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

In 30 seconds

This systematic review assessed radiomics-based models for identifying high-risk carotid plaques. It included 28 studies involving 6,108 patients, 72.4% of whom were male. Computed tomography angiography was the most common imaging modality, and manual segmentation was used in 27 studies. Internal-validation AUCs reached 0.984; combined models incorporating radiomics with clinical and/or conventional imaging features often achieved AUCs above 0.90. However, only four studies used external validation cohorts, limiting confidence in how well reported performance generalizes.

Key findings

• Across 28 studies and 6,108 patients, internal-validation AUCs for radiomics models reached 0.984.
• Combined models with radiomics and clinical and/or conventional imaging features often achieved AUCs above 0.90.
• Only four studies included external validation cohorts.
• Manual region-of-interest segmentation was used in 27 of 28 studies.

Why it matters

Plaque vulnerability may not be adequately represented by luminal stenosis alone. Radiomics-based assessment could provide non-invasive information to help identify high-risk carotid plaques, but the strong reported discrimination should be interpreted alongside limited external validation and widespread reliance on manual segmentation. Reproducibility and validation are important before these models can inform routine stroke-prevention decisions.

What to keep in mind

Only four studies included external validation cohorts, and manual segmentation was used in 27 of 28 studies, raising concerns about generalizability and reproducibility.