Thyroid Nodule Classification Using Two Levels Attention-Based Bi-Directional LSTM with Ultrasound Reports

Classification of thyroid nodules is a basic work for the diagnosis and treatment of thyroid nodules. The most common way to classified nodule now is the support vector machines (SVM), but this method reduces the accuracy of classification due to the defects of language processing. So we put forward a new method. We divide the report into two layers, namely, the word vector and sentence presentation layer, and we use attention mechanism and Bi-Directional Long Short-Term Memory (Bi-LSTM) in every layer. In the end, test data show that our model has good performance.