Establishment And Validation Of CT-Based Thyroid Nodule Radiomics Nomograms For The Prediction Of Benign And Malignant Lesions

Objective: To determine the benign and malignant thyroid nodules in clinical decision-making and management,this study aims to establish an imaging model based on computed tomography to predict benign and malignant thyroid nodules before surgery.Methods: The patients with pathologically confirmed thyroid nodules from the First Affiliated Hospital of Dali University(n=181),Yunnan Cancer Hospital(n=148)and Dali Prefecture People’s Hospital(n=41)from January 2015 to November 2021 were retrospectively collected,and the patients in the first two hospitals were divided into training set(n=152)and internal validation set(n=77)according to 7:3,and the patients in the last hospital were used as the external validation set.The radiomics features of each patient were extracted from the preoperative CT scan images,the variance threshold method,univariate selection method and the minimum absolute contraction and selection operator algorithm were used to screen the features with strong correlation with benign and malignant nodules,and the radiomics score was generated by combining the product of the selected features and their weighted coefficients,and univariate and multivariate logistic regression were used to select clinical risk factors.Logistic regression was used to establish models based on clinical model(model 1),radiomics model(model 2)and joint model(model 3),and a nomogram was drawn for the joint model.The performance of the model was evaluated by using the receiver operating characteristic curve(ROC)and calibration curve,and the clinical utility of the combined model was evaluated by decision curve analysis(DCA).Results: A total of 1409 radiomics features were extracted from the CT images of each patient,and 17 non-zero coefficient features associated with benign and malignant nodules were screened.Seven clinical variables(age,maximal diameter,margin,density,cystic change,aspect ratio,and blood FT3)were significantly associated with benign and malignant outcomes.The combined model obtained the best predictive performance,with AUCs of 0.965(95% [CI]: 0.944-0.982),0.938(95% [CI]: 0.893-0.972)and 0.907(95% [CI]: 0.816-0.976)in the training,internal and external validation sets,respectively.The AUCs of the radiomics models alone were 0.939(95% [CI]: 0.912-0.962),0.915(95% [CI]: 0.865-0.954)and 0.789(95%[CI]: 0.654-0.905)in the training,internal and external validation sets,respectively.The AUCs of the clinical model in the training,internal and external validation sets were 0.903(95% [CI]: 0.864-0.935),0.927(95% [CI]: 0.882-0.965),and 0.909(95%[CI]: 0.821-0.971),respectively.The calibration curve reflects the good agreement between the actual probability and estimated probability of the joint model,and the DCA curve shows that the joint model has good clinical application value.Conclusion: The radiomics model is a non-invasive preoperative tool with high diagnostic efficacy for benign and malignant thyroid nodules.