Machine Learning Detects Symptomatic Plaques in Patients With Carotid Atherosclerosis on CT Angiography

Area of Science:

• Vascular Medicine
• Radiology
• Artificial Intelligence in Medicine

Background:

• Carotid atherosclerosis poses a significant risk for cerebrovascular events.
• Identifying symptomatic carotid plaques is crucial for timely intervention.
• Current methods for plaque characterization have limitations.

Purpose of the Study:

• To develop and validate a machine learning model for detecting symptomatic carotid plaques.
• To utilize computed tomography angiography (CTA) derived plaque composition and stenosis degree.
• To enhance diagnostic accuracy beyond traditional methods.

Main Methods:

• A machine learning model was trained on CTA data, including plaque subcomponents (lipid, intraplaque hemorrhage, calcium) and stenosis degree.
• Internal validation was performed using 10-fold cross-validation.
• The model was tested on a dedicated cohort of 106 patients.

Main Results:

• The machine learning model achieved an area-under-receiver-operating characteristics curve of 0.89 on the testing cohort, significantly outperforming traditional analyses.
• Key predictors for symptomatic plaques included the ratio of intraplaque hemorrhage to lipid volume (≥50%) and percentage of intraplaque hemorrhage volume (≥10%).
• The model demonstrated high discrimination and calibration, with comparable performance on internal validation.

Conclusions:

• An interpretable machine learning model effectively identifies symptomatic carotid plaques using CTA-derived features.
• The model aids in clinical decision-making for patients with carotid atherosclerosis.
• Plaque composition, particularly intraplaque hemorrhage, is a vital indicator of plaque symptoms.