Disease Inference with Symptom Extraction and Bidirectional Recurrent Neural Network
Donglin Guo, Min Li, Ying Yu, Yaohang Li, Guihua Duan, Fang‐Xiang Wu, Jianxin Wang · 2018
Automatic disease inference is an important task in many medical applications. Many efforts have been made to predict patients' future health according to their full clinical texts, clinical measurements or medical codes. Symptoms reflect the onset of diseases and can provide credible information for diseases diagnosis. In this paper, we propose a new disease inference method by extracting symptoms. To reduce the uncertainty and irregularity of symptom descriptions, we use MetaMap to extract symptoms from the Electronic Medical Records (EMR), a comprehensive clinical knowledge database consisting of massive amount of data about diseases, symptoms, and their relationships. To take advantage of the complex relationship between symptoms and diseases to enhance the accuracy of disease inference, we present symptom representation model: term frequency-inverse document frequency (TF-IDF) for the representation of the relationship between symptoms and diseases. Based on the symptom representation, we employ bidirectional recurrent neural network (Bi-LSTM) to model symptom sequence in EMR. Our proposed model shows a significant improvement in disease inference, i.e., 0.853 AUC and 0.563 F1 for 50 diseases on MIMIC-III dataset.