Development of a Novel Contrast-Enhanced Ultrasound-Based Nomogram for Superficial Lymphadenopathy Differentiation:...

The aim of this study was to develop and prospectively validate a prediction model for superficial lymphadenopathy differentiation using Sonazoid contrast-enhanced ultrasound (CEUS) combined with ultrasound (US) and clinical data.

The training cohort comprised 260 retrospectively enrolled patients with 260 pathological lymph nodes imaged between January and December 2020. Two clinical US-CEUS models were created using multivariable logistic regression analysis and compared using receiver operating characteristic curve analysis: Model 1 included clinical and US characteristics; Model 2 included all confirmed predictors, including CEUS characteristics. Feature contributions were evaluated using the SHapley Additive exPlanations (SHAP) algorithm. Data from 172 patients were prospectively collected between January and May 2021 for model validation.

Age, tumor history, long-axis diameter of lymph node, blood flow distribution, echogenic hilus, and the mean postvascular phase intensity (MPI) were identified as independent predictors for malignant lymphadenopathy. The area under the curve (AUC), sensitivity, specificity, and accuracy of MPI alone was 0.858 (95% confidence interval [CI], 0.817-0.891), 86.47%, 74.55%, and 81.2%, respectively. Model 2 had an AUC of 0.919 (95% CI, 0.879-0.949) and good calibration in training and validation cohorts. The incorporation of MPI significantly enhanced diagnostic capability (p < 0.0001 and p = 0.002 for training and validation cohorts, respectively). Decision curve analysis indicated Model 2 as the superior diagnostic tool. SHAP analysis highlighted MPI as the most pivotal feature in the diagnostic process.

The employment of our straightforward prediction model has the potential to enhance clinical decision-making and mitigate the need for unwarranted biopsies.