Intel and the University of Pennsylvania Perelman School of Medicine (University of Pennsylvania School of Medicine) are forming an alliance, including 29 international medical and research institutions, using a privacy protection technology called “federal learning” to train artificial intelligence models that can identify brain tumors. This work is funded by the cancer research information technology (itcr) project of the National Cancer Institute (NCI) of the National Institutes of Health (NIH). It will provide research funds to Dr. Spyridon Bakas, chief researcher of the biomedical image computing and Analysis Center (cbica) of the University of Pennsylvania, with a total of US $1.2 million for three years.

AI has great potential in the early detection of brain tumors, but to give full play to its full potential, more data will be needed than any medical center. With the support of Intel Software and hardware and some Intel top talents, we are cooperating with the University of Pennsylvania and an alliance composed of 29 collaborative medical centers to promote the identification of brain tumors while protecting sensitive patient data.

Machine learning training needs a large number and rich and diverse data, which can not be held by a single institution, which has been widely recognized by our scientific community. We are coordinating an alliance composed of 29 collaborative international medical and research institutions. The alliance can use privacy protection machine learning technology including “federal learning”, and will train the most advanced AI medical model on this basis. This year, the alliance will begin to develop an algorithm for identifying brain tumors. The dataset of this algorithm comes from the significantly expanded dataset version of the international brain tumor segmentation (brats) challenge. The alliance will allow medical researchers to access much larger amounts of medical data than ever before, while protecting the security of these data.

Al t4518522638009344 Intel teamed up with the University of Pennsylvania to enable AI to recognize brain tumors

How did this happen? The University of Pennsylvania School of medicine and 29 medical and research institutions from the United States, Canada, Britain, Germany, the Netherlands, Switzerland and India use the technology of “federal learning”. This distributed machine learning method can enable organizations to cooperate in deep learning projects without sharing patient data.

Last year, the University of Pennsylvania School of medicine and Intel took the lead in publishing a paper on “federal learning” in the field of medical imaging, especially showing that the “federal learning” method can train a model to make its accuracy reach more than 99% of the accuracy of traditional non privacy training. The paper was originally published at the 2018 International Conference on medical image computing and computer-aided intervention (MICCAI) held in Granada, Spain. This new work will implement “federal learning” using Intel Software and hardware to provide additional privacy protection for models and data.

According to the American Brain Tumor Association (ABTA), nearly 80000 people will be diagnosed with brain tumors this year, including more than 4600 children. In order to train and establish a model for detecting brain tumors to help early detection and obtain better results, researchers need to obtain a large number of relevant medical data. However, maintaining data privacy and protecting data is very important, which is where “federal learning” using Intel technology comes into play. Through this method, researchers from all cooperative institutions will be able to work together to build and train an algorithm to detect brain tumors while protecting sensitive medical data.

In 2020, the University of Pennsylvania School of medicine and 29 international medical and research institutions will use Intel’s “federal learning” hardware and software to train on the largest brain tumor data set so far to generate a new AI model with the best performance, in which sensitive patient data will be stored separately in various cooperative institutions. It is expected that the cooperative organization groups involved in launching the first phase of the alliance include University of Pennsylvania Hospital, Washington University in St. Louis, University of Pittsburgh Medical Center, Vanderbilt University, Queen’s University, Munich Technical University, Bern University, King’s College London and Tata Memorial Hospital.

Tagged:

Leave a Reply

Your email address will not be published. Required fields are marked *