This is a retrospective study, which consecutively reviewed patients diagnosed with carotid atherosclerosis between November 2021 and April 2023. The patients were divided into primary and validation cohorts in a chronological order. The patients between October to November 2021 and December 2022 made up the primary cohort, and those between January 2023 and April 2023 were in the validation cohort. All patients with low and intermediate carotid artery stenosis who received magnetic resonance angiography (MRA) or computed tomography angiography (CTA) examinations and clinical follow-up using B-mode US or CEUS were included in this study. The study was approved by the medical ethics committee of Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine [No. (2022) 028].

The inclusion criteria included (1) ≥18 years of age, (2) low and intermediate carotid stenosis, and (3) no clinical contraindications for CEUS. The exclusion criteria included (1) severe stroke, (2) without MRA or CTA above the aortic arch, (3) carotid artery dissection, (4) Takayasu's arteritis, and (5) poor image quality of B-mode US and CEUS. We selected plaques with a thickness of ≥2 mm and that were mainly hypoechoic, as hyperechoic plaques with calcification cannot be clearly displayed by CEUS for sound shadow occlusion (14) (Figure 1).

Patients were divided into two groups based on the presence or absence of symptoms: asymptomatic group and symptomatic group. Patients with a history of IS within the previous 2 months belonged to the symptomatic group and had confirmed ischemic lesions in the ipsilateral side's carotid region (anterior circulation) by head CT/MRI. (15). On the other hand, the asymptomatic group consisted of patients without any prior history of stroke to the carotid artery.

All B-mode US and CEUS examinations were performed by vascular sonographers with at least 5 years of experience in B-mode US and at least 2 years of experience in CEUS of carotid plaques. The CEUS examinations were performed on Philips EPIQ Elite (Philips Healthcare, Amsterdam, the Netherlands) with a high-frequency probe (eL18-4, MHz). We adjusted the frame rate to 12 frames per second and the mechanical index to 0.13 in order to minimize the destruction of microbubbles. The focus location was fixed at the carotid artery's level, and the image depth was adjusted to 3–5 cm based on the carotid artery's size. The B-mode US also used LOGIQ-E9 (GE, Fairfield, USA), Aplio 500 (Toshiba, Japan), and Aplio i900 (Canon, Japan). The vascular sonographers were blinded to the neurologist's findings. All study participants were examined by B-mode US, and the C-IMT was measured. Thereafter, the maximum longitudinal section was considered for CEUS imaging. CEUS was performed after injection of a 1.0-ml bolus of SonoVue solution (Bracco, Milan, Italy) mixed with 5 ml of saline through a peripheral vein. A timer was started to continuously observe the enhancement of CP in real time and store the dynamic examination image. All videos (at least 2-min duration) and images were digitally stored on magnetic optical discs for offline analysis.

The selection of carotid artery plaques was based on the Mannheim consensus (16). According to the perspectives and guidelines of the ASNR Vessel Wall Imaging Study Group and Expert Consensus Recommendations of the American Society of Neuroradiology (17), we abstracted the US features, plaque morphology (irregular, regular) (Figure 2), and echogenicity (hyperechoic, isoechoic, hypoechoic, extremely hypoechoic) (Figure 3). IPN was identified by rapid movement of the echogenic reflectors of microbubbles within the plaque as follows: Grade 0 (no visible microbubbles within the plaque), Grade 1 (minimal microbubbles confined to the shoulder or adventitial side of the plaque), and Grade 2 (microbubbles present throughout the plaque core) (Figure 4). Perfusion pattern was divided into basal entry and surface entry (18) (Figure 5).

Interobserver consistency in IPN grading was analyzed by two independent radiologists (FL and L-NZ) who were blinded to each other's interpretations. To evaluate intra-observer consistency, the data were reanalyzed by the same radiologist (FL) after an interval of 1 month without reference to the initial results.

The specific rules of the ESRS include eight scoring risk factors, with a maximum score of 9 points (19) (Table 1).

A statistical analysis was performed using SPSS 22.0 and the rms packages in R 4.1.0 (The R Foundation for Statistical Computing, Vienna, Austria). Continuous variable data were represented as mean ± standard deviation, and normality testing was performed by using Student's t-test. The chi square test was used to assess the relationship between categorical variables. A binary logistic regression model was used to evaluate the relationship between cerebrovascular events and characteristic plaques and the influence of various patch characteristics on the results. The ESRS and C-IMT were analyzed by using the receiver operating characteristic (ROC) curve, and the best cutoff value was determined by the maximum Youden index (defined as sensitivity + specificity − 1). A p-value of <0.05 was considered statistically significant.

The prediction nomogram based on the multivariate logistic regression analysis results was developed using R software. The ROC analysis was used to evaluate the discrimination performance of the model. Calibration curves were plotted via bootstrapping using 1,000 resamples. The decision curve analysis (DCA) was performed to determine the clinical usefulness of the model by quantifying the net benefit at the different thresholds.

A total of 824 patients with carotid artery plaques were recruited in the primary cohort. Sixteen patients with severe stroke, 35 patients who did not undergo MRA or CTA above the aortic arch, two patients with carotid artery dissection, four patients with Takayasu's arteritis, 18 patients with poor US imaging quality, and 44 patients with incomplete clinical data were excluded. Finally, a total of 705 patients were included in the study, of which 394 patients belonged in the symptomatic group and 311 patients belonged in the asymptomatic group, as presented in Figure 1. The basic characteristics of the included patients are listed in Table 2.

The mean C-IMT was the mean thickness of bilateral common carotid arteries, carotid bifurcations, and the beginning of internal carotid arteries. The mean C-IMT of symptomatic and asymptomatic groups was 2.89 ± 1.14 mm and 2.55 ± 1.06 mm, respectively. The difference between the groups was statistically significant (p < 0.001). The optimal coefficient of variation (COV) of C-IMT was 2.25 and 4.55 mm, and the ROC curve analysis showed that the AUC of C-IMT was 0.621 (95% CI: 0.578–0.663).

The ESRS of the symptomatic group and the asymptomatic group were 4.46 ± 0.99 and 2.81 ± 1.09, respectively, with a statistically significant difference of p < 0.001. The best COV of the ESRS was 3.5, and the ROC curve analysis showed that the AUC of the ESRS was 0.634 (95% CI: 0.579–0.662).

We used univariable logistic regression analysis to determine the US features for multivariable logistic regression analysis at a significance level of 0.05. By rounding the mean of regression coefficients of the multivariable logistic analysis to obtain scores for CP risk stratification, we established the CP score according to the parameters from B-mode US and CEUS. Thereafter, we combined the CP score with the ESRS to establish a IS risk model. These variables were used to construct the predictive nomogram (Figure 6).

Based on the mean of regression coefficients of the multivariable logistic analysis of the data in Table 2, the B-mode US features such as hypoechoic, extremely hypoechoic, and irregular shape were assigned 1 point, a C-IMT between 2.23 and 4.55 mm was assigned 1 point, and a C-IMT of greater than 4.55 mm was assigned 2 points. In CEUS, Grade 1 was assigned 1 point, Grade 2 was assigned 2 points, and the perfusion pattern from surface perfusion was assigned 1 point. The ESRS between 3.5 and 8 was assigned 2 points (Table 3).

The sum of the points determined the ischemic stroke risk stratification (ISRS): 0–2 points corresponded to CERS 1 (low risk), 3–6 points indicated CERS 2 (intermediate risk), and 7–9 points indicated CERS 3 (high risk) (Figure 7).

The areas under the ROC curve of the model, CEUS, B-mode US, and ESRS in IS were 0.91, 0.82, 0.74, and 0.73, respectively, with statistically significant difference of p < 0.001 (Figure 8A). The sensitivity and specificity of the model were 72.6% and 89.3%, respectively. The Hosmer–Lemeshow test illustrated good calibration (p = 0.903). The calibration curve of the ischemic stroke risk model demonstrated good agreement between prediction and observation in the primary cohort (Figure 9A).

Internal validation: Using the 68 patients diagnosed with carotid atherosclerosis between January 2023 and April 2023 as the internal validation dataset. The areas under the ROC curve of the model, CEUS, B-mode US, and ESRS in IS were 0.84, 0.72, 0.67, and 0.62, respectively, with statistically significant difference of p < 0.001 (Figure 8B). The Hosmer–Lemeshow test yielded a non-significant statistics of p = 0.886. Good calibration via bootstrapping using 1,000 resamples was observed for the probability of the ischemic stroke risk in the validation cohort (Figure 9B).

The decision curve analysis for the IS risk model is presented in Figure 10. The decision curve showed that if the threshold probability of a patient or doctor is 8%, using the model to predict ischemic stroke adds more benefit than either the treat-all-patients scheme or the treat-none scheme.

The IPN grading and perfusion patterns in CEUS showed substantial inter-reader agreement (k coefficients = 0.744 and 0.721, respectively), and echogenicity, plaque morphology, plaque location, calcification, and C-IMT in B-mode US showed almost-perfect inter-reader agreement (k coefficients = 0.911, 0.901, 0.891, 0.903, and 0.881, respectively).