Multimodal ultrasound radiomics of carotid plaques is associated with ischemic heart disease: additional discriminative information from vector flow imaging and shear wave elastography

Ischemic heart disease (IHD) is associated with coronary atherosclerosis, but related atherosclerotic phenotypes may also be reflected in carotid plaques. Ultrasound radiomics, vector flow imaging (V-Flow), and shear wave elastography (SWE) characterize plaque texture, local hemodynamics, and tissue mechanics. We assessed associations between multimodal carotid plaque features and IHD and the additional discrimination provided by V-Flow/SWE beyond radiomics. We retrospectively analyzed 210 patients from a consecutively screened cohort of 215 potentially eligible patients with carotid plaques who underwent clinically indicated coronary angiography (CAG) and complete multimodal carotid ultrasound during the same hospitalization. IHD status was the routine cardiology diagnosis integrating CAG and ischemia-related manifestations; significant stenosis was defined as at least 50% luminal narrowing in one major coronary artery. Radiomics, V-Flow, and SWE measurements were obtained from the same index plaque. Four fixed-predictor logistic models were evaluated: clinical, clinical plus V-Flow/SWE, clinical plus radiomics, and integrated. Discrimination was assessed from out-of-fold predictions generated by stratified 10-fold outer/10-fold inner cross-validation. Fully selection-aware and expanded-clinical analyses were conducted as sensitivity analyses. Among the 210 patients, 115 had IHD and 95 did not. Rad-score was associated with IHD (OR = 10.85, 95% CI: 5.24–22.46, P < 0.001) and correlated with the coronary angiographic stenosis severity score (Spearman r = 0.731, P < 0.001). Upstream maximum wall shear stress, downstream turbulence index, and whole-plaque Young’s modulus were also associated with IHD. The original fixed-predictor integrated model yielded an AUC of 0.905 (95% CI: 0.862–0.942) and modestly outperformed the clinical plus radiomics model ( P = 0.003). In a separate fully selection-aware sensitivity analysis, the corresponding integrated model yielded an AUC of 0.896 (95% CI: 0.851–0.936). Across 20 repetitions of the selection-aware procedure, the mean AUC was 0.8823 ± 0.0140, and the integrated model ranked highest in all 20 repetitions. Multimodal carotid plaque features were associated with IHD and showed internal discriminative information in this selected CAG-referred cohort. Radiomics contributed most, while V-Flow/SWE provided a modest additional increment. External multicenter validation and recalibration are required before clinical application.