Research On Disease Classification Based On Patient Complaints And Disease History

As the demand for public medical services increasing,the imbalance between the supply and demand of medical resources in China is becoming seriously prominent.The development of medical information technology has enabled medical institutions to accumulate a large amount of electronic health records,providing data support for intelligent diagnosis technology.Utilizing the chief complaint and medical history information in electronic health records for intelligent diagnosis can help doctors improve diagnosis quality and efficiency,and is an important way to alleviate the contradiction between medical resources supply and demand.With the development of deep learning technology,research on disease classification based on chief complaint and medical history information has made some progress.However,there are still some challenges,such as ignoring the differences in different types of data in the representation of chief complaint and medical history information,insufficiency in integrating external medical knowledge,and insufficient feature extraction of chief complaint and medical history information.Therefore,in this thesis,we use deep learning technology to conduct exploration and research on some of the problems that currently exist in the task of disease classification based on chief complaint and medical history information,and summarize the main research work as follows:(1)Addressing the issue of ignoring differences in various types of data in the chief complaint and medical history information,this thesis proposes a disease classification model based on multi-type data in the chief complaint and medical history information.Firstly,numerical information is used to enhance the text representation in the chief complaint and medical history information.Secondly,Bi LSTM-CRF is used to extract entities from the chief complaint and medical history information,and CNN is used to obtain entity representations.Finally,the text representation and entity representation are fused to obtain the final representation of the chief complaint and medical history,and disease classification is performed.The experiment conducted on a real clinical dataset showed that the model can effectively represent chief complaints and medical history information and perform disease classification.(2)Regarding the insufficiency of integrating external medical knowledge,this work proposes a disease classification model that integrates external medical knowledge with chief complaints and medical history information.Firstly,using the document similarity to obtain external medical knowledge that is similar to the chief complaints and medical history information.Secondly,the chief complaints and medical history information,as well as external medical knowledge,are encoded using document encoders.An attention mechanism is utilized to integrate the external medical knowledge.Finally,the chief complaints and medical history information are fused with the integrated external medical knowledge to obtain the final representation for disease classification.The experiments demonstrate that the model can effectively integrate chief complaints and medical history information with external medical knowledge,resulting in improved accuracy in disease classification.(3)For the insufficient feature extraction in the chief complaint and medical history information,this work proposes a disease classification model based on local and global semantic information in the chief complaint and medical history.Firstly,BERT is used as the embedding layer to obtain the encoding of the chief complaint and medical history information.Secondly,CNN and Bi LSTM-Attention are employed to capture local semantic information and global semantic information in the chief complaints and medical history information respectively.Finally,the local and global semantic information is fused to obtain the final representation of the chief complaint and medical history information,and disease classification is performed.The experimental results indicate that the model can effectively capture feature representations of the chief complaints and medical history information,leading to improved accuracy in disease classification.