MRI images (magnetic resonance imaging) can be generated in just a few minutes using Facebook's artificial intelligence (AI). In collaboration with NYU Langone Health and Facebook's AI research team, the project called FastMRI was born.
Physicians and other medical experts had been using MRI since it was developed in 1970. This technology has been useful for garnering vital insights into the patients' organs.
Facebook and NYU use artificial intelligence to make MRI scans four times faster - The Verge: Facebook and NYU use artificial intelligence to make MRI scans four times faster The Verge https://t.co/wL7pRFv1ke #AI #artificialintelligence #Finperform pic.twitter.com/G3tgMQ9j5m — Suriya Subramanian (@SuriyaSubraman) August 18, 2020
However, this technology requires patients to stay perfectly still for an extended period since it operates at a plodding pace. The issue makes MRI not suited to be used for sick children and ill people who are experiencing time-critical medical emergencies, including strokes.
After two years of research, Facebook AI and NYU Langone Health have developed a neural network that can lessen the amount of time required to generate MRIS, which usually puts people in an MRI machine for more than an hour.
"fastMRI" only takes a few minutes to generate MRI because it only needs a quarter of the traditional MRI's required data. Magnetic resonance imaging works by creating a localized, intense magnetic field. Magnetic field exposure makes the atomic nuclei of different elements, such as hydrogen, absorb radio frequency (RF) energy and then reemit it as a measurable RF frequency.
"If you've been sitting in an MRI, you've been hearing that buzzing sound it makes when it gathers data," said a researcher at NYU Langone Health, Dr. Dan Sodickson.
He described the raw data, which a magnetic resonance image is derived, looks like a fascinating starburst.
How much data fastMRI needs?
The fastMRI does not need to wait for "k-space" to fill up; instead, it only needs 25% of the original data required. This neural network is actively generating MRI images, which are effectively identical to traditional scans, from the raw data.
NYU Langone and Facebook AI team for Fast MRI breakthrough. NYU Radiology Chair Dr. Michael Recht and Facebook AI's Nafissa Yakubova in Techstination interview. https://t.co/pzY5Bmj29O pic.twitter.com/L80xOun42s — Fred Fishkin (@ffishkin) August 18, 2020
Six radiologists, recruited by Facebook, examined two sets of MRI sequences of a patient's knee. The first one was created by traditional MRI, while the second image was generated using fastMRI. Five out of the six radiologists were unable to identify which one was developed using Facebook's AI.
Nafissa Yakubova, a researcher at Facebook AI, said that to avoid overfitting the MRI images, they need to begin with a large data set. The researcher also added that they generated thousands of MRI cases from the knee. MRI brain scans were also included, which contained as many as 800 still images in each of the scans used to improve the fastMRI model.
For more scientific innovation, always keep your tabs open here at TechTimes.
This article is owned by TechTimes,
Written by: Giuliano de Leon.