A Multi-label International Classification of Diseases Training Method Based on Curriculum Learning

Abstract

The invention discloses a multi-label international disease classification training method based on curriculum learning, which controls label distribution through three different small-batch sample sampling methods when classifying and automatically encoding a large-scale international disease data set. Firstly, the multi-label international disease training sample set is obtained, and the multi-label international disease training sample set is divided into multiple training sample subsets; in the first stage of training, the training sample subset is iteratively layered and sampled and the gradient is calculated. Update the model parameters; in the second stage of training, iteratively scramble the training sample set and calculate the gradient and update the model parameters in the second round; in the third stage of training, iterative probability sampling and calculation of the gradient are performed on the training sample subset and in the third stage round to update the model parameters. The invention improves the training stage of the current mainstream model, and the improved model greatly improves the model accuracy and generalization ability in the ICD coding multi-label classification task.

Description

technical field [0001] The invention relates to the technical field of natural language processing, in particular to a multi-label international classification of diseases training method based on course learning. Background technique [0002] The International Classification of Diseases (ICD) code is an important label in electronic health records. The ICD code is based on the etiology, pathology, clinical manifestations and other characteristics of the disease, and the same type of disease is classified into an ordered code combination. These codes are used to quantify important statistical data, facilitate the search for patient cohorts with similar diagnoses, and are of great value and significance as a means of standardized information exchange between hospitals. [0003] It is a very meaningful work to automatically classify ICD codes for electronic medical records. On the one hand, automatic classification saves a lot of manual classification costs, and on the other...

AI Technical Summary

Problems solved by technology

From the experimental results, it can be observed that the Fscore on the test set is generally low, while the Fscore on the training set is high. The generalization ability and accuracy of the three models are not ideal.

Images