Some say you’re only as old as you feel. Others say you’re as young as you think. But how old is your brain really?

Scientists have been developing and refining methods to accurately measure the age and health of our brains, without cutting into our skulls to have a look. Understanding brain health is critical to identifying and addressing disorders, including Alzheimer’s disease and other forms of dementia, multiple sclerosis and Parkinson’s disease.

Curtis Johnson, associate professor of biomedical engineering at the University of Delaware, has been in the thick of that study for more than a decade, using magnetic resonance elastography (MRE) to map and measure the mechanical properties of brains, with special focus on brain stiffness. The MRE technique he designed gently vibrates someone’s head while they are being scanned using magnetic resonance imaging. The stiffness map shows how tissue in each part of the brain would respond to a force like a light touch.

Brain stiffness is an important indicator of brain health. Knowing about brain stiffness in different regions of a healthy brain is important, because brains tend to lose stiffness and grow softer as we age or experience neurodegenerative diseases.

Johnson and his Mechanical Neuroimaging Lab have gathered enormous amounts of data on brain stiffness, and now, in collaborative research with Austin Brockmeier, assistant professor of electrical and computer engineering, new information is coming to light.

For the first time, Johnson and Brockmeier, along with three current and former UD students, have shown that combining artificial intelligence and MRE techniques is a reliable way to predict the age of a healthy brain and could be used to identify structural differences that indicate departure from the normal aging process.

In recent findings, Johnson and Brockmeier show that measuring both brain stiffness and brain volume produces the most accurate predictions of chronological age. The paper was published in a recent edition of the journal Biology Methods and Protocols.

“Brain volume is a common measure that we use to study the brain,” Johnson said. “But something has to be happening to cause a brain to shrink. Something is happening at the microscale that causes it to shrink — changes in the tissue that also cause stiffness to change. And that precedes whatever happens when the volume changes.”

“The stiffness maps all seem kind of random — until we see a large number of images and the randomness fades away and we start to see common patterns in stiffness,” Johnson said. “We sort of knew there was more [information] in there than what we were extracting. But the traditional way of interacting with the stiffness data is pretty crude.”

UD’s Center for Biomedical and Brain Imaging is anything but crude, with its cutting-edge magnetic resonance imaging (MRI) scanner.

“The machine we have at CBBI is really nice and ideal for developing new methods for brain scanning,” Johnson said. “Trying to take one of our datasets on an older scanner would not have been possible. It would have crashed the scanner.”

Brockmeier, a resident faculty member in UD’s Data Science Institute, and third-year doctoral student César Claros, the first author on the paper, brought high-dimensional data analysis and artificial intelligence models to the task through a collaboration initiated by Rebecca Clemens, an honors biomedical engineering alumna of UD, who is now working toward her doctoral degree at Northwestern University, and Grace McIlvain, who was mentored by Johnson during her doctoral work at UD and now is a professor at Columbia University.

Patterns and new insights emerged, and they brought their findings to Johnson.

“Beyond the more accurate predictions, we wanted to understand what the model is seeing, because it’s mathematically defined, a bunch of numbers in a function on the computer,” said Brockmeier, who leads the Computational Neural Information Engineering Laboratory. “Why is it making that prediction? What is it looking for? We started showing him the model’s regions of interest. It gives confidence to an expert like him that the model was picking up on what he was seeing before.”

Brockmeier brought his expertise in machine learning, applied to neural networks and neuroscience, with game-changing capacities for this study.

Neural networks are of keen interest in biology and in machine learning, which borrowed the term from brain scientists. For neuroscientists, neural networks are formed by the pathways among neurons that define how our brains work. For data analysts, neural networks refer to the mathematical machines and artificial intelligence that produce predictive models and contribute to problem solving.

“The neural network in your cortex — all of those little neurons connecting — that’s creating this matrix of stiffness in your brain,” Brockmeier said. “Every time you learn something new and your brain rewires that is what’s creating this stiffness. When many neurons are connected between brain areas they get stiff like a cable.”

On the artificial intelligence side, the brain maps were analyzed by three-dimensional “convolutional neural networks,” which — as the name suggests — are convoluted and complicated, incorporating many layers and dimensions.

“We’re very fortunate at UD to work with good collaborators, have good grant funding and an imaging center that allows us to collect and analyze a lot of data,” Johnson said. “And the methods we develop we share worldwide for other researchers interested in similar problems.”

About the researchers

Curtis Johnson is an associate professor of biomedical engineering at the University of Delaware. His research focuses on magnetic resonance imaging (MRI), elastography, brain tissue mechanics and neuroimaging. Johnson earned his bachelor’s degree in mechanical engineering at Georgia Institute of Technology, and his doctorate in mechanical engineering at the University of Illinois at Urbana-Champaign. Johnson joined the UD faculty in 2016. He is the director of the Mechanical Neuroimaging Lab and was the 2021 winner of UD’s Gerard J. Mangone Young Scholars Award, chosen by faculty members who have won the University’s highest competitive faculty honor, the Francis Alison Award.

Austin Brockmeier is an assistant professor of electrical and computer engineering, computer and information sciences, and a resident faculty member of UD’s Data Science Institute. His research focuses on artificial intelligence, machine learning, algorithms, signal processing and data science. He earned his bachelor’s degree in computer engineering at the University of Nebraska-Lincoln and his doctorate at the University of Florida. Before joining the UD faculty in 2018, he was a research fellow at the University of Manchester, United Kingdom, and a research associate at the University of Liverpool, United Kingdom. He is the director of the Computational Neural Information Engineering Laboratory, which covers research in both biological and artificial neural networks.