Research And Implementation Of Structured Thyroid Discharge Summary Based On Deep Learning

With the continuous development of hospital digital information construction,lots of hospitals have accumulated a large amount of clinical data,many of which have great medical research value.Among these medical digital data,the discharge summary contains relatively complete information about the patient’s visit,such as the patient’s symptoms,diagnosis,testing and treatment and other related aspects,and the discharge summary contains relatively complete information about the patient’s visit,so the study of the discharge summary has great significance.From the perspective of data analysis,discharge summary is a kind of unstructured text data,which cannot be directly used in existing data mining and analysis models,so it needs to be structured first in order to be better applied in downstream data analysis and mining models for clinical medical research.In this paper,we focus on the study of thyroid discharge summaries,and use the thyroid discharge summaries provided by a tertiary hospital in Shanghai as the data base to carry out the study,and propose a deep learning-based text structuring method for thyroid discharge summaries.The main contributions are as follows.1)Designed thyroid discharge summary field ontologyCombining the content characteristics of thyroid discharge summary text data with domain ontology knowledge,using a seven-step approach to build the ontology,and then full discussion with the doctor,determining the validity of the ontology.2)Realize structured text of thyroid discharge summaryThe structured approach consists of two phases.The first stage is based on the Bi LSTM-CRF model,with the help of the BERT model,the word vector,sentence vector and position vector of the text are superimposed as input to obtain the model BERT-Bi LSTM-CRF,which incorporates the global semantic information of the text into the character information while obtaining the contextual dependencies of the sequences,and the model has better effect through experiments.The second stage is based on the BERT model,which enhances the attention information of words by adding the relative positions of entities into the process of attention calculation of the model,and then uses a specific structure to input multiple pairs of entity relations at one time,and through experiments,it is verified that the model has a better effect of entity relation extraction.The third stage takes the extracted entity pairs and entity pair relationship triples and instantiates the domain ontology according to the structure of the domain ontology,thus structuring the thyroid discharge summary text.3)Designed and developed structured systemsFrom the needs of actual clinicians,a structured system of thyroid discharge summary text has been implemented to facilitate clinicians to quickly understand patient visit information and provide hospital clinicians with assistance in diagnosis.In summary,this paper uses thyroid discharge summary text data as the experimental data set,firstly designs domain ontology,then builds the domain ontology according to the ontology construction method,so as to complete the text structuring,and finally designs and develops the structured system.