Facebook and NYU Langone Health announced their commitment to making MRI scans more efficient through the use of AI more than one year ago, and a new story from Popular Science provides a glimpse into what the two have been up to.
Gina Ciavarra, a study participant and radiologist at NYU, along with others tried about 1,000 different AI software variations using real MRI data before landing on the one they tested. Their algorithm produces MRI knee scans using machine learning, which are then reviewed by experts such as Ciavarra. They’ve even injected some noise data into the images to test whether radiologists can pick out artificially generated images, according to the story.
The research—currently being prepared for peer-review—is especially unique, Maciej Mazurowsky, an associate professor at Duke University, told Popular Science. That's because it's attempting to create images using AI, rather than having the algorithm read real world scans.
"I think that this is a very exciting and important direction of study," said Mazurowsky, who specializes in AI in radiology but is not involved in the investigation. "It's different from what most of the radiology AI studies are."
Read more about the project below.