From the Brain, Through the Skull, to the Scalp

Whether it be a rough tackle on the football field or blast exposure on the battlefield, high-impact accidents can induce a condition that’s very difficult to treat, frequently resulting in devastating outcomes like long-term disability or death: traumatic brain injury (TBI).

TBIs are clinically challenging in part because hours to weeks after an initial injury, around half of patients experience secondary brain injuries, which often go undetected until after they’re over and further damage to the brain has already been done.

Secondary brain injuries are induced by “brain tsunamis,” waves of suppressed neural activity that creep slowly across the surface of the brain. Clinically, they’re known as cortical spreading depolarizations.

Luckily, brain tsunamis can be treated—but only if clinicians know they’re happening. “If you can stop these waves, then you can also stop these injuries,” says Pulkit Grover, professor of electrical and computer engineering. 

With collaborators at the University of Cincinnati and the University of Pittsburgh, Grover, a member of CMU's Neuroscience Institute, and former Ph.D. student Alireza Chamanzar have previously shown that their machine learning algorithm WAVEFRONT can detect brain tsunamis from electroencephalogram (EEG) data taken from electrodes on TBI patients’ scalps.

This was a major advance for neurocritical care. Before, brain tsunamis could only be detected through electrodes placed directly on the brain. That type of data is very limited in spatial coverage; only a few electrodes can be placed on the brain, so it’s hard to tell a local silencing of neural activity from a traveling one.

Scalp electrodes, though, are not only noninvasive, they can be placed anywhere and pick up data from the entire brain. “You can get much better inferences from the scalp if you use the right techniques and the right algorithms,” says Grover.

The WAVEFRONT algorithm flags potential brain tsunamis as they move and get picked up by different electrodes, which requires a massive amount of data to be synthesized. WAVEFRONT looks for waves that can be detected from the surface of the brain, and checks whether they are “consistent with the clinical understanding of these waves—that they are traveling at the right speed, that they maintain a certain steady direction,” Grover explains. “Because we know a lot about neurobiology—thanks to a lot of animal experiments, fundamental science, and clinical studies—we can utilize all of that information and embed it into the algorithm.”

Yet another new development from the team has revealed a superpower unique to WAVEFRONT: the ability to detect tsunamis even through intact skulls.

Through research presented at the 2025 annual meeting of the Neurocritical Care Society, for which they won the Best Science Abstract Award, the team demonstrated the first “robust, noninvasive” way to detect brain tsunamis since they were first characterized more than 80 years ago.

In the team’s prior work, scalp electrodes were placed on severe TBI patients while they were undergoing a procedure to relieve swelling in the brain. That procedure requires a large portion of the skull to be temporarily removed—which incidentally made the EEG data easier for WAVEFRONT to work with, since the waves did not need to travel through the skull.

The new study included patients from three different hospitals who had injuries that were still severe, but somewhat milder, and intact skulls. Despite this additional challenge, WAVEFRONT’s robustness was evident from its nearly 89% accuracy in making predictions from data that weren’t used to train it.

“We showed that when we trained the algorithm on one set of patients in one center, we could generalize it to another set of patients in another center, even though there are some differences in how they record the signal,” says Grover.

While the fact that institutions use different monitoring hardware and protocols produced unique challenges, partnering with leaders in the field to understand and address them was very beneficial. With the expertise of Eric Rosenthal, medical director of the neurosciences intensive care unit at Massachusetts General Hospital, and Jens Dreier at Charité University of Berlin, the team was able to put WAVEFRONT to the test at both institutions, as well as the University of Cincinnati.

“This diverse collaboration with clinicians has been instrumental in refining and tailoring the key components of WAVEFRONT for applications in translational medicine,” says Chamanzar, who developed the WAVEFRONT algorithm while a Ph.D. student in Grover’s lab and is now an assistant professor at the University of Pittsburgh.

Given how much work is yet to be done to better diagnose and treat brain injuries—from severe TBIs to milder injuries like concussions—and how widespread the availability of EEG is, Grover’s team is always thinking about the next questions to tackle with WAVEFRONT and EEG data. 

“If you can do it with EEG, you can enable so much more,” says Grover.