Machine learning detects symptomatic patients with carotid plaques based on 6-type calcium configuration...

Area of Science:

• Cardiovascular Imaging
• Machine Learning in Medicine
• Cerebrovascular Disease Research

Background:

• Carotid plaque composition is linked to cerebrovascular events, but the role of calcium configuration is not well understood.
• Accurate identification of symptomatic patients with carotid plaques is crucial for preventing adverse events.

Purpose of the Study:

• To develop and validate a machine learning (ML) model using CT angiography (CTA)-based carotid plaque calcium grading and clinical data.
• To identify patients with bilateral carotid plaques who are symptomatic for cerebrovascular events.

Main Methods:

• A multicenter, retrospective diagnostic study included 790 patients for training/internal validation and 159 for external testing.
• Four ML models and logistic regression were used to identify symptomatic patients based on clinical factors and 6-type plaque calcium grading.
• Internal and external validation assessed model discrimination and calibration.

Main Results:

• The ML model incorporating plaque grading (ML-All-G) achieved an area under the ROC curve of 0.71 with 80% sensitivity.
• Model performance was consistent in external validation.
• Calcified plaque, particularly the positive rim sign, along with advanced age and hyperlipidemia, significantly impacted the identification of symptomatic patients.

Conclusions:

• A CTA-based ML model utilizing carotid plaque calcium configuration and clinical data can identify symptomatic patients with reasonable diagnostic accuracy.
• The 6-type calcium grading combined with clinical variables effectively identifies symptomatic patients.
• This ML approach, leveraging fast CTA acquisition, can streamline the diagnosis of cerebrovascular symptoms.