Prediction of Central Lymph Node Metastasis in Papillary Thyroid Carcinoma Using Multi-Ultrasound Radiomics of Intratumoral and Peritumoral Regions.
Authors
Affiliations (2)
Affiliations (2)
- Department of Ultrasound, The Second Affiliated Hospital of Harbin Medical University, Harbin, China.
- Department of General Surgery IV, The Second Affiliated Hospital of Harbin Medical University, Harbin, China.
Abstract
The aim of this study was to integrate multimodal intratumoral and peritumoral ultrasound radiomic features with clinical data to construct a predictive model for central lymph node metastasis (CLNM) in patients with papillary thyroid carcinoma (PTC) and to explore its clinical utility. A total of 470 PTC patients who underwent 2-dimensional (2D) ultrasound, color Doppler flow imaging (CDFI), and strain elastography (SE) were enrolled between June 2021 and September 2023. Patients were randomly divided into a training group (n = 329) and a validation group (n = 141) in a 7:3 ratio. Imaging features were extracted from multimodal intratumoral and peritumoral ultrasound images, followed by feature selection. A multimodal intratumoral and peritumoral ultrasound radiomics model was constructed using machine learning algorithms. Univariate and multivariate logistic regression analyses identified independent clinical predictors, which were used to build a clinical model. The ultrasound radiomics model was then integrated with the clinical model to form a combined model (the RadClip model). The multimodal intratumoral and peritumoral radiomics model achieved AUCs of 0.936 and 0.843 in the training and validation groups. Multivariate analysis identified male, high elasticity score, irregular margins, hyperechoic areas, and the presence of suspicious lymph nodes as independent predictors of CLNM (P < .05). The clinical model achieved AUCs of 0.750 and 0.735 in the training and validation groups, respectively. The RadClip model demonstrated significantly superior predictive performance compared to both the clinical and radiomics models, achieving AUCs of 0.948 and 0.867 in the training and validation groups, respectively (Delong, P < .05). The RadClip model demonstrates excellent predictive efficacy for preoperative detection of CLNM in PTC.