Facebook, NYU use AI to speed up knee MRI scanning process

Facebook, NYU use AI to speed up knee MRI scanning process.   | Photo Credit: Facebook

(Subscribe to our Today's Cache newsletter for a quick snapshot of top 5 tech stories. Click here to subscribe for free.)

Magnetic resonance imaging (MRI) scanning process can be accelerated using artificial intelligence (AI), according to a joint research initiative, called fastMRI, between Facebook AI and NYU Langone.

The study found fastMRI’s AI-generated images were created with about 25% data from scanning machines, and they can be replaced with traditional MRI scans for diagnosis.

(L) Traditional MRI, (R) AI-enhanced MRI.

(L) Traditional MRI, (R) AI-enhanced MRI.   | Photo Credit: Facebook


“We are using AI to create complete images from far less raw data. Since collecting that data is what makes MRIs so slow, this has the potential to speed up the scanning process significantly,” Facebook said in a statement.

MRI scan help diagnose organs, muscle, and other soft tissue. Traditional MRI scanning process uses powerful magnets and radio waves to gather raw data necessary to create images. It usually takes about an hour to complete scanning.

The hour-long procedure can be difficult for some people, like small children. And movement during the scan will result in blurry images.

During a scan, the signals collected by the MRI scanner are known as k-space data. Upon collection of all the data, the system applies a complex mathematical formula to create detailed MRIs. The math cannot isolate individual signal origin points without a complete set of data points.

The fastMRI research team built a neural network and trained it using a large open-source data set of knee MRIs, provided by NYU Langone Health. The AI model was fed with only about a quarter of the raw data in each scan. And over a period of time, it learned to generate complete images from the limited data.

(L) Traditional MRI using fully sampled K-space. (R) FastMRI using undersampled K-space.

(L) Traditional MRI using fully sampled K-space. (R) FastMRI using undersampled K-space.   | Photo Credit: Facebook


“This study is an important step toward clinical acceptance and utilization of AI-accelerated MRI scans, because it demonstrates for the first time that AI-generated images are indistinguishable in appearance from standard clinical MR exams and are interchangeable in regards to diagnostic accuracy,” said Michael P. Recht, MD, Louis Marx Professor and Chair of Radiology at NYU Langone.

Two sets of MRI scans of 108 test patients in the 18 to 89 age range were generated for the study. One set used fastMRI AI model, and the other used standard imaging technique. These were reviewed by six expert radiologists, Facebook said.

The k-space data for about 1,200 scans in the data set was retrospectively under sampled for use in training the network while the clinical study was performed using 3 Tesla (3 T) machines, it added.

Facebook says the researchers want to expand the present study’s focus, and develop fastMRI for other vital organs, such as brain. As the data, models, and code have been published, other researchers, and MRI manufacturers can build or test fastMRI.

This article is closed for comments.
Please Email the Editor

Printable version | Mar 2, 2021 11:28:21 PM |

Next Story