Abstract: Objective Accuracy of predicting pN staging of papillary thyroid carcinoma (PTC) based on ultrasonography radiomics and deep neural network (DNN). Methods 375 patients with PTC (261 in the training set and 114 in the test set) were collected. No cervical lymph node metastasis was defined as pN0, central lymph node metastasis as pN1a, and lateral cervical lymph node metastasis as pN1b. The Region of Interest (ROI) of PTC was manually sketched by the ultrasound physician and 1899 radiomics features were extracted. The least absolute shrinkage and selection operator (LASSO) will be used for dimensionality reduction of the above radiomics features. A deep neural network (DNN) model was constructed to predict the pN staging of PTC in the training set by the H2O deep learning platform, and the accuracy of the optimal model was verified in the test set. Results There were 153 cases in pN0, 131 in pN1a and 91 in pN1b. Fifteen radiomics features were selected from each PTC by LASSO regression. The accuracy of the optimal DNN model based on the 15 radiomics features in the training set, and test set was 85.82% and 81.57%, respectively. Conclusion PTC ultrasound radiomics has a high accuracy in predicting pN staging, and has the potential to automate N staging for patients.