Research On Medical Text Semantic Analysis Method For Decision Support

The amount of massive heterogeneous medical data stored in medical information systems increase continuously with the rapid development of medical informatics.Medical text information is an important basis of data to facilitate relevant studies on decision-making support in the medical and health field.For example,electronic medical records contain many descriptions of symptoms,diagnostic information,health-related conditions,clinical notes,and telemedicine and other medical information;fully utilizing medical texts with rich knowledge in medical domains is important for decision-making in the field.However,valuable medical text is currently not applied effectively in actual scenarios,such as hospital management,clinical decision support,personal health management,and public health decision-making,because of the following reasons: non-structured medical text cannot be structured precisely,rich semantics in the text make analysis difficult,textual information without detailed medical scenarios hinders future uses,inconsistent medical information standards cause difficulty in exchanging medical data,and lack of data sharing mechanisms among medical institutions.These problems are caused by the reliance on medical text,which is a professional language in specific domains,which includes various semantic information and rich medical domain knowledge.Effective semantic analysis for medical text is the key to solving these issues.Most existing studies on semantic analysis focused on natural language processing techniques in common domains;although the techniques are innovative in their methodological perspective,they often fail to provide high-quality data sets for training and testing in machine learning and even in deep learning models.Meanwhile,the analysis of massive medical text in the distributed computing environment has hindered the analyses and medical decision-making because traditional medical text analysis methods have bottlenecks in methodology and performance in such environments.This study,which is supported by a key project of National Natural Science Foundation of China “Big Data Driven Innovation and Management of Intelligent Healthcare”(No.71532002),aims to facilitate semantic analysis for medical text in China and other countries on the basis of existing theories and techniques in the literature for decision-making support.This study includes text analysis methods,medical information standardization,domain modeling,machine learning,and big data-driven algorithms in medical scenarios.Accordingly,the study solves the research problems of medical text mining and analysis toward decision support and provides reference for the development of medical informatics in China.Research contents and results are as follows:(1)Medical text structuring and standardization methods based on natural language processingThe characteristics of existing text structures in Chinese and English electronic medical records are first analyzed,and a text semantics model based on the N-gram model is proposed.Then,for massive non-standardized free medical text,we study the techniques of word segmentation,part-of-speech tagging,semantic relation extraction,and other specific analysis tasks and accordingly propose an automated structuring method for Chinese medical text.With the use of semantics and contexts in medical scenarios and taking the analysis of narrative Chinese ultrasound reports as example,we perform analysis tasks,such as word segmentation,annotation,and re-organization of key information within the medical text.The results indicate that the semantics-based methods improved the performance of medical text structuring,word segmentation,annotation,and other analysis tasks.(2)Semantic analysis and knowledge discovery methods for medical text based on domain knowledge sourcesExisting knowledge sources in medical domains are first studied.Then,promising applications of medical text mining and knowledge discovery are analyzed.Subsequently,we propose a method for calculating the text similarity based on weighted Levenshtein distance and N-Gram models to solve the problem of semantic similarity estimation in medical domains.We propose medical text similarity and relatedness methods based on Unified Medical Language System to solve the similarity estimation between medical concepts.The results indicated that by combining the semantic features in medical domains and the textual definitions of medical concepts in the knowledge sources,the method can meet the actual needs of medical text analysis tasks.Finally,we propose a computer-assisted coding method based on the 11 th Revision of International Classification of Diseases(ICD-11)to facilitate the coding,mapping,and standardization of ICD-11 for medical records based on semantic analysis and the semantic dictionary in Word Net,improving the efficiency of medical coders in classifying medical records in hospitals.(3)Medical text analysis and its applications for decision-makingOn the basis of(1)and(2),we study the methods and applications for analysis of massive medical text and accordingly propose performance optimization techniques for big data analysis.First,we propose medical topic modeling methods for medical texts with complicated data structures and examine the characteristics of different machine learning techniques in various applications.Then,for the analysis of massive medical data,we propose novel techniques based on the Map Reduce model to facilitate the analysis of medical text and leverage medical text analysis tasks,such as data association,automated medical coding,and other applications in the distributed computing environment where experiments on distributed text analysis methods are conducted.On the basis of the big data analysis platform,we propose a self-tuning method based on derivative-free optimization theory to improve the efficiency of Map Reduce performance tuning,solving the problem of methodological design and performance optimization in distributed computing environments for various decision-making methods in the medical and health domain.