Automatic Ischemic Core Estimation Based on Noncontrast-Enhanced Computed Tomography

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Assessing ischemic stroke extent is crucial for guiding thrombolysis and thrombectomy.
• Current methods like the Alberta Stroke Program Early CT Score lack precision and inter-rater reliability.
• Need for a reliable, automated tool for ischemic core volume estimation.

Purpose of the Study:

• To develop and validate a fully automated machine learning model for ischemic core segmentation.
• To utilize only noncontrast-enhanced computed tomography (CT) images for the model.
• To improve the accuracy and consistency of ischemic core volume assessment in acute ischemic stroke.

Main Methods:

• Retrospective multicenter study including patients with anterior circulation acute ischemic stroke.
• Development of a deep learning (DL) model using CT data from 272 patients.
• Validation of the DL model on a separate cohort of 106 patients, comparing with MRI-derived core volumes.

Main Results:

• The DL model showed significant correlation with the reference standard for ischemic core volume (ICC=0.90).
• High accuracy was maintained across different time windows (≤4.5 hours and >4.5 hours).
• The model demonstrated excellent performance in distinguishing large ischemic cores (AUC=0.91, sensitivity=84.2%, specificity=97.7%).

Conclusions:

• A DL-based ischemic core segmentation model using noncontrast CT is highly accurate.
• This automated model offers a reliable alternative for assessing ischemic core volume in acute ischemic stroke.
• Potential to enhance treatment decisions by providing consistent and precise imaging analysis.