Electronic medical record text abstract extraction method

Abstract

The invention discloses an electronic medical record text abstract extraction method. By defining the semantic coverage loss function, abstract abstracts summarized by doctors in the past can be used for training an automatic extraction abstract model, and data annotation is avoided. The method comprises the steps of screening out relatively important sentences from an original electronic medical record text as a candidate set, narrowing a search range of an automatic extraction type abstract, and finding out a sentence combination with the highest semantic coverage degree in an original electronic medical record as an extraction type abstract by listing different combination modes of the sentences in the candidate set; when the semantic coverage degree is judged, making a judgment in combination with semantic information of all sentences in the candidate set, so the possibility of information redundancy and information loss in the automatic extraction type abstract is reduced, and the quality of the automatic extraction type abstract is improved.

Description

technical field [0001] The invention relates to the technical field of medical informatization, in particular to a method for extracting text abstracts from electronic medical records. Background technique [0002] Electronic medical records are the original records of patients in the whole process of diagnosis and treatment in hospitals, and they are also an important system to provide clinical decision support for doctors. Although electronic medical records replace paper medical records and facilitate the storage and search of relevant data, many important information of electronic medical records are still deeply buried in a large amount of text content. If a doctor needs to read a patient's electronic medical record comprehensively during clinical work, it often takes a long time. Therefore, extracting abstracts from electronic medical records is of great significance to increase the efficiency of doctors in clinical work, especially in the entry of discharged medical ...

AI Technical Summary

Problems solved by technology

[0004] However, the current general extractive summarization technology faces the following problems: 1) The commonly used extractive summarization technology generally takes the sentence in the text as the extraction unit and relies on the judgment of text similarity to obtain the more important sentences in the text, that is, from A subset is extracted from all the sentences in the original text, but the semantic connection between sentences in the extracted summary is weak, so the coverage of the extracted sentence combination on the original text is not necessarily the highest, and information redundancy will still occur. Situations where residual or critical information is missing

2) The performance of general unsupervised summary extraction models is often unsatisfactory. If you want to obtain a high-performance extraction model, you need data labeling. The purpose of data labeling is to mark important sentences from the original electronic medical records. Data labeling of medical texts is a A highly specialized, costly, time-consuming, and difficult-to-manage job

Images