自然語言處理於胸腔X光文本病歷報告的情感分析分類器及句子相似度計算

Our research project is to apply the most advanced natural language processing technology, Sentence-BERT model, to analyze the real clinical free-text diagnosis notes. Chest X-ray is currently the biggest dataset available for training artificial intelligence systems for automatic medical image diagnosis. However, for the machine learning purpose, an accurate ground truth labeling strategy is needed. Natural language processing (NLP) is a hot topic nowadays that has been applied to big data, data mining, and automatic document classification. BERT is a state-of-the-art NLP technology which can get a better embedding result for many downstream tasks. In order to understand its potential in biomedicine text analysis, this study applied Sentence-BERT on chest X-ray (CXR) reports dataset for labeling normal and abnormal sentences. It will be applied to Mimic’s two hundred thousand of clinical notes and the results will be compared to the NegBio method.