Facebook and NYU want to use AI to make MRI exams faster

Donna Miller
August 21, 2018

Scientists at Facebook's artificial intelligence labs think they have found a way to use a method called deep learning to make the process more efficient by training an algorithm to recognize the specific configuration of bones, ligaments and other body parts, allowing the program to create a portion of the image to save time. People who are claustrophobic, in pain or young have trouble with MRI scans that can take more than an hour.

How's it work? MRI scans will be enhanced using what's known as a generative model. "The key is to train artificial neural networks to recognize the underlying structure of the images in order to fill in views omitted from the accelerated scan", said researchers in a blog post about the work. "It would be a real game-changer", said Daniel Sodickson, NYU School of Medicine's vice chair for radiology research and director of the Center of Advanced Imaging Innovation and Research.

Flash floods in Italy leaves at least 8 hikers dead
Tourists can visit without a guide, making it hard to know exactly how many people were in the canyon when the floods hit. Rescue squads worked through the night with spotlights, and helicopters took off a first light to survey the area.

According to NYC and Facebook, the initial focus of fastMRI is to change how the MRI machines operate. The project will use image data from 10,000 clinical cases with roughly 3 million MRIs of the knee, brain and liver. It will be vital to make sure patient data is protected, to account for algorithmic bias, and to find ways to explain algorithmic reasoning.

In theory, new techniques for MRIs could treat more patients and lower costs as well as radiation exposure from X-rays and CT scans. The NYU and FAIR researchers want to speed up scans by collecting less raw data and having trained neural networks fill in the gaps. NYU Langone Health, which counts the NYU School of Medicine as a department, has been working on imaging and AI. But as the area that needs to be scanned gets larger, so does the scan time.

Other reports by

Discuss This Article