Related Experiment Video For Thyroid cancer (TC)
Updated: Jun 17, 2026
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Diagnostic value of the nomogram model based on multimodal ultrasound features combined with PS-Tg for
Zhenhao Zheng1, Yang Yu1, Zelin Xu1
1Department of Ultrasound Medicine, The First Affiliated Hospital of Shihezi University, Shihezi, China.
Background:
Thyroid cancer (TC) is the most common malignant disease occurring in the neck region, exhibiting an increasing incidence trend year by year. Fine-needle aspiration (FNA), as an invasive diagnostic procedure, carries risks such as puncture failure and local hemorrhage. Therefore, there is a need to explore early, accurate, and noninvasive or minimally invasive diagnostic methods for TC. This study aimed to investigate the diagnostic value of a nomogram model based on multimodal ultrasound features combined with preoperative serum thyroglobulin (PS-Tg) for differentiating benign and malignant thyroid nodules.
Methods:
A total of 435 patients with 572 thyroid nodules undergoing FNA or thyroidectomy at The First Affiliated Hospital of Shihezi University between January 2023 and January 2025 were enrolled. All patients underwent preoperative thyroid function tests, two-dimensional ultrasound (2D-US), microvascular flow imaging (MV-Flow), strain elastography (SE), and elasticity contrast index (ECI) examinations. PS-Tg levels, 2D-US features, microvascular flow vascular index (VI), and elastography indices [strain ratio (SR), ECI] were collected. Based on final pathological results, nodules were divided into the benign group (n=248) and the malignant group (n=324). Multivariate logistic regression analysis was employed to explore the diagnostic value of a model combining multimodal ultrasound features and serological indicators for differentiating benign and malignant thyroid nodules.
Results:
Multivariate logistic regression analysis identified the following as independent risk factors for malignant thyroid nodules (P<0.05): hypoechoic [odds ratio (OR) =5.843, 95% confidence interval (CI): 3.391-10.075, P<0.001], anteroposterior/transverse (A/T) ratio ≥1 (OR 2.838, 95% CI: 1.821-4.424, P<0.001), irregular or lobulated margin (OR 2.234, 95% CI: 1.444-3.456, P<0.001), microcalcifications (OR 2.267, 95% CI: 1.458-3.525, P<0.001), PS-Tg (OR 1.005, 95% CI: 1.001-1.010, P=0.03), ECI-Max (OR 1.379, 95% CI: 1.150-1.653, P=0.001), SR-Max (OR 1.325, 95% CI: 1.036-1.696, P=0.03), and SR-Min (OR 1.652, 95% CI: 1.024-2.667, P=0.04). Among the evaluated models, the multimodal ultrasound combined with serological model demonstrated the best diagnostic performance, with a sensitivity (SEN) of 0.867, specificity (SPE) of 0.698, accuracy (ACC) of 0.794, and area under the curve (AUC) of 0.864. The calibration of the nomogram was good, showing no significant deviation (χ2=4.43, P=0.816).
Conclusions:
The nomogram based on the multivariate logistic regression model incorporating multimodal ultrasound features combined with PS-Tg demonstrates high diagnostic value in differentiating benign and malignant thyroid nodules.
