Abstract
Abstract
Background
Carotid plaque vulnerability is a significant risk factor for ischemic stroke. Traditional imaging methods are limited by high subjectivity, whereas radiomics and deep learning (DL) facilitate automated quantitative assessment.
Objective
This study aims to systematically assess the diagnostic effectiveness of radiomics and deep learning technologies in evaluating plaque vulnerability and predicting stroke risk, as well as to explore their potential for clinical translation.
Methods
A systematic search was conducted until June 12, 2026. Twenty-eight studies were included. Methodological quality was appraised using the QUADAS-AI and the Radiomics Quality Score (RQS 2.0).
Results
The pooled area under the curve (AUC) for plaque assessment was 0.84 (95% CI: 0.80–0.88) for radiomics and 0.92 (95% CI: 0.89–0.94) for DL. For stroke risk prediction, the pooled AUC was 0.84 (95% CI: 0.74–0.95). MRI-based radiomics demonstrated superior diagnostic consistency (I² =0.00%), while ultrasound (US) showed the highest numerical efficacy (AUC: 0.87) among radiomics subgroups.
Conclusion
Both radiomics and DL exhibit good diagnostic efficacy in research settings, but notable challenges in standardization and model interpretability limit their clinical translation.