Facebook AI Backed fastMRI Speeds Up MRI Scans with Less Data
- By John K. Waters
Facebook AI Research (FAIR) and NYU Langone Health this week announced the results of a joint clinical study demonstrating that artificial intelligence (AI) technologies can be used to generate detailed and accurate Magnetic Resonance Imaging (MRI) scans using about one-fourth of the raw data traditionally required for a full MRI.
Through a joint initiative dubbed fastMRI, researchers from the two organizations are investigating the potential of AI to make MRI scans up to 10 times faster. Currently, the scans can take up to an hour--a long time for a patient to be stuck, immobile, in a tight, noisy tube. Speeding up the process would not only improve the patient experience, it would free up the machine to serve more patients.
Faster scans could also enable expanded uses for MRIs, the researchers pointed out in a blog post, potentially allowing doctors to use MRIs in place of X-rays and CT scans in some cases. This possibility is especially exciting, they said, because, unlike those other technologies, MRIs don't use ionizing radiation.
In the study, radiologists found that fastMRI's "AI-accelerated" images, which were created with about four times less data from the scanning machine, were "diagnostically interchangeable" with traditional MRIs. They also arrived at the same diagnoses using those images and could not distinguish which were produced with an AI-assist and which came from the slower traditional scans. "All examiners rated the AI-generated images as being higher in quality than traditional scans," the researchers said.
"This study is an important step toward clinical acceptance and utilization of AI-accelerated MRI scans," said Michael P. Recht, Chair of Radiology at NYU Langone, in a statement, "because it demonstrates for the first time that AI-generated images are indistinguishable in appearance from standard clinical MR exams and are interchangeable in regards to diagnostic accuracy."
The results of the study ("Using Deep Learning to Accelerate Knee MRI at 3T: Results of an Interchangeability Study"), were published in the American Journal of Roentgenology. Those results represent the culmination of nearly two years of open research by FAIR and NYU Langone Health, which is a group of academic medical centers in New York City. "Open" has been a byword of this collaboration; Langone has released fully anonymized knee and brain MRI datasets that can be downloaded from the fastMRI dataset page.
"Since fastMRI is an open source project, other researchers and the companies that build MRI scanners can build on our work and test our code on their machines," the researchers said. "Our hope is that hardware vendors will get FDA approvals to bring these algorithms into production."
The two organizations also announced the "2020 fastMRI Reconstruction Challenge," which follows on last year's "2019 Challenge on Knee MRIs." The 2020 challenge will feature brain MRIs; the brain dataset at available at fastmri.med.nyu.edu. The challenge is tentatively set to open on October 1st. More details and challenge rules will be available soon at fastmri.org.
John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS. He can be reached at firstname.lastname@example.org.