Dual-modal ultrasound radiomics model enhances identification of symptomatic carotid plaque: a multicenter retrospective study.

We aimed to identify the most valuable variables in multi-modal ultrasound technologies and compare the performance of radiomics models in detecting symptomatic carotid plaques. From August 2013 to April 2021, we retrospectively enrolled 112 carotid plaque patients and randomly divided them into training and validation cohorts at a 3:1 ratio. Additionally, 42 patients from two other centers were enrolled as an independent test cohort. A traditional model (T model) was built using independently significant variables from multivariate analysis of clinical and multi-modal ultrasound characteristics. Radiomics models based on B-mode ultrasound, contrast-enhanced ultrasound (CEUS), and a combined approach (USR, CEUSR, and US-CEUSR models) were then developed. Model performance was assessed using area under the receiver operating characteristic curve (AUC) and decision curve analysis. Only intraplaque contrast enhancement from CEUS modality was independently associated with clinical symptoms [odds ratio = 3.328; 95% confidence interval (CI), 1.230-9.003; p = 0.018] and was used to construct the T model, which achieved an AUC of 0.658 (95% CI, 0.502-0.815) in the external test cohort. The radiomics model showed good and repeatable diagnostic performance for identifying symptomatic plaques, particularly the US-CEUSR model, which had an AUC of 0.775 (95% CI, 0.607-0.915) in external test cohort. This model also provided high clinical benefits in detecting symptomatic plaques. Intraplaque contrast enhancement, a qualitative variable by CEUS, is an independent risk factor for symptomatic carotid plaques, and radiomics features from B-mode ultrasound and CEUS can effectively identify symptomatic carotid plaques.