Overview
- Facebook and NYU’s School of Medicine have joined hands to design an AI system that will make MRI scans up to 10 times faster
- Early studies indicate that artificial neural networks are the way forward, as they have produced promising results
- The dataset being used for the research consists of 10,000 cases and 3 million MRI images
Introduction
Getting a Magnetic Resonance Imaging (MRI) scan is a notoriously long process. For anyone who has been through the wringer, you know how harrowing that experience can be. It can take up to an hour for the entire process to work out while you sit inside that claustrophobic environment.
Facebook (yes, Facebook!) and the NYU School of Medicine have joined hands with the aim of designing an AI system that will make these MRI scans up to 10 times faster! Not only that, the researchers are hoping that this will make MRI technology available to even more people. The project, called FastMRI, was started back in 2015 by the NYU researchers.
In the current scenario, a MRI scan can take from anywhere between 15 minutes to an hour. This poses a significant problem for children, elderly people and especially those who suffer from claustrophobia. Additionally, the patient needs to hold his breath for prolonged periods (when scanning of the heart, liver, and other organs is ongoing). You might be getting an idea as to why this research has garnered everyone’s attention.
MRI scanners operate by creating a series of 2D images which need to be converted to 3D. This is done by stacking up the 2D images. The more in-depth the data needs to be (depending on what has to be scanned), the longer the scanning process lasts. This is the part where the AI system being designed will be aiming to make the difference.
The AI system should be able to capture less data, and hence accelerate the scanning process. The key, according to this blog post by Facebook, is to train artificial neural networks (ANNs) in order to recognize the underlying structure of the scanned images.
The dataset being used for this research contains 10,000 cases and around 3 million MRIs of various patients’ liver, knee and brain. Training neural networks on these might yield good results, but there are a few major challenges that could derail the whole project, like a few wrongly modeled pixels. It’s a fine line between success and failure in this field and one that the researchers will be well aware of.
Our take on this
It’s wonderful that machine learning and artificial intelligence are the heart of a truly revolutionary period in healthcare. When big companies like Facebook enter the fray, they garner the world’s attention. Let’s hope that this research comes to fruition soon and sees a quick deployment in hospitals worldwide.
Though as I mentioned, using neural networks to accelerate scanning has it’s own set of obstacles. I suspect it will a bit of time for them to perfect and open source their approach like they’re promised, but it should be worth the wait.
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!
Senior Editor at Analytics Vidhya.Data visualization practitioner who loves reading and delving deeper into the data science and machine learning arts. Always looking for new ways to improve processes using ML and AI.
