VGBNet: a disease diagnosis model based on local and global information fusion.

There are significant differences in the data volume of different types of diseases in the electronic medical record (EMR) data. In addition, the mainstream auxiliary diagnosis and prediction models have two shortcomings. They either only pay attention to the local features of medical records and ignore the global features, or only pay attention to the global features and ignore the local features. In response to these problems, we propose a method of fusion random resampling to balance the data set, Using graph convolutional neural network (GCN) to extract global features, combined with a bidirectional self-attention network (BERT), a VGBNet model is proposed to link local and global features to achieve diagnosis and prediction of diseases. A large number of experiments show that compared with the state-of-the-art models, the VGBNet model has improved F1 value and accuracy. This is of great significance to the precise auxiliary diagnosis.