Can AI Create Faster, More Reliable MRI Scans?

When it comes to healthcare, artificial intelligence and deep learning has become very much a part of it in recent years. AI in MRI is not a surprising leap, but it could end up being massively transformational, not only for individual patients, but for the healthcare industry as a whole. AI is already being used in CT scans to detect COVID-19 cases, detecting 17 out of 25 positive cases, all of which were missed by physicians.

Why MRIs Are Not Ideal

MRI scans, or magnetic resonance imaging, brought some impressive changes to the medical industry back when the technology first appeared. The ability to see inside the body in clear detail makes it possible to diagnose and treat some otherwise unnoticeable issues inside the body, but there are a few downsides to the technology.

MRI scans can take anywhere from 15 to 90 minutes, and they can cause quite a bit of distress. For patients who are claustrophobic, the machine is downright terrifying, but anyone will feel a bit put out by lying in one position for up to an hour while getting scanned. Often, the patient is already in pain and feeling unwell and lying still for a long period of time is even more uncomfortable.

The Benefits of Faster MRI Scans

There are a lot of reasons to want faster scans.

Less Artifacting: Since people are often in pain and distress when inside the MRI machine, they may move around. When they move, that image can be partially ruined and that causes more issues with diagnosis.

Shorter Wait Times: People usually have to wait for a long time to get an MRI because they are so slow and technicians have to wait for each person to go through the scan before they can move them and get the next one in. With shorter scans, it would be possible to scan more people in the same amount of time.

Less Patient Distress: Patients who are in pain or distressed by the scan will be able to leave sooner, which is much better for their mental health. For those who can’t stay still for an hour, a much shorter scan may be doable.

Saved Money: Hospitals and clinics can save a lot of money by doing faster scans, as well. They don’t need to spend as much personal time per patient and that means more earnings and less spending.

It’s obvious that reducing the amount of time an MRI takes would be greatly beneficial to everyone involved, so is it actually possible to make this happen?

How AI Can Help

Researchers at Stanford University have teamed up with professors from the Technical University of Munich and Rice University to come up with new techniques for MRIs. Using AI, they realized that it was possible to speed up the entire MRI process. What if they just lowered the resolution or the amount of data the MRI machine collected and used AI to replace the missing information? It was a novel concept, but one that made a lot of sense.

The way this would work would be to collect lower-resolution images. While there would be less data collected, the AI would take over at that point to reconstruct the MRI images and replace any small details. The mathematical process would work to fill in the gaps and then doctors could still use them as diagnostic devices.

Originally, it was thought that AI wasn’t capable of this methodology. Researchers looked at both trained and untrained neural networks, as well as non-AI image reconstruction options that are already in use. Then they decided to test the different methods and see just how effective they were at piecing together the many images that were taken in an MRI.

It was rapidly apparent that trained networks had been taught over the years and required very high quality example images in order to train. They also required a lot of hands-on supervision, which defeated the purpose of shorter MRI sessions.

As it turned out, the untrained AI networks are the most advanced options in AI and they don’t need to be supervised or trained, as they learn on their own. However, like the other methods, they fell prey to certain issues that the researchers knew might be a problem. These included:

• Recovering small image details.
• Distribution shift, when the algorithm trained on one type of the body and now must scan a different part.
• Sensitivity to perturbations or disruptions in data collection, such as when a patient moved or the machine glitched in one way or another.

In all cases, all types of AI found it difficult to handle these problems, and that means that while there is promise, they aren’t able to replace a high-quality image taken by the slower version of the MRI. Sadly for the patients, the original method of scanning is still necessary and it probably won’t change anytime soon.

However, the neural networks greatly outperformed traditional models when it came to determining fine detail in the images.

While the researchers were surprised by this last discovery, it wasn’t impressive enough to change the way MRIs work.

The End Result

At this point in time, AI isn’t infallible when it comes to restoring MRI images. However, with more training and further improvements to the technology, it’s possible that AI will eventually be ready to speed up MRIs. After all, artificial intelligence and deep learning are constantly improving and growing. This means there will be advancements and eventually it should be possible.

The big question is whether or not MRI capabilities will have improved or been replaced by that point in time. After all, AI isn’t the only technology that is improving in the healthcare industry.