Cardiovascular disease detection using machine learning and carotid/femoral arterial imaging frameworks in rheumatoid...

Area of Science:

• Cardiology
• Machine Learning
• Biomedical Informatics

Background:

• Cardiovascular disease (CVD) remains a leading cause of mortality globally.
• Accurate risk stratification is crucial for timely intervention in individuals with conditions like rheumatoid arthritis, diabetes mellitus, and arterial hypertension.
• Existing CVD risk prediction models often lack precision in diverse patient populations.

Purpose of the Study:

• To develop and validate a novel machine learning (ML) paradigm for early cardiovascular disease (CVD) detection.
• To assess the efficacy of ML classifiers using conventional, laboratory, and imaging-based phenotypes in predicting CVD.
• To compare the performance of the proposed ML framework against classical CVD risk scores.

Main Methods:

• A Greek cohort of 542 individuals with medium to high cardiovascular risk was studied over two time points (3-year interval).
• Data included conventional risk factors, laboratory biomarkers, and carotid/femoral ultrasound phenotypes.
• Three ML classifiers (Random Forest, SVM, LDA) were employed with a two-fold cross-validation and SMOTE data augmentation.

Main Results:

• The ML paradigm achieved a mean accuracy of 98.40% and an Area Under the Curve (AUC) of 0.98 for CVD detection at baseline.
• Performance remained high at the 3-year follow-up with 98.39% accuracy and 0.98 AUC.
• The developed cardiovascular framework demonstrated significantly superior performance compared to classical CVD risk scores.

Conclusions:

• The proposed ML paradigm is a powerful tool for predicting cardiovascular disease in at-risk individuals.
• Integrating diverse data types (clinical, lab, imaging) enhances ML model accuracy for CVD detection.
• This novel approach offers a promising advancement over traditional risk assessment methods for cardiovascular health.