Image reconstruction in a 52-year-old man referred for clinical knee MRI with an anterior cruciate ligament tear. The ligament tear, indicated by the yellow arrows, can be seen equally well in (A) the conventional protocol images and (B) the deep learning protocol images.
Artificial intelligence (AI) can reconstruct coarsely-sampled, rapid magnetic resonance imaging (MRI) scans into high-quality images with similar diagnostic value as those generated through traditional MRI, according to a new study by the NYU Grossman School of Medicine and Meta AI Research.
Reconstructing MRI scans with AI, which is four times faster than standard scans, promises to expand MRI access to more patients and reduce wait times for appointments, the study says.
The study, published January 17 in Radiology, is part of the fastMRI initiative established by NYU Langone Health and Meta AI Research (formerly Facebook) in 2018. This initiative, aimed at using AI to make MRI faster, resulted in an AI model jointly developed by Meta AI researchers and NYU Langone imaging scientists and radiologists. It also produced the largest-ever publicly-available collection of raw MRI data, which has been used by scientists and engineers around the world.
In an earlier “proof-of-principle” study, the NYU Langone team simulated accelerated scans by removing about three-fourths of the raw data acquired by conventional, slow MRI scans. The fastMRI AI model then generated images that matched those created from the slower scans. In this new study, the researchers performed accelerated scans with only one-fourth of the total data and used the AI model to “fill in” the missing information, similar to the way the brain builds images by filling in missing visual information from local context and previous experiences. In both studies, the fastMRI scans were found to be as accurate as traditional scans, with better quality.
“Our new study translates the results from the earlier laboratory-based study and applies it to actual patients,” says Michael P. Recht, MD, the Louis Marx Professor of Radiology and chair of the Department of Radiology at NYU Grossman School of Medicine. “FastMRI has the potential to dramatically change how we do MRI and increase accessibility of MRI to more patients.”
In the new study, a total of 170 participants received a diagnostic knee MRI using a conventional MRI protocol followed by an accelerated AI protocol between January 2020 and February 2021. Each examination was reviewed by six musculoskeletal radiologists, who looked for signs of meniscal or ligament tears and bone marrow or cartilage abnormalities. The radiologists evaluating the scans were not told which images were reconstructed with AI, and to limit the potential for recall bias, the evaluations of the standard images and AI-accelerated images were spaced at least four weeks apart.
The radiologists judged the AI-reconstructed images to be diagnostically equivalent to the conventional images for detecting tears or abnormalities, and found the overall image quality of the accelerated scans to be significantly better than the conventional images.
“This research represents an exciting step towards translating AI accelerated imaging to clinical practice,” says Patricia M. Johnson, PhD, assistant professor in the Department of Radiology. “It truly paves the way for more innovation and advancements in the future.”
Expanding Access to MRI
FastMRI, the researchers note, does not require special equipment. Technicians can program standard MRI machines to gather less data than is usually required, and the fastMRI initiative has made its data, models, and code available as an open-source project for other researchers, as well as manufacturers of commercial MRI systems.
Using fastMRI, an MRI examination that may take as long as 30 minutes can be completed in less than 5 minutes, making examination time for MRI comparable to X-rays or CT scans. However, unlike these imaging techniques, MRI provides incredibly rich information, from enabling views of soft tissues from different perspectives to highlighting small cartilage abnormalities to locating tumors in the abdomen.
“The price we have paid traditionally for all of that rich information is time,” says Daniel K. Sodickson, MD, PhD, chief of innovation in the Department of Radiology and director of the Center for Advanced Imaging Innovation and Research. “If we can supercharge MRI to approach the speed of CT scans, we can make all of that important information available to everybody.”
This study was funded by National Institutes of Health grants R01EB024532 and P41EB017183.
Media Inquiries
Allison Clair
Phone: 917-301-5699
allison.clair@nyulangone.org
