Radiomics performance to identify high-risk carotid artery plaques: a systematic review

Abstract

Stroke is a leading cause of mortality worldwide, with 15–20% of ischemic strokes caused by extracranial carotid atherosclerotic disease. Risk stratification has traditionally relied on luminal stenosis severity, which inadequately reflects plaque vulnerability. Radiomics offers an emerging solution by extracting high-dimensional quantitative data from standard medical images. This systematic review evaluated the performance of radiomics-based models for identifying high-risk carotid plaques. A systematic search was conducted across multiple databases. Studies were included if they developed or validated radiomics-based models for carotid plaque assessment and reported performance metrics, specifically the area under the curve (AUC). Data were extracted on study design, imaging modalities, segmentation methods, and modeling techniques. Twenty-eight studies involving 6,108 patients (72.4% male) were included. Computed tomography angiography was the most commonly used imaging modality (n = 16), and manual region-of-interest segmentation was performed in 27 of 28 studies. Radiomics models showed strong predictive performance. Internal validation AUCs were as high as 0.984, whereas combined models incorporating radiomics with clinical and/or conventional radiological features achieved the best predictive accuracy, often exceeding 0.90. However, only four studies included external validation cohorts. Radiomics is a powerful, non-invasive tool for identifying high-risk carotid plaques, offering superior discriminative power over conventional stenosis-based assessments. Despite these promising results, the field faces significant hurdles, and future research must prioritize prospective multicenter trials, the development of automated segmentation workflows, and the correlation of radiomic features with histopathological findings to ensure clinical reproducibility and transition into routine stroke prevention. Question Standard carotid atherosclerotic disease assessments fail to capture biological plaque complexity and vulnerability, necessitating objective, reproducible, and non-invasive methods to improve stroke prevention. Findings Radiomics and combined models demonstrate superior capability in identifying high-risk carotid plaques compared to traditional assessments. With four external validations, the field remains exploratory. Relevance statement Radiomics offers a non-invasive tool for personalized stroke prevention. However, technical barriers currently prevent its transition to routine clinical use, including reliance on manual segmentation and low compliance with international standardization guidelines.