Unlikely Friends: NYU and Facebook Tackle MRI Research


via flickr.com

Facebook is partnering with NYU Langone.

Kristina Hayhurst, News Editor

An unlikely pairing between NYU Medicine’s Department of Radiology and the social media giant Facebook has led to a revolutionary research initiative to improve MRIs. Combining artificial intelligence and radio-imaging, teams from both organizations aim to speed up one of the world’s most essential medical scans.

MRI, or magnetic resonance imaging, is a technique that utilizes strong magnetic fields and radio waves to create detailed images of the human body. The technology is employed for many diagnoses, from scanning for pelvic cancer to looking for a small tear in a tendon. Additionally, it is the least invasive and painful imaging procedure, mitigating the harmful radiation from X-rays while providing fast answers for doctors.

The collaboration between NYU and Facebook started a few months ago, two years after NYU first delved into the problem. Combining artificial intelligence expertise from Facebook’s AI Research, known as FAIR, and three million anonymous magnetic resonance images of the knee, brain and liver provided by NYU, the pair is working to make the machine faster and more accessible to patients across the globe. Daniel Sodickson, vice chair for research in radiology at NYU School of Medicine, said that the biggest obstacle to widespread MRI use was how long each individual scan can take.

“Most MRIs can be an hour long, and for some it can be a very challenging proposition,” Sodickson said in an interview with WSN. “If you can get a 10 times faster MRI, you can get the 60 minutes down to five. This is great for the patient experience, but it’s also great to increase accessibility for the MRI in areas where there are limited scanners — in the developing world and even in the states.”

The other issue with MRI, explained NYU Department of Radiology Chairman Michael Recht, is that the slow speed causes doctors to rely on X-rays that can provide immediate results, along with
harmful radiation.

“If some people injure themselves in sports, the first thing they do is get an X-ray. In most cases, the X-ray doesn’t show anything, but we do it because it’s fast and it’s much less expensive than the MRI,” Recht told WSN. “If we were able to get [MRI results] faster, we wouldn’t have to radiate people and we could get concrete answers faster.”

The two groups combined to perfect deep learning, an AI process that mirrors the way humans learn new information. This process allows AI to become familiar with the tissue and cellular makeup of the human body so that it can complete the partial MRI scans using previously stored knowledge. This will allow the MRI technology to work faster as it has to collect much less data thanks to the use of these partial pictures.

“AI will learn all of the common, underlying features [in the MRI images] and the doctors will focus data acquisitions on the stuff that’s new and particular to an individual patient,” Recht said. “We’re going to be working very carefully so we can get the patient-specific things accurately quoted, and we’re working on different methods to make sure we don’t lose the true
patient information.”

While the partial MRIs are instrumental in speeding up the imaging process, they also pose one of the biggest potential problems with the new technology. If the deep learning network creates a partial image in place of a tumor, doctors could miss the problem altogether. Yvonne Lui, the associate chair for artificial intelligence in the Department of Radiology, said she is looking forward to overcoming these obstacles.

“A few missing or incorrectly modeled pixels could mean the difference between an all-clear scan and one in which radiologists find a torn ligament or potential tumor,”  Lui said. “Conversely, capturing previously inaccessible information in an image can quite

A version of this article appeared in the Tuesday, Sept. 4 print edition. Email Kristina Hayhurst at [email protected].