Int J CARS (2009) 4:509–516 DOI 10.1007/s11548-009-0362-6
ORIGINAL ARTICLE
Radiology information system: a workflow-based approach Jinyan Zhang · Xudong Lu · Hongchao Nie · Zhengxing Huang · W. M. P. van der Aalst
Received: 29 January 2009 / Accepted: 13 May 2009 / Published online: 9 June 2009 © CARS 2009
Abstract Purpose Introducing workflow management technology in healthcare seems to be prospective in dealing with the problem that the current healthcare Information Systems cannot provide sufficient support for the process management, although several challenges still exist. The purpose of this paper is to study the method of developing workflow-based information system in radiology department as a use case. Method First, a workflow model of typical radiology process was established. Second, based on the model, the system could be designed and implemented as a group of loosely coupled components. Each component corresponded to one task in the process and could be assembled by the workflow management system. The legacy systems could be taken as special components, which also corresponded to the tasks and were integrated through transferring non-workflow-aware interfaces to the standard ones. Finally, a workflow dashboard was designed and implemented to provide an integral view of radiology processes. Result The workflow-based Radiology Information System was deployed in the radiology department of Zhejiang Chinese Medicine Hospital in China. The results showed that it could be adjusted flexibly in response to the needs of changing process, and enhance the process management in the department. It can also provide a more workflow-aware J. Zhang · X. Lu (B) · H. Nie · Z. Huang The Key Laboratory of Biomedical Engineering, College of Biomedical Engineering and Instrumentation of Zhejiang University, Hangzhou, China e-mail:
[email protected] W. M. P. van der Aalst Department of Information and Technology, Eindhoven University of Technology, P.O. Box 513, 5600 MB, Eindhoven, The Netherlands
integration method, comparing with other methods such as IHE-based ones. Conclusion The workflow-based approach is a new method of developing radiology information system with more flexibility, more functionalities of process management and more workflow-aware integration. The work of this paper is an initial endeavor for introducing workflow management technology in healthcare. Keywords Workflow management · Radiology information system · Workflow modeling · System integration Purpose The healthcare practitioners are now facing challenges of increasing cost and growing demand for higher efficiency and quality. The quickly evolving technologies of workflow management provide solutions to these challenges. With functionalities of workflow automation, monitoring, and configuration, it could be prospective to improve the efficiency and quality of healthcare, so as to reduce the cost. Nevertheless, current medical information systems cannot provide sufficient support to these functionalities. Typical medical information systems are data-oriented and composed of a group of data Create, Read, Update and Delete (CRUD) transactions. The processes are driven by the codes encapsulated in the systems which could hardly be configured corresponding to changes in demand. Meanwhile, due to the heterogeneous environment in healthcare, the processes are divided into several segments for different systems and it is difficult to monitor and connect them as a whole. Integrating the Healthcare Enterprise (IHE) has addressed the problem of workflow integration through integration profiles [1] that aim at connecting processes in heterogeneous medical information
123
510
systems. But this bottom-up approach only concerns with the linkage of the processes among different systems other than managing the process as a whole. As pointed in [2], it not only makes the process inflexible to modify but also shields the process from audition and manipulation. Therefore, developing medical information system through a workflow-based approach seems to be a better solution. Since the radiology process is typical in healthcare, the following of this paper will introduce an approach to develop workflow-based information system in radiology department as a use case. Workflow management [3] is a state-of-the-art technology focusing on the process itself. Initially emerging as management research consideration of office automation, the workflow management technology has experienced several evolution stages, and has spawned various workflow management systems (WfMSs). Early workflow management systems appeared as document-driven systems automating office work processes. Then they evolved as embedded software components supporting business processes and standalone software suits providing workflow management functions [4,5]. The Workflow Management Coalition (WfMC) was founded in 1993 to clarify the confusions in WfMS product market by specifying the desired functions of workflow management system, and proposing a reference model of workflow management system and corresponding Workflow Application Programming Interfaces (Workflow APIs) [6]. Recently, workflow management system has evolved as the execution kernel of business process management (BPM) discipline [7]. As specified in the reference model, it can take control of designation, execution and analysis of processes. It is assumed that workflow management system is a promising complement of current healthcare information system development, as proposed in [8] and [2]. However, there exist certain challenges. First, the complex and flexible radiology workflow makes modeling difficult. Second, the heterogeneous systems in radiology environment have to be integrated in the perspective of workflow. In our study, a workflow-based radiology information system was designed and implemented to tackle these challenges, and the results of its deployment in the radiology department of Zhejiang Chinese Medicine Hospital in China were discussed. The remaining parts are organized as follows: “Method” describes the method of system design and implementation, “System implementation” presents the results after the system was deployed. Finally we conclude this paper with some remarks in “Conclusion”. Method Overview The workflow-based radiology information system takes WfMS as its backbone, which controls the system behavior
123
Int J CARS (2009) 4:509–516
according to a workflow model prescribing radiology processes. The first step of this approach is workflow modeling. It depicts the typical radiology process using a workflow modeling language. The second step is to design and implement the workflow-based system. Most tasks defined in the model are designed as independent components so that they could be easily controlled by the WfMS. The other tasks which stand for legacy systems such as modalities from different vendors need to be integrated. The existing non-workflowaware interfaces of these systems need to be transferred to workflow-aware ones. At runtime, a workflow management system interprets and executes the input workflow model to produce process instances; the newly designed components and the legacy systems could be controlled so as to enact the whole process. Finally, a workflow dashboard is designed to provide an integral process view of the reality for better process management. The modeling language and corresponding WfMS must be selected at the beginning. Petri Net is a mathematical modeling language widely used in describing distributed systems, and is deemed also proper for process modeling [9]. Compared to other process modeling languages, it has exact mathematical definition of execution semantics and hence well-formed theories in process analysis. A workflow language named Yet Another Workflow Language (YAWL) [10] is proposed as a variant of Petri Net and expresses workflow patterns [11] more explicitly. In this study YAWL language is selected as workflow modeling language and hence the supporting WfMS -YAWL system [12].
Workflow modeling Workflow modeling abstracts the tasks and their logical relations in radiology process. The modeling effort starts from requirement analysis, including some process statements with management rules or some drawings that illustrate routine workflows in the radiology domain. Then the statements and initial drawings are interpreted to workflow notations using model elements—YAWL semantic elements in this case. In this approach, the initial model got from the requirement analysis is taken as the basic one. It is used to guide the system development since the tasks inside could be taken as the counterpoints of the system components. After the system being deployed, the model may be dynamically adjusted to reflect the changing need. The result model is presented in Fig. 1. The dashed rectangle indicates the trunk of the routine radiology process, while the remainders represent the ad-hoc operations and dynamic decisions in the process. The process starts from the operation of examination scheduling or registering. A routing task with XOR-split is
Int J CARS (2009) 4:509–516
511
Fig. 1 The workflow model of radiology information system
used to separate the examinations that need scheduling from the ones that do not. A Deferred Choice pattern [11] is applied to represent the choice between the task of scheduling modification before the registration, and the task of registration directly. Once the Scheduling Modification task is started the Registration task is disabled until the former task is completed. Meanwhile, the scheduling information can be modified any time as long as the Registration task has not been started. After registration, another routing task with AND-split enables two possible routes, both of which are of Deferred Choice pattern. One of them indicates the choice between the task of Registration Modification before Examination and the task of Examination directly. The other indicates the choice between the task of Fee Modification and the task of Examination Canceling before it is completed. The sub-process within the dashed ellipse is denoted as a Cancelation set of Examination task, which means that all the tasks in the set will be disabled if the Examination task is completed. After examination, a XOR-split routing is used to differentiate the ordinary exams that can be reported after exam and the ones that need Post-Processing before Reporting. Postprocessing task is carried out by technicians to perform image processing such as image enhancement and reconstruction, while the Reporting task is done by the radiologist. After reporting, the task of Review will be started for senior ones to further review and approve the report. In case of problems, it will be returned to the reporting radiologists for modifications; otherwise, the report will be delivered to the referring physician and the whole process completes. YAWL shows its capability of explicitly representing flexible and complex healthcare workflow patterns. Without its semantic features, it is difficult to model the Deferred Choice
or Cancelation patterns, since the tasks inside the pattern may be enabled, disabled or even removed dynamically. Compared with a scrambled model by other language, it ends up in a concise and precise representation with YAWL.
System implementation General structure The system architecture of the workflow-based radiology information system is illustrated in Fig. 2. The Workflow Management System (WfMS) acts as the backbone in the architecture. It interprets and executes the established workflow model and controls all the components in the system to enact the whole process. The YAWL system clearly supports YAWL modeling language and is
Fig. 2 System structure of workflow-based radiology information system
123
512
open-sourced with a flexible and extensible service-oriented architecture. It has been selected as the WfMS in the design. The functionalities of the system are organized as several components corresponding to tasks. A Worklist Handler is designed as a task dispatcher of WfMS, which drives and controls these components in a de-coupled manner. Meanwhile, an Integration Engine is designed to integrate legacy systems such as acquisition modalities and post-processing workstations. Adapters may be used in case the interfaces are proprietary other than standard ones like Health Level 7(HL7) and Digital Images Communication of Medicine (DICOM). Finally, a Workflow Dashboard is developed as the monitoring tool to show the information of the running process instances and resources participating in radiology process. Worklist handler YAWL system provides Representational State Transfer (REST)-style web services as its interfaces to other components [13], and the Worklist Handler interacts with it through two kinds of services, the worklist service and the custom service. The worklist service, which is an instantiation of Interface 2 of the WfMC, allows the Worklist Handler to retrieve and manipulate work items, which means the instantiation of tasks [14], while the custom service [15], which is an instantiation of Interface 3 of the WfMC Workflow Reference Model [14], enables the Worklist Handler to be invoked by the WfMS. Through this interaction, the Worklist handler can retrieve or receive work items and dispatch them to the working queue of users or the automatically invoking system components. With a predefined organization model, it can also dispatch work item according to the departmental work planning strategies. For example, when a work item of Report task of CT examination is instantiated, the Worklist handler is invoked by WfMS to be aware of this new work item. Through the organization model, the Worklist handler can dispatch it to the working queue of the radiologist of CT examination on duty. The system components correspond to the tasks in the model one by one. In order to achieve the system flexibility, they are developed as atomic standalone modules so that it Is easy to attach or detach them from the Worklist Handler using a configuring tool. Two types of components are developed: manual task components and automatic task components. The work items of manual tasks will be listed in the user’s worklist page provided by the Worklist Handler, and the corresponding component can be invoked manually through selecting one work item in the page. To take the CT reporting as an example, the radiologist on duty can look up his work items when he opens the worklist page of the system. To start reporting, the radiologist selects one reporting work
123
Int J CARS (2009) 4:509–516
item and then the corresponding manual task component is invoked, so the radiologist can work on the component for structured reporting. The component will interact with the Worklist Handler through two interfaces. The “check-out” interface is called at the beginning of reporting, and the “check-in” interface is called after reporting is completed. Through these two interfaces the Worklist Handler knows the status of the work item, so that it can refresh the results appearing on the worklist page as soon as possible. The automatic task components are invoked by the Worklist Handler directly. When an automatic work item is enabled, the Worklist Handler locates its associated system component, “check-outs” the work item and passes it to the system component through the component’s invocation interface, so that the component can be processed automatically. After processing completes, the component will “check-out” the work item to the Worklist Handler to inform about the status. For example, a Queuing task is implemented as an automatic system component. After exam registration, the Queuing work item is enabled. At this time, according to configuration, the Worklist Handler finds the Queuing system component which is automatically running at a server, and checks out the work item from the WfMS. Then the Queuing component is invoked and the work item is passed to it from the Worklist Handler by calling its “invocation” interface. It is in this component that the actual task function is executed, where the exam is allocated to a specific exam room with a queue number. After that the work item is checked in back to the Worklist Handler, and finally checked in to the WfMS. Through this mechanism, the system is made very extensible for future changing needs. Appending new tasks or adjusting current tasks can be done by simply replacing the components, while the workflow re-engineering can be performed quickly by adjusting the workflow model. Integration engine Apart from the RIS components implemented with workflow-aware interfaces, there are some heterogeneous systems that fulfill the tasks in radiology workflow but do not comply with the workflow-based specification, such as modalities that perform examination tasks and post-processing workstations that perform image processing tasks such as image reconstruction in CT exams. These tasks should also be integrated to the workflow since they may have relationships with other tasks. For example, the post-processing task must be performed before reporting. The Integration Engine is implemented to achieve this goal. Although legacy systems do not support workflow-aware interfaces, most of them use message interfaces complying with IHE profiles and the data exchange standards such as HL7 and DICOM. The Integration Engine is designed as
Int J CARS (2009) 4:509–516
Fig. 3 DICOM Adapter implementation of Examination task
513
are redirected to the engine. When it receives the storage messages, the corresponding Post-Processing task is checked out and immediately checked in to the WfMS through the Worklist Handler. Then the messages are routed to PACS. In this manner, the status of the post-processing task is acknowledged by the WfMS and the Reporting task could be enabled as soon as the post-processing completes. Though the above example only deals with the system integration based on DICOM standards, the Integration Engine could also integrate systems with appropriative interfaces through developing specific adapters to transfer them to standard ones. Monitoring and controlling tools
a bridge between workflow-aware interfaces and IHE-based ones, as shown in Fig. 2. For those with non-IHE-based interfaces, specific adapters will be used to translate them to IHEbased ones. To take the modality as an example, it has delegated the Examination task in radiology workflow, and uses message interfaces in compliance with the IHE SWF profile. The modality implements the Acquisition Modality actor as specified in IHE profiles [16–19]. As shown in Fig. 3, The Integration Engine receives messages from modality when the modality queries Modality Work List (MWL), when the modality Performed Procedure Step (SPS) is executed and when it is completed. Then it transfers these messages to workflow-aware interfaces of “Get Task Worklist”, “Start Work Item” and “Complete Work Item”, respectively, to the Worklist Handler, so that the Examination task could be enacted. The post-processing workstation only sends data-oriented DICOM Storage messages to PACS for image archiving when post-processing is completed. To integrate this task, the engine is implemented as an interceptor between this system and PACS, where DICOM messages originally sent to PACS
Since the status of each work item is recorded in the WfMS, it is feasible to provide an integral view of what is going on in the department. A workflow dashboard is helpful to monitor processes running in the radiology environment. The dashboard provides comprehensive knowledge of processes and resources from different but related perspectives. Figure 4 shows a screenshot of the dashboard about an integral view for one process instance. In this figure, the examination process of the specific patient is depicted, where the currently proceeding work item is marked with highlight. The waiting time and performing time of each task and its associated performers are also presented in the middle of the view. Moreover, integrated patient-and-exam information is shown in the bottom, enabling quick and comprehensive knowledge acquisition about this specific patient. Results The system has been deployed in the radiology department of Zhejiang Chinese Medicine Hospital. Before the system has been deployed there were several isolated systems in
Fig. 4 Screenshot of the workflow dashboard
123
514
Int J CARS (2009) 4:509–516
Fig. 5 The gained workflow and associated systems
the department, such as modalities and reporting systems, supporting only fractions of their routine work. Under such condition the departmental workflow is neither complete nor smooth. The workflow-based RIS supports most of the routine works, and the legacy isolated systems are also integrated with the workflow. It has streamlined and automated the processes of the radiology department as a result. Figure 5 illustrates the results of system deployment, along with the corresponding workflow model. In the figure, several screenshots of the desktop user interfaces of the system components are presented. These components correspond to Registration task, Fee modification task and Report task, respectively. Based on workflow management technology, the system shows the following advantages compared to traditional RIS.
• System flexibility. By de-coupling and assembling the system components according to the workflow model, the system achieved the capability of flexibility. The case of introducing Queuing task into the model is an example to show system flexibility. Before the system was deployed, the hospital was experiencing the problems caused by a large number of registered patients waiting for examination. In order to channelize the examinations, an automatic task of Queuing was introduced into the model, as the dashed ellipse shows in Fig. 6. Correspondingly,
123
the queuing application was implemented as a system component and was deployed and attached to Worklist Handler. When the work item of Queuing task is enabled, the system component is invoked by the Worklist Handler to handle the queue of patients for different examination rooms. Hence, the workflow-based RIS could be easily adjusted to this changing need. • Process improvement. With the workflow-based radiology information system, the department benefits not only from the automation of its process with the system components controlled according to the workflow model, but also from being provided with comprehensive knowledge of what is going on in this department by the workflow dashboard. With the status of the whole process known at a glance, the exceptions can be identified more easily and handled instantly. It is reported from the questionnaire of the radiology department that since the deployment of this workflow-based system, the examination delays caused by exception have been dramatically reduced because of timely interventions. • Workflow-aware integration. Traditional workflow integration though IHE-based approach mainly copes with the connection among heterogeneous systems based on the guidelines from some IHE integration profiles. However, the one-to-one connection introduces complexity and inflexibility when the integration scale expands. Comparatively, the workflow-based approach integrates
Int J CARS (2009) 4:509–516
515
Fig. 6 Adjustment of workflow when introducing queuing
system components at process level, and the many-toone mechanism makes system more flexible, since all components are integrated into the workflow management system rather than integrated with each other. In addition, to integrate legacy systems as tasks in the process gives way to process monitoring and improvement from an integral perspective.
Conclusion In this paper, a workflow-based approach of radiology information system has been presented. Radiology process is typical of the routine departmental process in healthcare. The results of this case study have shown that workflow technology could bring system flexibility and departmental process management, and can be deemed as starting point of healthcare process management endeavor. The modeling effort in the paper is intuitively dependent on mostly manual requirement collection and analysis. Since there do exist workflow solutions in healthcare such as IHE integration profiles that describe typical radiology workflow, a more systematic modeling method is required, which takes IHE profiles as the initial input for the comprehensive model. Although radiology process exhibits some flexible features, most parts are in well-formed structure. As the scope extends to multiple departments or even the whole hospital, this topic becomes more complex and needs further investigations. Acknowledgments The authors wish to thank W.M.P van der Aalst, Thomas Wendler for their effort in supporting this project named “BrainBridge-Project”. They contributed excellent methods and provided good tools to the project. Thanks also go to Chenhui Zhao and Zhaodan Chen
for their joint efforts in completing this project. Last but not least, the authors would also say thanks to Maosheng Xu, Shiwei Wang and other supporting staffs in Zhejiang Chinese Traditional Medicine Hospital.
References 1. HIMSS/RSNA (2004) IHE Radiology Technical Framework white Paper 2004–2005 For Trial Implementation Departmental Workflow Revision 2.4 2. Wendler T, Loef C (2001) Workflow management—integration technology for efficient radiology. Medicamundi 45:41–48 3. WfMC (1999) Terminology & Glossary. Document Number WFMC-TC-1011 4. Georgakopoulos D, Hornick M, Sheth A (1995) An overview of workflow management: from process modeling to workflow automation infrastructure. J Distrib Parallel Datab Syst 3:119–153. doi:10.1007/BF01277643 5. van der Aalst WMP, van Hee K (2002) Workflow management models, methods, and systems. MIT Press, Cambridge 6. WfMC (1995) The Workflow Reference Model. Document Number TC00-1003 7. van der Aalst WMP, ter Hofstede AHM, Weske M (2003) Business process management: a survey. In: Business process management, pp 1–12 8. Graber S (1997) The impact of Workflow Management Systems on the design of hospital Information Systems. AMIA ANNUAL FALL SYMPOSIUM 9. Van der Aalst WMP (1998) The application of Petri Nets to Workflow Management. J Circuits Syst Comput 8:21–66 10. van der Aalst WMP, ter Hofstede AHM (2005) YAWL: yet another workflow language. Inf Syst 30(4):245–275. doi:10.1016/j.is.2004. 02.002 11. van der Aalst WMP, ter Hofstede AHM (2002) Workflow patterns: on the expressive power of (Petri-net-based) workflow languages. In: Jensen K (ed) Proceedings of the fourth workshop on the practical use of coloured petri nets and CPN Tools (CPN 2002), vol 560. Aarhus, Denmark, pp 1–20 12. van der Aalst WMP, Aldred L, Dumas M, ter Hofstede AHM (2004) Design and implementation of the YAWL system. In: Proceedings
123
516 of the 16th international conference on advanced information systems engineering (CAiSE 04). Riga, Latvia 13. Aldred L (2006) YAWL core classes cheat sheet: a short architectural introduction. In: Online Documents of YAWL system. http:// yawlfoundation.org/yawldocs/YAWLArchitectsCheatSheet_ Beta7.pdf. Accessed at 4 17 2009 14. WfMC (1998) Workflow management application programming interface (Interface 2&3) Specification. Document Number WFMC-TC-1009 15. Dumas M, Fjellheim T, Aldred L (2006) Connecting custom services to the YAWL engine. In: Online documents of YAWL system. http://yawlfoundation.org/yawldocs/CustomWebServices_ Beta7.pdf. Accessed at 4 17 2009
123
Int J CARS (2009) 4:509–516 16. HIMSS/RSNA(2002) Integrating the Healthcare Enterprise IHE Technical Framework, vol I Integration Profiles Revision 5.3 17. HIMSS/RSNA(2005) Integrating the healthcare enterprise IHE technical framework, vol II Transactions Revision 6.0 18. HIMSS/RSNA(2005) Integrating the Healthcare Enterprise IHE Technical Framework, vol III Transactions (Continued) Revision 6.0 19. HIMSS/RSNA (2005) Integrating the Healthcare Enterprise IHE Technical Framework, vol IV National Extensions Revision 6.0
