- Facebook and NYU School of Medicine’s Department of Radiology have announced a new collaborative research project that will study the use of artificial intelligence (AI) to make MRI scans up to 10 times faster to complete.
The project, called fastMRI, represents one of Facebook’s first major forays into healthcare.
Although MRIs show a greater level of detail than other medical images, these scans can take anywhere from 15 minutes to over an hour, which is far longer than other imaging modalities.
“These long scan times can make MRI machines challenging for young children, as well as for people who are claustrophobic or for whom lying down is painful,” Facebook and NYU noted in a blog post.
“Additionally, there are MRI shortages in many rural areas and in other countries with limited access, resulting in long scheduling backlogs. By boosting the speed of MRI scanners, we can make these devices accessible to a greater number of patients.”
Initially, fastMRI will focus on changing how MRI machines operate. Current scanners gather raw numerical data and turn it into cross-sectional images of body structures for doctors to evaluate. The more data there is for the machine to collect, the longer the scan will take.
Artificial intelligence could allow MRIs to capture less data and scan faster while still maintaining the quality of images. Researchers will train neural networks to recognize the underlying structure of medical images and fill in the missing parts of the accelerated scans.
Past research conducted at NYU School of Medicine demonstrated that neural networks can generate high-quality images from much less data than was previously thought necessary, the two organizations said.
However, reconstructing images from partial information is an extremely difficult task in practice. The neural networks will have to effectively scan the data without sacrificing accuracy. A mistake as simple as a few missing or moderated pixels could mean the difference between an all-clear scan and one in which radiologists find possible tumors or other issues.
fastMRI will use an imaging dataset collected by NYU School of Medicine that consists of 10,000 clinical cases and contains approximately 3 million MRIs of the knee, brain, and liver.
NYU has scrubbed all potential distinguishing features from the MRIs, and all imaging and raw scanner data have been deidentified.
Facebook and NYU will also allow the wider research community to access this project and further build on their advancements. As the project develops, Facebook will share AI models, baselines, and metrics of the research, while NYU will make the image dataset publicly available.
fastMRI builds on previous collaborative work from NYU School of Medicine and Facebook.
The Radiology Department at NYU includes the Center for Advanced Imaging Innovation and Research, a team of scientists, engineers, radiologists, and other clinicians with expertise in image reconstruction and rapid image acquisition.
Since 2016, the center has worked with the Facebook Artificial Intelligence Research (FAIR) group to pursue the idea of using AI to accelerate MRI scanning.
Both Facebook and NYU expect that this AI project can extend to other forms of medical imaging, making this area of care more effective and accessible to all patients.
“We believe the fastMRI project will demonstrate how domain-specific experts from different fields and industries can work together to produce the kind of open research that will make a far-reaching and lasting positive impact in the world,” the organizations concluded.