A Preliminary Study Of Ultrasonographic Omics For The Expression Of Molecular Markers In Papillary Thyroid Carcinoma

Objectives: The purpose of this study was to construct and validate an ultrasonics-based radiomics model to evaluate the expression of immunohistochemical molecular markers(CK19,Galectin-3,TPO,MC)in papillary thyroid carcinoma.Materials and Methods : In this study,the dataset of 158 patients with papillary thyroid carcinoma,confirmed with pathological and immunohistochemical results after surgical resection from July 2017 to November 2020,was retrospectively analyzed.All the included patients completed preoperative thyroid ultrasonography.According to the immunohistochemical molecular markers,all the patients were firstly divided into four groups,including cytokeratin 19(CK19)in 158 cases,156 patients with galectin-3,148 patients with thyroid peroxidase(TPO),and 122 patients with mesothelial cell(MC).Then,each immunohistochemical molecular markers group was divided into training group and test group at a ratio of 7:3.The clinical characteristic data of training dataset and test dataset were analyzed by chi-square test and Mann-Whitney U test.The 3D-Slicer was used to manually delineate the region of interest(ROI)on the maximum cross section ultrasonic image of thyroid nodule,and 837 radiomics features were extracted from the maximum cross section of the lesion in each patient.On the basis of the training group,intraclass correlation coefficient(ICC),pearson correlation coefficients,and KruskalWallis were filtered the features.Then three methods of Support Vector Machine(SVM),Native Bayes(NB)and Logistic Regression(LR)were constructed to predict immunohistochemical markers in papillary thyroid carcinoma.The optimal model was determined based on the accuracy,specificity and sensitivity of the model on the test set,and the receiver operating curve(ROC)was evaluated the diagnostic performance of the model.RESULT:Based on optimal model performance,4,6,19 and 11 texture features were extracted from CK19,Galectin-3,TPO and MC groups,respectively.In the end,the more robust model in Galectin-3 group 、 TPO group and MC group was LR model,the AUC of the test group was 0.751、0.781、0.841,the accuracy was 0.739、0.773、0.838,the sensitivity was 0.759、0.800、0.950,and the specificity was 0.706、0.750、0.706.While,the optimal model of CK19 group is SVM model.The AUC of the test dataset was 0.750,the accuracy was 0.708,the sensitivity and specificity were 0.625,0.875.CONCLUSION: In this study,the best model based on ultrasonic omics for predicting the immunohistochemical parameters of thyroid nodules was obtained among the three models.Thus,it is suggested that the ultrasound based radiomics model might be non-invasively predicting the cytokeratin 19,Galectin-3,thyroid peroxidase and MC in thyroid papillary carcinoma.The model based on ultrasonic omics has the potential to assist in the differential diagnosis of PTC and thyroid benign nodules,and to achieve more accurate personalized diagnosis and treatment.