Machine Learning for Carotid Atherosclerosis Detection Using Optimized AI Models on Doppler Ultrasound Features

抄録

Carotid Atherosclerosis (CAS) is a key risk factor for cardiovascular diseases and stroke, requiring early and accurate detection. This study evaluates machine learning (ML) models to classify CAS severity using Doppler ultrasound-derived features. A dataset of ultrasound signals was processed using MATLAB, extracting key biomarkers such as spectral entropy, peak frequency shifts, spectral energy, and estimated blood flow velocity. Three classification models?Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN)?were trained and optimized through hyperparameter tuning with five-fold cross-validation. Feature standardization and noise augmentation were applied to improve model generalization. Results indicate that SVM and Random Forest achieved the highest classification accuracy (83.33%), outperforming KNN (66.67%). The optimized SVM model with a linear kernel and fine-tuned hyperparameters demonstrated superior robustness in distinguishing CAS cases. Additionally, the study highlights the significance of frequency-domain features and velocity estimations in CAS diagnosis. The best-performing model was saved for future deployment in clinical decision support systems. This research underscores the potential of AI-driven diagnostic tools in enhancing non-invasive screening, facilitating early detection, and aiding in stroke prevention. Future work will explore deep learning approaches and multi-modal integration for improved diagnostic accuracy.