An explainable CT-based machine learning model integrating carotid plaque and perivascular adipose tissue for...

Area of Science:

• Medical Imaging
• Machine Learning
• Cardiovascular Disease

Background:

• Accurate identification of symptomatic carotid plaques is challenging due to limitations in conventional imaging.
• Current methods primarily focus on luminal stenosis, neglecting plaque vulnerability and perivascular inflammation.

Purpose of the Study:

• To develop and validate an explainable machine learning model for identifying symptomatic carotid plaques.
• The model integrates CT-based radiomics from carotid plaque and perivascular adipose tissue (PVAT).

Main Methods:

• Retrospective analysis of 324 patients with carotid atherosclerosis and stenosis.
• Extraction of 3D radiomics features from carotid plaque, PVAT, and combined regions.
• Development and internal validation (5-fold cross-validation) of a combined model integrating radiomics and clinical factors.
• Model interpretability assessed using Shapley Additive Explanations (SHAP).

Main Results:

• The combined model achieved excellent discriminative performance (mean AUCs of 0.903 and 0.904 in training and testing sets).
• The model significantly outperformed models using only plaque, PVAT, or clinical factors.
• SHAP analysis confirmed radiomics features enhance model interpretability and clinical relevance.

Conclusions:

• An explainable radiomics model integrating combined plaque and PVAT features with clinical factors shows promise for identifying symptomatic carotid plaques.
• This tool can support individualized cerebrovascular risk assessment.