Radiomic Carotid Plaque Features Integrated into Machine Learning Models for Cardiovascular Risk Prediction.

Area of Science:

• Cardiovascular imaging and diagnostics
• Machine learning in medicine
• Biomedical data analysis

Background:

• Carotid plaque visualized by ultrasound is linked to major adverse cardiovascular events (MACE).
• Radiomic analysis, both manual and automated, can characterize carotid plaque.
• Machine learning (ML) can model complex interdependencies within plaque characteristics for predictive purposes.

Purpose of the Study:

• To develop an ML model for predicting MACE.
• To utilize clinical variables, focused vascular ultrasound (FOVUS) measurements, and semi-automated radiomic ultrasound features for prediction.
• To evaluate the predictive performance of different data combinations.

Main Methods:

• Carotid ultrasound scans from 493 patients were analyzed, with MACE outcomes tracked over 5 years.
• ML models were built incorporating clinical data, manual FOVUS measurements, and radiomic features.
• Feature selection using ReliefF identified 11 top features; four ML classifiers were trained and evaluated using 10-fold cross-validation.

Main Results:

• Over 5 years, 144 patients (29%) experienced a MACE.
• The best performing model (x-gradient boost) integrating all data types (clinical, FOVUS, radiomics) achieved the highest prediction accuracy (0.958 ± 0.023).
• Models combining FOVUS and radiomic data (0.935 ± 0.043) also showed strong predictive power.

Conclusions:

• Integrating clinical, FOVUS, and radiomic data offers robust predictive capability for distinguishing MACE.
• Both automated and manual plaque quantification methods yield valuable predictive features.
• ML models demonstrate significant potential in predicting cardiovascular events from ultrasound-derived plaque characteristics.