An automatic matching method for ICD surgery and operation codes based on deep learning

Abstract

An automatic matching method for ICD surgery and operation codes based on deep learning, using a modular modeling method, each module only completes a relatively simple task, which greatly reduces the search space for model parameters and reduces the required amount of data. This method uses a bidirectional autoregressive language model to model the natural language sequence, uses each operation description and each ICD code to combine, calculates the semantic space weight between them, and uses the semantic space weight to reconstruct the operation description , and finally use the reconstructed surgery description to perform ICD code classification matching to solve the problem of concept splitting. In the calculation, the inherent hierarchical structure of ICD surgery and operation codes is also used for bidirectional autoregressive model modeling, incorporating business prior test knowledge. It solves the problems encountered in clinical practice, and can quickly and accurately perform ICD code matching.

Description

technical field [0001] The invention relates to the field of medical technology, in particular to an ICD operation and operation code automatic matching method based on deep learning. Background technique [0002] The International Classification of Diseases Surgery and Operation Code (ICD-9-CM-3) is an important tool for the summary and statistics of hospital medical record information, and plays an important role in hospital medical treatment, research, and management. In actual clinical application, how to match the operation description entered by the medical personnel in the electronic medical record to the ICD operation and operation code is a time-consuming and labor-intensive task, requiring a lot of medical record reading work and code review work. Moreover, in the actual electronic medical record, the operation and operation description entered by the medical staff may be relatively short, that is, there are several surgical operation categories in a short descript...

AI Technical Summary

Problems solved by technology

In actual clinical application, how to match the operation description entered by the medical personnel in the electronic medical record to the ICD operation and operation code is a time-consuming and labor-intensive task, requiring a lot of medical record reading work and code review work

Moreover, in the actual electronic medical record, the operation and operation description entered by the medical staff may be relatively short, that is, there are several surgical operation categories in a short description, so how to conceptually split the operation description entered by the medical staff into the medical record and matching to standard ICD surgery and procedure codes is a lengthy and error-prone affair

However, general statistical learning, machine learning and deep learning classification models are often unable to cope with super-large-scale classification problems such as ICD coding, because the classification space is too large, and direct training using labeled data often results in serious overfitting or underfitting. And it is unable to solve the problem of conceptual separation of surgical descriptions. For example, the surgical description of "head and face laceration debridement and suture" needs to be split into two ICD surgery and operation codes, namely '86.2201 Excisional debridement of skin wounds' and '86.5900x006 skin suture', the general classification model cannot find a reasonable split method, and the general algorithm requires a large amount of labeled data. Under actual conditions, it is often difficult to obtain a large amount of labeled data due to various conditions. , and in clinical applications, because the error tolerance rate of medical work is relatively low, the errors caused by over-fitting and under-fitting of the model are unacceptable

Images