A fellow at Mayo Clinic is using artificial intelligence to pinpoint seizure-causing areas in patients’ brains. And it might not end there.
Yogatheesan Varatharajah, a Ph.D. candidate at the University of Illinois, researches ways to apply AI to medicine — specifically, to disorders in our brains.
Using data from nonseizure times, artificial intelligence can identify seizure-generating parts of the brain much more quickly than usual.
Then, when an epileptologist looks at a patient’s records and MRI scans, the AI’s conclusion also would pop up — "here are the sensors within the seizure-generating region, according to these EEG readings."
"Looking at those, if there is a good concordance within the different ways of identifying the seizure emitting regions, they can go ahead and do this surgery instead of waiting for seizures to occur," Varatharajah said. "For the patient, the most important thing is you save a lot of trauma."
Varatharajah, a graduate student in electrical and computer engineering, is also a fellow in the Mayo Clinic & Illinois Alliance for Technology-Based Healthcare since 2016.
"We know that there are a lot of issues, a lot of areas where you can bring AI technology," said Varatharajah’s advisor, Dr. Ravi Iyer. "It was a leap of faith, if you will, to think we could work well together."
How it works
Ordinarily, in order to find seizure-causing areas in the brain of an epileptic patient, doctors have to wait until a seizure occurs. The patient wears EEG sensors, and providers attempt to identify the sensors in which the earliest seizure onset occurred.
After that, the seizure is mapped back to a specific location in the brain, which can be surgically removed.
Varatharajah’s work focuses on finding the seizure-generating regions while they’re inactive.
"When we find this region that generates seizures, they are still generating abnormal EEG activity, or electric activity, even during nonseizure times," he said. "But these are not so prominent as the ones you see during a seizure, these are very small in magnitude, and it’s not so visible. So, we need to be able to identify these patterns using a computer program or some kind of tech which is able to recognize these very small but abnormal patterns. So, that is where the AI comes in."
Previous studies had tried to zoom in to EEG readings and find the patterns manually. It did work, in a rough sense — some patterns were identifiable to the naked eye. But the accuracy was not sufficient to bring it into clinical practice.
"We will give the tech two hours of EEG recording, during an unseizured time, and then the technology will be able to identify these patterns," Varatharajah said. "The different abnormal patterns. Then it will look for their temporal evolution and the spatial relationships across different sensors. And using all of this, it will be able to tell us which of these sensors is on a seizure-generating region, and then a surgeon can go and remove that region."
AI: a tool?
The AI can’t replace a traditional physician, Varatharajah said.
"I see an understanding of the brain that can be developed through analyzing and pinpointing a region of the disease," he said. "AI is a set of methods. What makes it useful is, of course, the user."
Public perception of artificial intelligence still is mixed. Popular culture, in particular, still views AI as more of a threat than a helping hand.
"It’s a wrong perception," Varatharajah said. "Some people just go and tell that AI beats humans in jeopardy … but that’s not the whole point. In my perspective, how AI can be used to better humanity is that you can use AI for the tasks in which humans are not that good. By just looking at these discrete patterns, you know, just manually counting them, it did not result in good accuracies. But by using the AI technologies, we can basically capture other information, which we may be able to see, but we may not be able to quantify, so that it results in a better accuracy."
Varatharajah also is involved in a project to identify patients who are at risk of developing Alzheimer’s disease in the very near future, so those patients can receive cognitive training or other health improvements to prolong the period of time before the onset of the disease.
"We’ve seen some early success, but it is early to tell," Iyer said.