Even if you’ve never been in an MRI machine, chances are you might have seen it in use in television shows and movies. Usually this involves doctors telling patients that they’ll have to remain perfectly still in order for the machine to capture images, which can be a frustratingly slow process that can take as long as an hour.
This can be problematic for some patients who might be claustrophobic, plus there is also the noise generated by the machine that can feel disconcerting. The good news is that thanks to the efforts of researchers at Facebook AI and NYU Langone Health, they have developed a neural network that can apparently cut the amount of time it takes to capture MRI images from an hour down to just minutes.
How this works is that instead of relying on the MRI machine to capture 100% of the data, this neural network, known as fastMRI, will only require 25% of the data and it can fill out the rest of the information itself. So far based on tests of fastMRI against the more traditional methods, it seems that for the most part, there were no significant differences and that five out of six radiologists could not tell the AI generated version apart from the regular scan.
The best part is that this can be retrofitted onto existing MRI machines, which means that hospitals don’t need to purchase brand new equipment. Also, given how fast it scans and generates images, it can also reduce the amount of time people spend waiting to get scanned, which can potentially lead to better health care.