Predicting central lymph node metastasis in papillary thyroid microcarcinoma: a study of ultrasound and clinical features.

Background: Papillary thyroid microcarcinoma (PTMC) generally has a favorable prognosis. Early central lymph node metastasis (CLNM) can significantly impact treatment strategy and prognosis. However, CLNM lacks typical ultrasound features. Accurate preoperative prediction of CLNM remains challenging. This study aims to develop and validate a high-accuracy tool for preoperatively assessing the risk of lymph node metastasis in PTMC patients. Methods: We retrospectively analyzed clinical and ultrasound data from 534 PTMC patients who underwent initial thyroidectomy with central lymph node dissection. Patients were randomly divided into training (n=373) and validation (n=161) cohorts. We calculated high-throughput radiomics features, including tumor size, tumor shape, margin, capsular contact, microcalcifications, and peritumoral echogenicity features. A combined feature selection strategy was then used to identify features with the greatest discriminatory power for lymph node status. A Logistic Regression machine classifier was employed to build and validate the prediction model. Additionally, ultrasound ACR TI-RADS and clinical variables were evaluated. Univariate and multivariate logistic regression was used to identify independent predictors, which were further incorporated into a nomogram model. The area under the operating characteristic curves (AUCs) was used to draw comparisons between different models and the decision curve analysis was conducted to assess their clinical utility. Results: In the clinical model based solely on clinical and conventional ultrasound features, multivariate analysis identified five independent predictors of CLNM: age <46.5 years, male sex, capsular contact ≥50%, peritumoral hyperechogenicity and heterogeneous echotexture (AUC: 0.857 in the training set and 0.840 in the validation set). By further integrating a radiomics score with all univariately significant clinical variables, a combined clinical-radiomics nomogram was developed. In this combined model, age, transverse diameter of tumor, capsular contact, peritumoral echo changes, and the radiomics score were identified as independent predictors. The combined model achieved an improved AUC of 0.900 in the validation set, demonstrating superior predictive performance and higher clinical net benefit than the clinical model alone. Conclusion: The proposed clinical-radiomics nomogram, which incorporates conventional ultrasound features and radiomics signatures, outperforms the standalone clinical model in predicting CLNM. This non-invasive approach provides superior pre-operative risk assessment in optimizing treatment strategies for PTMC patients.