Each year, nearly 800,000 people in the United States have a stroke: a serious medical condition where blood supply to the brain is blocked or a blood vessel in the brain bursts. In the latter case, emergency surgery is necessary to remove any blood from the brain. Speed is of the essence: In either type of stroke, parts of the brain become damaged, which can lead to permanent brain damage and long-term disability.
“With stroke, imaging is critical to guide treatment,” says Alan McMillan, PhD. “It’s important to make treatment decisions quickly.”
Dr. McMillan is one of a six-person research team investigating ways to streamline image-guided stroke intervention. In these procedures, surgeons rely on magnetic resonance imaging (MRI) of the brain to provide information about the size and location of the clot.
“Getting a segmentation is time-consuming, and you need to know the answer quickly,” says Dr. McMillan (“segmentation” refers to the time it takes for a radiologist to process the image and delineate the various types of tissues present within). “It could take a human 20-30 minutes, maybe even longer, to segment a single case.”
UW-Madison’s research team is developing a machine-learning network that can cut down on the time it takes to process an image.
The multi-disciplinary team includes members of the Departments of Radiology, Medical Physics, Biomedical Engineering, Neurosurgery and Electrical Engineering. Oh, and one of the authors is Denver Broncos defensive end Matt Henningsen, who contributed to the project while attending UW-Madison (where he received both his bachelor’s and masters degree in electrical engineering).
Dr. McMillan contributed his expertise in artificial intelligence (AI) to the research team. The first author, Thomas Lilieholm, a PhD candidate in the Department of Medical Physics, also took the Machine Learning for Medical Imaging (ML4MI) Initiative’s Summer Bootcamp, which gives participants a hands-on introduction into the principles and application of machine learning for medical imaging. Lilieholm even credited ML4MI in the article’s acknowledgements.
The research team trained a machine learning algorithm to identify where a blood clot is in the brain. Their findings – which were recently published in Magnetic Resonance Imaging – show that their machine-learning network was able to identify and segment MRI images of blood clots swiftly and accurately.
“It makes it more feasible to do an image-guided intervention,” says Dr. McMillan. “If you know where the clot is with good certainty, you can use that to inform the intervention.”
With more information about the size and location of the blood clot, neurosurgeons can select an intervention that’s suitable for that patient. Optimizing the treatment, in turn, can mitigate the extent of brain damage due to stroke.
While a peer-reviewed article is an important first step, there’s more to be done before automated segmentation is standard practice.
In the event of a stroke, every minute counts. While research like this aims to accelerate imaging and intervention, being able to quickly recognize the signs and symptoms of stroke is crucial. The acronym F.A.S.T. (coined by Centers for Disease Control and Prevention) is a useful tool to identify stroke:
- Face: Ask the person to smile. Does one side of the face droop?
- Arms: Ask the person to raise both arms. Does one arm drift downward?
- Speech: Ask the personal to repeat a simple phrase. Is the speech slurred or strange?
- Time: If you see any of these signs, call 9-1-1 right away.
Review other signs and symptoms of a stroke here.
Read the full article in Magnetic Resonance Imaging here.