Intelligent Image Management in a Distributed

Intelligent Image Management in a Distributed PACS and Telemedicine Environment M. Tsiknakis1, D. Katehakis1, S. Orphanoudakis1,2

1

Institute of Computer Science, Foundation for Research and Technology - Hellas, P.O. Box 1385, 711 10 Heraklion, Greece 2

Department of Computer Science, University of Crete, Heraklion, Greece Tel.: +30-81-391600, Fax.: +30-81-391601

IEEE Communications Magazine vol. 34(7), pp. 36-45 1996

Intelligent Image Management in a Distributed PACS and Telemedicine Environment M. Tsiknakis1, D. Katehakis1, S. Orphanoudakis1,2 1

Institute of Computer Science, Foundation for Research and Technology - Hellas, Heraklion, Crete, Greece.

2

Department of Computer Science, University of Crete, Heraklion, Crete, Greece.

Abstract Advances in information technology in the past decade have resulted in a proliferation of clinical information systems dedicated to individual user groups and clinical functional areas. This, in turn, has led to the need for hospital-wide management and integration of information, and has triggered major efforts towards the development of integrated hospital information systems. A framework for developing an integrated regional health telematics system is presented, based on the functional and data integration of federated autonomous information systems. The concept of the patient meta-record is introduced, as a central element in achieving integration in terms of content, structure, and access to information resources. Intelligent image management at the level of either a single health care provider (hospital) or at the level of a regional health telematics network is addressed and is achieved through event and model-driven strategies enabled by a distributed hierarchical storage management system. Intelligence, in this context, refers to methods of efficient and effective real-time resource management in the timely delivery of health care. Keywords: computerized patient record, PACS, intelligent image management, telemedicine, hierarchical storage management, patient meta-record.

1.

Introduction

The health care organizational structure is naturally distributed, consisting of a geographical spread of medical centers in a hierarchy of regional hospitals down to individual general practitioners. The objective of this structure is to offer comprehensive medical care at a local and regional level with continuity across different levels of the hierarchy. Although each medical center is autonomous and devoted to the delivery of a particular set of services, the desirable continuity of care requires that different medical centers, offering complementary services or different levels of expertise, exchange relevant patient data and operate in a cooperative working environment. The sharing of information resources is generally accepted as the key to substantial improvements in productivity and to better quality of service [[23] ]. The diversity of hospital organizations, the complexity of clinical protocols and procedures, as well as the different preferences of various user groups make it extremely difficult for a single monolithic information system to effectively serve the needs of an entire health care organizational structure. As a consequence, users must be allowed to select the applications most suitable for their needs and requirements. In addition, a number of applications are already available on the market that address specific aspects of the health care structure. Thus, information and -1 -

telecommunications systems must primarily provide the infrastructure to permit the effective integration of distributed and heterogeneous components, ensuring overall integrity in terms of functional and information interworking. The physically distributed resources of the health care sector and the diverse requirements of different medical facilities and clinical departments require that specialized autonomous information systems are used to support different functionalities, while they interact transparently to the user as a federation of autonomous systems. This approach to developing and managing regional health telematics applications ensures the transfer and integration of consistent information throughout a network of health care facilities, without imposing constraints on the operation of individual units. From the technological viewpoint, the adoption of standards and open architectures represents the only viable solution to the problem of integrating diverse system components through an incremental approach, consistent with evolving requirements, while securing the investments already made. The definition and adoption of standards, not only as these concern the technical aspects, but mainly with respect to the functional interfaces between the various system components, represents a fundamental requirement which must be satisfied before different applications, developed by different suppliers, can be integrated and used effectively in health care. The purpose of this paper is to present the key design principles of a regional health telematics network and, specifically, to address the issue of intelligent image management in such a system. Section 2 introduces the patient meta-record concept as a means of integration of all patient related data and the activity concept as a formalism capable of modeling a broad spectrum of healthcare processes. Section 3 presents the architecture of a hospital information system, integrated, through the patient metarecord. Section 4 presents the architecture of TelePACS 2.0, an extensible image management system, and its efficient and intelligent management and communication of images, achieved through the implementation of a distributed hierarchical storage management system. Section 0 presents some advanced added-value telematic services to be implemented within the framework of the regional health telematics network of Crete. The final section concludes this phase of our R&D work and presents our future work. A discussion of work to date and future work is included in Section 6.

2.

The Patient Meta-Record Concept

The integration of information and knowledge from different sources is increasingly becoming a key to better quality of care. Achieving this integration, however, is a challenging problem mainly because the logic, knowledge and data structures used in various systems are complex and often incompatible. Many tasks require a large volume of data processing and communications across heterogeneous and distributed environments [[8] ]. Today, the problem of harnessing disparate information resources remains one of the most intensely contested information technology issue in the international research arena [[14] ]. A promising approach to this integration problem is to gain control of the organization’s information resources at a meta-data level, while allowing autonomy of individual subsystems at the data instance level [[11] ]. The objective of the meta-2 -

Figure 1. Multi-level patient record integration at a regional level.

Health Care Region

Regional PMR

Hospital Institutional PMR

Departmental System

...

Departmental System

...

Institutional PMR

Institutional PMR

Departmental System

... Physical CPR Segments

database model is to achieve enterprise information integration over distributed and potentially heterogeneous systems while allowing these systems to operate independently and concurrently [[12] ]. In order to turn a large, heterogeneous and distributed set of Computerized Patient Record (CPR) components into a seemingly integrated and homogeneous patient record, the concept of the Patient Meta-Record (PMR) has been introduced [[18] , [21] ]. The PMR concept permits the integration of CPR components, and also provides aggregation and navigation facilities (which, in turn, can be fully configured and customized) in order to assist in the appropriately personalized utilization of the CPR by the respective functional units or health care providers. The PMR manages, at a single logical point, references to all of the physical information related to a patient throughout the Integrated Hospital regardless of where such information may reside. Therefore, it is an indexing system, providing access to all stored data for a particular patient. Physical CPR segments, distributed among several autonomous departmental information systems are integrated by means of the institutional PMR. This PMR notion can be extended at a regional, national or even transnational level (Figure 1). The PMR may be seen as a central element in the provision of integrated computer assisted health care. The general approach relies (i) on the activity concept [[3] ] associated with the requester-performer paradigm (Figure 2), since act-based information systems allow for the gradual utilization of software agents to achieve specific goals [[9] ], and (ii) on the semantic model whose exploitation ensures the overall integrity of the information system.

-3 -

Figure 2. The requester-performer paradigm.

Requester

Performer Header Request Form

MRI

Header

The requester-performer paradigm supports all medical activities in the hospital, since a single functional unit’s information system may simultaneously play both the requester and the performer role.

report

2.1. The Activity Concept In developing the architecture of an integrated hospital information system, our objective is to introduce a formalism that is capable of modeling a broad spectrum of structured and unstructured healthcare tasks, in a rigorous and comprehensive manner. The Activity Manager System (AMS) formalism [[3] ] is particularly suitable as a representation tool to describe the characteristics of the complex hospital tasks. Acts and intelligent act management are central concepts in our approach. An act is any professional activity, which influences a patient’s status or augments the knowledge about a patient’s status. Acts have a life-cycle consisting of different successive states: demanded, established, accepted, canceled, performed, etc. As analyzed in [[3] ], three types of information are encapsulated by the activity concept: •

the start-state, describing the precondition for performing the activity (e.g. authorization of user x to request examination y),

•

the terminal-state, describing the effect (or the reached goal) caused by the execution of the act, and

•

the body, describing the way the activity is to be performed.

Figure 3 shows the two different types of an actual medical act, as well as its structure. Acts can be distinguished into (1) terminal or elementary activities whose bodies are terminal actions not capable of being decomposed further, and (2) compound or complex activities whose bodies are sequential descriptions capable of further decomposition. A set of functionalities enables end-users to access the medical content of the actual patient record according to authorization and confidentiality rules, and to browse it efficiently, independently of data location. The search criteria may be based on -4 -

Figure 3. Structure of an actual medical act.

Elementary Medical Act • Act Header Header

Medical Content

Patient ID, Act ID, Class of Act, Requester ID, Performer ID, Date, Current Status, History, ...

• Medical Content

Compound Medical Act Decomposition of a compound act into its components.

Image, Bio-signal, Lab Results, Report, ...

CT check up

ECG

information contained in an act’s header or in its medical content. A search may also exploit the semantic model, which provides information on the structure of an act, its possible links to other acts, etc.

2.2. The Semantic Model We base our PMR on the Semantic Indexing System (SIS) [[7] ]. The SIS is a tool for describing and documenting large evolving varieties of highly interrelated data, concepts and complex relationships. The SIS consists of a persistent storage mechanism based on an object-oriented semantic network data model, and a generic interactive user interface to insert and retrieve information in various ways. The SIS offers significantly richer referencing mechanisms than relational or ordinary objectoriented systems. Together with the very high query speed, these mechanisms allow the data and schema to be kept free of redundancies. An interesting feature of the SIS is that its data entry mechanism, query system and user interface treat data and schema in a uniform manner. The SIS will also be employed for the development of a context mediator in the sense described in [[22] ]. The context mediator is an agent that directs the exchange of values, from one autonomous information system to another, using domain knowledge, and provides services such as local-to-global schema translation, intersystem attribute mapping, and consistency checking. All data exchange goes through the context mediator. The context mediator in our approach is an example of the mediator concept of Wiederhold [[26] ]. As we can see in Figure 4, each subsystem provides pairs of conversion functions for the two-way mapping between its local values and the global values, as used by the PMR. This way subsystems export communications methods and allow for dynamic schema definition and modification at runtime. Moreover, the schema is visible to the user, supporting the explanation and exploration of its structure.

-5 -

Figure 4. Conversion routing implemented by means of the SIS.

SIS

HL7

LIS

HL7/SCP-ECG

PMR

mediator

,&8

DICOM 3.0

PACS Our approach enforces an important trend towards decoupling of applications from other applications and data sources, thus simplifying the interface among system components [[22] ].

3.

Architecture of an Information System

Integrated

Hospital

A hospital is by far the most complex organization in the health care hierarchy. Therefore, a primary objective in developing a regional health telematics network is to design and develop an Integrated Hospital Information System (IHIS) capable of effectively and efficiently supporting all patient related clinical processes within a hospital. The IHIS development is based on the definition and implementation of an open architecture where the individual modules: •

are autonomous and self-consistent, supporting specific functional units;

•

interwork through stable, public interfaces;

•

are configurable, able to operate in a distributed environment, and can adapt to the specific requirements and characteristics of an individual organization.

The key characteristic of the proposed architecture is its layered structure, based on four fundamental layers, as shown in Figure 5. •

The Technological Platform is responsible for the transparent integration of distributed heterogeneous technological environments.

•

The Generic Communication Environment (GCE) is responsible for ensuring the transparency of the actual technological (both hardware and -6 -

Figure 5. Generic architecture of an integrated hospital information system.

...

Nursing

Radiology

Administration

... User-Oriented Applications

Authority & Authentication

Mediation

PMR

Workflow Management

Distributed Application Environment (DAE) Generic Communication Environment (GCE)

Local Network

Technological Platform

Local Network

software) configuration. It permits the interaction of individual system modules through mechanisms which are independent of the adopted network products and physical locations. It provides a flexible connection-based communication mechanism that isolates the application from details of lower-level communication protocols.

4.

•

The Distributed Application Environment (DAE) represents the central element of the whole architecture and is responsible for the functional and data integration of the system, according to the organizational, logistic and clinical requirements of the health care institution. To accomplish this goal, the DAE consists of a set of services capable of supporting the interaction of the applications from individual units, the management of data and the execution of activities relevant to the whole health care institution. All these services are facilitated through the use of SIS.

•

The User-Oriented Applications are autonomous and provide support to specific activities of the various units of a health care organization.

Departmental Information Systems

As stated in previous sections an important requirement in the development of an IHIS is the development of departmental information systems, customized for addressing the specific needs of different hospital units. In the following sections, one such autonomous information system will be presented in terms of its requirements and architecture, as well as the advanced features implemented to allow intelligent performance. The system to be described is the PACS, which is primarily associated with diagnostic imaging. By being open and extensible, the proposed distributed and -7 -

modular architecture provides the basis for developing not only a hospital wide PACS, but ultimately an integrated regional Image Management and Communication System (IMACS) covering all of medical imaging including microscopy, endoscopy, etc.

4.1. The Diagnostic Process

Imaging

and

Communication

Medical images are one of the most important sources of diagnostic information. The information flow involved, from the actual request for a radiological examination by the referring physician to obtaining the radiologist’s report, requires considerable movement of people, paper, and film with all of the attendant possibilities for delay and loss. The actual acquisition of diagnostic images is only one element in the chain. The most important elements from the viewpoint of patient care are the following : •

the referral process, in which the attending physician requests an examination,

•

the acquisition process, in which diagnostic imaging takes place,

•

the reporting process, in which the radiologist reads the films and reports his findings,

•

the image management process, in which diagnostic images and other patient related data are properly stored for future reference, and

•

the consultation process, in which the radiologist and one or more attending physicians review the case.

4.2. PACS Storage and Communication Requirements In designing a PACS network architecture, selecting the appropriate type of network and developing adequate communication mechanisms, the PACS image communication pattern has to be determined. Estimates of the anticipated image communication load and its temporal patterns can be found in [[17] ], where representative figures, derived from observations and statistical evaluations at several hospitals, are presented. A first estimate of the mean image data transmission load was based on the total image production and image consumption at reporting sites versus time. For example, the radiology department of the Aachen University Hospital (1500 beds) produces about 20 Tbytes of image data per year. Similarly, the annual volume of images generated in a medium-size hospital of 600 beds is reported as 1.15 million images or 1.9 Tbytes in [[15] ]. In order for PACS to be effective in its role as a added-value service for the improvement of patient care, image data must be delivered to their destination with negligible delay. An image latency time of 1-2 seconds or data rates of 65-335 Mbps are considered adequate for typical clinical situations [[13] ]. In a more recent publication on a PACS testbed installation [[27] ], the effective transfer rate was reported to be 3 Mbps with an Ethernet (disk-to-disk) and 14.4 Mbps with an FDDI network (RAM-to-RAM). Therefore, FDDI with its 100 Mbps physical data rate may also result in a communication bottleneck. Intelligent image management -8 -

[[16] ] provides solutions to the problem of timely delivery of image data in a PACS environment, given such a communication bottleneck. In this context, intelligence or intelligent behavior is characteristic of a system capable of achieving near real-time performance by selective exploitation of tools and strategies, given limited communication and computational resources.

4.3. The TelePACS Image Management System The Medical Information Systems Laboratory of the Institute of Computer Science (ICS), Foundation for Research and Technology - Hellas (FORTH), has designed and implemented TelePACS 1.0, a system for the acquisition, management, and communication of medical images and related patient information (see Figure 6). This system has been installed at the Radiology Clinic of the University of Athens, various imaging clinics of the University Hospital in Heraklion, Crete and the Venizelion Regional Hospital, also in Heraklion. TelePACS 1.0 forms the basis of a pilot telemedicine network connecting these three hospitals, which is currently undergoing clinical evaluation. In this Section, we present the architecture of TelePACS 2.0 [[21] ], which is currently being implemented and supports the intelligent management of images at a local (hospital) and regional level.

Figure 6. Typical screen from TelePACS 1.0.

-9 -

4.3.1. TelePACS 2.0 Servers The architecture of TelePACS 2.0 is open, distributed and modular, based on a large number of file servers which are distinguished according to their function. The TelePACS 2.0 servers and their corresponding functions are: •

The Acquisition Server, which is connected to medical imaging modalities and is responsible for the acquisition of multimedia medical data. A single acquisition server can be configured to serve one or multiple imaging modalities.

•

The Archive Server, which manages the permanent storage of images in a transparent hierarchical storage structure. One such server is normally present in each hospital department.

•

The Central Hospital Server, that holds meta-information on patient data and acts as a gateway among heterogeneous information systems. This is where the context resolution takes place. One such server is normally present in each hospital.

•

The Departmental Server, which is the front end to the intra-hospital network. One per department or cluster of servers.

•

The Display Server, which is connected to a set of display workstations with advanced Graphical User Interfaces (GUI). Caches patient related data. Schedules examination requests on acquisition servers.

•

The Key Server, which is responsible for encryption and authentication. One per Hospital.

•

The Name Server, which holds the directory of both the local TelePACS 2.0 clusters and all remote TelePACS 2.0 systems. Manages both incoming and outgoing communication. Each TelePACS 2.0 System has one primary name server.

4.3.2. TelePACS 2.0 Clusters TelePACS 2.0 clusters form the basic components of the system in the intra-hospital environment. Each cluster usually covers a hospital department and consists of more than one servers. Departmental servers are responsible for cluster coordination. Thus, all intra-hospital communication is managed by departmental servers (placement of service requests, monitoring of outstanding requests, etc.). 4.3.3. TelePACS 2.0 Systems TelePACS 2.0 systems form the nodes of the regional medical image management system and are intended for autonomous hospital environments. Inter-hospital communications are managed by the central hospital server that holds the most important information (e.g. PMR, local TelePACS addresses, availability of resources, etc.).

-10 -

4.4. Work Flow Scenarios The AMS formalism as a theoretical tool for the development of knowledge-based systems has been demonstrated in [[1] ], where an office model, using office operators (e.g. request, approve, perform, answer, circulate) as basic building blocks to construct office procedures, has been presented. These procedures can then be defined by creating activities, which are then implemented by using these operators and the requester-performer matching mechanism. The AMS formalism together with the requester-performer paradigm was employed for analyzing and modeling several medical acts to be supported by TelePACS 2.0. The description of some acts, and the resulting information flow among information systems involved follows. 4.4.1. A Single Examination Request Act The referring physician places a request for a Magnetic Resonance Imaging (MRI) examination at his display station. This exam request is forwarded by the local display server to the departmental server (Figure 7), which in turn transmits it to the central hospital server. The context mediator agent, part of the DAE (see Section 3), by utilizing its domain knowledge and the information encapsulated by the appropriate activity concept in terms of start-state, terminal-state and body of the activity, undertakes the task of the overall coordination of its execution. This involves request translation (e.g. HL7 to DICOM) and routing to the appropriate information system, gathering of information related to the state of the execution, translating and transmitting the terminal-state of the activity back to the requesting information system.

Figure 7. A specific example for a single examination request act.

Central Hospital Server 3

7 forward exam request

acknowledge exam scheduling

4

Service Department

8

Archive Server

5a 5b


6 acknowledge exam scheduling

2 9

place exam forward exam request request 1

Display Server

MRI Display Server

acknowledge data prefetching

Radiology

Requesting Department

Departmental Server

Departmental Server forward exam request request data prefetching 5

forward exam request

Pathology

-11 -

10


referring physician

Figure 8. Image acquisition and archiving during a radiological examination execution act.

Central Hospital Server Service Department (Acquisition Act)

place image capture request

MRI Display Server forward image capture request

acknowledge image capture

7

3

acknowledge image archiving


forward image archive request 4

acknowledge image capture


Departmental 4 Server

MRI Acquisition Server

6

forward image archive request 5

Radiology (Image Acquisition Act)


Display Server

8

Service Department (Archiving Act)

Departmental Server 2b forward image archive request

2a

Archive Server

Pathology

2 forward image MRI archive request place image archive request


9

operator

1

MRI Display Server

10 acknowledge image archiving

Radiology (Archiving Act)

4.4.2. The Examination Execution Act Once the context mediator transmits an examination request to the Radiology Information System (RIS) a corresponding examination execution act is initiated locally with RIS playing both the requester and performer roles simultaneously. The examination execution act is a compound activity, whose body contains scheduling, image acquisition, reporting and archiving as shown in Figure 8. Also the body of the examination execution act contains the pre-fetching act (dotted lines in Figure 7). During the execution of this act all previous, relevant, patient exams are moved up the storage pyramid and are dispatched to the appropriate display servers, so they are present locally when needed. 4.4.3. The Compound Examination Request Act The referring physician places a request for study including a Computed Tomogram (CT) and a laboratory examination for his/her patient (Figure 9). Once again the context mediator agent undertakes the task of the overall coordination of its execution. In addition to its response to a single examination request, a compound act requires request decomposition to its components and result composition into a single terminalstate.

-12 -

Figure 9. The scenario of a compound act decomposition.

Central Hospital Server forward exam request acknowledge exam scheduling

Service Department 1



Archive Server


acknowledge Departmental exam scheduling

Server

Departmental Server forward exam request request data prefetching



forward exam place exam request request

Display Server



CT Display Server


referring physician

Pathology acknowledge exam scheduling

Radiology Service Department 2

Departmental Server forward exam request request data prefetching

Archive Server


Bioch. Analyzer Display Server


LIS

4.5. Distributed Hierarchical Storage Management The IMACS system is in essence a content provider handling real-time storage, retrieval and communication of multimedia data. As such, it is viewed as an active information system. As stated in [[5] ] an active image information system should meet the following general user requirement of: timely delivery and easy accessibility of image and associated information, at a resolution appropriate for the intended task(s) [[2] ]. Thus, the ultimate performance objective for an IMACS system is Images Where Needed When Needed. Given the storage and communication requirements of an IMACS system (see Section 4.2) intelligent image management strategies are necessary [[4] ] for the achievement of the performance objective stated above. Efficient storage management, which includes pre-loading and pre-fetching, represents such a strategy. The Distributed Hierarchical Storage Management (DHSM) [[24] ] module has been developed for the implementation of such a strategy.

-13 -

4.5.1. Hierarchical Storage Management The key characteristic of storage media necessitating the introduction of a storage hierarchy are access, speed, capacity and cost [[6] ]. The overall objective in introducing a storage hierarchy is to provide average access speed almost as fast as the fastest level, the High Speed Buffer (HSB), with an average cost per bit of on-line data almost as low as the least expensive level, the Direct Access Storage Device (DASD). This performance requirement can only be achieved if the majority of on-line data are resident on DASD while almost all of the Central Processing Unit’s (CPU) storage accesses are satisfied by high-speed buffers. Figure 10 shows such a storage Figure 10. The hierarchical storage management pyramid with respect to the physical medium.

Large Storage Capacity

User

Short Access Time

HSB

DASD

pyramid. 4.5.2. Image Management Strategies The DHSM allows for the intelligent management of storage, according to pre-defined models (model-driven) or in response to trigger events (event-driven). Model-Driven Image Management: Image data is automatically moved downward according to a Least Recently Used (LRU) algorithm and the corresponding settings (e.g. a third of the HSB capacity must always be empty to satisfy requirements for event driven upward migration of required data). Event-Driven Image Management: As stated in Section 0, an examination execution act initiates the prefetching act which has as a precondition the successful scheduling of the requested examination. The DHSM prefetching algorithm identifies all previous patient data managed by the IMACS system and transfers them to the appropriate display server, to be available during the reporting phase of the process together with the currently generated examination image data. Additionally, the DHSM pre-fetching algorithm not only retrieves past examinations into the highest storage level of the archive server but also dispatches them to distributed caches at the corresponding display servers.

-14 -

Figure 11. Data sharing through the use of the DHSM module. Site 1

Site 2

Application 1 (shares its data with site 2)



RDBMS 1

RDBMS 2

RDBMS 3

dhsm-deamon

Site 3

dhsm-deamon

dhsm-deamon

Cache of A1

Table A1

Hierarchical Storage Management

Cache of A3

Cache of A2

Table A2

Table A3



4.5.3. The DHSM System Architecture The DHSM system introduces an additional layer of cache memory on disk between the application and the information managed by distant, slow-to-access, archive servers (Table 1). Upward and downward migration through the storage hierarchy is based on reference activities. The management approach is based on an LRU algorithm for replacing an inactive data element by a more active one. The element replaced is moved downward in the hierarchy, while the requested element rises to the top of the hierarchy. Figure 11 illustrates the proposed architecture through a possible scenario of three applications running in three different locations (e.g. departments) and sharing images of the same modality, stored in tables A1, A2, and A3. The role of the dhsmdeamon is to maintain the loose-consistency among the distributed caches.

Table 1. The DHSM pyramid.

Layers of Storage Management User Application Cache Storage Medium

5.

Location

Maximum Capacity

Display Server Limited by the size of RAM Local Archive Server A fraction of the total locally kept data Remote Archive Server Practically unlimited

Telematic Services

Health care is an important telematics application domain in the emerging information society. In recent years, we have all been witnesses to the gradual transformation of health informatics into health telematics, a process which continues. For this transformation to be successful a strategy is needed for the creation of an integrated -15 -

health care information infrastructure. A fundamental problem for the establishment of a scaleable regional health telematics network is the development of an architecture and tools for the integration of specialized autonomous applications that, together with a shared patient record, will support the interoperability of functions and services within a health care institution, the interconnection of different institutions, and the intelligent management of medical data within such a network. With this in mind the Medical Information Systems Laboratory of ICS - FORTH is developing a regional health telematics network based on an open, expandable and extensible architecture with various added-value autonomous systems supporting clinical decision making, medical training, and clinical research, which will eventually extend the functionalities and services supported. Examples of such subsystems which are currently at various stages of development are: a computer supported conferencing application, an advanced image processing module and the Image Indexing by Content (I2C) system described below. Synchronous teleconsultation is possible via a Computer Supported Cooperative Work (CSCW) application [[25] ], which allows interactive real-time cooperation among several conference participants. This application provides a shared workspace for the display, discussion and annotation of multimedia medical data which is multicasted to conference participants prior to the consultation session. A replicated architecture and techniques for caching, data compression and screening of events allow its successful operation over heterogeneous, relatively slow networks. Version 1.0 of TelePACS (see Section 4.3) has already been integrated with the I2C system [[20] ], an information system providing an object oriented environment for the indexing and retrieval of medical images by pictorial content (Figure 12). The I2C architecture integrates a set of tools and algorithms for the extraction, indexing and storage of image descriptions. The interoperability of the two systems provides addedvalue capabilities to both systems, thus creating a powerful environment for clinical, educational and research activities. I2C is currently being extended to I2Cnet [[19] ], so that it will operate over the evolving regional health telematics network. I2Cnet is a network of image description servers based on I2C, which operate over the World Wide Web. Through the WWW browser, I2Cnet users are provided with reliable, network transparent content-based access to a geographically distributed image collections maintained by I2C servers. The architecture of I2Cnet brought out several new issues and challenges. Peer-to-peer server communication, agent based computing, query formulation, incremental query modification, query decomposition and result composition are but a few of them. A Pre-Hospital Health Emergency Management System (PHEMS) is also being developed as a distributed real time system. The system allows prompt and efficient management of the response to a health emergency and the efficient mobilization of the operational resources at a regional level. Functional integration of the basic health telematics system, through the PMR, with the PHEMS allows the efficient transmission of the physical patient record contents to relevant health care providers and/or experts, as required in intelligently and efficiently managing health emergencies.

-16 -

Figure 12. The I2C user interface.

6.

Discussion

The ultimate goal of our R & D efforts is to design and develop an integrated health telematics network in the region of Crete (Figure 13), by developing and validating telematic services connecting health care professionals at all levels of care. A fundamental problem for the establishment of such a network is the development of an architecture and tools for the integration of specialized autonomous applications, that together with a shared patient record will support interoperability of function and services within the health care institution, the interconnection amongst institutions, and the intelligent management of medical data within such an integrated network. This paper presents the PMR concept as a mean for integrating a large, heterogeneous and distributed set of CPR components (direct observations) into an unified and homogeneous patient record. This allows for the medical record of a patient to be viewed in a variety of different ways, according to the needs and capabilities of the individual customized information system. Through this approach, individual information systems can be autonomously customized, or even optimized as far as their independent operation is concerned, according to the specific requirements of the corresponding functional units. The architecture of such a system, i.e. TelePACS 2.0, is presented. Effective and efficient image management strategies, which in the context of this application domain are

-17 -

Figure 13.The proposed regional health telematics network of Crete.

KRITIKON PELAGOS Chania

Dia Rethymnon

Heraklion Agios Sitia Nikolaos

Ierapetra Gavdos

LIVIKON PELAGOS

Koufonisi

Chrysi

Regional Hospitals District Hospitals Primary Health Centers

Ambulances, Mobile Screening Units

Isolated Sites

Community Doctors

referred to as intelligent, have been presented as a means for achieving optimal performance of the specific system. In addition, the existence of an integrated patient record at a regional level, and the integration and interoperability of individual information systems allow for the development of software agents to manage image data and other resources of a regional health telematics network efficiently, thus exhibiting an apparently intelligent behavior as perceived by users. A software agent that exhibits intelligent nature [[9] , [10] ] is being developed to coordinate the activities of the intelligent management of image data and the execution of the preloading and prefetching strategies within the distributed IMACS architecture presented. This agent is designed as an Intelligent Personal Service Agent (IPSA) [[26] ] to fit the individual needs and preferences of the user. The IPSA, located at the client’s sites, utilizes domain knowledge coupled with user preferences and customized data utilization patterns for each user to select the most relevant image data, in cooperation with the DHSM servers. Subsequently, the IPSA transmits image data to the client workstation or judiciously caches the data at neighboring servers so that it is where needed when needed.

References [1] M. Ader, “The Dynamic Model”, Bull Technical Report, 1990. [2] R. Alonso, D. Barbara, H. Garcia-Molina, “Data Caching Issues in an Information Retrieval System”, ACM Transactions on Database Systems, Vol. 15, No. 3, September 1990, pp. 359-384. [3] J. S. K. Ang, J. Hong, “AMS Formalism: An Approach to Office Modeling and OIS Development”, Data Base, Vol. 25, No. 4, November 1994, pp. 25-38. [4] A. R. Bakker, et al , "Pre-Fetching and Pre-Loading Strategies", Chapter 5.3, in M. Osteaux, editor. "Hospital Integrated Picture Archiving and Communication Systems - A Second Generation PACS Concept". Berlin: Springer-Verlag, 1991; pp. 145-170. -18 -

[5] S. Chang, A. Hsu, “Image Information Systems: where do we go from here?”, IEEE Transactions on Knowledge and Data Engineering, Vol. 4, No. 5, October 1992, pp. 431-442. [6] E. I. Cohen, G. M. King, J. T. Brady, "Storage Hierarchies", IBM Systems Journal, Vol. 28, No. 1, 1989. [7] P. Constantopoulos, M. Doerr, “The Semantic Indexing System”, Technical Report, Institute of Computer Science, Foundation for Research and Technology Hellas, 1995. [8] D. M. Dilts, W. Wu, "Using Knowledge-Based Technology to Integrate CIM Databases", IEEE Expert, Vol. 3, No. 2, June 1991, pp. 237-245. [9] Oren Etzioni, D. S. Weld, "Intelligent Agents on the Internet: Fact, Fiction, and Forecast", IEEE Expert, Vol. 10, No. 4, August 1995, pp. 44-49. [10] M. R. Genesereth, S. P. Ketchpel, "Software Agents", Communication of the ACM, Vol. 37, No. 7, July 1994, pp. 48-53. [11] C. Hsu, M. Bouziane, W. Cheung, L. Rattner, L. Yee, "Meta-Database Modeling for Enterprise Information Integration", Journal of Systems Integration, Vol. 2, No. 1, 1992, pp. 5-37. [12] C. Hsu and L. Rattner, "Metadatabase Solutions for Enterprise Information Integration Problems", Data Base, Vol. 24, No. 1, Winter 1993, pp. 23 - 35. [13] G. Irie, T. Miyamoto, T. Kojim, I. Yamamoto, T. Kudo, "PACS experience at the University of Hokkaido Medical School", SPIE 1990, No. 1234, pp. 26 - 32. [14] V. S. Krishnamurthy, H. Lam, M. Mitchell , E. Barkmeyer, "IMDAS, An Integrated Manufacturing Data Administration System", Journal of Data and Knowledge Engineering, Vol. 3, No. 2, 1988, pp. 109-131. [15] D. F. Leotta, Y. Kim, “Requirements for Picture Archiving And Communications”, IEEE Engineering in Medicine and Biology, Vol. 12, No. 1, March 1993, pp. 62-80. [16] R. Mattheus, J. G. Tiberglien , "Network Management", Chapter 5.2 in M. Osteaux, editor, "Hospital Integrated Picture Archiving and Communication Systems- A Second Generation PACS Concept", Berlin: Springer-Verlag, 1991, pp. 120-140. [17] D. Meyer-Ebrecht, "Digital image communication", European Journal of Radiology, Vol. 17, No. 93, pp. 47-55. [18] M. Ohly, L. Kleinholz, H. Oswald, B. Mahr, R. Felix, E. Fleck, "Integration in the BERMED Project: The Multimedia Patient Record", Proceedings of the CAR ‘95, Berlin, June 1995, pp. 542-547. [19] S. Orphanoudakis, C. Chronaki, D. Vamvaka: "I2Cnet: Content-based Similarity Search in Geographically Distributed Repositories of Medical Images", Technical Report, Institute of Computer Science, Foundation for Research and Technology Hellas, 1996. (Also to appear in the Journal of Computerized Medical Imaging and Graphics). [20] S. Orphanoudakis, C. Chronaki, S. Kostomanolakis, “I2C: A System for the Indexing, Storage, and Retrieval of Medical Images by Content”, Journal of Medical Informatics, Vol. 19, No. 2, April-June 1994, pp. 109-122. -19 -

[21] S. Orphanoudakis, M. Tsiknakis, C. Chronaki, S. Kostomanolakis, M. Zikos, Y. Tsamardinos, “Development of an Integrated Image Management and Communication System on Crete”, Proceedings of the CAR ‘95, Berlin, June 1995, pp. 481-487. [22] E. Sciore, M. Siegel, A. Rosenthal, “Using Semantic Values to Facilitate Interoperability Among Heterogeneous Information Systems”, ACM Transactions on Database Systems, Vol. 19, No. 2, June 1994, pp. 254-290. [23] J. T. C. Teng, W. J. Keffinger, “Bussiness Process Redesign and Information Architecture: Exploring the Relationships”, Data Base, Vol. 26, No. 1, February 1995, pp. 30-42. [24] Y. Tsamardinos, “DHSM: A Distributed Hierarchical Multimedia Database Management System”, Graduate Thesis, Department of Computer Science, University of Crete, Heraklion, July 1995. [25] M. Tsiknakis, “CoMed - Cooperation in Medicine”, ERCIM - News, No. 21, April 1995, page 23. [26] G. Wiederhold, “Mediators in the Architecture of Future Information Systems”, IEEE Computer Magazine, Vol. 25, No. 3, March 1992, pp. 38-49. [27] A. W. K. Wong, H. K. Huang, "Subsystem Throughputs of a Clinical Picture Archiving and Communication System", Journal of Digital Imaging, No. 5, 1992, pp. 252-261.

-20 -

About the authors: Manolis Tsiknakis (S’85 - M’89) received a B.Eng. degree in electronic engineering, a M.Sc. in microprocessor engineering and a Ph.D. in control systems engineering from Bradford University, U.K. He is at the Institute of Computer Science of Foundation for Research and Technology - Hellas, where he is currently coordinating the activities of the Medical Information Systems Laboratory. His current research interests are in the areas of multimedia systems with emphasis on medical information systems, Picture Archive and Communications Systems, CSCW and Telematics Engineering. Dimitrios G. Katehakis (S’89 - M’93) received his Diploma in electrical engineering from the Technical University of Patras, Greece in 1991 and the M.S. degree also in electrical engineering from the University of Maryland at College Park in 1993. He is a Telecommunications Engineer of the Medical Information Systems Laboratory at the Institute of Computer Science, Foundation for Research and Technology - Hellas. He works on data integration analysis and the design of objectoriented systems and distributed architectures for medical telematics applications. His research focuses in the functional integration of heterogeneous information systems, as well as the intelligent management and communication of multimedia data. Stelios C. Orphanoudakis (S’72-M’76-SM’83) received the B.A. degree in engineering sciences from Dartmouth College, Hanover, NH, in 1971, the M.S. degree in electrical engineering from M.I.T., Cambridge, MA, in 1973, and the Ph.D. degree in electrical engineering from Dartmouth College in 1976. He is Director of the Institute of Computer Science, Foundation for Research and Technology - Hellas, and Professor of Computer Science, University of Crete, Greece. He held a faculty appointment in the Departments of Diagnostic Radiology and Electrical Engineering at Yale University, USA from 1975 until 1991. Prof. Orphanoudakis has many years of academic and research experience in the fields of computer vision and robotics, intelligent image management and retrieval by content, and medical imaging. He has served on various committees and working groups of the European Commission and has been active in European R&D programs. He currently serves on the Board of Directors and is Vice President of the European Research Consortium for Informatics and Mathematics (ERCIM). He is also a member of the Natioal Telecommunications Commission and the National Advisory Research Council of Greece.

Authors’ Present Address: ICS-FORTH Science and Technology Park of Crete Vassilika Vouton, P.O. Box 1385 GR 711 10, Heraklion, Crete, Greece. Email: {orphanou, tsiknaki, katehaki}@ics.forth.gr -21 -