Traceable medical record classification method

Abstract

The invention discloses a traceable medical record classification method, and the method comprises the steps: carrying out the semantic integration on the context information of all the characters in the medical record by using the language model, and finally suppressing the influence of meaningless characters on the classification result through a gating mechanism, so the signal-to-noise ratio of intermediate data of the classification model is improved, and then the classification accuracy of medical record texts is improved. Meanwhile, meaningful characters for the classification result can be traced back through a gating value. Finally, while high-accuracy medical record classification is achieved, the gating values of the characters are output, and tracing of the character basis of medical record classification is achieved through the gating values of the characters.

Description

technical field [0001] The invention relates to the technical field of medical information processing, in particular to a traceable medical record classification method. Background technique [0002] The classification of medical records has a wide range of applications in the fields of current medical and health statistics, disease coding quality control, DRGs and medical insurance audits. But nowadays, the classification of medical records cannot be 100% accurate, and the medical industry is due to its rigor and other characteristics. The human-machine collaborative experience in product development is particularly important. There are two main technical solutions for traditional medical record classification methods: [0003] Methods based on statistical language processing: such as TF-IDF, BM25, N-GRAM, hidden semantic analysis, topic models, etc. The common feature of these methods is to extract text feature information from medical record texts from a statistical poi...

AI Technical Summary

Problems solved by technology

These methods have many obvious shortcomings: 1. When the text features of medical records are extracted by statistical language processing methods, the sequence structure of the original text is destroyed. Although the process of classification is more explanatory, the characteristics of medical records and the text information of medical records are only There is a statistical correlation, and there is no one-to-one correlation, so it cannot be traced back

2. Since the text features of medical record texts extracted based on statistical language processing methods are all literal statistics, it is impossible to go deep into the semantic information of each word, so the classification accuracy is not high

[0004] Methods based on deep learning: This type of method uses deep neural networks, such as CNN, RNN, Transformer, etc., to classify text. Although its classification accuracy is high, because deep learning methods are obvious black-box mechanisms, and recently There are a lot of literature and research pointing out that the traceability of deep learning, especially the traceability based on the Attention mechanism, has problems such as unexplainable problems. Although methods such as sensitivity analysis can be used for traceability, such methods can only be used for single words in the text. For traceability, full-text information cannot be combined during the traceability process