Research & Innovation

AI-Powered Imaging Could Help Defuse Cerebral Aneurysms

Stevens faculty-student research offers new insights into devastating brain disease

Cerebral aneurysms, which affect one in 50 people in the United States, are like ticking bombs: bulging arteries in the brain that can burst without warning. Each year about 30,000 people in the United States suffer aneurysm ruptures, which kill 60% of sufferers and leave half of survivors with devastating brain injuries.

While many ruptures can be prevented if they’re treated in time, predicting which aneurysms will burst is tricky: markers such as size, shape, and location offer only imperfect hints as to any given aneurysm’s risk of rupturing. Now, a student-led research team from Stevens Institute of Technology, working in collaboration with researchers at Stanford University and the University of Auckland, has developed a new artificial intelligence (AI)-powered imaging tool that could allow doctors to identify and treat high-risk aneurysms far more quickly and accurately.

By analyzing minute vibrations in the walls of cerebral blood vessels, the new technology allows clinicians to spot high-risk aneurysms before they burst, potentially enabling them to dramatically reduce the toll taken by the devastating disease. “This research will save lives,” said Mehmet Kurt, an assistant professor in the Department of Mechanical Engineering at Stevens Institute of Technology, who supervised the project. “Cerebral aneurysms are monsters, and we urgently need to know whether they’re going to rupture so that surgeons can intervene in time.”

Looking for warning signs

Javid AbderezaeiStevens doctoral student Javid Abderezaei

Researchers have suspected for some time that the motion of blood vessel walls might offer clues as to an aneurysm’s likelihood of rupturing, but reliably detecting the tiny vibrations deep within the living brain has proven all but impossible.

To overcome that challenge, the Stevens team used 4D flow MRI, a sophisticated brain imaging technique that combines multiple MRI scans into a video-like sequence showing blood flowing through the brain. Even using such scans, though, it’s ordinarily impossible to detect the subtle flickering of aneurysm walls, so the Stevens researchers built upon previous research to develop an advanced image-processing algorithm called aFlow to isolate and amplify the tiny movements.

“The brain is a noisy place, so you can’t just look at scans and identify vibrations in specific blood-vessel walls,” explained Ph.D. student Javid Abderezaei, the study’s lead author. “But with aFlow, we were able to home in on the exact movements we were looking for.”

To test aFlow, Abderezaei and Kurt’s team used digitally simulated MRI scans in which they could predetermine the vibrations of a simulated, or “phantom,” aneurysm. By using AI to cancel out background noise and amplify desired signals, aFlow was able to pinpoint phantom vibrations extremely accurately.

The team also tested aFlow on real MRI data from five healthy subjects, to show that the vibrations it detected correlated to real physiological deformation. “We were able to show that the vibrations we were detecting correlated with local deformations of the brain tissue, and were caused by variations in blood flow,” Abderezaei explained.

Finally, the students partnered with researchers at Mount Sinai to process scans previously collected from two aneurysm patients—one high-risk and one low-risk—over a one-year period. Using aFlow, the team determined that over the course of 12 months, the high-frequency wall movements of the high-risk patient’s aneurysm increased almost four times more than that of the low-risk patient.

That’s significant because it shows that aFlow can rapidly differentiate high-risk and low-risk aneurysms based on previously undetectable motion in blood-vessel walls, Abderezaei explained. “In the long run, we’d like to use machine learning to automatically detect these warning signs,” he said. “But for now, we’re thrilled to have shown that aFlow can yield diagnostically useful data.”

The brain in motion

The Stevens team’s findings are part of a new wave of research, driven in part by Kurt, that views the brain as a dynamic, moving organ. “People used to think of the brain as static, but it’s actually in constant motion,” Kurt explained. “Blood flow, the movement of cerebrospinal fluid, and ripples of force as the head moves or suffers impacts can have a big effect on its operation.”

In recent years, Kurt and his research team have pioneered the use of imaging, experimental, and computational tools to explore the brain’s motion, unlocking new insights into the biomechanics of neural function. Such methods will ultimately help clinicians understand and treat conditions ranging from concussions to vascular diseases, Kurt explained.

First, though, there’s work to be done—and Stevens’ students are a critical part of that process. While many researchers use students to conduct routine lab work, Kurt brings undergraduate and graduate students into his planning sessions from the very beginning. “We hold weekly meetings where we brainstorm problems, and think about strategies to move things forward,” he explained. “The energy and fresh perspectives that students bring are a vital part of my lab’s research process.”

In fact, Kurt said, an undergraduate was one of the masterminds behind the aneurysm imaging project. During an early brainstorming session, John Martinez (B.E. / M. Eng. ’18) realized the lab’s work on brain biomechanics could be applied to cerebral aneurysms. “Because of the education they’re receiving at Stevens, these students think about problems in completely new ways,” Kurt said. “As a researcher, it’s shocking how much you can learn from your students.”

From the lab to the clinic

Kurt said he’ll continue to rely on undergraduate and graduate students as he works to understand the brain’s motion and create new clinical tools. For now, Abderezaei’s team is preparing to test aFlow on larger data sets, and to start to establish diagnostic benchmarks. That could set the stage for widespread clinical use in coming years.

“Because aFlow is simply an algorithm, doctors won’t need expensive new hardware to gain better clinical insights,” Abderezaei explained. “It can be rolled out quickly once we finish our work.”

Better yet, added Kurt, the same computational imaging used to study aneurysms could also be applied to many other neurological conditions, including strokes and traumatic brain injuries.

“The key is getting the patient data we need to adapt the method to other use cases,” Kurt said. “But this approach has enormous potential. In five to ten years, thanks to the work being done now by Stevens students, we’ll see surgeons using these tools to treat their patients.”

Learn more about academic programs and research in the Department of Mechanical Engineering: