An Economical, Personal Computer based Picture Archiving and Communication

inforad An Economical, Personal Computer based Picture Archiving and Communication System 1 Tain-Chen Wu, MD San-Kan Lee, MD Chen-Hsing Peng, PhD Chia-Hsien Wen, PhD Shu-Kun Huang Taichung Veterans General Hospital has been developing a hospital-wide picture archiving and communication system (PACS) since 1993. A personal computer based environment was implemented to reduce costs (only $2,500 for each view station) and take advantage of distributed system techniques. Other features of the PACS are automatic image acquisition, hierarchic storage management, efficient image transmission, robust fault tolerance, and user-friendly image manipulation. The system is integrated with the hospital information system so that Chinese-language patient data can be automatically transferred. A four-tier storage hierarchy and a multipath search strategy are used to improve reliability and efficiency. Image compression and efficient image transmission techniques (autorouting and prefetching) are used to reduce the response time. Robust fault tolerance is achieved with fault-tolerant hardware, image replication, and a system watchdog. User-friendly image manipulation features include easy adjustment of the brightness, contrast, or quality of the displayed image; several windows for image display; and image measurement capability. The PACS currently supports computed tomography, ultrasound, magnetic resonance imaging, computed radiography, and digital fluoroscopy; almost all appropriate personal computers in the hospital can be used as view stations. Users are satisfied with the quality, reliability, and performance of the system. Abbreviations: CD-ROM = compact disk, read-only memory; DICOM = Digital Imaging and Communication in Medicine; HIS = hospital information system; PACS = picture archiving and communication system; PC = personal computer Index terms: Computers Images, storage and retrieval Picture archiving and communication system (PACS) RadioGraphics 1999; 19:523 530 1 From the Administrative Office (T.C.W.), Department of Radiology (S.K.L., S.K.H.), and Computer and Communication Center (C.H.P., C.H.W.), Taichung Veterans General Hospital, 160 (section 3) Taichung Kang Rd, Taichung 40705, Taiwan; and the Department of Diagnostic Radiology, National Defense Medical Center, Taipei, Taiwan (S.K.L.). Presented as an inforad exhibit at the 1997 RSNA scientific assembly. Received April 20, 1998; revision requested July 15 and received August 21; accepted September 8. Supported by joint grant VGHTH 85-024-3 in memory of Chi-shuen Tsou, MD, from the Medical Research Advancement Foundation to the research programs of Taichung Veterans General Hospital and National Tsing-Hua University. Address reprint requests to S.K.L. RSNA, 1999 523

INTRODUCTION There has been considerable interest in the development of picture archiving and communication systems (PACS) since the 1980s (1). Some controversies still remain, particularly the issue of whether PACS are ready for routine use (2,3). In our opinion, PACS are ready for routine use for consultations between clinicians and radiologists. Nevertheless, further technical innovations to improve PACS and reductions in equipment costs must await the development of more practical applications (2). Reliability, response time, user-friendliness, connectivity, and upgradability are the main concerns of end users (4). With these guidelines in mind, we started to develop a PACS at Taichung Veterans General Hospital in 1993 (5). The first PACS implemented at this institution was the image capture and communication system reported by Yang et al (6). The success of this system prompted us to extend it to provide hospital-wide service. The system is based on personal computers (PCs) and is integrated with the existing hospital information system (HIS). The goal of the system is to ensure image quality, system reliability, service availability, and good performance. More important, the system provides a Chinese-language interface environment for medical personnel at our hospital so that training costs can be greatly reduced. Details of the design and performance of the system are discussed later in this article. Although implementing a PACS at a large hospital requires enormous resources, it is misleading to assume that developing a PACS requires a costly investment. A large storage system may be needed due to an extremely high image volume. High-quality and expensive view stations may also be required for clinical purposes. Further, processing images in real time may require high performance and communication power for the view stations and high-speed networks. All of these costs make developing even a small-scale PACS appear not to be cost-effective. In this article, we demonstrate that it is possible to reduce these costs while maintaining the quality, reliability, and performance of the system. DESIGN OF THE SYSTEM A PC-based System Our PACS uses PCs as major view stations and image acquisition stations. All of the required servers are Pentium (Intel, Santa Clara, Calif) based computers. There are several reasons why we adopted a PC-based system. First, use of PCs significantly reduces the cost of the system. A view station used in our system, such as a Pentium II central processing unit (CPU) with 64 Mbytes of random-access memory (RAM) and a 17-inch (43-cm) monitor, costs only $2,500. This price is affordable for any hospital that would like to establish a low-cost hospitalwide PACS. In particular, there is no need to worry that the investment would be wasted. Second, the open architecture of PCs makes it easy to get sufficient support for both hardware and software and allows future upgrades of the system. With Windows 95 (Microsoft, Redmond, Wash) used as the platform, a PACS user can easily and simultaneously integrate other systems and applications (eg, the HIS, office automation packages, Internet applications) on the PC. Third, because most of the hospital staff is familiar with Windows 95, the costs of training and of maintaining the system are considerably reduced. Some may question the image quality of a PC-based PACS. The quality of an image is generally determined by the spatial resolution and gray-scale resolution (ie, brightness). Table 1 lists the imaging modalities available with our PACS, the acquisition devices and techniques, and the suggested spatial resolution and brightness of the images (1,7). To ensure the quality of displayed images, the monitor of each view station is approved by experienced technologists and radiologists. Users who perform image-based diagnosis, such as radiologists and physicians in the emergency department, are provided with high-resolution ImageSystem (Minnetonka, Minn) M21P monitors. Seventeen-inch (43-cm) ViewSonic (Walnut, Calif) PT775 monitors are provided to other clinicians for image reference. Although the latter monitors are much cheaper than the former, our users are satisfied with the quality of the displayed images. To make the system more reliable and efficient, some distributed system techniques (8) are used in our PACS. For example, we distrib- 524 inforad Volume 19 Number 2

Table 1 Imaging Modalities Available on the PACS Spatial Method of Resolution Brightness Image (pixels (bits per Imaging Modality Imaging Unit* Acquisition per image) pixel) Computed tomography (CT) Siemens Somatom DR3 Frame grabbing 512 512 8 Picker 1200 SX Frame grabbing 512 512 8 Picker PQ2000 DICOM 512 512 12 Ultrasound (US) Aloka SSD650 Frame grabbing 512 512 8 Advanced Technology Frame grabbing 512 512 8 Laboratories HDI Diasonics SPA 1000 Frame grabbing 512 512 8 Magnetic resonance (MR) Picker Vista Magnetic tape 256 256 12 imaging GE Signa 1.5T DICOM 256 256 12 Computed radiography Fuji 9000C DASM 2,048 2,048 10 Fuji 9000HQ DASM 2,048 2,048 10 Digital fluoroscopy Philips Diagnost 97 DICOM 512 512 10 GE Legacy DICOM 1,024 1,024 8 *Full names and locations of the manufacturers are as follows: Advanced Technology Laboratories, Bothell, Wash; Aloka, Tokyo, Japan; Diasonics Ultrasound, Santa Clara, Calif; Fuji Systems, Tokyo, Japan; GE Medical Systems, Milwaukee, Wis; Philips Medical Systems International, Best, the Netherlands; Picker International, Cleveland, Ohio; and Siemens, Erlangen, Germany. DICOM = Digital Imaging and Communication in Medicine. DASM is an external small computer systems interface (SCSI) device of Fuji DMS workstations for exporting images. ute the data and images into local computers and local servers to improve the fault tolerance of the system and eliminate the transmission bottleneck. In addition, we dispatch tasks to cooperative PCs to increase the computing power of the system. These techniques are explained in detail later in this article. Automatic Image Acquisition One of the requirements for implementation of our PACS was that images must be acquired automatically (ie, without human intervention). Different imaging modalities may require different acquisition techniques (Table 1). For an imaging modality without the capability for digital image exporting, the images are captured by connecting the video cable of the console to a frame-grab card installed on an acquisition station (6,9). Otherwise, an acquisition PC is connected to one of the output ports of the imaging unit to receive the output images directly. Because the DICOM standard protocol (10) has been well supported by most manufacturers in recent years, we have requested that all newly purchased imaging units comply with the DICOM protocol. Therefore, DICOM will be the only acquisition method for our PACS in the near future. The PACS is integrated with the HIS at our institution to provide seamless service. A patient s demographic data and examination information are stored in the HIS. As soon as a physician issues an image examination order, the HIS generates a unique serial number, which is termed the order number. The demographic data and examination information related to the order are automatically transferred to the order database server of the PACS via the HIS gateway. For every examination, a March-April 1999 Wu et al RadioGraphics 525

Figure 1. Flow of image acquisition for a computed radiography system. (1) When an order number is given at the identification terminal (IDT) gateway, the patient s demographic data and examination information are transferred into the computed radiography (CR) system. (2) The images are read at the computed radiography reader, then automatically routed to a data management system (DMS) workstation and the acquisition station. (3) The images and patient data are automatically combined by the acquisition station and stored in the image center. DASM = an external small computer systems interface device of Fuji DMS workstations for exporting images, ID = identification. technologist only has to enter the order number at the identification terminal (IDT) gateway. The acquisition station automatically links each image of the examination to the order and records them in the index database of an image center. By this means, the patient data in the PACS are consistent with those in the HIS. Figure 1 shows the flow of image acquisition for a computed radiography system. Coupling the PACS with the HIS can obviate reentering of order data and patient data, simplify imaging operations, and eliminate possible typing errors. Furthermore, although our computed radiography system does not manipulate Chinese characters directly, we can print a patient s Chinese name on film images with the Japanese kanji code by converting the patient s name from the Chinese Big-5 code into the Japanese Industrial Standards (JIS) code via the identification terminal gateway. Hierarchic Storage Management A well-structured storage hierarchy improves the reliability and efficiency of a PACS. For our PACS, we designed a four-tier storage hierarchy. Replicated temporary images are stored on local 4-Gbyte hard disks of the view stations or in 16-Gbyte cache servers. Current acquired images are stored in the system image database, which is served by a Digital Equipment (Maynard, Mass) Prioris HX 5166 server with a Novell (San Jose, Calif) 3.12 operating system and a 200-Gbyte Storage Computer (Nashua, NH) RAID-7 disk subsystem. Image archiving is done in a group of 100-piece Pioneer (Tokyo, Japan) DRM-1004X compact disk, read-only memory (CD-ROM) jukeboxes. When an image is captured by an acquisition station, the image is sent to the system image server and stored there (Fig 2). The image location and data related to the order are recorded in the index database simultaneously. As soon as the newly stored images exceed 650 Mbytes, they are automatically copied onto a CD-ROM in the 100-piece jukebox writer. When all of the CD-ROMs in the writer are recorded, they are manually moved to a jukebox reader. The group of CD-ROM readers forms an image archive. Some images may be replicated and stored locally at view stations or a cache server for fast retrieval. To find a requested image, a strategy called multipath search is implemented for view 526 inforad Volume 19 Number 2

Figure 2. Architecture of the image center. (1) Acquisition stations send image files to an image server that has a redundant array of inexpensive disks (RAID) device. (2) Incoming images are automatically copied into the CD-ROM jukebox writer. (3) When all of the CD-ROMs in the jukebox writer are recorded, they are manually moved into the jukebox readers. (4) The index database records the patient data and the location where the corresponding images are stored. stations in our PACS. When an image is requested, the view station used for retrieval searches the local hard disk first. If the image is not on the local disk, the retrieval process shifts to the related cache server. If the image is not in the cache server, the index database is used to find the image. If it is a current image, it will be retrieved from the central image server; otherwise, it will be retrieved from one of the CD-ROM jukebox readers. Several advantages can be realized from this four-tier storage hierarchy. This storage hierarchy makes the system more reliable because at least two copies of a current image are stored in the system at any time. The multipath search strategy relieves the central image server from access congestion and makes image retrieval more efficient. Instead of an expensive terabyte jukebox, we use CD-ROM jukebox readers as the image archive so that the storage can be expanded progressively. In addition, the highcapacity digital video disk (DVD) ROM drivers can accept the formats of all current CD-ROMs. This capability allows us to upgrade drivers without changing the jukeboxes whenever new DVD-ROM writers are available. Efficient Image Transmission A digitized radiographic image generally requires up to tens of megabytes and several seconds for transmission. This fact may lead one to believe that only an expensive high-speed network can support PACS. However, this belief is not necessarily true because the performance of a PACS is greatly determined by the user response time. In this section, we describe how performance can be improved by reducing the required network bandwidth without sacrificing much response time. Instead of the expensive asynchronous transfer mode (ATM) network, a 100-Mbit/sec fast Ethernet is used to transmit images in our PACS. The average response time is 0.5 seconds for retrieval of a CT image and 3.3 seconds for retrieval of a computed radiograph. Image replication, compression, and transmission techniques can be used to reduce the response time. In this section, we focus on image compression and transmission. Image compression involves reduction of image data for storage and transmission. Two types of compression are used: lossless and lossy. Lossless compression of gray-scale images generally results in no more than 3:1 data reduction, whereas lossy compression of grayscale images may result in as much as 30:1 data reduction (11). However, lossy compression is usually not preferred for image archiving because a reconstructed image is only an approximation of the original. As a result, a lossless March-April 1999 Wu et al RadioGraphics 527

Figure 3. CT image of the head displayed on a view station of the PACS. The display window can be adjusted by clicking on an icon or changing the settings on the histogram. compression technique called GZIP (12) is used in our PACS. The compression ratio achieved with GZIP ranges from 2.4:1 to 3.3:1 according to the imaging modality. We are also studying a region of interest based compression technique, which allows a higher compression ratio to be achieved by compressing regions of interest with a lossless technique and compressing other areas with a lossy technique (13). Methods of image transmission include autorouting and prefetching. Autorouting is usually applied to images with known distributed destinations. For example, as soon as images of an inpatient are generated, they are automatically transferred to and stored in the view stations of the ward where the patient is hospitalized. Whenever the image file is referenced, it is retrieved from the hard disk of the view stations instead of the system image server. Prefetching is applied to images with unknown distributed destinations. If a patient makes an appointment for a return visit, images from the patient s recent examinations are sent to the outpatient image cache server before the return visit. The requested images can then be retrieved directly from the cache server. Some improvements related to image transmission have made the system more userfriendly. For instance, it may be desirable to transmit a set of images from an acquisition station to the image database server in background execution. As soon as an operator submits the request, the system will be ready to accept the next command. Although this feature may not significantly accelerate image access, the prompt return of the system can free users to perform the next task. Robust Fault Tolerance Because our PACS is designed to reduce the user s intervention as much as possible, robust fault tolerance is necessary (14). The easiest way to achieve this feature is by using fault-tolerant hardware. For example, our PACS uses RAID-7 (Storage Computer) as the image database, and RAID-7 is itself fault tolerant. The system implements a store-and-forward strategy for acquisition stations to ensure that no image is lost during acquisition. An incoming image is stored in the local hard disk of an acquisition station and is not deleted until successfully sent to the image center. If any problem occurs in the image center or the network, the acquisition station retains the image and 528 inforad Volume 19 Number 2

Table 2 Cost Breakdown by Year for the PACS Investment Year (dollars) 1993 86,000 1994 143,000 1995 201,000 1996 198,000 1997 283,000 1998 245,000 Table 3 Images Acquired on July 20, 1998 Amount of Imaging No. of No. of Storage Modality Patients Images (Mbytes) Radiography 266 533 1,593 CT 76 2,648 1,555 MR imaging 26 2,130 279 US 19 336 110 Total 387 5,647 3,537 continues trying to send it until the image is transmitted successfully. Other strategies of image replication were discussed in previous sections. In addition, there is a system watchdog that monitors every server and acquisition station periodically. Each server and acquisition station is required to issue a status report to the watchdog every 5 minutes. If a server or acquisition station misses a status report, that server or acquisition station is assumed to be out of order and is automatically reset by the watchdog. The system operator is also notified about what has happened. User-friendly Image Manipulation Finally, our PACS provides many user-friendly functions for image manipulation. These functions can be used to support image-based diagnosis. A user can easily change the brightness or contrast of the displayed image by moving the mouse left and right or up and down. The user can also adjust the quality of the image by specifying an appropriate convolution mask or selecting an image enhancement filter. Because a monitor generally displays images with only 256 gray levels, it is important to define an appropriate window level for image display for diagnosis when more than 256 gray levels are used. The system offers three icons for quick selection of an appropriate window for displaying the lungs, soft tissue, or bone (Fig 3). Simply clicking on one of these icons causes the image to be displayed within the selected window. The window used for display varies with the imaging modality and is predefined by experienced radiologists and physicians. Clicking on another icon produces a histogram of the image; with this histogram, the window for image display can be defined by the user. In addition, the system provides icons for image measurement. For example, a user may measure the length of a specified segment; the degree of a specified angle; or the area of a specified rectangle, ellipse, or random region. PERFORMANCE OF THE SYSTEM Our PACS has evolved in several stages since late 1993. Our total investment in the system has been approximately $1,156,000, which includes the costs of storage devices (30.1%), view stations (20.4%), system software (16.5%), acquisition stations (14.1%), servers (12.1%), and diagnosis stations (6.9%). The cost breakdown by year is shown in Table 2. In addition, the total cost for setup of a hospital-wide highspeed network was approximately $332,000. Currently, there are 77 view stations in service: nine in the emergency department, six in the radiology department, nine in intensive care units, 28 in inpatient nursing stations, and 25 in the offices of clinical departments. Four diagnosis stations are installed in the radiology department and emergency department. To keep our PACS current, we continuously make efforts to improve the system. We began by implementing CT image acquisition for the emergency clinicians. US, inpatient CT, and MR imaging were gradually integrated into the system. In late 1995, the computed radiography system was connected to the PACS. Since then, all newly purchased image devices are required to be DICOM compatible, a feature that makes image acquisition easier. To keep the view stations in optimal working condition, they are upgraded or replaced every 2 years. Most important, all programs are reviewed on a regular basis to ensure that the daily process runs smoothly. Currently, Taichung Veterans General Hospital has 1,359 beds and treats 3,800 outpatients daily. All radiologic imaging except outpatient radiography is supported by the PACS. The average number of images acquired per day is approximately 5,500 (Table 3); the total number March-April 1999 Wu et al RadioGraphics 529

Table 4 Total Images Acquired as of July 30, 1998 Amount of No. of No. of Storage Period of Type of Storage Patients Images (Gbytes) Image Acquisition RAID-7* 21,221 251,543 177 May 1998 to July 1998 CD-ROM jukebox 85,600 1,012,000 715 July 1997 to May 1998 Off-line CD-ROM 23,300 279,000 195 March 1997 to June 1997 Total 130,121 1,542,543 1,087 March 1997 to July 1998 Note. The CD-ROM jukebox writer was installed in February 1997; images acquired before then were purged. *Manufactured by Storage Computer. of images in the system is approximately 1.5 million (Table 4). CONCLUSIONS Our PACS was designed and initiated in 1993 for emergency CT service. Since then, the PACS has been improved and now includes US, MR imaging, computed radiography, and digital fluoroscopy. Methods of image acquisition consist of frame grabbing, magnetic tape, DICOM, and DASM (an external small computer systems interface device of Fuji DMS workstations for exporting images). All images except plain radiographs acquired in outpatients can be accessed with the PACS. The system will be completed after digitization hardware is purchased. Tremendous benefits of the PACS in the consultation process have been noted by both clinicians and radiologists. Better image quality and rapid access to film images are appreciated by clinicians. The presence of the PACS also reduces the number of trips between the darkroom and imaging suite and decreases the circulation of film jackets. We expect new benefits to be generated as long as we continue to improve the system. In the future, filmless radiology will be possible. REFERENCES 1. Huang HK. Picture archiving and communication systems in biomedical imaging. New York, NY: Wiley-VCH, 1996; 5 7. 2. Steckel RJ. The current applications of PACS to radiology practice. Radiology 1994; 190(3): 50A 52A. 3. Friedenberg RM. PACS in the clinical setting: revisited. Radiology 1994; 190(3):53A 54A. 4. Gur D, Fuhrman CR, Thaete FL. Requirements for PACS: users perspective. RadioGraphics 1993; 13:457 460. 5. Yang CW, Chung PC, Chang CI, Lee SK, Wen CH, Kung LY. A hierarchical model for PACS. Comput Med Imaging Graph 1997; 21:29 37. 6. Yang CW, Chung PC, Lee SK, Chang CI, Wen CH, Kung LY. An image capture and communication system for emergency computed tomography. Comput Methods Programs Biomed 1996; 52:139 145. 7. Rompelman O. Medical image compression: possible applications of subband coding. In: Woods JW, ed. Subband image coding. Norwell, Mass: Kluwer Academic, 1991; 319 352. 8. Tanenbaum AS. Distributed operating system. Paramus, NJ: Prentice Hall, 1995. 9. Lee SK, Huang SK, Yang CW, Peng CH, Wen CH. A design of ultrasound image acquisition for picture archiving and communication system. J Med Ultrasound 1998; 6:68 73. 10. Hindel R. Implementation of the DICOM 3.0 standard: a pragmatic handbook. Oak Brook, Ill: Radiological Society of North America, 1994. 11. Gonzalez RC, Woods RE. Digital image processing. Reading, Mass: Addison-Wesley, 1992. 12. Deutsch P. GZIP file format specification version 4.3. Internet Engineering Task Force RFC 1952, May 1996. Available at: ftp://ftp.isi.edu/ in-notes/rfc1952.txt. 13. Chen ZD. Adaptive predictive MAR models for the medical image compression. Thesis. National Chung Cheng University, Chiayi, Taiwan, 1996. 14. Tucker DM, McEachern M. Quality assurance and quality control of an intensive care unit picture archiving and communication system. J Digit Imaging 1995; 8:162 167. 530 inforad Volume 19 Number 2