Fig. 5
From: Master clinical medical knowledge at certificated-doctor-level with deep learning model
Architecture of our proposed deep learning framework Med3R. Med3R is consists of three modules: Free Reading, Guided Reading, and Multi-layer Reasoning. Firstly, primary medical knowledge is coarsely captured from medical text via the “Free Reading” module which is trained over large medical corpus with Unsupervised Learning (UL) methods. Then, Supervised Learning (SL) methods are conducted in the “Guided Reading” module where a “Fast Reading” strategy is first used to collect a small digest (strongly related with given medical questions) from large medical corpus, then “Deep Reading” strategies are employed to analyze the digest and the given medical questions in deep manners. In the Reasoning phase, the “Multi-layer Reasoning” module is used to produce robust decision-makings by integrating reasoning at key-points level (“KeyPoint Reasoning”), sentence-context level (“Context Reasoning”) and global-digest level (“Global Reasoning”), respectively
