Semantic Interoperability for Data Integration Framework using Semantic Web Services and Rule-based Inference: A case study in healthcare domain

Transcription

1 Semantic Interoperability for Data Integration Framework using Semantic Web Services and Rule-based Inference: 1 Suphachoke Sonsilphong, 2 Ngamnij Arch-int Semantic Mining Information Integration Laboratory (SMIL) Department of Computer Science, Faculty of Science, Khon Kaen University, Thailand 1 suphachoke.son@gmail.com, 2 ngamnij@kku.ac.th Abstract This paper proposes a Semantic Interoperability for Data Integration-SIDI framework to integrate information from heterogeneous databases of difference providers in the same domain. A framework is designed to incorporate with important procedures based on ontology and semantic web services technologies. The semantic web services annotation is imperative to cope with the semantic service discrepancies with the help of ontology. The research also proposes the ontology mapping technique to determine the correspondences between information concepts with semantic bridges description to automatically construct the semantic rule-based inference. As a result, the framework has been applied to the healthcare domain to enable semantic interoperability among independently developed health information system-his for integrating healthcare data. Keywords: Data Integration, Semantic Interoperability, Semantic Web Services, Health Information, Rule-based Inference 1. Introduction A classic problem in the information integration among organizations is that the providers develop information systems which are appropriate to their needs, differ from place to place. These systems as a whole have become heterogeneous and dependent on a variety of applications or database management systems [1-3]. This leads to the interoperability problems that can be investigated in two categories: Interoperability of the data exchanged and interoperability between heterogeneous systems. Although the metadata standards are invented to be the solution for handling the data interoperability, these standards are still developed appropriately to only local area systems and differ from other standards in the global scale. Furthermore, most of the metadata standards still lack a formal semantics and a common standard between heterogeneous metadata descriptions across domains. Thus, the heterogeneity still occurs when we need the information integration from difference standards. For example, in the healthcare domain, many health information systems or Electronic Patient Records (EPRs) are developed proprietary and often only serve one specific requirement. Each EPR is designed through different standard bodies like HL7 [4], CEN TC251 [5], ISO TC215 [6] and other local standards. It is impossible to easily share data across other healthcare organizations, hence, making it difficult for doctors to capture a complete clinical history of a patient. This includes prescription information that will enable the clinician to make an accurate primary decision or diagnosis. On the other hand, Web services [7][8] have been identified as the salient technology in providing a flexible solution for integrating the heterogeneous applications and enabling the dynamic interoperability between different systems. However, the current service description [9] and discovery mechanism are not powerful enough for computer-interpretation to enable automatic Web service discovery and invocation. Although, the research of semantic Web services [10][11] introduced technique to enhance flexibility and dynamically of Web services management and invocation, using semantic web services for information integration [12-15] still has a limitation of parameter extraction for fully supporting data inference processing and solving semantic service discrepancies. This research proposes the Semantic Interoperability for Data Integration (SIDI) framework for solving the interoperability problems across system heterogeneity. This work applies a Journal of Convergence Information Technology(JCIT) Volume8, Number3,Feb 2013 doi: /jcit.vol8.issue3.18

2 framework to the healthcare domain for integrating patient data stored in heterogeneous EPR systems and supporting health services such as diagnostics and counseling for the treatment of a doctor or specialist. In the context of semantic health data exchange and integration, the research employs an ontology-based model for health data integration [16][17] and rule-based inference for automated ontology mapping with the Health Level 7 (HL7) standard. The research also employs semantic Web services to facilitate maximal automation and dynamism in data retrieval. The SIDI framework is designed into various components to support the resolution as follows: (1) Solving the platform independence of the heterogeneous information systems. (2) Solving the semantic conflicts of service descriptions by a means of Web services annotation and rule-based inference. (3) Collecting the partitioned data from individual Web services using Web services composition. The rest of the article is structured as follows: The next section explain motivating example to this research. Section 3 present the SIDI framework architecture and the design processes of the SIDI is presented in section 4-7. In section 8, we show the experimental and evaluation, and the final section is the conclusion. 2. Motivating example In the context of the health information systems, these systems have become heterogeneous and dependent on a variety of applications or database management systems, as well as the database structures. Examples of such heterogeneity between two databases of different healthcare systems are shown in Figure 1-A and 1-B. The database heterogeneity leads to the problems of schema conflicts and data restriction conflicts. In Figure 1-A, the table Patient in database A is semantically equivalent with the table p_ptdata in database B, and the attributes NAME and SURNAME of table Patient in database A are semantically equivalent with the attributes fname and lname of table p_ptdata in database B, respectively. Another conflict is attributes aggregation which is also an example of schema conflicts. The attribute NAME of table Doctor in database A is semantically equivalent with the aggregation of attributes fname and lname of table p_drdx in database B. For the data restriction conflicts, Figure 1-B illustrates the data restriction of attributes sex in database A and B which contain different data values. Figure 1. An example of databases heterogeneity; A: schema conflicts B: constrain conflicts 3. The Semantic Interoperability for Data Integration-SIDI framework The SIDI framework is designed as a layer of collaborating stakeholders, as shown in Figure 2, consisting of (1) the Resources Layer which is the layer of the provider system (2) the Mediator Layer which acts as a broker system, and (3) the Application Layer which is a layer of the data requester. The details of the framework are described as follows.

3 Figure 2. Semantic Interoperability for Data Integration Framework The Resources Layer consists of heterogeneous database systems that are required data integration between them. This layer consists of the following processes: Web services creation is a process of creating the code program and WSDL of Web services. This framework enables the services provider to create Web services manually, and also provides a tool for Web services creator to create Web services from the extracted schema of database. Ontology creation is a process of creating the local ontology used for annotating the Web services parameters. For this process, the semi-automatic ontology extraction tool has been developed and provided to construct the consistent ontology structure from the extracted database schema automatically. However, the extracted ontology is a kind of loose-defined ontology because a database schema usually lacks of description of entities and attributes. Therefore, this process needs the ontology editor to modify the extracted ontology before exporting to the SIDI broker. For this research, the local ontology extraction is expressed in the form of OWL (Ontology Web Language) [18]. Examples of Web services creation and local ontology extraction from local databases of different health information systems are illustrated in Figure 3. In order to describe the semantic of Web services explicitly and unambiguously, each Web service provider is able to create a local ontology with any terminologies standards in the healthcare domain. Figure 3 Example of web services creation from local database The Mediator Layer consists of a broker-based system which is responsible for integrating data collected from Web services in the resource layer. The Mediator layer provides main components as follows: Web services registration is used for the service providers to register their services and local ontology into the SIDI broker to be used for the next steps. Semantic Web services generation aims to create a semantic description of services which are automatically generated from converting of the WSDL into the descriptive section of

4 Service Profile, Service Model and Services Grounding according to the OWL-S [19] structure. The initial semantic descriptions of services are then passed to the Web services annotation process to describe the parameters of services with conceptual terms. Web service annotation is a process for describing the input/output parameters of Web service derived from the initial description of Service Model and Grounding with terms defined in local ontology through the parameterstype property of OWL-S. This procedure requires Web services annotator to annotate parameters manually. The annotated services which are atomic services will be placed in Service Profile repository. Ontology mapping is a process to link the local ontology with the global ontology defined as the domain information schema. The mapping process defines the semantic bridges, expressed in OWL, to describe the mapping scheme, and requires a domain expert to map the concepts manually. Consequently, these semantic bridges will be used for generating the semantic inference rules. Rule generation is a process of semantic rules generation expressed by the Semantic Web Rule Language (SWRL) [20]. The semantic rules are automatically generated from the semantic bridges and stored in the repository to be used for the data inference process. Web services composition is a process of combining atomic processes from dispersed Web services that are registered in the Service Profile repository to construct the composite processes. The Web services composition process can be accomplished through the ontology editor tools to compose the services. Service execution is a process for invoking the selected composite service to access data stored in the local databases. In addition, this process includes the sub-process Web services parameter transformation to transform the xml-based message into the ontology-based data structure conforming to the local ontology schema. Data inference is a process to infer data derived from the Web service invocation of the service execution process. The data inference process employs the semantic rules for accessing and collecting the complete data corresponding to the global ontology schema automatically. The Application Layer consists of client applications which are developed for the information search. This layer is connected to the SIDI broker to receive a request from user and return response data from the data inference process of the mediator layer. The format of response data conforms to the standard metadata defined according to the global ontology of the SIDI system. 4. Web services annotation, ontology mapping and rules generation In the process of Web services annotation, the ServiceModel consists of classes that can be described with the capability of Web services, such as Input, Output, Precondition, and Effect. For this research, we scope on input and output parameters. The parametertype property is used as a means to annotate the Web service parameters with classes or properties defined in the local ontology. An example of Web services annotation with the local ontology is shown in Figure 4. The message schema of the getpatient operation in the WSDL is annotated with classes and properties defined in the local ontology. Figure 4. Semantic annotation of services parameters with local ontology

5 This work presents a global ontology (see Figure 5), which contains the global information concepts conforming to the standard HL7 [4] terminologies. These global information concepts indicate scope of data elements to be retrieved from the Web services structure. Once the application client needs to search data from the SIDI broker, the application client has to send a request with search condition corresponding to the scope of the global ontology. The response data returned from SIDI broker are integrated and conformed to the global ontology structure. Figure 5. Example of global information concepts in Healthcare domain To resolve the conflicts which occur when each local ontologies defines terms under its own standard, this research proposes the semantic bridges (SBs), expressed in OWL, to map terms defined in the local ontology to terms defined in the global ontology through the ontology mapping process. The semantic bridge has been proposed with three different bridge types, namely, the EquivalentBridge, the AggregationBridge and the DataMediatorBridge as shown in Figure 6. Each bridge type is defined as a subclass of the DataPropertyBridge. Hence, the SBs are intended to be used to define the linkage rules that are mentioned as the follows: Figure 6. Schema of semantic bridges based-on OWL EquivalentBridge is used to resolve the synonymous conflicts which are concerned with the semantically equivalent classes or properties defined by different names. For example, the property family_name defined in global ontology and the property last_name defined in local ontology are the synonymous properties, since both of them refer to the same fact. The EquivalentBridge will bridge the last_name and the family_name as the source and destination properties, respectively. AggregationBridge is used to resolve the aggregation conflicts which arise when a class or a property of global ontology is mapped to a group of concepts or properties, respectively, in local ontology. For example, the property name of the concept Person in global ontology is equivalent to the aggregation of property first_name and last_name in local ontology. AggregationBridge is given first_name and last_name as source properties and given the property name as the destination property. Figure 7 illustrates the AggregationBridge creation expressed in OWL format.

6 <sb:aggregationbridge rdf:id="aggregationbridge_1"> <sb:sourceproperty rdf:resource=" <sb:sourceproperty rdf:resource=" <sb:destinationproperty rdf:resource=" <sb:aggregateproperties> <sb:aggregatelist rdf:id="aggregatelist_1"> <sb:first rdf:resource=" <sb:rest> <sb:aggregatelist rdf:id="aggregatelist_2"> <sb:first rdf:resource=" </sb:aggregatelist> </sb:rest> </sb:aggregatelist> </sb:aggregateproperties> </sb:aggregationbridge> Figure 7. Example of AggregationBridge description in OWL DataMediatorBridge is used to resolve data restriction conflicts which concern semantically equivalent properties defined with different property data values. For example, a local ontology defines a value of the property sex to be an enumeration value of m or f, another local ontology defines a value of sex as 1 or 2, whereas the global ontology defines a value of property sex as male or female. For solving this kind of conflict, the DataMediatorBridge is proposed by setting a bridge to map the source property sex, with the source value of m and f (or 1 and 2 ) to the destination property gender, with the destination value of "male" and "female" respectively. An example of the whole graph of semantic mapping is shown in Figure 8. The Web service parameters are annotated with terms defined in local ontology, whereas the local ontology terms are mapped with the global ontology terms via the designate semantic bridges. Figure 8. An example of global to local ontology mapping In addition, the proposed semantic bridges can be deduced into the semantic rules, expressed by the SWRL syntax, through the rules generation module. This paper presents the semantic rules generation deduced from the semantic bridge description as shown in Table 1. These rules will be employed in the next data inference process.

7 Table 1. Example of SWRL rules for data inference No Semantic bridge SWRL from Rule generation Description 1 EquivalentBridge: 1) thcc:patient(?x) thcc:last_name(?x,?c) Mapping the local property EquivalentBridge(?b) destinationproperty(?b, family_name) dom:person(?x) dom:family_name(?x,?c) last_name to the global property family_name sourceproperty(?b, last_name) 2)dom:Person(?x) dom:family_name(?x,?c) 2 AggregrationBridge: AggregationBridge(?b) destinationproperty(?b, name) sourceproperty(?b, first_name) sourceproperty(?b, last_name) 3 DataMediatorBridge: DataMediatorBridge(?b) destinationproperty(?b, gender) sourceproperty(?b, sex) destinationvalue(?b, male ) sourcevalue(?b, m ) 5. Web services composition thcc:patient(?x) thcc:last_name(?x,?c) 1) thcc:patient(?x) thcc:first_name(?x,?a0) thcc:last_name(?x,?a1) swrlb:stringconcat(?c,?a0,?a1) dom:person(?x) name(?x,?c) 1) thcc:patient(?x) loc1:sex(?x, "m") dom:person(?x) dom:gender(?x, "male") 2) dom:person(?x) dom:gender(?x, "male") thcc:patient(?x) loc1:sex(?x, "m") Mapping the local property by aggregate first_name and last_name to the global property name Mapping the local property sex to the global property gender with value transformation from m to male The Web services composition is a process of creating a composite process from atomic processes that are stored in the services profile repository. Figure 9 shows an example of a composite process that is derived from composing of the three atomic processes, such as getpatient, getdiagnosis and getdoctor. The getpatient process contains citizen_id as an input parameter and contains patient_code, given_name, and family_name as the output parameters. The patient_code output derived from the getpatient process will be used as an input of the getdiagnosis process which returns icd10_code, and physician_code as the outputs and will be used as inputs to the next getdoctor process that is finally return name as an output of the process. Figure 9. Example of Composite Process 6. Web services execution & Data inference In the run-time procedures, the application sends a data request and selected services to the SIDI broker which will execute the composite process according to the request. For this research, we implemented the service execution engine to invoke each atomic process from the composite process structure. Once the SOAP response message is returned from Web service execution, the data message part is then transformed to be instance of the associated local ontology via the Web services parameter transformation process. For this process, each output parameter will be assigned a value id to be an instance of a class in the local ontology that is associated with that output parameter through the parametertype property. Each child element of the output parameter is then mapped with the predicate of that associated class. The value of each child element is transformed to be value of each predicate. For example in Figure 10, the element p_ptdata of selectp_ptdata response is generated a value id to be an instance of the class &loc2#patient. The child elements of p_ptdata, such as personid, hn, title, etc. and their values are being mapped to be the predicates and predicate values of the class &loc2#patient, respectively.

8 Figure 10. Example of complex parameters to ontology transformation For the data inference, we implemented the rule inference engine which requires SWRL to infer instances from local ontologies to be integrated with instances of the global ontology standard. An example of data inference is shown in Figure 11. An instance of the class loc2:patient in a local ontology is transformed to be an instance of the class dom:patient in the global ontology. Figure 11. The result of local ontology to global ontology data inference 7. Experimental and evaluation In order to evaluate the experimental results, this research employs existing two databases of the health information systems from local hospitals in Thailand. These systems have different platforms, such as the databases and application systems. The Web services were created for these systems consisting of 372 atomic services for querying the request data from all tables in databases. In this experiment, we set up the 20 patterns of data request conformed to the global ontology. A portion of data query request is given in Table 2. Table 2 Example of data request and selected service No Data query request (SPARQL) Composite Service 1 select?p Split-Join{&loc1;#selectPatient, where {Patient(?p). citizen_id(?p, )} &loc2;#selectp_ptdata} 2 select?p,?s,?d where {Patient(?p). citizen_id(?p, ). Sequence{&loc1;#selectPatient, &loc1;#selectdiagreg, &loc1;#selectdoctor} Diagnosis(?s). subject_of(?p,?s). Physicial(?d). perform_by(?s,?d)} In this work, we considered the accuracy of data retrieved from the selected services in two cases: (1) the accuracy of data instances. As shown in example of Figure 12-A, if a data request needs every instance of class Patient, but the data output retrieved from the service execution returns only some instances. This case is considered to be false negative. (2) The accuracy of data properties. As shown in example of Figure 12-B, if there is a request to find instances of class Patient with properties citizen_id and gender, the data output returns not only these two properties but also other properties. This case is considered to be false positive.

9 Figure 12 A. request patient s instances, B. request patient s instances with citizen_id, and gender properties To evaluate the results, this research calculated the average recall and precision score of the data retrieved from the service execution. To compare the accuracy of the output results returned from system with reference data results set, this research considered cases of true positive (TP), false positive (FP) and false negative (FN) with the reference data. Equation 1 is used for calculating the recall score of the accuracy to indicate the false negative rate of the data retrieval and the equation 2 is used for calculating the precision score to indicate the false positive rate. TP recall = (1) TP + FN TP precision = (2) TP + FP This research illustrated graphs of average recall and precision scores associated with the n- request patterns as in Figure 13. According to the graph, the error rate of false negative which is shown on recall functions line is not more than 0.03, while the false positive error rates of data retrieve which is shown on precision functions line is between 0.05 to Conclusion Figure 13 Recall & Precision of data retrieval This paper presents a framework for data integration using semantic Web services technologies which can be used for developing the information integration system through a broker based approach. The main components of the proposed framework include: (1) Web services annotation is the process designed for associating the input and output parameters of services to the local ontologies to share understanding of their service information. (2) Ontology mapping is the process for providing semantic bridges to generate reasoning rules for resolving the semantic conflicts of different local ontologies and mapping to the global ontology. (3) Web services composition is the process for creating composite processes from atomic processes by verifying the input/output parameters between atomic processes to generate the combination of services. The approach has been applied to a practical case of healthcare domain to enable the interoperability of health information systems which are platform-independent to enable sharing and exchanging of patient data. The approach provides the maximum benefit for healthcare

10 organizations to link and capture a complete clinical history of patients stored in dispersed systems to enable the clinician to make an accurate primary decision and medical diagnosis. 9. Acknowledgements In this research, we deeply thank to Nong Bua Rawae Hospital, Chaiyaphum Province, Thailand for data support and the Royal Golden Jubilee Ph.D. Program of the Thailand Research Fund for financial support of the research due to research grant: PHD/0278/ References [1] Sheth, A.P., Larson, J.A., "Federated database systems for managing distributed, heterogeneous, and autonomous databases", ACM Comput. Surv., vol.22, no.3, pp , [2] Kossmann, D., "The state of the art in distributed query processing", ACM Comput. Surv., vol.32, no.4, pp , [3] Halevy, A., Rajaraman, A., Ordille, J., "Data integration: the teenage years", In Proceedings of the Proceedings of the 32nd international conference on Very large data bases, pp.9-16, [4] HL7, "Health Level 7 (HL7)", 2009, Available at: [5] CEN, "CEN/TC 251", Available at: [6] ISO, "TC 215 Health informatics", Available at: [7] Yu, Q., Liu, X., Bouguettaya, A., Medjahed, B., "Deploying and managing Web services: issues, solutions, and directions", The VLDB Journal, [8] Alonso, G., Casati, F., Kuno, H., Machiraju, V Web Services - Concepts, Architectures and Applications. Springer. [9] W3C, "Web Services Description Language (WSDL)", [10] McIlraith, S.A., Son, T.C., Zeng, H.L., "Semantic Web services", Ieee Intelligent Systems & Their Applications, vol.16, no.2, pp.46-53, [11] Sycara, K., Paolucci, M., Ankolekar, A., Srinivasan, N., "Automated discovery, interaction and composition of Semantic Web services", Web Semantics, vol.1, no.1, pp.27-46, [12] Kotsiopoulos, I., Keane, J., Turner, M., Layzell, P., Zhu, F., "IBHIS: Integration Broker for Heterogeneous Information Sources", In Proceedings of the Proceedings of the 27th Annual International Conference on Computer Software and Applications, pp.378, [13] Dogac, A., Laleci, G.B., Kirbas, S., Kabak, Y., Sinir, S.S., Yildz, A., Gurcan, Y., "Artemis: Deploying semantically enriched Web services in the healthcare domain", Information Systems, vol.31, no.4-5, pp , [14] Domingue, J., Cabral, L., Galizia, S., Tanasescu, V., Gugliotta, A., Norton, B., Pedrinaci, C., "IRS-III: A broker-based approach to semantic Web services", Web Semant., vol.6, no.2, pp , [15] Zhonggui, M., Chengyao, W., Zongjie, W., "Semantic model based on three-layered metadata for oil-gas data integration", Advances in Information Sciences and Service Sciences, vol.3, no.7, pp , [16] Beydoun, G., Low, G., Tran, N., Bogg, P., "Development of a peer-to-peer information sharing system using ontologies", Expert Syst. Appl., vol.38, no.8, pp , [17] Xu, Y., "Research on method of ontology-based knowledge representation and integration for drilling risk decision", International Journal of Digital Content Technology and its Applications, vol.6, no.16, pp , [18] Smith, M.K., Welty, C., McGuinness, D.L., "OWL Web Ontology Language Guide", 2004, Available at: [19] Martin, D., Burstein, M., Mcdermott, D., Mcilraith, S., Paolucci, M., Sycara, K., Mcguinness, D.L., Sirin, E., Srinivasan, N., "Bringing Semantics to Web Services with OWL-S", World Wide Web, vol.10, no.3, pp , [20] Horrocks, I., Patel-Schneider, P.F., Boley, H., Tabet, S., Grosof, B., Dean, M., "SWRL: A Semantic Web Rule Language Combining OWL and RuleML", 2004, Available at: