Objectives To evaluate the value of Ultrasound-based machine learning and Ultrasound-based machine learning combined with clinical featuers in predicting lymph node metastasis in occult papillary thyroid carcinoma.Methods Retrospectively analyzed the clinical data and ultrasound images of 81 occult papillary thyroid carcinoma patients with lymph node metastasis and 103 patients without lymph node metastasis,and divided randomly into the training and validation groups at the ratio of 7:3. Extracted the features of ultrasound images and used Support Vector Machine SVM to build models.Furthermore, built combined model based on ultrasound featuers and clinical features.Then observe the efficacy of each model.Results 11 ultrasound features and 2 clinical featuers were ultimately selected to build machine learning models and combined models.The AUC、accuracy、sensitivity、specificity、positive predictive value and negative predictive value in the training group of machine learning model was 0.78[95%CI(0.74,0.82)]、79%、76%、80%、78%、81%,while in the validation group was 0.72[95%CI(0.68,0.75)]、71%、69%、69%、74%、74%.The AUC、accuracy、sensitivity、specificity、positive predictive value and negative predictive value in the training group of combined model was 0.87[95%CI(0.83,0.90)]、87%、85%、87%、87%、88%,while in the validation group was 0.81[95%CI(0.78,0.83)]、80%、83%、86%、73%、79%.Conclusions Ultrasound-based machine learning was effective at predicting the lymph node metastasis of occult papillary thyroid carcinoma patients,and the combined model achieved a much higher diagnostic accuracy.