Abstract: Background and purpose: The ultrasonic grading method of thyroid nodules has been partially reported. However, most of the reports are focused on assigning values directly to each characteristic, and few studies are focused on weight scoring. In this study, the independent risk factors of thyroid cancer in ultrasonic signs were screened, and the ultrasound characteristic diagnostic model was established by using the weighted scoring method based on multivariate logistic regression β value of ultrasonic signs. Then we evaluated its application value and verified its work efficiency. Methods: A total of 1 749 patients with 1 988 thyroid nodules underwent ultrasound diagnosis followed by pathological examination in Affiliated Hospital of North China University of Technology from Jan. 2015 to Aug. 2018. The conventional sonographic features and pathological results of these nodules were analyzed. Ultrasonic characteristics included composition, echogenicity, shape, margin, extension beyond the thyroid border, echogenic foci and S/L ratio≥1. Then we screened the ultrasound features with significant difference identified by the univariate analysis. The multivariate logistic regression analysis was performed to screen independent risk factors for thyroid cancer. The partial regression coefficient beta of each risk sign was used as the weighted score, and the total score of nodules was used to establish the risk prediction model of thyroid cancer. Finally, the conventional ultrasound characteristic diagnostic model of thyroid nodules based on the composite score was established, and we drew the receiver operating characteristic (ROC) curve to evaluate the model. A total of 150 thyroid nodules confirmed by pathological results in Affiliated Hospital of North China University of Technology from Sep. 2018 to Dec. 2018 were selected as the validation data. Results: The ultrasound characteristic diagnostic model of thyroid nodules based on the composite score method had high diagnostic accuracy of 95.3% (95% CI: 0.942-0.964). The best cut-off for the diagnosis of malignant lesions was 24.2 with a sensitivity of 88.6%, a specificity of 93.3%, a positive predictive value of 86.8% and a negative predictive value of 94.4%, and the accuracy of the verification study was 83.3%. Conclusion: The ultrasound characteristic diagnostic model of thyroid nodules based on ultrasound characteristic multivariate logistic regression coefficient β composite score method has relatively high efficiency in differentiating benign from malignant thyroid nodules.

Key words: Ultrasonography, Thyroid cancer, Differential diagnosis, Logistic regression, Composite score