- Politics
- Diversity, equity and inclusion
- Financial Decision Making
- Telehealth
- Patient Experience
- Leadership
- Point of Care Tools
- Product Solutions
- Management
- Technology
- Healthcare Transformation
- Data + Technology
- Safer Hospitals
- Business
- Providers in Practice
- Mergers and Acquisitions
- AI & Data Analytics
- Cybersecurity
- Interoperability & EHRs
- Medical Devices
- Pop Health Tech
- Precision Medicine
- Virtual Care
- Health equity
Machine Learning Automates Analysis of CTA for Stroke Patients
This method could improve access to screening for stroke treatments.
A prediction model helped physicians outside of major stroke treatment centers assess whether a patient suffering from ischemic stroke would benefit from an endovascular procedure to remove a clot blocking an artery, researchers at the University of Texas Health Science Center at Houston (UTHealth) found.
The machine-learning model identified the intracerebral vasculature — narrowing of the arteries at the base or inside the brain — on a computed tomography angiogram (CTA) and detected large vessel occlusion with an area under the curve of 0.88.
Endovascular thrombectomy is a procedure where a catheter can mechanically remove the clot. The procedure can improve outcomes for stroke patients, but only if there is minimal brain tissue injured at the time of the treatment. But technology and expertise to quickly detect the amount of brain tissue injured are not available at most hospitals and primary stroke centers.
“Unfortunately, the advanced imaging techniques used currently to identify which patients benefit from this procedure are not widely available outside of large referral hospitals,” said corresponding author Sunil Sheth, M.D., assistant professor of neurology at the McGovern Medical School at UTHealth. “As a result, most stroke patients don’t have access to guideline-based screening for these treatments.”
The research team created and trained a machine-learning model, called DeepSymNet, to identify large vessel occlusion and infarct core from CTA source images.
The model automatically learned subtle image patterns that can be used as a proxy for more advanced imaging modalities.
Researchers used a stroke registry and an electronic health record to identify patients with acute ischemic stroke and stroke mimics with contemporaneous CTA and computed tomography perfusion with Ischema View’s Rapid. Of 224 patients who had stroke, 179 had blocked blood vessels.
The algorithm identified the blockages from the CTA images and trained the software to use the images to define the area of the brain that died using concurrent acquired CT perfusion scans.
“The advantage is you don’t have to be at an academic health center or a tertiary care hospital to determine whether this treatment would benefit the patient,” said Sheth. “And best of all, CT angiogram is already widely used for patients with stroke.”
DeepSymNet determined infarct core from the CTA source images with an area under the curve of 0.88 and 0.90.
“These results demonstrate that the information needed to perform the neuroimaging evaluation for endovascular therapy with comparable accuracy to advance imaging modalities may be present in CTA and the ability of machine learning to automate the analysis,” the study authors concluded.
It will be important to further study images from multiple centers to assess the generalizability of the method, the researchers suggested.
The study can be found in the journal Stroke.
Get the best insights in digital health directly to your inbox.
Related
Medtronic and Viz.ai Partner to Improve Stroke Care with AI
IBM Develops Virtual Assistant to Aid Stroke Patients
Digital Health Program Helps Reduce Blood Pressure, Study Finds
