MR scans could soon be done 10 times faster, thanks to a a large-scale MR dataset just released to the public from fastMRI, a collaboration between Facebook AI Research and NYU Langone's Department of Radiology.
“We hope that the release of this landmark data set, the largest-ever collection of fully-sampled MR raw data, will provide researchers with the tools necessary to overcome the challenges inherent in accelerating MR imaging,” Dr. Michael P. Recht, chair and the Louis Marx Professor of Radiology at NYU Langone said in a statement following his announcement of the news in a plenary address at the 2018 annual meeting of the Radiological Society of North America (RSNA).
Recht also shared baseline results from the collaboration, made up of 1.5 million MR images of the knee from 10,000 scans, plus raw measurement data from almost 1,600 scans. The collaboration demonstrates that acceleration of MR imaging by a factor of four “is already possible".
The data set is fully anonymized, HIPAA-compliant information from NYU's medical school – and no Facebook data. Future releases will add data from liver and brain scans.
In addition, the open source tool is expected to boost the development of AI systems that are capable of deciphering MR scans, boost research reproducibility, and open the door for more consistent evaluation methods. Plans call for the collaboration to develop a suite of tools and baseline metrics to compare results in an organized challenge that will be announced “in the near future,” according to the NYU report.
“This collaboration focuses on applying the strengths of machine learning to reconstruct high-value images in new ways. Rather than using existing images to train AI algorithms, we will radically change the way medical images are acquired in the first place,” Dr. Daniel K. Sodickson, director of NYU's Center for Advanced Imaging Innovation and Research, added in a statement. “Our aim is not merely to enhance data mining with AI, but rather create new capabilities for medical visualization, to benefit human health.”
MR scans can generate a huge amount of valuable information but are slow in nature. By applying AI to exams, researchers believe they can cut down on the amount of data captured, while maintaining and even boosting the richness of the information in images.
Such features are expected to benefit patients who cannot tolerate the typical length of the process, including very young children, elderly adults and those who are claustrophobic. It may also reduce the need for drug administration to calm jittery patients.
“fastMRI not only could have an important impact in the medical field, it’s also an interesting research challenge that will help to advance the field of AI,” said Larry Zitnick, research manager, Facebook AI Research. “To be medically useful, our AI-reconstructed images need to be more than just good looking, they must also be accurate representations of the ground truth, even though they're created from significantly less data. The data set NYU Langone is releasing and the baseline models we've open-sourced will enable other researchers to join us in working on this challenging problem, and we believe this open approach will bring about positive results more quickly.”
The next step for the team is to explore AI-based image reconstruction techniques using the data set, according to Dr. Yvonne W. Lui, associate professor of radiology and associate chair of artificial intelligence at NYU. “Additionally, any progress made at NYU School of Medicine and FAIR will be part of a larger effort that spans multiple research communities,” she advised, explaining that “results will be compiled and compared on a fastMRI leaderboard, as well as in research papers and workshops to come.”
When the collaboration was announced in August
, Lui stressed the approach, which uses AI to reconstruct views skipped in a scan from underlying image structure, similar to the way people interpolate sensory information.
“MR is the gold standard imaging technology for soft tissues of the human body. However, its main limitation is the amount of time an exam takes,” Lui told HCB News at the time. “Using AI, our aim is to acquire significantly less data than typically needed for a high-quality medical image, allowing the examination to be completed in a significantly shorter period of time while maintaining diagnostic imaging quality.”