Predicting internal carotid artery system risk based on common carotid artery by machine learning

Area of Science:

• Vascular Medicine
• Medical Imaging
• Machine Learning

Background:

• Early identification of internal carotid artery (ICA) diseases is crucial for stroke prevention.
• Current diagnostic methods are often costly and require specialized expertise, limiting accessibility.
• Developing efficient screening tools is essential for timely intervention.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model for predicting ICA disease risk.
• To utilize common carotid artery (CCA) features for accessible and efficient screening.
• To identify key predictors of ICA disease from CCA data.

Main Methods:

• Analysis of clinical data from 1612 patients (806 high-risk, 806 low-risk ICA disease).
• Training five ML models using CCA features: blood flow, intima-media thickness, internal diameter, age, and gender.
• Evaluation of model performance using accuracy, sensitivity, specificity, AUC-ROC, and F1 score; SHAP analysis for predictor identification.

Main Results:

• The support vector machine (SVM) model achieved the highest performance (accuracy 84.9%, AUC 92.6%).
• SVM outperformed neural networks (accuracy 81.4%, AUC 89.8%).
• SHAP analysis identified CCA blood flow (negative correlation) and intima-media thickness (positive correlation) as dominant predictors.

Conclusions:

• CCA hemodynamic and structural features, analyzed with interpretable ML, can effectively predict ICA disease risk.
• The SVM-based framework provides a cost-effective screening tool for early intervention, especially in resource-limited settings.
• Further validation in multi-center cohorts is recommended.