Data-driven method for extracting information from electronic medical record

Abstract

The invention relates to a data-driven method for extracting information from an electronic medical record, which comprises the following steps of: S1, effectively expanding a collected core vocabulary library, and constructing a comprehensive cross-domain vocabulary library; and S2, extracting structured information containing a time-medical event-description triple from the electronic medical record by using the vocabulary library. According to the method, a vocabulary library for automatic recognition is expanded by adopting a data-driven enrichment mode, so that the vocabulary library is remarkably superior to a supervised learning model which is most popular in the same period in recognition of related medical terms, meanwhile, a support vector machine (SVM) which takes normalized Google Distance (NGD) as a feature is adopted as a matching model, and the correlation of events and corresponding descriptions thereof of the method is also superior to that of other feasible schemes, in addition, manual annotation is almost not needed, large-scale extraction is easy to achieve, and good stability is achieved when a large number of data faces are processed for increasing variation and noise.

Description

technical field [0001] The invention relates to the technical field of information extraction from electronic medical records, in particular to a data-driven method for extracting information from electronic medical records. Background technique [0002] Electronic medical records (EMRs) record a lot of useful information, such as descriptions of diseases, examination results, specific treatment options and curative effects, etc., which will help medical experts understand the disease more clearly. The development process, so as to find more effective treatment methods or discover the deficiencies of certain medical needs, and then automatically extract these useful information from unstructured electronic medical record texts is a rather intricate process. It is recorded during the patient's diagnosis and treatment process, so it has the characteristics of simplicity and individuality. Although medical records have corresponding practical writing standards, due to the diffe...

AI Technical Summary

Problems solved by technology

[0002] Electronic medical records (EMRs) record a lot of useful information, such as descriptions of diseases, examination results, specific treatment options and curative effects, etc., which will help medical experts understand the disease more clearly. The development process, so as to find more effective treatment methods or discover the deficiencies of certain medical needs, and then automatically extract these useful information from unstructured electronic medical record texts is a rather intricate process. It is recorded during the patient's diagnosis and treatment process, so it has the characteristics of simplicity and individuality. Although medical records have corresponding practical writing standards, due to the differences in writing habits or prior knowledge of doctors, the form of records will also be significantly different, and sometimes even possible. error log

[0003] In addition, the language complexity of Chinese has increased significantly compared to English, and Chinese physicians have also experienced traditional Chinese medicine (TCM) training, and their writing habits may even be closer to ancient Chinese. They want to effectively extract information from Chinese electronic medical records. Will face greater challenges. Therefore, the existing common methods for extracting information from news or general literary works are not suitable for extracting information from Chinese electronic medical records. A large number of manually annotated Chinese electronic medical records are used to train a It is indeed a feasible method to use the new model to identify, but it requires a lot of time for personnel with professional knowledge to annotate, which is obviously not easy to operate and promote. Therefore, a data-driven method for extracting data from electronic medical records is proposed. method of information

Images