Methods of spreading depolarization scoring and model development. Credit: Scientific Reports (2025). DOI: 10.1038/s41598-025-91623-7
A University of Cincinnati study found machine learning models can aid in the automation and detection of abnormal brain activity sometimes referred to as a “brain tsunami.”
UC’s Jed Hartings, Ph.D., is corresponding author of the study published March 12 in the journal Scientific Reports detailing how automation can aid clinicians treating patients with spreading depolarizations (SDs).
Hartings said SDs are believed to occur in patients with virtually any type of acute brain injury, including different kinds of strokes and traumatic brain injuries (TBI). Approximately 60% to 100% of all patients in these different disease categories are believed to experience SD.
Just like a battery, brain cells have a stored, or polarized, charge that enables them to send signals to one another. During SD, brain cells become depolarized and unable to send these electrical signals, which Hartings said essentially turns brain cells into a “big bag of saltwater that’s not functional anymore.”
“This happens en masse in a local area of tissue and then spreads out like a wave, like ripples in a pond, and it interrupts every aspect of cell function,” said Hartings, professor and vice chair of research in the Department of Neurosurgery in UC’s College of Medicine.
SDs can occur continuously in patients for several days, but they can also continue on and off up to two weeks after a severe brain injury.
Study details and results
Research has focused on patients who have had an electrode strip surgically placed in the brain to monitor for SDs. Physicians also need to receive specific training on reading the brainwave recordings so they can constantly monitor the data.
“This is time-intensive and expensive, and few physicians have the highly specialized expertise to diagnose SDs reliably,” Hartings said. “Therefore, we wanted to automate SD diagnosis to make this monitoring more accessible and widely available.”
Hartings and his colleagues used more than 2,000 hours of brain monitoring data from 24 patients who were hospitalized for severe TBI, and experts manually reviewed and identified more than 3,500 unique SD events in the data set.
Half of this patient data was used to train a machine learning model how to accurately recognize and classify SD events. Once the model was trained, researchers used the other half of data to see how accurately it could identify SDs in “new” data it hadn’t seen before.
“We showed that the method is able to identify SDs with a high degree of sensitivity and specificity,” Hartings said. “Overall, the performance was similar to an expert human scorer.”
Unexpectedly, the team found the algorithm could detect many SD events that were not identified using human scoring, likely due to a higher degree of objectivity. Testing the limits of the algorithm, researchers found it could achieve a high degree of performance using one voltage reading per 10 seconds, compared to the typical method of collecting 256 data points per second.
“If we could achieve a high degree of performance with minimal information, this would leave a lot of ‘headroom’ to improve performance even further by adding in the additional information,” Hartings said. “That work is ongoing now.”
Study impact
Hartings said when the algorithm is fully realized and implemented, automated SD detection would allow any neurosurgical center to monitor patients for SDs even if they do not have a physician on staff with this specialized training.
“Many neurosurgical centers are interested in monitoring SDs for patient care but simply don’t have the knowledge or resources to implement it,” he said. “Having automated SD reading will lower these barriers and hopefully make this technology more accessible, and thus accelerate research and hopefully patient benefit.”
While the study results are promising, Hartings cautioned there is further development and validation needed before automated detection fully replaces human expertise and detection of SDs.
“I think we are headed in that direction. But even if not, automated detection would, at a minimum, considerably reduce workload and increase response times, since alarms could alert physicians to review data or take action earlier than they might otherwise, following usual intervals to review patient progress,” he said. “This is another significant benefit that should not be overlooked.”
Limitations of the study include the need for the electrode strip to be placed on the brain during neurosurgery, limiting the number of patients who can be monitored to those undergoing surgery. However, Hartings said research is ongoing to develop noninvasive detection methods that could be used to monitor a larger population of patients.
Moving forward, Hartings and his colleagues are continuing to refine the algorithm using larger data sets and testing software implementation, with the plan for other institutions to trial the software and to begin using it for patient care and research.
Additionally, the team is conducting clinical trials like the INDICT trial to determine optimal treatments for SDs. Having a more precise detection method coupled with better tools to treat SDs could have a significant impact on patient care.
More information:
Sreekar Puchala et al, Automated detection of spreading depolarizations in electrocorticography, Scientific Reports (2025). DOI: 10.1038/s41598-025-91623-7
Provided by
University of Cincinnati
Citation:
The ‘brain tsunami’: Machine learning aids in detection of abnormal activity (2025, March 19)
retrieved 19 March 2025
from https://medicalxpress.com/news/2025-03-brain-tsunami-machine-aids-abnormal.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
