Research On The Integration Of Medical Information System Based On Grid Technology

Proposed in 1990s, Grid intends to incorporate networking, communication, computation and information resources to make a virtual platform for some important research projects. As an important branch of grid in data management, data grid technology attempts to hide heterogeneity of distributed data resources to provide a uniform logical view for users, thus they can access heterogeneous data resources in a convenient and immediate way. Therefore, data grid provides a new solution for the integration of heterogeneous databases.This dissertation makes a deep research on integration of medical information systems based on data grid technology. Firstly, a metadata model is proposed for the integration of heterogeneous medical databases in grid environment. Based on the metadata model and some functional modules, an integration model of medical databases is put forward, and a protocol system is set up in experiment environment to test its availability. Finally, this dissertation attaches importance to the problem of semantic matching of database attributes, and proposes an attribute matching method based on artificial neural network (ANN), which deepens the research of database integration to semantic integration.The main work and contributions of this article are as following:1) Metadata modeling for medical databases integration in grid environment. Metadata is one of the most important technical factors in data management. According to the specialty of data stored in medical database, a metadata model is put forward for integration of medical information systems. Different types of metadata are defined in the model for meeting different access requirements, which expand the scope of data management, enhance the security of data access, and provide a more natural expression to relationship of data objects. Meanwhile, a layered metadata management framework is adopted which accomodates to grid environment. All this work lays a foundation for the design of database integration model.2) Research on integration model of medical informatiom systems in grid environmentBased on corresponding functional components and grid services, an integration model for medical informatiom systems is proposed in this dissertation that provides a uniform access interface to the underlying heterogeneous medical databases. After introducing to the architecture of the model, there is a detailed analysis to those functional components respectively, including data integration, query processing, data transfer and metadata management. Finally, a protocol system is set up in lab environment which is used to testify availability of the model through some query instances.3) Research on semantic integration of heterogeneous medical databasesThe main task of semantic integration of medical databases is to identify corresponding attributes in different databases that represent same concept in real world, which can resolve data conflict problem and provide a uniform logical view for users. Semantic integration is very important to establishing more effective metadata model for dynamic grid environment, which can improve the effectiveness of medical databases integration system, increase its usability and expand its application scope. Artificial neural network (ANN) is applied to the semantic integration of medical database attributes in this dissertation. Characteristic vectors which are composed of corresponding metadata and content value of attributes extracted from databases are used to train neural networks to find semantic corresponding attributes from other databases. To improve the accuracy of attributes matching, the idea for establishing multi classifiers is proposed, in which the Self Organizing Map (SOM) algorithm serves as a clustering algorithm to classify attribute patterns. The output of the classifier is then used as training data for a back-propagation network. Classification lowers the number of nodes in the network output layer and reduces the computational complexity as well as the training time, while increases the accuracy of attribute matching. Experiments in simulated environment show its availability and efficiency.