Atherosclerosis risk classification with computed tomography angiography: A radiologic-pathologic validation study.

Area of Science:

• Cardiovascular imaging and pathology
• Artificial intelligence in medicine
• Atherosclerosis research

Background:

• Machine learning (ML) is increasingly applied to cardiovascular CT angiography (CTA) for plaque risk assessment.
• Studies correlating ML models with histologic definitions of plaque risk are limited.
• Histologic validation is crucial for ML accuracy in cardiovascular plaque analysis.

Purpose of the Study:

• To evaluate the accuracy of ML software in determining plaque risk phenotypes.
• To compare ML-based plaque risk assessment against expert pathologists' histologic ground truth.
• To assess ML performance in classifying plaque stability using CTA data.

Main Methods:

• Prospective collection of atherosclerotic plaque sections paired with CTA from carotid endarterectomy patients.
• Histologic annotation of plaque components (lipid-rich necrotic core, calcification, matrix, intraplaque hemorrhage).
• Classification of plaques into minimal disease, stable, or unstable phenotypes using a modified AHA definition and ML classifier.

Main Results:

• Excellent agreement between ML software and histologic ground truth in the validation cohort (weighted kappa 0.82).
• High area under the receiver operating curve for identifying plaque types: unstable (0.97), stable (0.95), and minimal disease (0.99).
• Poor correlation between diameter stenosis and histologically defined plaque type (weighted kappa 0.25).

Conclusions:

• ML software trained on histologic data demonstrates high accuracy in identifying plaque stability phenotypes.
• The ML tool shows strong concordance with expert pathologist assessments.
• This approach offers a promising method for non-invasive plaque risk stratification using CTA.