Artificial Intelligence-Driven Imaging Tool for Assessing Strokes Is Launched

Posted in News Releases | Tagged artificial intelligence, brain, brain imaging, MRI, rehabilitation, stroke, stroke imaging
Media Contact
Karen Teber
km463@georgetown.edu
(January 16, 2025) — A new first-of-its-kind, artificial intelligence-driven brain imaging platform launched by stroke physician-scientists at Georgetown University Medical Center in collaboration with MedStar Health is expected to revolutionize the way stroke recovery research is conducted. In turn, the new tool could better guide treatment options, as well as assist in a greater understanding and detection of stroke health disparities.
Every stroke is slightly different in its size and location in the brain. These differences make it difficult to study recovery from stroke because each survivor will face different challenges depending on the size and location of the stroke, such as weakness on one side of the body or difficulty speaking.
Launched in January, the Acute Stroke Imaging Database (AStrID) will scan MRI images in order to automatically identify the stroke’s type and location in approximately 5,000 acute stroke cases seen in the MedStar Health system annually. AStrID is part of a broader stroke registry that parses electronic medical records of stroke patients to facilitate research. AStrID will allow researchers to invite groups of stroke survivors to join clinical trials based on the attributes of their stroke, offering the most advanced treatment options to patients and greatly enhancing the validity of the studies.

“Certain treatments to improve recovery may only help people who have a stroke to a specific part of the brain,” says Peter E. Turkeltaub, MD, PhD, director of the Cognitive Recovery Lab at Georgetown University and the Aphasia Clinic at MedStar National Rehabilitation Hospital. “In the past, we had no way to quickly identify people based on the exact size and location of a stroke to test these treatments. AStrID makes this possible for the first time.”
Annually, about 800,000 people in the U.S. have strokes, which disproportionately affect marginalized and minoritized people. MedStar Health’s patient base, which comes from 10 hospitals in Maryland, Washington, D.C., and Virginia, is an ideal population to represent the U.S. as it spans a wide spectrum of socioeconomic, cultural and racial backgrounds at academic and community hospitals.
“Disparities in access to services affect every phase of stroke care,” Turkeltaub says. “AStrID and the broader stroke registry can help us understand how these disparities affect outcomes. We can identify groups of people who we’d expect to have similar recoveries because they have very similar strokes, and then assess if access to rehabilitation or other stroke services changes their recovery. That will help us find the people who would benefit most from better stroke services to improve equity and outcomes.”
AStrID was developed by Turkeltaub; Andrew T. DeMarco, PhD, an assistant professor in the Department of Rehabilitation Medicine; and Kyle Shattuck PhD, an instructor in the Department of Rehabilitation Medicine. It and the stroke registry (started by the late Alex Dromerick, MD) are projects of the Center for Brain Plasticity and Recovery, a joint collaboration between Georgetown University and MedStar National Rehabilitation Hospital.
Turkeltaub says he expects the registry to grow exponentially within only a year.
“We started with hundreds of training images, where we manually identified where the strokes were,” he says. “We fed that information into the AStrID learning algorithm that, based on the training sets we provided, can identify the stroke in new MRI images. We use a search tool to click on a specific part of the brain and find all the images in AStrID that have strokes at that location.
“Within the next year, we expect to have up to 20,000 stroke images in AStrID, which may make it the biggest stroke imaging repository in the world,” Turkeltaub predicts.
To ensure confidentiality, AStrID removes all identifying information from MRIs before processing them and only stores digital images showing where the strokes are, not the actual MRI images. These digital images are stored separately from the associated electronic medical record information in the broader stroke registry, and can only be linked using anonymizing codes.
Turkeltaub’s clinic and research is primarily focused on recovering language abilities after stroke. About a third of stroke survivors have difficulties using language, called aphasia. Brain plasticity after stroke allows people to relearn language over time as new connections form in the brain. Boosting plasticity might help people recover better from aphasia, but the pattern of brain plasticity differs in each person depending on the size and location of their stroke.
“This has made it hard to learn how brain plasticity supports recovery,” Turkeltaub says.
Using AStrID, he and his colleagues can identify groups of people with very similar strokes, which may help them to uncover previously hidden patterns of brain plasticity that could be enhanced with new treatments to improve recovery.
“Addressing aphasia is only one example of how this database can benefit research, but we expect it to have many other applications,” Turkeltaub concludes.
Turkeltaub and his AStrID colleagues report having no personal financial interests related to this research. This work was supported in part by a grant from the NIH (R01DC020446) and funds from the GUMC Executive Vice President’s office.