This study compared radiomics models constructed from carotid artery CTA subtraction images. Two ROI delineation strategies were employed: the global vascular wall and the semi-global vascular wall, which focuses on the plaque enrichment side at the vascular plaque site. The objective was to differentiate between symptomatic and asymptomatic internal carotid artery plaques. Additionally, the study aimed to investigate the differences between the two strategies to assess their potential clinical applications and pathways for transformation. This retrospective study included 129 patients with carotid plaques, comprising 65 symptomatic and 64 asymptomatic cases. CTA and non-contrast images were coregistered, subtracted, and subsequently fused to generate the final CTA subtraction images. Using the carotid slice with the largest plaque as the reference, the full-circumference vascular wall, the half-circumference vascular wall (plaque-enriched side), and adjacent layers were delineated to generate two ROIs, which were then combined into a three-dimensional volume ROI. Radiomics features were extracted from each ROI, and feature selection was performed using LASSO regression. SVM models were constructed based on the selected features. Model performance in distinguishing symptomatic versus asymptomatic plaques was evaluated using AUC and accuracy. A total of 1,834 features were extracted from each ROI method. Eight features from the full-circumference ROI and 33 from the semi-circumference ROI were selected for modeling. The SVM models demonstrated strong discriminatory performance. In the training set, AUCs reached 0.871 (95% CI: 0.80–0.93) for the full-circumference ROI and 0.979 (95% CI: 0.95–0.99) for the semi-circumference ROI. Corresponding test-set AUCs were 0.893 (95% CI: 0.77–1.00) and 0.899 (95% CI: 0.78–1.00), respectively. Model accuracies were 0.806 and 0.913 in the training set and 0.846 for both ROI models in the test set. Radiomics models utilizing half-week and full-week vascular wall ROIs demonstrate strong discriminatory performance in distinguishing between symptomatic and asymptomatic carotid plaques, with overall performance being comparable. The delineation of full-week vascular wall ROIs is relatively straightforward, suggesting greater potential for clinical application.
Deng et al. (Tue,) studied this question.