Predicting Carotid Artery Stenosis Progression: A Comprehensive Machine Learning Approach

Panagiotis Siogkas, Dimitrios Pleouras, Vassilis Tsakanikas, Vassiliki Potsika, Fragiska Sigala, Dimitrios I. Fotiadis

Confirmation: I have read and agree with the IEEE BHI 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: Machine Learning in Carotid Artery Disease, Stenosis Progression, Prognosis

Abstract: Cardiovascular diseases remain a leading global health burden, with carotid artery stenosis progression being a critical determinant of adverse cerebrovascular outcomes. This study aims to enhance the predictive accuracy of carotid stenosis progression by leveraging advanced machine learning algorithms, thereby advancing precision medicine in vascular care. A comprehensive evaluation of six classification models was conducted, including Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Random Forest, Gradient Boosting, Logistic Regression, and Decision Tree. Each model was trained on a curated set of hemodynamic, imaging, and clinical features, and assessed using a 70%–30% stratified train-validation split. The study emphasizes the integration of high-fidelity simulation outputs and ultrasound-derived features in constructing predictive models. Among the classifiers tested, XGBoost demonstrated the highest performance, achieving an AUC of 0.741, accuracy of 70.5%, precision of 70%, and specificity of 88.9% on the validation dataset. Additionally, a 5-fold cross-validation strategy was performed to assess generalizability, with XGBoost achieving a mean AUC of 0.732 and improved overall robustness across folds. These metrics underline its superior capacity to distinguish between stable and progressive stenosis cases. SVM and Gradient Boosting also yielded competitive results, while simpler models lagged in performance. The findings underscore the value of machine learning—particularly ensemble-based approaches such as XGBoost—in predicting stenosis progression. By incorporating rich hemodynamic data and patient-specific imaging features, the model offers a viable tool for early intervention planning. Future work should focus on longitudinal datasets and further validation to support clinical translation and personalized therapeutic strategies.

Track: 4. Clinical Informatics

Registration Id: V3N4PKQ6DZ3

Submission Number: 284