Deep Learning-Based Collateral Scoring on Multiphase CTA in Patients with Acute Ischemic Stroke in the MCA Region

Area of Science:

• Neuroradiology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Collateral circulation is crucial for acute ischemic stroke (AIS) patient outcomes and endovascular therapy selection.
• Assessing collateral status traditionally involves subjective methods with observer variability.
• Efficient and objective collateral assessment is needed to improve patient management.

Purpose of the Study:

• To develop and validate an automated deep learning (DL) method for collateral circulation assessment.
• To quantify collateral status on multi-phase CT angiography (mCTA) in AIS patients.
• To reduce observer variability and enhance diagnostic efficiency in stroke care.

Main Methods:

• Retrospective analysis of mCTA scans from 420 AIS patients.
• Development of a DL classification model with a preprocessing module.
• Comparison of DL model performance against manual Menon method evaluations (ground truth).
• Performance evaluation using accuracy, F1 score, precision, sensitivity, specificity, AUC, ICC, and Kappa via five-fold cross-validation.

Main Results:

• The DL framework achieved high accuracy (87.6% for 3-class, 94.0% for 2-class good vs. non-good, 97.1% for 2-class poor vs. non-poor).
• Excellent performance metrics including high AUC (up to 97.1%) and inter-rater reliability (ICC up to 0.882).
• Multi-phase CTA demonstrated superior performance compared to single or two-phase CTA for collateral assessment.

Conclusions:

• The proposed DL framework accurately and consistently evaluates collateral circulation on mCTA in AIS patients.
• This automated method shows potential as a valuable tool for clinical decision-making.
• The AI approach can reduce variability and streamline the diagnostic workflow for stroke treatment.