CT radiomics model for differentiating malignant and benign thyroid nodules

Abstract

Objective To investigate the value of CT radiomics mode in differential diagnosis of benign and malignant thyroid nodules. Methods The clinical and imaging data of 179 patients with thyroid nodules confirmed by pathology from May 2017 to August 2018 were retrospectively analyzed in the Affiliated Huaian First People′s Hospital of Nanjing Medical University. Among the patients, 89 cases were benign nodules and 90 cases were malignant nodules. All patients underwent unenhanced and enhanced CT scan before operation. The stratified random sampling method was used to divide patients into a training group (143 cases) and a testing group (36 cases) according to a ratio of 8∶2. The A.K software was used to extract 378 imaging omics features based on preoperative CT images, and then Spearman correlation analysis and least absolute shrinkage and selection operator regression analysis were used for feature selection and model construction. The receiver operating characteristic (ROC) curve was used to verify the model in the training group and the testing group, and the efficacy of imaging omics features to predict benign and malignant thyroid nodules was evaluated. Results After feature screening, 16 radiomics features were used to construct an identification model between benign and malignant thyroid nodules. In the training group, the area under the ROC curve (AUC) was 0.92 [95% confidence interval (CI): 0.88-0.97], the sensitivity and specificity were 88.7%, 82.0%, and the diagnostic accuracy of the model was 91.1%. In the testing group, AUC was 0.90 (95%CI: 0.81-0.98), sensitivity and specificity were 88.5%, 84.6%, and the diagnostic accuracy of the model was 88.2%. Conclusion The CT radiomics mode has a good diagnostic performance in the identification of benign and malignant thyroid nodules. Key words: Tomography, X-Ray computed; Thyroid nodule; Diagnosis, differential; Radiomics