Facebook and NYU turn to AI to make MRI scans up to 10 times faster
The NYU School of Medicine's Department of Radiology is joining forces with Facebook's Artificial Intelligence Research group on a project aiming to utilize AI to dramatically speed up the time it takes to complete a magnetic resonance imaging (MRI) scan.
MRI scans are one of the best, and safest, diagnostic tools currently available, offering doctors an incredibly detailed insight inside a patient's body. However, as anyone who has been subjected to one of these scans can attest, they are time-consuming, claustrophobic and generally unpleasant experiences. Depending on the target, scans can take up to an hour and involve extensive stretches where a subject must hold their breath and remain still for long periods of time.
Researchers at NYU have been working on ways to speed up MRI scans for several years now. An artificial neural network was revealed in 2016 that was capable of producing a detailed MRI scan from less data than is regularly needed, but the researchers faced a roadblock. They needed more computing resources and AI expertise than they had access to.
This is where the Facebook Artificial Intelligence Research (FAIR) group came into the picture. FAIR's mission statement is to work with academics and the research community to collaboratively advance the field of machine intelligence with pragmatic, real-world outcomes. The new collaborative project is called fastMRI and the goal is to make MRI scans up to 10 times faster.
The idea is to speed up scans by collecting less data and then using a trained neural network to fill in the gaps. This, of course, is not an easy task, as anomalies on a MRI scan can often be incredibly small. Whole chunks of data can't easily be skipped while still offering clear results, as scans could then miss small tumors, dangerously delivering false all-clear messages.
The fastMRI project will involve NYU supplying FAIR with a massive volume of data, including three million MRI images from 10,000 clinical cases. This data, stripped of patient names and identifying information, will be used to train the new algorithms.
Speaking with Forbes, Larry Zitnick from the FAIR group estimates useable results should appear within the next 12 months, "In six months we should be able to make good progress on this. It could take less with a breakthrough, or it could take a year."
This isn't the first time an algorithmic solution to the problem of long MRI scans has been proposed – in 2011, MIT researchers developed an algorithm that could reportedly reduce the longest MRI scan down to just 15 minutes. And it's no surprise many researchers are working hard to solve this problem. The benefits of faster MRI scans would revolutionize doctors' diagnostic processes, as the scans could become more accessible to a larger volume of patients, and they could become more effective for patients that are simply unable to be effectively scanned due to the extreme duration.
A statement from Facebook sums up the overriding aim of the project, "With the goal of radically changing the way medical images are acquired in the first place, our aim is not simply enhanced data mining with AI, but rather the generation of fundamentally new capabilities for medical visualization to benefit human health."
Source: Facebook Code