A method, system and device for constructing a medical record model based on deep learning

Abstract

The invention discloses a method, system and device for constructing a medical record model based on deep learning. The method includes obtaining medical records; performing aggregation training on the obtained medical records to generate word-level vectors, record-level vectors and patient-level vectors; according to the generated Word-level vectors, record-level vectors, and patient-level vectors of , using recurrent neural networks to model medical records. The system includes an acquisition module, an aggregation training module and a model building module. The device includes memory and a processor. The present invention makes full use of free text records including patient personal data, test values, and medical records to construct models, which improves the prediction accuracy of medical record models; in addition, the present invention uses cyclic neural networks to construct medical record models, compared with existing There are topic models or methods of learning direct word representations, which reduce the amount of computation and improve the efficiency of building models. The invention can be widely applied in the field of natural language processing.

Description

technical field [0001] The present invention relates to the field of natural language processing, in particular to a method, system and device for constructing a medical record model based on deep learning. Background technique [0002] In recent years, with the advent of electronic health records (EHRs), many attempts have been made to apply machine learning methods to patient data to address problems such as survival analysis, causal inference, and mortality prediction. The data-heavy records in EHR databases usually contain a large number of numerical features, such as patient demographics (age, gender, ethnicity, etc.), laboratory measurements (such as blood gases, fluid balance, vital signs, etc.), and secondary data of diseases and medical procedures. meta-indicators, and free-text medical records, etc. Although numerical indicators can be directly applied to most machine learning models, for medical records written by doctors and nurses, researchers use existing tech...

AI Technical Summary

Problems solved by technology

Although numerical indicators can be directly applied to most machine learning models, for medical records written by doctors and nurses, researchers use existing techniques to select existing topic models or learn direct word representations, but due to the The amount of data is huge and irregular, resulting in a large amount of calculation for these existing technologies, and the efficiency of model construction and deep learning is very low; in addition, for the medical records of patients composed of disordered words, the existing medical records The processing techniques cannot make full use of free-text medical record words (such as condition descriptions in medical records), resulting in low prediction accuracy of the final model

Images