The Application of Data Imputation and Deep Learning Network in the Papillary Thyroid Carcinoma Classification

Missing values, diversified data types and insufficient sample size have become obstacles to clinical data analysis. In this work, we proposed a machine learning pipeline to analyze the metastasis rate of papillary thyroid carcinoma using B mode ultrasound images and incomplete clinical features simultaneously. Missing values in the clinical features were imputed by the multiple imputation algorithm and US images were analyzed by a deep transfer learning network. We applied the support vector machine to concatenate two types of features and made final predictions. Our proposed method achieves an AUC of 0.76, a sensitivity of 0.67, a specificity of 0.75 and an accuracy of 0.72 under 10-fold cross validation. These results are better than the transfer learning network based on US images (AUC=0.74) and the SVM method based on the clinical features (AUC=0.73).