Automatic medical record coding method capable of learning in real time

Abstract

Provided is an automatic medical record coding method capable of learning in real time. The method comprises: utilizing the structural characteristics of ICD coding, using a comparative learning method to train an embedded model of a medical record, adopting a KNN reasoning method to realize automatic medical record coding in an embedded set formed by the embedded model, collecting the use behaviors of a coder in the use process, modifying the coding space and the embedding space in a customized manner, and realizing real-time learning of a single sample level. The method is matched with the use of coding software, self-evolution can be realized in the actual use process of a coding person until the coding habit of the coding person is almost completely met, and the working efficiency of hospital medical record coding can be greatly improved.

Description

technical field [0001] The invention relates to, in particular, an automatic encoding method for medical records capable of real-time learning. Background technique [0002] Coding of medical records is a basic task of manually reading the patient's medical records and selecting appropriate codes for summarization in accordance with the ICD coding standards. It is the basis of today's medical record management, medical insurance settlement, hospital performance appraisal, DRG payment and other scenarios. In the current hospital management practice, most of the medical record coding work is carried out by dedicated medical record coders. Because the medical record coding work requires long-term experience accumulation, and the medical record coding is the key to medical insurance settlement, hospital performance appraisal, DRG payment, etc. The accuracy of coding will greatly affect the work of these hospitals. [0003] At present, many medical record system software manufa...

AI Technical Summary

Problems solved by technology

In the current hospital management practice, most of the medical record coding work is in charge of a dedicated medical record coder, because the medical record coding work requires long-term experience accumulation, and the medical record coding is the basis of medical insurance settlement, hospital performance appraisal, DRG payment, etc. The accuracy of coding will greatly affect the work of these hospitals

[0003] At present, many medical record system software manufacturers and research papers have proposed many methods for automatic medical record coding. These methods are divided into the following categories: 1. End-to-end methods based on deep learning: all use deep learning methods to directly input medical records. Output coding result data, this method has no way to perform active learning at the single-sample level according to the coder's usage habits

2. Method based on knowledge map: This method is difficult to combine the content information of medical records. Most of them are coded based on the diagnostic text given by doctors in medical records. Moreover, for data with wrong coding, special technical personnel are required to maintain and maintain The cost is very high, and at the same time, real-time level self-learning is not possible