Completing an MR scan in a fraction of the time may soon be possible thanks to a new collaboration underway between the NYU School of Medicine and ubiquitous social media giant, Facebook.

The New York-based med school’s department of radiology has tapped Facebook Artificial Intelligence Research – a group dedicated to advancing the state of artificial intelligence – to support an NYU project called FastMRI that aims to reduce the time associated with MR scans without compromising image quality.

“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,” Yvonne Lui, associate professor and associate chair of artificial intelligence in the department of radiology at NYU School of Medicine, told HCB News. “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.”

MR scans range from 15 minutes to over an hour and require patients to remain very still within a narrow bore which can sometimes induce claustrophobia. The exams can be uncomfortable experiences for young children and patients suffering from chronic illness and pain, requiring holding one’s breath for a length of time when imaging certain organs.

Speeding up the process may alleviate some of these discomforts while also yielding practical benefits to providers, particularly in regions where MR access is limited and scheduling backlogs prevent patients from undergoing needed diagnostics.

The length of time associated with MR scans is due largely to the collection and transformation of raw numerical data into cross-sectional images of internal body structures for evaluating patient health, with larger data sets tacking on more time to exams.

Researchers in the NYU radiology department’s Center for Advanced Imaging Innovation and Research (CAI²R) are planning to use artificial intelligence networks to recognize the underlying structure of images, filling in views omitted from accelerated scans, similar to how humans process sensory information.

Although reconstructing images from smaller amounts of information is difficult due to the risk of a few missing or incorrectly modeled pixels leading to inaccurate diagnoses, previous work at the medical center has found that artificial neural networks are capable of generating high-quality images from far less data, and may be able to capture previously inaccessible information that could help save lives.

The project will apply the deep learning and AI training expertise of Facebook's FAIR initiative to an imaging data set of 10,000 clinical cases collected exclusively by NYU that comprise approximately three million MR images of the knee, brain and liver.

“Our goals are to open-source this project to allow the wider research community to build on our developments in the interest of advancing the state of the art in medical imaging as quickly as possible,” Lui said. “Longer-term, we hope one day that because of this project, MR will be able to replace X-ray or CT for certain applications, also leading to decreased radiation exposure to patients. However, we hope to apply these techniques to other imaging modalities such as CT, enabling ultra-low-dose scans and decreasing patient exposure to radiation.”

Facebook will share AI models, baselines and evaluation metrics as research progresses. NYU School of Medicine will open-source the image data set.

The project is fully HIPAA-compliant and approved under NYU Langone’s Institutional Review Board, which oversees all human subject research at the medical center. It is filed under strict human data protection regulations.

No Facebook data will be used in the research.