Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
743
Review
.
Infrastructure Communication H. K. Huang1
and
Ricky
Design of a Picture System
A picture archiving and communication system (PACS) infrais the necessary framework to integrate distributed and #{149}eterogeneous imaging systems, provide intelligent data-base nanagement of all radiology-related information, arrange an eficient means of viewing, analyzing, and documenting study re3u1t5, and furnish a mechanism for effectively communicating study results to the referring physician. The PACS infrastructure consists of a basic skeleton of hardware components integrated by standardized, flexible software subsystems. This review describes these concepts and basic building blocks drawn from our
investigation,
system
in our radiology
.
Archiving
.
:
“:
past experience,
and the current
clinical
department.
A PICTURE ARCHIVING AND COMMUNICATION SYSTEM (PACS) has many definitions, depending on the user’s perspective. It can be as simple as a film digitizer connected to a display station and a small image data base, or it can be as complex as a total hospital image management system. Generally speaking, a PACS consists ofacquisition, storage, and display subsystems integrated by various digital networks. From 1 986 to 1 989, we designed three PACS modules that serviced a small subset of the total operations of the radiology department. These PACS modules were independently implemented within our institution in pediatric radiology [1 ], the coronary care unit [2], and neuroradiology [3]. The modules demon-
and
strated the PACS concept and worked adequately for each of the different radiology and clinical services. However, our piecemeal approach did not address all the intricacies of connectivity and cooperation between modules. This weakness surfaced as more PACS nodes were added to the network. The maintenance, routing decisions, coordination of machines, fault tolerance, and expansibility of the system became increasingly difficult problems. The inadequacy in our design was partially due to a lack of understanding of the complexity of a large-scale PACS and to the unavailability of certain PACS-related technologies. In the late 1 980s, two new high-level PACS implementation strategies emerged. Both emphasize system connectivity. The first uses a top-down engineering approach to integrate various hospital information systems [4-7]. The second relies on building a foundation (i.e., PACS infrastructure) for a general multimedia data management system that is easily expandable, flexible, and versatile in its programmability. A hospital-wide PACS is attractive to administrators because it provides economic justification for implementing a PACS. Proponents of PACS are convinced that the cost benefit of the system should not be evaluated as a resource of the radiology department alone but should extend to the entire hospital. This concept has gained momentum during the past 2 years. A few large hospitals have obtained necessary
Received September 30, 1991 : accepted after revision November 13, 1991. This work was supported in part by Public Health Service Grant P01 CA 51 198, awarded by the National Cancer Institute, Services. I Both authors: Medical Imaging Division, Department of RadiOlOgical Sciences, University of Califomia, Los Angeles, School Los Angeles, CA 90024-1721 . Address reprint requests to H. K. Huang. AJR 158:743-749,
#{149}
K. Taira
tructure
original
.
Article
April 1992 0361-803X/92/1584-0743
0 American Roentgen Ray Society
Department of Medicine,
of Health 1 0833
and Human
Le Conte
Ave.,
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
744
HUANG
funding by using this strategy and are at various stages of implementation [4-7]. This review describes the second implementation strategy of the PACS infrastructure method. The PACS infrastructure is the basic design concept to ensure that PACS includes features such as standardization, open architecture, capability for future growth, connectivity, and reliability. This design philosophy can be constructed in a modular fashion. We used the PACS infrastructure designed and implemented in our department as an example to illustrate the importance of these features.
PACS
Infrastructure
The goals of a PACS infrastructure are to provide the necessary framework to integrate distributed and heterogeneous imaging systems; provide intelligent data-base management of all radiology-related information; arrange an efficient means of viewing, analyzing, and documenting study results; and furnish a method for effectively communicating study results to the referring physician. The PACS infrastructure consists of a basic skeleton of hardware components (acquisition interfaces, storage devices, host computers, communication networks, display systems) integrated by standardized, flexible software subsystems. The important software subsystems include communication, data-base management, storage management, job scheduling, interprocessor communication, error handling, and network monitoring. The infrastructure as a whole is versatile and can incorporate rules to reliably perform not only basic PACS management operations but also more complex research job requests. The software modules of the infrastructure have sufficient understanding and cooperation at a system level such that the components work together as a team rather than as individual computers connected in a network. In our design, the PACS infrastructure is physically cornposed of five classes of computer systems connected by various network circuits: (1) radiologic imaging systems, (2) study acquisition computers, (3) cluster controllers, (4) database servers, and (5) display workstations.
Acquisition
Computers
The most troublesome PACS task to date has been the reliable and timely acquisition of image and associated study support text (information on patients, description of the study, and parameters of acquisition and image processing) from a radiologic imaging system (e.g., GE 9800 CT scanner, Milwaukee, WI). Although the ACR/NEMA (AMERICAN COLLEGE OF RADIOLOGY/NATIONAL TION) standard [8]
ELECTRICAL
MANUFACTURERS
ASSOCIA-
proposes to solve these difficulties, the interface is not yet mature. For example, for many imaging systems, especially computed radiography (CR), sonography, digital subtraction angiography, and film digitizer systems, this standard is not used. In our design, we impose a generalpurpose computer, a CAPTURE COMPUTER, between the radiologic imaging system and the rest of the PACS network. The idea is to isolate the radiologic imaging host computers from the PACS via these capture computer gateways. The isolation
and
TAIRA
AJR:158,
April 1992
is done because specialized imaging device computers lack the necessary communication and coordination software that is standardized within the PACS infrastructure. Furthermore, the radiologic imaging computers lack general PACS system knowledge that enables the PACS computers to cooperatively recover from various error conditions. The capture computer has three primary tasks: it acquires image data from the radiologic imaging system, converts the data from a manufacturer specific to a PACS standard format (header format, byte-ordering, matrix sizes) that is compliant with the proposed ACR/NEMA data formats, and forwards the image study to a CLUSTER CONTROLLER MACHINE (described later). Two interface types exist between general-purpose PACS capture computers (e.g., SUN 4/370, Sun Microsystems, Mountain View, CA) and radiologic imaging systems (e.g., GE 9800 CT scanner). The first are PEER-TO-PEER network interfaces. For example, our GE Signa MR scanners (Milwaukee, WI) communicate with SUN 4/370 PACS capture computers by using the TCP/IP (TRANSMISSION CONTROL PROTOCOL/INTERNET PROTOCOL) ETHERNET protocol. When peer-to-peer interfaces are used, image transfers can be initiated either by the radiologic imaging system (a PUSH OPERATION) or by the destination PACS capture computer (a PULL oPERATIoN). The pull mode is advantageous because images may queue in the radiologic imaging system if a capture computer is down. Once the capture computer becomes operational, these queued images can be pulled and normal image flow resumed. The pull mode is the preferred mode of operation because a capture computer can be programmed to reschedule study transfers if a failure occurs (to itselfor to the radiologic imaging system). This assumes sufficient data buffering is available on the imaging system. If a delay in acquisition is not acceptable, studies can be rerouted to a designated backup capture computer on the network when the primary capture computer is unavailable. We developed both types of interfaces for CT and MR with assistance from G.E. Medical System engineers. The second interface type is a MASTER-SLAVE device-level connection such as the de facto industry standard, DR-i i W. This parallel-transfer direct memory access connection is a point-to-point board-level interlace. Recovery mechanisms again depend on which machine (capture computer or imaging system) can initiate a study transfer. In the case of acquiring images from a Philips PCR/SP-90i (Shelton, CT) CR system, data transfer can be initiated from the CR system only. If the capture computer is down, data may be lost. An alternative image acquisition method then must be used to acquire these images (e.g., the technologist manually sends individual images stored in the CR computer after the capture computer is brought back up, or the technologist laser scans the digital hard-copy film image).
Cluster
Controller
Studies acquired by PACS capture computers are sent to machines known as cluster controllers. A cluster is logically defined as a group of computers with statistically similar case and data access patterns. For example, a pediatric radiology cluster, a neuroradiology cluster, or a chest radiology cluster
INFRASTRUCTURE
AJR:158,
April 1992
TABLE
1: Operations
DESIGN
OF
745
PACS
of a PACS Cluster Controller
Receives images
of a study from capture computers text information describing the received studies Updates a network-accessible data-base management system
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
Determines
the
destination
erated studies Automatically retrieves uted
optical
disk
workstations
necessary
library
archive
to forward
comparison
images
the
newly
Image
New
Extracts
__.
Destination
Routing
Image
Workstation
Algorithm
gen-
from a distrib-
P
system
Automatically
corrects the orientation of computed radiographs Determines optimal contrast and brightness parameters for image display [9] Performs image data compression Archives new studies onto optical disk library Deletes images stored on remote capture computers after they have been archived Services archive retrieval requests from workstations and other cluster controllers
p.ti.nt $9S
Fig. 1.-Image
b.d location
routing
decisions
stats of PACS computsrs
usr class
Imglng modality
from cluster
controller
to workstations.
over image routing. The decisions on routing from the cluster controllers to the workstations are shown in Figure 1 Note .
that a cluster
could exist as a logical entity within the PACS infrastructure. The cluster concept provides a more statistical basis for determining routing of studies and for archival strategies. At the center of each cluster is an intelligent machine termed a cluster controller. It has a central role in image acquisition, image routing, archive management, and system reliability.
Cluster
controllers
within
the University
of California,
NY) and 4 gigabytes
Los
Data-Base
capture
By using
only
2: Important Guidelines
a few
can autoComparison
image, etc.)
Servers
The data-base
schema
design
incorporates
the local user views
mation, processing
a PACS Data-Base
requirements,
Sharing
information
hardware
Management
System
among
and software
heterogeneous
systems
Performance
Access to user efficiently
Scalability
Easy to accommodate growth as the application scope increases With hardware redundancy and automatic data-base recovery capabilities Ability to interface to other data-base management systems like hospital and radiol-
Open architecture
ogy development
tools
include data
three levels model,
for each stage
resources
Description
Reliability
should
The first level is the external
which
of the ra-
diology process (Fig. 2). For each radiology process, the following information should be obtained: a description of the transaction or report, data items involved, priority of transaction, frequency of requests, confidentiality, routing infor-
central
for Designing
Integration
Flexibility
cluster.
for the patient. The correlation is imaging technique, date and time, image, discharge image, preoper-
of modeling.
Feature
Advanced
computers
into an alternative
cluster
The architecture and design structure of the PACS data base is critical for the efficient access and integration of radiology operations. Table 2 provides the important guidelines for designing a good data-base management system.
cache cluster
of studies to workstations, we reduce the routing improve security, and have more central control
TABLE
disabled,
ative image, postoperative
cache of 2.6 gigabytes, Storage Concepts, Irvine, CA.) Forwarding of imaging studies to the workstation has been delegated to the two cluster controllers in our PACS. This is opposed to sending images directly from the capture com-
to the workstations.
reconfigure
previous studies available based on organ system, and study type (admission
controllers allows the most recently acquired images to be quickly available for any PACS workstation. (Each workstation also supports a local parallel-transfer magnetic-disk
puters
becomes
matically
and if a given
of PARALLEL-TRANSFER
IPI (INTELLIGENT PERIPHERAL INTERFACE) MAGNETIC-DISK storage [3]. The large magnetic-disk cache on these
dispatchers complexity,
controller
definition,
images are also automatically retrieved and forwarded to the workstation for returning patients. We use an algorithm to calculate the degree of correlation between a new study and
Angeles (UCLA), PACS perform many tasks (Table i). Currently, we have developed two clusters-an inpatient cluster and an outpatient cluster. Each cluster controller has a Kodak optical-disk library unit with 1 -terabyte archive ca-
pacity (Rochester,
is only a logical
information
and
program-requested
data
systems
Upgradable to high-level data-base language interfaces and object-oriented tools [10] Easy to make changes in data-base schema
required,
person-
HUANG
746
AND
TAIRA
AJR:158,
Fig. 2.-External data model incorporating local user views for each stage of radiology process.
Radiology Registration
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
I
‘nd
all lmsg.s Ui. p..t two y..rs dl.gno..d wfth rfght pI.ursI .ffuslons”
I I
I
Patlint Jon.. will hay. a bons a9#{149} p.rform.d nsxt w..lc. Mak. aura all n.c..sary historical data Is r..dy for rsvl.w.
Research
April 1992
PACS(RIS)
I I I
Technologists ‘Flnd th. atorag. location of all lmag.s for patl.,it 123-4547. m. this auba.t of lmag.s
7
V
I I I I I
Case Management
Transcription “H.r. Is a finish.d docum.ntlng pali.nt Bon. ag. r.sults. avallabl. 10 your
nel required,
gies. Once
I
Report
I
I
processing
and routing
user requirements
or conceptual
“List th. pati.nts and imag.s most sctivs during th. past w..k”
and alternative
the high-level
second-level
I
Jon.a Mak. 1 PACS
model
strate-
are known,
the
the data and processing needs of the various service centers can be developed to produce an integrated global data structure capable of supporting all PACS applications. The synthesis of the conceptual data model involves determining what entities (e.g., patients, cases, studies, folders, sequences, archives) are required, what relationships exist among these entities, and what properties the entities should have.
The third level, the physical
that consolidates
implementation
model,
must
logically implement the conceptual model and simultaneously satisfy user performance and reliability requirements. In this model, large data items (images) and smaller data items (text)
are managed differently because of differences in their storage, access, reliability, and processing requirements. In the UCLA system, small data items (e.g., patient demographics, clinical history, study description, image description, diagnostic reports) are maintained by a commercial relational data-
base management system (Sybase Inc., Emeryville, CA). This system uses a client-server architecture [1 i ]. All text records are centrally stored in tables maintained by mirrored data-
All nodes in the PACS network are data-base (text) clients. Client nodes send remote procedure calls from an application program over a network to the data-base server machine.
The data-base
language
used
is
SQL
(STRUCTURED
QUERY
LANGUAGE).
For large data items (images), it is important to place active files in a location where the study is most likely to be viewed. The large image files from these studies can then be allocated onto the fastest storage devices, which are local to the workstation. Images can be moved to other storage locations as their access characteristics change (such as when a patient
moves from one clinical service
to another
is discharged).
Image
files may reside
workstations.
Factors
that determine
or when a patient
on several
image
diagnostic
file placement
include radiology service of study, supervising radiologist, clinical service attending the patient, patient’s status (inpatient, outpatient, emergency patient), type of examination,
and whether
the patient’s
case is currently
active at a partic-
ular workstation.
Display
Stations
perform data integrity checking; data-base query processing; data inserts, deletes, and updates; retrieval speed optimiza-
PACS display stations should fully use the resources and processing power of the entire PACS network. A station includes communication, data base, display, resource management, and processing software. The fundamental workstation operations are listed in Table 3. We use three types of display stations: (i) high-resolution 2.5K x 2K for primary diagnosis, (2) medium-resolution iK x
tion; cluster
i K stations
base
server
machines
[i 2]. Data
tables
may
span
several
physical storage devices. The centralized data-base server cuts down on data redundancy and inconsistency while still maintaining fast response to user queries server machines are dedicated processors
table
indexing;
and backup
for text data. The tuned to efficiently
and recovery
of data.
for referring
physicians
and conferences,
and (3)
AJR:158,
INFRASTRUCTURE
April 1992
TABLE
3: Operations
Performed
DESIGN
Description
Accumulation
of all relevant
to a patient
images
Allows the selection of cases for a given subpopulation Tools for arranging and grouping images for easy review Measurement tools for facilitating the diagnosis
Documentation
Tools
for
Case presentation
Tools
for a comprehensive
image
annotation,
text, case
addition
thermore, transferring
to being a client to the global PACS text data base
and distributed global image archive, the primary workstation is also a client to a centralized voice management system. Access to the voice server is instantaneous over the public telephone system. All dictation and playback functions are interfaced directly to the workstation. We integrated the primary diagnosis station by using components from various vendors and developed our own display software. In the case of medium-resolution stations, we designed and built the
System
Networking
A basic function of any computer network is to provide an access path by which end users (e.g., radiologists and clinicians) at one geographic location can access information (e.g., images, reports) at another location. The most obvious way to characterize a radiologic PACS network is to examine traffic of information between various locations and users.
The important
networking
data
needed
for system
Digital communication
in our infrastructure
design is based
on a three-tiered network architecture consisting of low-speed (i 0 megabits/sec signaling rate) Ethernet circuits, mediumspeed (i 00 megabits/sec) FIBER DISTRIBUTED DATA INTERFACE (Fool) circuits, and high-speed (i gigabit/sec) ULTRANET net-
work circuits (Ultra Network
Technologies,
are adequate system
computers
for this application have relatively
voice
reports
presentation
networking
capabilities.
FDDI
images from the capture
computers
A faster
is used
image
network
is used
to transmit
to the cluster controllers.
between
capture
computers
and cluster controller machines because several imaging tems may be connected to a single capture computer.
several local Ethernet circuits may be involved in data from imaging systems to capture computers.
The UltraNet cluster stations
sysFur-
system
is used to transmit
images
controllers and from cluster controllers when long delays are not tolerable.
Process
coordination
between
tasks
running
between to
display
on different
PACS machines is an extremely important issue in system networking. This coordination of processes running either on the same computer or on different computers is accomplished by using interprocessor communication methods implemented by using socket-level interfaces to TCP/IP. Commands are exchanged as American Standard Code for Information Interchange (ASCII) messages to ensure standard encoding of messages. Jobs requests are stuffed into disk resident priority queues, which are serviced by various DAEMON processes. The queue software has a built-in job scheduler that is programmed to retry a job several times by using either a default set of resources or alternative resources if a
hardware
error is detected.
Some
Important
PACS
Design
Concepts
When we planned a totally digital radiology department in i 989 and i 990, we realized the importance of system stand-
ardization, We
were
purpose
open architecture, fortunate
became
that
available
connectivity,
all necessary
and reliability technologies
at that time, allowing
[i 6].
for this
us to imple-
ment this concept in a clinical environment. We firmly believe that the concept of a robust PACS infrastructure is essential for small- and large-scale implementations alike.
San Jose, CA) [i 4,
i 5]. All three types of networks are commercially available. These networks together span three buildings housing the radiology department. All three use the TCP/IP communication protocol and are accessed by the same software calls. The local Ethernet circuits are used as a low-cost network to connect all PACS nodes. Primarily, the Ethernet circuits transmit messages, data-base query results, and images from radiologic imaging systems to capture computers. Low-speed imaging
and
and
[i 3]
design
include location and function of each node, frequency of information passed between any two nodes, cost for transmission between nodes with various speed lines, desired reliability of the communication, and required throughput. The variables in the design include the network topology, line capacities, and flow assignments.
networks
belonging
Case selection Image arrangement Interpretation
tion
in-house.
and information
examination
high-resolution hard-copy print stations. At the primary diagnosis display workstation, images are stored on fast-access parallel-transfer magnetic disks. Each workstation maintains a local data base for managing current cases. It also has access to the global PACS data base for historical images. In
system
747
PACS
at a PACS Workstation
Operation
Case preparation
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
OF
because slow
data
most genera-
Standardization
The first important
rule in building
a PACS infrastructure
is
to incorporate as many industry de facto standards as possible that are consistent with the overall PACS design schema. The philosophy is to minimize the developing of your own software. Furthermore, use of industry standard hardware and software increases the portability of the system to other computer platforms. For example, the following industry standards are integrated into our PACS: (i) UNIX operating
HUANG
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
748
system, (2) TCP/IP communication protocol, (3) SQL as the data-base query language, (4) ACR-NEMA standard (higher layers of International Standards Organization model only), (5) C programming language, (6) x WINDOWS user interface, and (7) ASCII text representation for message passing. The implications of PACS standardization include the following: Implementation of all future PACS installations is straightforward. System maintenance is easier because each module looks similar to others, if not physically, then logically. By
defining amount
the PACS of redundant
The code
therefore
operations, we minimize the computer code within the PACS system. is easier to debug, understand, and
among all levels of developers.
AJR:158, April1992
used include automatic resources and algorithms
If two PACS modules in the same department cannot communicate with each other, they become two isolated systems and can never be combined with other systems to
form a total departmental
PACS. An example
is the pediatric
radiology PACS module and the coronary care unit PACS module we developed earlier [i 2]. These two modules have no means of communicating with each other. Images and patient information are isolated within each module. An open network design architecture allows a standardized method ,
for data and message
exchange
between
heterogeneous
systems. Because computer and communications technology changes rapidly, a closed architecture would hinder system upgradability. As an example, an independent imaging console from a given manufacturer is a good addition to an MR or a CT scanner for viewing images; however, it always has a closed proprietary architecture design. As a result, no other
those
specified
by the same manufac-
turer can be augmented to the system. This limits potential upgrading and improvement of the system. Consideration of the connectivity is important even when a small-scale PACS is planned. A yes to questions like “Can we transmit images from this PACS to other systems and vice-versa?” “Does this module use a standard data and image format?” and “Does the machine use a standard communication protocol?” means that the PACS is well designed and will allow future connec-
tivity.
How to minimize the cost requires further study.
Security
it cannot be down for
extended periods. In designing a PACS, it is therefore important to use fault-tolerant measures, including error detection and logging software, external auditing programs (i.e., network management processes that check network circuits, magnetic-disk space, data-base status, processes status, and queue status), hardware redundancy, and intelligent software recovery blocks. Some recovery mechanisms we have
alternative ROUTINES
while
maintaining
high
reliability
is an important
consideration
because
of the need
trol, and the use of views. Most sophisticated data-base management systems have identification and authorization mechanisms that use accounts and passwords. Application programs may also supply additional layers of protection.
control
refers to granting
and revoking
the user’s
access to specific tables, columns, or views. These security measures provide the PACS infrastructure with a mechanism for access control to clinical and research data. With these mechanisms, the system designer can enforce departmental policy as to which persons have access to clinical studies. For example, referring clinicians may be granted image study access only after a preliminary radiology reading has been performed and attached to the image data.
Glossary AcR/NEMA
AMERICAN
MANUFACTURERS
COLLEGE
OF
RADIOLOGY/NATIONAL
ELECTRICAL
The members in the ACR-NEMA cornfor writing a standard for image information
ASSOCIATION.
are responsible
mittee
exchange
acceptable
BOOTSTRAP
by
ACR and NEMA.
both
A software
ROUTINE
block executed
by a computer
when
it is restarted. CAPTURE
Computer
COMPUTER
interposed
radiologic rnIts main task is
between
a
aging device (e.g., CT scanner) and PACS network. to standardize
the image
file format.
The computer that maintains the optical archive for a cluster of computers (e.g., neurology cluster, pediatric cluster). Its functions are similar to those of the film library CLUSTER
CONTROLLER
in a conventional
MACHINE
film-based
system.
A background computer program that runs continuously. It does not affect the system operation unless a particular circumstance arises. In that case, the program will automatically change the information flow in the system. Usually each daemon performs a specific task. DAEMON
FDDI
and displays critical patient information,
with
BOOTSTRAP
for patient confidentiality and medicolegal issues. Three major security mechanisms include account control, privilege con-
The most common baseband communication it is a relatively slow network.
connecting
Reliability is a major issue in a PACS for two reasons. First, a PACS has many components, so the probability of a component failing is high. Second, because the PACS manages
jobs
Security
ETHERNET
Reliability
retry of failed and intelligent
that allow a PACS computer to automatically continue operations after a power outage [17]. Improving reliability is costly.
Privilege
Open Architecture
except
TAIRA
primitive
search. Standardizing terminology, design concepts, and so forth, facilitates system understanding and documentation
components
AND
FIBER
DISTRIBUTED
DATA
that uses a ring topology. IPI
network
for
computers.
INTELLIGENT
PERIPHERAL
INTERFACE.
A communication
It is a medium-speed network. INTERFACE used to connect
network a computer
to a disk drive. MASTER-SLAVE
which
the
INTERFACE
master
always
The connection between initiates and requests
two
machines
operations
in
to be
performed by a slave device. PACS
PICTURE
ARCHIVING
AND
COMMUNICATION
SYSTEM.
device typically used for images. Special magnetic-disk controllers allow the bytes of a single image to be distributed over several disks. Access to the data can be done in parallel, and this is much faster than conventional disks. PARALLEL-TRANSFER
MAGNETIC
DISK
A magnetic-disk
AJR:158,
INFRASTRUCTURE
April 1992
DESIGN
5.
INTERFACE A network interface in which the communicating machines can perform redprocal operations. PULL OPERATION A network data transfer operation in which the computer that desires the data initiates the data transfer. PUSH OPERATION A network data transfer operation in which the computer with the data to send initiates the data transfer. SQL STRUCTURED QUERY LANGUAGE. It is commonly used in a later generation data base.
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
PEER-TO-PEER
TCP/IP
TRANSMISSION
standard puters.
CONTROL
and popular
method
A commercial
ULTRANET
communication
PROTOCOL/INTERNET
for
data communication
high-speed
gigabit-per-second
7.
A
PROTOCOL.
between
6.
corn8.
fiber-optic
network.
A computer operating system commonly used in computers introduced in the late 1980s. x WINDOWS A software subsystem that allows developers to conUNIX
struct
a graphical
accessing
text,
user
interface
image-processing
(menus,
buttons,
icons,
etc.)
for
9.
10.
modules. 11.
1
2.
REFERENCES 13.
1 . Taira RK, Mankovich and implementation pediatric radiology.MR
NJ, Boechat MI, Kangarloo
H, Huang, communication
HK. Design
of picture archiving and system for 1988;150:1117-1121 2. Cho PS, Huang HK, Tillisch J, Kangarloo H. Clinical evaluation of a radiologic picture archiving and communication system for a coronary care unit. MR 1988:151:823-827 3. Lou SL, Loloyan M, Weinberg W, et al. Image delivery performance of a CT/MR PACS modu applied in neuroradiology. Proc SPIE 1991:1446:302-311 4. me G, Miyasaka K. Clinical experience: 16 months of HU-PACS. comput Med Imaging Graph 1991;15: 191 -1 95
14.
15.
16. 1
7.
OF
PACS
749
Masser H, Mandl A, Urban M, Hradil H, Hruby W. The Vienna Project SMzO in picture archiving and communication systems (PACS). In: Huang HK, Ratib 0, Bakker AR, Witte G, eds. NATO AS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag. 1991:247-250 Glass HI. Slark NA. PACS and related research in the lkiited Kingdom in picture archiving and communication systems (PACS). In: Huang HK, Ratib 0, Bakker AR, Witte G, eds. NATO AS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag, 1991:319-324 Goeringer F. Medical diagnostic imaging support systems for military medidne in picture archiving and communication systems (PACS). In: Huang HK, Ratib 0. Bakker AR, Witte G. eds. NATO AS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag. 1991:213-230 ACR-NEMA Digital Imaging and Communication Standards Committee. Digital imaging and communications ACR-NEMA 300-1988. Washington, DC: National Electrical Manufacturers Association, 1988 McNitt-Gray MF, Taira RK, Eldredge S, Razavi M. Brightness and contrast adjustments for different tissue densities in digital chest radiographs. Proc SPIE 1991:1445:468-478 Parsaye K, Chignell M, Khoshaflan 5, Wong H. Intelligent databases. New York: Wiley, 1989 Malamud C. !NGRES: tools for building an information architecture. New York: Von Norstrand Reinhold, 1989 Taira RK, Stewart BK, Sinha U. PACS database architecture and design. Comput Med Imaging Graph 1991:15: 171 -176 Ratib 0, Ugier Y, Funk M, et aL PACS workstation: user interface design. In: Huang HK, Ratib 0, Bakker AF, Witte G, eds. NATO AS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag, 1991:57-61 Huang HK, Lou SL, Cho PS, et al. RadiOIOgIC image communication methods. MR 1990:155:183-186 Stewart BK, Lou SL, Wong WK, Huang HK. An ultrafast network for communication of radiologic images. MR 1991:156:835-839 Huang HK, Kangarloo H, Cho PS, Talra RK, Ho BKT, Chan KK. Planning a totally digital radiology department. MR 1990:154:635-639 Taira RK. Chan KK, Stewart BK, Weinberg WS. PACS reliability issues. In: Huang HK, Ratib 0, Bakker AR, Witte G, eds. NATOAS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag, 1991:149-156
743
Review
.
Infrastructure Communication H. K. Huang1
and
Ricky
Design of a Picture System
A picture archiving and communication system (PACS) infrais the necessary framework to integrate distributed and #{149}eterogeneous imaging systems, provide intelligent data-base nanagement of all radiology-related information, arrange an eficient means of viewing, analyzing, and documenting study re3u1t5, and furnish a mechanism for effectively communicating study results to the referring physician. The PACS infrastructure consists of a basic skeleton of hardware components integrated by standardized, flexible software subsystems. This review describes these concepts and basic building blocks drawn from our
investigation,
system
in our radiology
.
Archiving
.
:
“:
past experience,
and the current
clinical
department.
A PICTURE ARCHIVING AND COMMUNICATION SYSTEM (PACS) has many definitions, depending on the user’s perspective. It can be as simple as a film digitizer connected to a display station and a small image data base, or it can be as complex as a total hospital image management system. Generally speaking, a PACS consists ofacquisition, storage, and display subsystems integrated by various digital networks. From 1 986 to 1 989, we designed three PACS modules that serviced a small subset of the total operations of the radiology department. These PACS modules were independently implemented within our institution in pediatric radiology [1 ], the coronary care unit [2], and neuroradiology [3]. The modules demon-
and
strated the PACS concept and worked adequately for each of the different radiology and clinical services. However, our piecemeal approach did not address all the intricacies of connectivity and cooperation between modules. This weakness surfaced as more PACS nodes were added to the network. The maintenance, routing decisions, coordination of machines, fault tolerance, and expansibility of the system became increasingly difficult problems. The inadequacy in our design was partially due to a lack of understanding of the complexity of a large-scale PACS and to the unavailability of certain PACS-related technologies. In the late 1 980s, two new high-level PACS implementation strategies emerged. Both emphasize system connectivity. The first uses a top-down engineering approach to integrate various hospital information systems [4-7]. The second relies on building a foundation (i.e., PACS infrastructure) for a general multimedia data management system that is easily expandable, flexible, and versatile in its programmability. A hospital-wide PACS is attractive to administrators because it provides economic justification for implementing a PACS. Proponents of PACS are convinced that the cost benefit of the system should not be evaluated as a resource of the radiology department alone but should extend to the entire hospital. This concept has gained momentum during the past 2 years. A few large hospitals have obtained necessary
Received September 30, 1991 : accepted after revision November 13, 1991. This work was supported in part by Public Health Service Grant P01 CA 51 198, awarded by the National Cancer Institute, Services. I Both authors: Medical Imaging Division, Department of RadiOlOgical Sciences, University of Califomia, Los Angeles, School Los Angeles, CA 90024-1721 . Address reprint requests to H. K. Huang. AJR 158:743-749,
#{149}
K. Taira
tructure
original
.
Article
April 1992 0361-803X/92/1584-0743
0 American Roentgen Ray Society
Department of Medicine,
of Health 1 0833
and Human
Le Conte
Ave.,
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
744
HUANG
funding by using this strategy and are at various stages of implementation [4-7]. This review describes the second implementation strategy of the PACS infrastructure method. The PACS infrastructure is the basic design concept to ensure that PACS includes features such as standardization, open architecture, capability for future growth, connectivity, and reliability. This design philosophy can be constructed in a modular fashion. We used the PACS infrastructure designed and implemented in our department as an example to illustrate the importance of these features.
PACS
Infrastructure
The goals of a PACS infrastructure are to provide the necessary framework to integrate distributed and heterogeneous imaging systems; provide intelligent data-base management of all radiology-related information; arrange an efficient means of viewing, analyzing, and documenting study results; and furnish a method for effectively communicating study results to the referring physician. The PACS infrastructure consists of a basic skeleton of hardware components (acquisition interfaces, storage devices, host computers, communication networks, display systems) integrated by standardized, flexible software subsystems. The important software subsystems include communication, data-base management, storage management, job scheduling, interprocessor communication, error handling, and network monitoring. The infrastructure as a whole is versatile and can incorporate rules to reliably perform not only basic PACS management operations but also more complex research job requests. The software modules of the infrastructure have sufficient understanding and cooperation at a system level such that the components work together as a team rather than as individual computers connected in a network. In our design, the PACS infrastructure is physically cornposed of five classes of computer systems connected by various network circuits: (1) radiologic imaging systems, (2) study acquisition computers, (3) cluster controllers, (4) database servers, and (5) display workstations.
Acquisition
Computers
The most troublesome PACS task to date has been the reliable and timely acquisition of image and associated study support text (information on patients, description of the study, and parameters of acquisition and image processing) from a radiologic imaging system (e.g., GE 9800 CT scanner, Milwaukee, WI). Although the ACR/NEMA (AMERICAN COLLEGE OF RADIOLOGY/NATIONAL TION) standard [8]
ELECTRICAL
MANUFACTURERS
ASSOCIA-
proposes to solve these difficulties, the interface is not yet mature. For example, for many imaging systems, especially computed radiography (CR), sonography, digital subtraction angiography, and film digitizer systems, this standard is not used. In our design, we impose a generalpurpose computer, a CAPTURE COMPUTER, between the radiologic imaging system and the rest of the PACS network. The idea is to isolate the radiologic imaging host computers from the PACS via these capture computer gateways. The isolation
and
TAIRA
AJR:158,
April 1992
is done because specialized imaging device computers lack the necessary communication and coordination software that is standardized within the PACS infrastructure. Furthermore, the radiologic imaging computers lack general PACS system knowledge that enables the PACS computers to cooperatively recover from various error conditions. The capture computer has three primary tasks: it acquires image data from the radiologic imaging system, converts the data from a manufacturer specific to a PACS standard format (header format, byte-ordering, matrix sizes) that is compliant with the proposed ACR/NEMA data formats, and forwards the image study to a CLUSTER CONTROLLER MACHINE (described later). Two interface types exist between general-purpose PACS capture computers (e.g., SUN 4/370, Sun Microsystems, Mountain View, CA) and radiologic imaging systems (e.g., GE 9800 CT scanner). The first are PEER-TO-PEER network interfaces. For example, our GE Signa MR scanners (Milwaukee, WI) communicate with SUN 4/370 PACS capture computers by using the TCP/IP (TRANSMISSION CONTROL PROTOCOL/INTERNET PROTOCOL) ETHERNET protocol. When peer-to-peer interfaces are used, image transfers can be initiated either by the radiologic imaging system (a PUSH OPERATION) or by the destination PACS capture computer (a PULL oPERATIoN). The pull mode is advantageous because images may queue in the radiologic imaging system if a capture computer is down. Once the capture computer becomes operational, these queued images can be pulled and normal image flow resumed. The pull mode is the preferred mode of operation because a capture computer can be programmed to reschedule study transfers if a failure occurs (to itselfor to the radiologic imaging system). This assumes sufficient data buffering is available on the imaging system. If a delay in acquisition is not acceptable, studies can be rerouted to a designated backup capture computer on the network when the primary capture computer is unavailable. We developed both types of interfaces for CT and MR with assistance from G.E. Medical System engineers. The second interface type is a MASTER-SLAVE device-level connection such as the de facto industry standard, DR-i i W. This parallel-transfer direct memory access connection is a point-to-point board-level interlace. Recovery mechanisms again depend on which machine (capture computer or imaging system) can initiate a study transfer. In the case of acquiring images from a Philips PCR/SP-90i (Shelton, CT) CR system, data transfer can be initiated from the CR system only. If the capture computer is down, data may be lost. An alternative image acquisition method then must be used to acquire these images (e.g., the technologist manually sends individual images stored in the CR computer after the capture computer is brought back up, or the technologist laser scans the digital hard-copy film image).
Cluster
Controller
Studies acquired by PACS capture computers are sent to machines known as cluster controllers. A cluster is logically defined as a group of computers with statistically similar case and data access patterns. For example, a pediatric radiology cluster, a neuroradiology cluster, or a chest radiology cluster
INFRASTRUCTURE
AJR:158,
April 1992
TABLE
1: Operations
DESIGN
OF
745
PACS
of a PACS Cluster Controller
Receives images
of a study from capture computers text information describing the received studies Updates a network-accessible data-base management system
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
Determines
the
destination
erated studies Automatically retrieves uted
optical
disk
workstations
necessary
library
archive
to forward
comparison
images
the
newly
Image
New
Extracts
__.
Destination
Routing
Image
Workstation
Algorithm
gen-
from a distrib-
P
system
Automatically
corrects the orientation of computed radiographs Determines optimal contrast and brightness parameters for image display [9] Performs image data compression Archives new studies onto optical disk library Deletes images stored on remote capture computers after they have been archived Services archive retrieval requests from workstations and other cluster controllers
p.ti.nt $9S
Fig. 1.-Image
b.d location
routing
decisions
stats of PACS computsrs
usr class
Imglng modality
from cluster
controller
to workstations.
over image routing. The decisions on routing from the cluster controllers to the workstations are shown in Figure 1 Note .
that a cluster
could exist as a logical entity within the PACS infrastructure. The cluster concept provides a more statistical basis for determining routing of studies and for archival strategies. At the center of each cluster is an intelligent machine termed a cluster controller. It has a central role in image acquisition, image routing, archive management, and system reliability.
Cluster
controllers
within
the University
of California,
NY) and 4 gigabytes
Los
Data-Base
capture
By using
only
2: Important Guidelines
a few
can autoComparison
image, etc.)
Servers
The data-base
schema
design
incorporates
the local user views
mation, processing
a PACS Data-Base
requirements,
Sharing
information
hardware
Management
System
among
and software
heterogeneous
systems
Performance
Access to user efficiently
Scalability
Easy to accommodate growth as the application scope increases With hardware redundancy and automatic data-base recovery capabilities Ability to interface to other data-base management systems like hospital and radiol-
Open architecture
ogy development
tools
include data
three levels model,
for each stage
resources
Description
Reliability
should
The first level is the external
which
of the ra-
diology process (Fig. 2). For each radiology process, the following information should be obtained: a description of the transaction or report, data items involved, priority of transaction, frequency of requests, confidentiality, routing infor-
central
for Designing
Integration
Flexibility
cluster.
for the patient. The correlation is imaging technique, date and time, image, discharge image, preoper-
of modeling.
Feature
Advanced
computers
into an alternative
cluster
The architecture and design structure of the PACS data base is critical for the efficient access and integration of radiology operations. Table 2 provides the important guidelines for designing a good data-base management system.
cache cluster
of studies to workstations, we reduce the routing improve security, and have more central control
TABLE
disabled,
ative image, postoperative
cache of 2.6 gigabytes, Storage Concepts, Irvine, CA.) Forwarding of imaging studies to the workstation has been delegated to the two cluster controllers in our PACS. This is opposed to sending images directly from the capture com-
to the workstations.
reconfigure
previous studies available based on organ system, and study type (admission
controllers allows the most recently acquired images to be quickly available for any PACS workstation. (Each workstation also supports a local parallel-transfer magnetic-disk
puters
becomes
matically
and if a given
of PARALLEL-TRANSFER
IPI (INTELLIGENT PERIPHERAL INTERFACE) MAGNETIC-DISK storage [3]. The large magnetic-disk cache on these
dispatchers complexity,
controller
definition,
images are also automatically retrieved and forwarded to the workstation for returning patients. We use an algorithm to calculate the degree of correlation between a new study and
Angeles (UCLA), PACS perform many tasks (Table i). Currently, we have developed two clusters-an inpatient cluster and an outpatient cluster. Each cluster controller has a Kodak optical-disk library unit with 1 -terabyte archive ca-
pacity (Rochester,
is only a logical
information
and
program-requested
data
systems
Upgradable to high-level data-base language interfaces and object-oriented tools [10] Easy to make changes in data-base schema
required,
person-
HUANG
746
AND
TAIRA
AJR:158,
Fig. 2.-External data model incorporating local user views for each stage of radiology process.
Radiology Registration
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
I
‘nd
all lmsg.s Ui. p..t two y..rs dl.gno..d wfth rfght pI.ursI .ffuslons”
I I
I
Patlint Jon.. will hay. a bons a9#{149} p.rform.d nsxt w..lc. Mak. aura all n.c..sary historical data Is r..dy for rsvl.w.
Research
April 1992
PACS(RIS)
I I I
Technologists ‘Flnd th. atorag. location of all lmag.s for patl.,it 123-4547. m. this auba.t of lmag.s
7
V
I I I I I
Case Management
Transcription “H.r. Is a finish.d docum.ntlng pali.nt Bon. ag. r.sults. avallabl. 10 your
nel required,
gies. Once
I
Report
I
I
processing
and routing
user requirements
or conceptual
“List th. pati.nts and imag.s most sctivs during th. past w..k”
and alternative
the high-level
second-level
I
Jon.a Mak. 1 PACS
model
strate-
are known,
the
the data and processing needs of the various service centers can be developed to produce an integrated global data structure capable of supporting all PACS applications. The synthesis of the conceptual data model involves determining what entities (e.g., patients, cases, studies, folders, sequences, archives) are required, what relationships exist among these entities, and what properties the entities should have.
The third level, the physical
that consolidates
implementation
model,
must
logically implement the conceptual model and simultaneously satisfy user performance and reliability requirements. In this model, large data items (images) and smaller data items (text)
are managed differently because of differences in their storage, access, reliability, and processing requirements. In the UCLA system, small data items (e.g., patient demographics, clinical history, study description, image description, diagnostic reports) are maintained by a commercial relational data-
base management system (Sybase Inc., Emeryville, CA). This system uses a client-server architecture [1 i ]. All text records are centrally stored in tables maintained by mirrored data-
All nodes in the PACS network are data-base (text) clients. Client nodes send remote procedure calls from an application program over a network to the data-base server machine.
The data-base
language
used
is
SQL
(STRUCTURED
QUERY
LANGUAGE).
For large data items (images), it is important to place active files in a location where the study is most likely to be viewed. The large image files from these studies can then be allocated onto the fastest storage devices, which are local to the workstation. Images can be moved to other storage locations as their access characteristics change (such as when a patient
moves from one clinical service
to another
is discharged).
Image
files may reside
workstations.
Factors
that determine
or when a patient
on several
image
diagnostic
file placement
include radiology service of study, supervising radiologist, clinical service attending the patient, patient’s status (inpatient, outpatient, emergency patient), type of examination,
and whether
the patient’s
case is currently
active at a partic-
ular workstation.
Display
Stations
perform data integrity checking; data-base query processing; data inserts, deletes, and updates; retrieval speed optimiza-
PACS display stations should fully use the resources and processing power of the entire PACS network. A station includes communication, data base, display, resource management, and processing software. The fundamental workstation operations are listed in Table 3. We use three types of display stations: (i) high-resolution 2.5K x 2K for primary diagnosis, (2) medium-resolution iK x
tion; cluster
i K stations
base
server
machines
[i 2]. Data
tables
may
span
several
physical storage devices. The centralized data-base server cuts down on data redundancy and inconsistency while still maintaining fast response to user queries server machines are dedicated processors
table
indexing;
and backup
for text data. The tuned to efficiently
and recovery
of data.
for referring
physicians
and conferences,
and (3)
AJR:158,
INFRASTRUCTURE
April 1992
TABLE
3: Operations
Performed
DESIGN
Description
Accumulation
of all relevant
to a patient
images
Allows the selection of cases for a given subpopulation Tools for arranging and grouping images for easy review Measurement tools for facilitating the diagnosis
Documentation
Tools
for
Case presentation
Tools
for a comprehensive
image
annotation,
text, case
addition
thermore, transferring
to being a client to the global PACS text data base
and distributed global image archive, the primary workstation is also a client to a centralized voice management system. Access to the voice server is instantaneous over the public telephone system. All dictation and playback functions are interfaced directly to the workstation. We integrated the primary diagnosis station by using components from various vendors and developed our own display software. In the case of medium-resolution stations, we designed and built the
System
Networking
A basic function of any computer network is to provide an access path by which end users (e.g., radiologists and clinicians) at one geographic location can access information (e.g., images, reports) at another location. The most obvious way to characterize a radiologic PACS network is to examine traffic of information between various locations and users.
The important
networking
data
needed
for system
Digital communication
in our infrastructure
design is based
on a three-tiered network architecture consisting of low-speed (i 0 megabits/sec signaling rate) Ethernet circuits, mediumspeed (i 00 megabits/sec) FIBER DISTRIBUTED DATA INTERFACE (Fool) circuits, and high-speed (i gigabit/sec) ULTRANET net-
work circuits (Ultra Network
Technologies,
are adequate system
computers
for this application have relatively
voice
reports
presentation
networking
capabilities.
FDDI
images from the capture
computers
A faster
is used
image
network
is used
to transmit
to the cluster controllers.
between
capture
computers
and cluster controller machines because several imaging tems may be connected to a single capture computer.
several local Ethernet circuits may be involved in data from imaging systems to capture computers.
The UltraNet cluster stations
sysFur-
system
is used to transmit
images
controllers and from cluster controllers when long delays are not tolerable.
Process
coordination
between
tasks
running
between to
display
on different
PACS machines is an extremely important issue in system networking. This coordination of processes running either on the same computer or on different computers is accomplished by using interprocessor communication methods implemented by using socket-level interfaces to TCP/IP. Commands are exchanged as American Standard Code for Information Interchange (ASCII) messages to ensure standard encoding of messages. Jobs requests are stuffed into disk resident priority queues, which are serviced by various DAEMON processes. The queue software has a built-in job scheduler that is programmed to retry a job several times by using either a default set of resources or alternative resources if a
hardware
error is detected.
Some
Important
PACS
Design
Concepts
When we planned a totally digital radiology department in i 989 and i 990, we realized the importance of system stand-
ardization, We
were
purpose
open architecture, fortunate
became
that
available
connectivity,
all necessary
and reliability technologies
at that time, allowing
[i 6].
for this
us to imple-
ment this concept in a clinical environment. We firmly believe that the concept of a robust PACS infrastructure is essential for small- and large-scale implementations alike.
San Jose, CA) [i 4,
i 5]. All three types of networks are commercially available. These networks together span three buildings housing the radiology department. All three use the TCP/IP communication protocol and are accessed by the same software calls. The local Ethernet circuits are used as a low-cost network to connect all PACS nodes. Primarily, the Ethernet circuits transmit messages, data-base query results, and images from radiologic imaging systems to capture computers. Low-speed imaging
and
and
[i 3]
design
include location and function of each node, frequency of information passed between any two nodes, cost for transmission between nodes with various speed lines, desired reliability of the communication, and required throughput. The variables in the design include the network topology, line capacities, and flow assignments.
networks
belonging
Case selection Image arrangement Interpretation
tion
in-house.
and information
examination
high-resolution hard-copy print stations. At the primary diagnosis display workstation, images are stored on fast-access parallel-transfer magnetic disks. Each workstation maintains a local data base for managing current cases. It also has access to the global PACS data base for historical images. In
system
747
PACS
at a PACS Workstation
Operation
Case preparation
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
OF
because slow
data
most genera-
Standardization
The first important
rule in building
a PACS infrastructure
is
to incorporate as many industry de facto standards as possible that are consistent with the overall PACS design schema. The philosophy is to minimize the developing of your own software. Furthermore, use of industry standard hardware and software increases the portability of the system to other computer platforms. For example, the following industry standards are integrated into our PACS: (i) UNIX operating
HUANG
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
748
system, (2) TCP/IP communication protocol, (3) SQL as the data-base query language, (4) ACR-NEMA standard (higher layers of International Standards Organization model only), (5) C programming language, (6) x WINDOWS user interface, and (7) ASCII text representation for message passing. The implications of PACS standardization include the following: Implementation of all future PACS installations is straightforward. System maintenance is easier because each module looks similar to others, if not physically, then logically. By
defining amount
the PACS of redundant
The code
therefore
operations, we minimize the computer code within the PACS system. is easier to debug, understand, and
among all levels of developers.
AJR:158, April1992
used include automatic resources and algorithms
If two PACS modules in the same department cannot communicate with each other, they become two isolated systems and can never be combined with other systems to
form a total departmental
PACS. An example
is the pediatric
radiology PACS module and the coronary care unit PACS module we developed earlier [i 2]. These two modules have no means of communicating with each other. Images and patient information are isolated within each module. An open network design architecture allows a standardized method ,
for data and message
exchange
between
heterogeneous
systems. Because computer and communications technology changes rapidly, a closed architecture would hinder system upgradability. As an example, an independent imaging console from a given manufacturer is a good addition to an MR or a CT scanner for viewing images; however, it always has a closed proprietary architecture design. As a result, no other
those
specified
by the same manufac-
turer can be augmented to the system. This limits potential upgrading and improvement of the system. Consideration of the connectivity is important even when a small-scale PACS is planned. A yes to questions like “Can we transmit images from this PACS to other systems and vice-versa?” “Does this module use a standard data and image format?” and “Does the machine use a standard communication protocol?” means that the PACS is well designed and will allow future connec-
tivity.
How to minimize the cost requires further study.
Security
it cannot be down for
extended periods. In designing a PACS, it is therefore important to use fault-tolerant measures, including error detection and logging software, external auditing programs (i.e., network management processes that check network circuits, magnetic-disk space, data-base status, processes status, and queue status), hardware redundancy, and intelligent software recovery blocks. Some recovery mechanisms we have
alternative ROUTINES
while
maintaining
high
reliability
is an important
consideration
because
of the need
trol, and the use of views. Most sophisticated data-base management systems have identification and authorization mechanisms that use accounts and passwords. Application programs may also supply additional layers of protection.
control
refers to granting
and revoking
the user’s
access to specific tables, columns, or views. These security measures provide the PACS infrastructure with a mechanism for access control to clinical and research data. With these mechanisms, the system designer can enforce departmental policy as to which persons have access to clinical studies. For example, referring clinicians may be granted image study access only after a preliminary radiology reading has been performed and attached to the image data.
Glossary AcR/NEMA
AMERICAN
MANUFACTURERS
COLLEGE
OF
RADIOLOGY/NATIONAL
ELECTRICAL
The members in the ACR-NEMA cornfor writing a standard for image information
ASSOCIATION.
are responsible
mittee
exchange
acceptable
BOOTSTRAP
by
ACR and NEMA.
both
A software
ROUTINE
block executed
by a computer
when
it is restarted. CAPTURE
Computer
COMPUTER
interposed
radiologic rnIts main task is
between
a
aging device (e.g., CT scanner) and PACS network. to standardize
the image
file format.
The computer that maintains the optical archive for a cluster of computers (e.g., neurology cluster, pediatric cluster). Its functions are similar to those of the film library CLUSTER
CONTROLLER
in a conventional
MACHINE
film-based
system.
A background computer program that runs continuously. It does not affect the system operation unless a particular circumstance arises. In that case, the program will automatically change the information flow in the system. Usually each daemon performs a specific task. DAEMON
FDDI
and displays critical patient information,
with
BOOTSTRAP
for patient confidentiality and medicolegal issues. Three major security mechanisms include account control, privilege con-
The most common baseband communication it is a relatively slow network.
connecting
Reliability is a major issue in a PACS for two reasons. First, a PACS has many components, so the probability of a component failing is high. Second, because the PACS manages
jobs
Security
ETHERNET
Reliability
retry of failed and intelligent
that allow a PACS computer to automatically continue operations after a power outage [17]. Improving reliability is costly.
Privilege
Open Architecture
except
TAIRA
primitive
search. Standardizing terminology, design concepts, and so forth, facilitates system understanding and documentation
components
AND
FIBER
DISTRIBUTED
DATA
that uses a ring topology. IPI
network
for
computers.
INTELLIGENT
PERIPHERAL
INTERFACE.
A communication
It is a medium-speed network. INTERFACE used to connect
network a computer
to a disk drive. MASTER-SLAVE
which
the
INTERFACE
master
always
The connection between initiates and requests
two
machines
operations
in
to be
performed by a slave device. PACS
PICTURE
ARCHIVING
AND
COMMUNICATION
SYSTEM.
device typically used for images. Special magnetic-disk controllers allow the bytes of a single image to be distributed over several disks. Access to the data can be done in parallel, and this is much faster than conventional disks. PARALLEL-TRANSFER
MAGNETIC
DISK
A magnetic-disk
AJR:158,
INFRASTRUCTURE
April 1992
DESIGN
5.
INTERFACE A network interface in which the communicating machines can perform redprocal operations. PULL OPERATION A network data transfer operation in which the computer that desires the data initiates the data transfer. PUSH OPERATION A network data transfer operation in which the computer with the data to send initiates the data transfer. SQL STRUCTURED QUERY LANGUAGE. It is commonly used in a later generation data base.
Downloaded from www.ajronline.org by 182.71.157.122 on 10/02/15 from IP address 182.71.157.122. Copyright ARRS. For personal use only; all rights reserved
PEER-TO-PEER
TCP/IP
TRANSMISSION
standard puters.
CONTROL
and popular
method
A commercial
ULTRANET
communication
PROTOCOL/INTERNET
for
data communication
high-speed
gigabit-per-second
7.
A
PROTOCOL.
between
6.
corn8.
fiber-optic
network.
A computer operating system commonly used in computers introduced in the late 1980s. x WINDOWS A software subsystem that allows developers to conUNIX
struct
a graphical
accessing
text,
user
interface
image-processing
(menus,
buttons,
icons,
etc.)
for
9.
10.
modules. 11.
1
2.
REFERENCES 13.
1 . Taira RK, Mankovich and implementation pediatric radiology.MR
NJ, Boechat MI, Kangarloo
H, Huang, communication
HK. Design
of picture archiving and system for 1988;150:1117-1121 2. Cho PS, Huang HK, Tillisch J, Kangarloo H. Clinical evaluation of a radiologic picture archiving and communication system for a coronary care unit. MR 1988:151:823-827 3. Lou SL, Loloyan M, Weinberg W, et al. Image delivery performance of a CT/MR PACS modu applied in neuroradiology. Proc SPIE 1991:1446:302-311 4. me G, Miyasaka K. Clinical experience: 16 months of HU-PACS. comput Med Imaging Graph 1991;15: 191 -1 95
14.
15.
16. 1
7.
OF
PACS
749
Masser H, Mandl A, Urban M, Hradil H, Hruby W. The Vienna Project SMzO in picture archiving and communication systems (PACS). In: Huang HK, Ratib 0, Bakker AR, Witte G, eds. NATO AS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag. 1991:247-250 Glass HI. Slark NA. PACS and related research in the lkiited Kingdom in picture archiving and communication systems (PACS). In: Huang HK, Ratib 0, Bakker AR, Witte G, eds. NATO AS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag, 1991:319-324 Goeringer F. Medical diagnostic imaging support systems for military medidne in picture archiving and communication systems (PACS). In: Huang HK, Ratib 0. Bakker AR, Witte G. eds. NATO AS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag. 1991:213-230 ACR-NEMA Digital Imaging and Communication Standards Committee. Digital imaging and communications ACR-NEMA 300-1988. Washington, DC: National Electrical Manufacturers Association, 1988 McNitt-Gray MF, Taira RK, Eldredge S, Razavi M. Brightness and contrast adjustments for different tissue densities in digital chest radiographs. Proc SPIE 1991:1445:468-478 Parsaye K, Chignell M, Khoshaflan 5, Wong H. Intelligent databases. New York: Wiley, 1989 Malamud C. !NGRES: tools for building an information architecture. New York: Von Norstrand Reinhold, 1989 Taira RK, Stewart BK, Sinha U. PACS database architecture and design. Comput Med Imaging Graph 1991:15: 171 -176 Ratib 0, Ugier Y, Funk M, et aL PACS workstation: user interface design. In: Huang HK, Ratib 0, Bakker AF, Witte G, eds. NATO AS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag, 1991:57-61 Huang HK, Lou SL, Cho PS, et al. RadiOIOgIC image communication methods. MR 1990:155:183-186 Stewart BK, Lou SL, Wong WK, Huang HK. An ultrafast network for communication of radiologic images. MR 1991:156:835-839 Huang HK, Kangarloo H, Cho PS, Talra RK, Ho BKT, Chan KK. Planning a totally digital radiology department. MR 1990:154:635-639 Taira RK. Chan KK, Stewart BK, Weinberg WS. PACS reliability issues. In: Huang HK, Ratib 0, Bakker AR, Witte G, eds. NATOAS! series F, PACS in medicine, vol. 74. New York: Springer-Verlag, 1991:149-156
