Facebook and NYU speed up MRIs to make them more bearable.

Facebook (NASDAQ: FB) and NYU Langone Health collaborated on research to incorporate Artificial Intelligence (AI) into medical imaging technology. The objective: to speed up MRI scanning so that patients have to spend less time cooped up, completely still, inside an MRI machine for as long as an hour. The project, fastMRI, found AI can help make these agonizing scans up to four times faster. (The Verge)

The new technique needs only a quarter of the usual data

The researchers trained a machine learning model by feeding it pairs of both low-resolution and high-resolution MRI scans. After training, the algorithm could extrapolate a high-resolution image from just a quarter of the usual input data. This meant that patients were in and out of the MRI machine much quicker.

“It’s a major stepping stone to incorporating AI into medical imaging,” Nafissa Yakubova, a visiting biomedical AI researcher who worked on the project, told The Verge.

After thorough training, the neural network was able to predict the final output, by incorporating individual data from the patient to produce AI-enhanced final images.

“The neural net knows about the overall structure of the medical image,” said Dan Sodickson, professor of radiology at NYU Langone Health. “In some ways what we’re doing is filling in what is unique about this particular patient’s [scan] based on the data.”

As a test, the researchers submitted both conventional MRI scans and the AI-enhanced scans to doctors. They found that the doctors arrived at the exact same medical assessments and made similar recommendations from both types of scans.

Doctors and radiologists, therefore, found that fastMRI’s AI-generated images were diagnostically interchangeable with traditional MRIs.

Is AI “imagining” the MRIs, and could it be dangerous?

According to the researchers, even the shorter scanning completely covers the targeted area required to be imaged. The algorithm has all the information required for a dependable scan with one crucial difference – it has it all but simply at a lower resolution.

The researchers also built in a control check during the process of final image creation. The algorithm ensures that its output is within the physical parameters of an MRI machine.

“We don’t just allow the network to create any arbitrary image,” says Sodickson. “We require that any image generated through the process must have been physically realizable as an MRI image.

Advantages

Faster scanning can make an MRI machine available to more number of patients thereby expanding its access.

The faster process also means that doctors could use an MRI instead of an X-Ray or CT scan in some cases. This would avoid exposing the patient to the radiation emitted by the latter, and potentially deliver a better diagnosis.

The findings are being published in the American Journal of Roentgenology.

Related Story:  Israeli Start-Up Nanox, Which Is Reinventing X-Ray, gets $20M