Related Experiment Video
Updated: Dec 29, 2025
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
Machine Learning for Detecting Early Infarction in Acute Stroke with Non-Contrast-enhanced CT
Wu Qiu1, Hulin Kuang1, Ericka Teleg1
1From the Calgary Stroke Program, Departments of Clinical Neurosciences (W.Q., H.K., E.T., J.M.O., M.G., M.D.H., A.M.D., B.K.M.), Radiology (M.G., M.D.H., A.M.D., B.K.M.), and Community Health Sciences (M.D.H., B.K.M.), University of Calgary, 239 Strathridge Pl SW, Calgary, AB, Canada T3H 4J2; Hotchkiss Brain Institute, Calgary, Alberta, Canada (M.G., M.D.H., A.M.D., B.K.M.), Department of Neurology, Keimyung University, Daegu, South Korea (S.I.S.); and Division of Neuroradiology, Clinic of Radiology and Nuclear Medicine, University Hospital Basel, University of Basel, Basel, Switzerland (J.M.O.).
Machine learning accurately quantifies brain infarction on CT scans for acute ischemic stroke (AIS) patients, aiding treatment decisions. This automated method shows strong agreement with MRI, improving stroke care.
Area of Science:
- Radiology and Medical Imaging
- Neurology
- Artificial Intelligence in Medicine
Background:
- Accurate assessment of brain infarction extent is critical for acute ischemic stroke (AIS) treatment selection.
- Patients with large infarcts may not benefit from reperfusion therapies like thrombolysis or thrombectomy.
Purpose of the Study:
- To develop and validate an automated machine learning (ML) approach for detecting and quantifying brain infarction.
- Utilize non-contrast-enhanced computed tomography (CT) scans for infarction assessment in AIS patients.
Main Methods:
- Retrospective study using non-contrast-enhanced CT and diffusion-weighted (DW) MRI scans from AIS patients (<6 hours symptom onset).
- Ischemic lesions on DW MRI served as the reference standard for training and testing an ML segmentation algorithm.
- Quantitative comparison using Bland-Altman plots and Pearson correlation between ML-derived CT volumes and DW MRI volumes.
Main Results:
- The ML algorithm demonstrated strong correlation (r = 0.76, P < .001) between algorithm-detected lesion volume and reference DW MRI lesion volume.
- The mean difference in volume between algorithm-segmented CT and DW MRI was 11 mL (P = .89), indicating good agreement.
- The study included 257 AIS patients, with 100 in the independent testing dataset.
Conclusions:
- An ML-based approach for segmenting infarction on non-contrast-enhanced CT images in AIS patients is effective.
- The automated method shows good agreement with stroke volumes measured by diffusion-weighted MRI.
- This technology has the potential to improve baseline infarction assessment and guide treatment decisions in AIS.
