Even if you’re not claustrophobic, magnetic resonance imaging (MRI) scans can be discomforting. It’s important that you lie very still and breathe normally for anywhere from 10 to 80 minutes while loud noises explode, sounding like you imagine a space capsule sounds when breaking up in deep and dark outer space. You know you’re not in space, but that doesn’t help. MRIs are wonderful technology for the precise, minute detail they can render. Compared to X-rays and CT(CAT) scans which take just a second and a few minutes, respectively, however, MRI sessions can seem to be interminable.
Speedier MRIs alone could be reason enough to applaud the collaboration between the NYU School of Medicine’s Langone Department of Radiology and Facebook AI Research (FAIR). The joint effort supports fastMRI, a project to speed up MRI scans by applying artificial intelligence (AI). According to NYU Langone’s Michael P. Recht MD, the collaboration has already been able to accelerate MRIs four times faster than normal, but the goal is 10X speed. The increased speed is made possible by collecting less data and using AI to produce equivalent or better MRI images.
To quickly answer concerns about data security and trust, a press release about the project stresses that it does not use Facebook data of any kind. The initial dataset is made up of 1.5 million anonymized MRI images derived from 10,000 scans as well as raw data from 1,600 scans. The researchers are sharing the AI models through open source. It will encourage others to become involved by providing a suite of tools and baseline metrics to encourage reproducibility of the test results.
Now the fastMRI group is focused on validating the results of the 4X faster scans. They are also using machine learning to determine methodologies for obtaining the very best scans. The initial release, already the largest public release of raw MRI data, consisted only of knee scans. Subsequent releases will also include liver and brain scans. The most desirable end result is the discovery of new AI and MRI capabilities to benefit human health, but even just decreasing the time people have to spend in the tube will be a big win.
