Relation of Carotid Plaque Features Detected with Ultrasonography-Based Radiomics to Clinical Symptoms

Carotid plaque is one of the predominant causes of stroke. We sought to build a nomogram using ultrasonography (US)-based radiomics and clinical features for identification of symptomatic carotid plaques. We prospectively enrolled 548 patients (mean age +/- standard deviation, 63 +/- 10 years; 373 men) were randomly divided into training and test cohorts. Clinical and conventional US features of carotid plaques were used to generate a clinical and conventional US model. US-based radiomics model was constructed by extracting radiomics features from grayscale and strain elasticity images. Multivariate logistic regression was performed using the radiomics scores together with clinical and conventional US data, and a final nomogram was subsequently developed. The performance of the final nomogram was assessed with respect to discrimination and clinical usefulness in the training of the test cohorts and contrast-enhanced US test cohort. All the radiomics scores were significantly higher in patients with symptomatic carotid plaques. The US-based radiomics model [area under the curve (AUC) = 0.930 and 0.922 for training and test cohorts, respectively] and final nomogram (AUC = 0.927 and 0.919, respectively) outperformed the clinical and conventional US model (AUC = 0.723 and 0.580, respectively). The decision curve analysis indicated that the final nomogram was clinically useful. In patients undergoing the contrast-enhanced US, the prevalence of plaque enhancement was higher in high-risk patients than in low-risk patients based on the final nomogram-score (P = 0.008). Nomogram has a high diagnostic performance for identification of symptomatic carotid plaques.

Automated summary

By utilizing ultrasonography and clinical features, a nomogram was developed to identify symptomatic carotid plaques, demonstrating higher diagnostic performance compared to traditional clinical and conventional ultrasound models through analysis and evaluation of data.