Facebook says AI can help you spend less time getting an MRI scan
The social network teams up with NYU to create faster MRIs that don't sacrifice accuracy for speed.
- 2022 Eddie award for consumer analysis
A woman lies in a MRI machine in Germany.
The two images from an MRI scan of an injured knee look similar, but there's a major difference: One was created more quickly with the help of artificial intelligence.
Facebook researchers and doctors from NYU Langone Health say they were able to speed up MRI scans with the help of AI. The pictures were so clear that most doctors couldn't distinguish an AI-created scan from a traditional one.
Doctors use detailed images of organs, muscles and other soft tissues from MRI scans to help diagnose problems inside the human body from tumors to knee injuries. But getting a scan can be uncomfortable for some patients and kids because you have to remain still in a confined space for sometimes more than an hour. Quicker MRI scans would not only improve a patient's experience and potentially increase the use of this technique, but it could also cut down on wait times in the doctor's office and the cost of medical bills.
The MRI image on the left was created with the help of AI. The image on the right was created using the traditional approach.
Facebook doesn't manufacture MRI machines, but the partnership with NYU is a way for the company to advance AI research. Artificial intelligence is playing a larger role in health care, including in robotic surgery, cancer risk predictions and medical image analysis.
Larry Zitnick, who helped lead the project at Facebook AI research, said the company's scientists are interested in working in the medical space because they want to tackle difficult and impactful problems that AI could help solve. In order to cut down on the time it takes to conduct an MRI scan, the AI model that Facebook developed used less data to create an image. But these images have to be accurate, otherwise it could result in a doctor misdiagnosing a patient.
Dr. Michael Recht, a professor and chair of the Department of Radiology at NYU Langone Health, worked with Facebook on a study that involved 108 patients who got an MRI scan of their knees last year. Using the conventional approach, their MRI scans took between roughly 8 minutes to 11 minutes. With the help of AI, Facebook and NYU were able to cut down the time for these scans to between 4 to 6 minutes, according to a study set to be published in the American Journal of Roentgenology.
That time saved might not seem like a lot, but it adds up. The time it takes to conduct an MRI scan varies depending on what body part doctors are trying to examine and the condition they're trying to diagnose. When a patient is suffering from a stroke -- when the blood supply to their brain is interrupted -- minutes can mean the difference between life and death, said Recht, who focuses on musculoskeletal imaging.
"We think in the brain we can accelerate much faster than we were able to do in the knee. And if we can accelerate...three times as fast or four times as fast, then you're really saving significant amounts of time," he said.
An MRI, which stands for magnetic resonance imaging, uses strong magnetic fields and radio waves to create images of organs and other soft tissues inside a person's body. Most of the human body is made up of water. A molecule of water contains two hydrogen atoms and one oxygen atom. The hydrogen atoms react to the magnetic field and emit a certain radio frequency signal, generating data collected by the MRI scanner. That data is then used to create an image of a human body part.
The study also showed that the faster MRI scans were also accurate despite using less data than the traditional approach. Most radiologists couldn't tell which images were created by AI and they also thought those scans resulted in a clearer picture of the knee. When a patient moves, for example, that can make an image from an MRI scan blurry.
Facebook compared the process conducted with the help of AI to accurately completing a 1,000-piece jigsaw puzzle with just 250 pieces.
The algorithms, Zitnick said, has looked at thousands of images of knees, so it knows how a knee is structured and is able to create an image of that body part "in a way that's clear."
MRI is just one way doctors can peek into the human body without slicing a patient open. Unlike other methods such as CT scans and X-rays, MRIs don't expose patients to radiation. By cutting down on the time and cost it takes to conduct MRIs, Recht said that doctors including in developing countries might use MRIs over other imaging techniques. Researchers still have to make sure that AI-accelerated MRI scans of the brain, liver and other body parts are accurate.
"We're getting the same or more information in less time. So we should be able to use this sequence for all types of pathology if it holds up," Recht said.