Construction and validation of a diagnostic model for benign and malignant differentiation of TI-RADS category 4 thy-roid nodules based on sonographic features and clinical features

WANG Lang; XIONG Lingling; YANG Hanhui

doi:10.20258/j.cnki.1006-9011.2026.01.007

Abstract：Objective To construct a diagnostic model for benign and malignant differentiation of thyroid imaging reporting and data system(TI-RADS)category 4 thyroid nodules based on sonographic features combined with clinical features,and to verify its differential diagnostic value.Methods A total of 112 patients with TI-RADS category 4 thyroid nodules were selected as the training set,and divided into the malignant nodule group(n=66)and benign nodule group(n=46)based on pathological results.According to a 7∶3 ratio,another 48 patients with TI-RADS category 4 thyroid nodules were enrolled as the validation set,including 30 cases of malignant nodules and 18 cases of benign nodules.Clinical data,in terms of sex,age,body mass in-dex(BMI),smoking,drinking,clinical manifestations,thyroid stimulating hormone(TSH),thyroglobulin(TG),TSH/TG,thyroglobulin antibody(TgAb),thyroid peroxidase antibody(TPOAb),serum adiponectin(ADPN),and ultrasound features,such as,structure,number,echogenicity,echo texture,shape,margin,presence of calcification,aspect ratio,Adler blood flow grading,ultrasonic elastography score,enhancement intensity,enhancement pattern,enhancement uniformity,and pres-ence of peripheral rim enhancement of the lesion,were compared between the malignant and benign nodule groups in the training set.The LASSO algorithm was used to screen variables with non-zero coefficient features affecting the benign and malignant na-ture of TI-RADS category 4 thyroid nodules.Logistic regression model was applied to analyze the influencing factors of malig-nancy in TI-RADS category 4 thyroid nodules and to evaluate the diagnostic efficacy of the model.Three models,i.e.,clinical feature model,sonographic features model,and combined model,were established based on clinical data and ultrasound fea-tures.The predictive efficacy of the models was assessed using receiver operating characteristic(ROC)curves,calibration curves,and clinical decision curves.Results In the training set,statistically significant differences were observed between the malignant and benign nodule groups in TSH,TG,TSH/TG,TgAb,TPOAb,ADPN,lesion shape,Adler blood flow grading,elastography score,enhancement intensity,enhancement pattern,and peripheral annular enhancement(all P<0.05).The LASSO algorithm screened out 11 most relevant characteristic variables with non-zero coefficients,including 6 clinical feature variables,including TSH,TG,TSH/TG,TgAb,TPOAb,ADPN and 5 ultrasound feature variables,including lesion shape,Adler blood flow grading,elastography score,enhancement pattern,peripheral annular enhancement.Logistic regression analy-sis showed that all 11 variables were independent influencing factors for malignancy in TI-RADS category 4 thyroid nodules(all P<0.05).The clinical feature model,sonographic features model,and combined model were constructed using these variables,and significant differences were found in the model scores between the malignant and benign nodule groups(all P<0.05).ROC curve analysis revealed that the area under the curve(AUC)of the combined model was higher than that of the clinical feature model and sonographic features model in both the training and validation sets.Calibration curves demonstrated high predictive ac-curacy of the combined model.Clinical decision curve analysis indicated that the combined model was superior to the other two models in terms of safety,net benefit,and clinical practicality.Conclusion A diagnostic model for benign and malignant dif-ferentiation of TI-RADS category 4 thyroid nodules based on sonographic features and clinical features is successfully con-structed,which has high clinical differential diagnostic value.