labs, we are collaborating with the University of Pennsylvania and a federation of 29 medical centers to advance brain tumor identification, while protecting information. patient staff ”.
How does it work?
Penn Medicine and 29 other health and research organizations from the US, Canada, UK, Germany, the Netherlands, Switzerland and India will employ federated learning, a distributed machine learning approach that enables organizations collaborate on deep learning projects without the need to share patient data.
Penn Medicine itself and Intel Labs were the first to publish a federated learning study in the field of medical imaging, specifically demonstrating that the federated learning method was capable of training a model with an accuracy equivalent to 99% of a model trained using the traditional approach without privacy protection. This study was initially presented during the International Conference on Medical Imaging Computing and Computer Aided Intervention held in Granada in 2018. The new work will leverage Intel software and hardware to implement federated learning in a way that provides additional privacy protection. both to the model and to the data.
Machine learning
The scientific community understands that machine learning training requires large and diverse data sets that no single institution can collect on its own. Algorithms to identify brain tumors will begin this year using an expanded version of the data set from the International Brain Tumor Segmentation Challenge. This federation will grant medical researchers access to much larger amounts of health data, while protecting the security of this data.
According to the American Brain Tumor Association (ABTA), almost 80,000 people will be diagnosed with a brain tumor throughout this year, of which more than 4,600 will be children. In order to build and train a model capable of detecting a brain tumor that could help in the early detection and improve the results of these patients, researchers require large amounts of relevant medical data. However, maintaining the privacy and security of this data is essential, and that is where federated learning using Intel technology comes in. Using this method, researchers from all participating organizations will be able to collaborate in the creation and training of an algorithm to detect brain tumors while protecting sensitive medical data.
In 2020, Penn and the 29 international health and research organizations will use federated learning hardware and software provided by Intel to produce a new advanced model of AI trained using the largest brain tumor data set ever collected, and all without need for each individual contributor to share their sensitive patient data. A subset of agencies is expected to participate in the beginning of the first phase of this federation, including the University Hospital of Pennsylvania, Washington University of St. Louis, the University of Pittsburgh Medical Center, Vanderbilt University, the Queen’s University, the Technical University of Munich, the University of Bern, King’s College London and the Tata Memorial Hospital.
