NYU and Facebook Release Database to Teach AI to Read MRI Scans

Imaging scans, such as those taken using MRI machines and CT scanners, are hot targets for artificial intelligence research. Improving how well and how fast radiologists can spot lesions and other tissue characteristics can significantly improve the effectiveness of medicine overall. One problem regularly faced by researchers in this field is a shortage of data to work with. To help with this, the NYU School of Medicine has partnered with Facebook to make public a giant database of knee MRI scans and related data. The collaborators of the project, called fastMRI, hope that other researchers can use the data stash to create software to spot different characteristics on MRI scans for different conditions and diseases. NYU and Facebook is working on their own AI software to analyze imaging scans, but they’re also building tools to help others developing similar applications. The database includes over 1.5 million images from about 10,000 scans. 1,600 of the scans come with various anatomical measurement data. All the scans have patient information permanently removed and the database is fully HIPAA compliant. To soothe any recently rattled nerves, the announcement from NYU School of Medicine states that “No Facebook data of any kind are used in the project.”