CTA-derived radiomics features associated with prediction of ipsilateral stroke recurrence in retrospective carotid atherosclerosis cohort.

This retrospective cohort study evaluated the predictive value of CTA-derived radiomics features for ipsilateral stroke recurrence. The population consisted of 162 patients with unilateral carotid atherosclerosis, with a mean age of 69.28 ± 8.30 years and 83.95% male representation. Patients were followed for a median duration of 1 year to assess recurrence events over the study period.

The primary outcome was prediction of ipsilateral stroke recurrence. During the study period, 63 patients (38.9%) experienced recurrence. The combined machine learning model incorporating radiomics features and clinical factors achieved an AUC of 0.87 (95% CI: 0.74–0.97) for the primary outcome. This performance was higher than the radiomics-only model, which yielded an AUC of 0.80 (95% CI: 0.63–0.94), and the conventional clinical model based on stenosis degree, which yielded an AUC of 0.77 (95% CI: 0.58–0.91) for recurrence prediction. These metrics indicate the combined approach offered the highest predictive accuracy among the evaluated strategies. Statistical significance was not explicitly reported for these comparisons.

Safety data were not reported in this publication. The study design is retrospective, establishing only predictive associations rather than causality for the observed outcomes. Validation utilized repeated 10-fold cross-validation and an independent testing cohort to ensure robustness. P-values for model comparisons were truncated in the source text and unavailable for review. Clinicians should interpret these findings cautiously as observational data cannot confirm clinical utility without prospective validation in broader populations or settings before routine implementation.