Diagnosis prediction method of bidirectional recurrent neural network based on attention

Abstract

The invention discloses a diagnosis prediction method of a bidirectional recurrent neural network based on attention, and relates to the technical field of prediction diagnosis. The diagnosis prediction method firstly embeds high-dimensional medical codes (ie, a clinical variable) into a low code layer space, and then inputs code representation into a bidirectional recurrent neural network based on attention to generate hidden state representation. The diagnosis prediction method inputs the hidden representation through a softmax layer so as to predict medical codes accessed in future. The experimental data shows that, by using the diagnosis prediction method of a bidirectional recurrent neural network based on attention, when predicting future access information, the attention mechanism can assign different weights to previous accesses, thus effectively completing the diagnosis and prediction task, and reasonably explaining the prediction result.

Description

technical field [0001] The invention relates to the technical field of predictive diagnosis, in particular to a diagnostic predictive method based on an attention-based bidirectional recurrent neural network. Background technique [0002] Electronic Health Records (EHR) consist of longitudinal patient health data. Patient EHR data includes a sequence of visits over time, where each visit contains multiple medical codes, including diagnosis, medication, and procedure codes. Applied to several predictive modeling tasks in healthcare. EHR data consists of a set of high-dimensional clinical variables (ie, medical norms). One of the key tasks is to predict future diagnoses based on past EHR data of patients, i.e. diagnosis prediction. [0003] The timing of each patient visit and the medical code for each visit may have different importance in predicting a diagnosis. Therefore, the most important and challenging issues in diagnostic prediction are: 1. How to correctly model th...

AI Technical Summary

Problems solved by technology

Therefore, the most important and challenging issues in diagnostic prediction are: 1. How to correctly model these temporal and high-order secondary EHR data to significantly improve prediction performance; 2. How to rationally account for visits and medical norms in the predicted results the importance of

However, RNN-based methods may not fully remember all previous visit information, leading to wrong predictions

Images