Evolution of Technical and Organizational Concepts in the Innsbruck PACS Project: a Review of the First Year of Operation

Raimund Vogl1, Ph.D., Siegfried Peer2, M.D., Werner Jaschke2, M.D.
1 IT Department, Hospital Holding Company TILAK
2 Department of Radiology, University of Innsbruck

Introduction

Following about half a year of preparatory analysis of the workflow processes and organizational structures of the traumatology outpatient department and the traumatology branch of the radiology department, installation of a very homogenous large scale PACS started in March of 1997 in the newly commissioned traumatology building of Innsbruck University Hospital. In June 1997, the traumatology outpatient department started with film free operation in the new building, relying heavily on the PACS and its sophisticated interfaces to the traumatology information system (TIS), the radiology information system (RIS) and the imaging modalities, which were designed for optimum support of the workflow processes. Following a consolidation phase, PACS was expanded to further branches of the radiology department. To provide for adequate and even improved PACS performance with rising data volumes to be processed, tuning of the network was necessary and adapted concepts for image distribution had to be found.

Configuration of the Innsbruck PACS System

The initial setup of the Innsbruck PACS was concentrated in the traumatology outpatient department and consisted of 14 reporting stations, 12 viewing stations, and 2 archive server systems (one node dedicated as a database server, the other as a file server), all systems being UNIX based IBM RS/6000 systems. 3 Agfa ADC 70 phosphor plate systems, 1 GE CT system, 1 ATL ultrasound unit and 1 Philips Integris angiography unit were initially connected to the PACS, producing an average of 3.6 GB of data daily. Since then, 7 additional workstations, 1 GE CT system, 2 more angiography units and 2 fluroscopy units have been added. Currently, efforts are under way to connect 2 Siemens MR systems, 2 additional phosphor plate systems and 15 additional workstations to the PACS. The original design for image access was to route all newly acquired images to the reporting stations, which themselves can function as image servers to their peers, thus avoiding the bottleneck of only one image instance on the central archive system with all reporting and viewing stations accessing them (this distributed image database design allows to fully exploit the advantages of a switched high speed data network like ATM and was the key argument for the decision for the IBM/Tiani PACS). It showed that, especially for the reporting radiologists, the slowing down of their workstations due to remote image loading was cumbersome, so that the design of a dedicated cluster of 4 image server systems, handling the up to 5 GB of new and up to 10 GB of prefetching data daily, relying on some 200GB of fast SSA (IBMs serial storage architecture) disks, was created. These systems also provide the means to attached the growing number of CD-R jukeboxes (each containing 500 media of 650MB capacity) used for archiving. CT, XA and MR images are losslessly compressed and in this way archived and transferred via the network, thus reducing the demand for archive media and networking bandwidth. For CT images, average compression rates of 56% could be achieved; CR images are imported in 8 bit resolution from the phosphor plate systems and show a much lower compressibility and are thus stored uncompressed. The data network originally consisted of 2 Fore Systems ATM switches to attach the PACS workstations and archive systems with 155Mbit/s links and 2 ethernet switches to attach the imaging modalities and to provide a service network for the PACS workstations. Since then, 5 more ATM switches have been added, occasionally using 622Mbit/s links (see figure 1). Initially, only LAN emulation 1.0 was employed as a means to transmit the TCP/IP traffic over the ATM network, but the network throughput of about 5Mbyte/s (memory to memory, benchmarked between two RS/6000 workstations using the ”ttcp” program) showed to far from optimum and the software drivers for the RS/6000 machines showed occasional failures under heavy load. The migration to classical IP proved to be of great effect, giving a very stable network with average transfer rates of 10 Mbyte/s and peak rates of up to 15 Mbyte/s. For the further expansion of PACS to the wards, a Windows NT based version of the PACS viewing software will be used and the networking technology of choice is 100 Mbit/s ethernet, allowing to cut costs at only a small performance penalty. To provide good support for the workflow processes, interfaces between PACS and the RIS (MEDORA), the traumatology informations system (TIS MEDDOC) and the central hospital information system (MEDAS) have been created based on the industry standards DICOM and HL7. Worklist functionality for the imaging modalities, based on DICOM worklist or on proprietary protocols, is employed, improving data integrity and working efficiency (see figure 2).

Result

Despite the very limited time for installation, completely film free operation in the traumatology outpatient department with its high demand for rapid availability of images and continuous operation 24 hours a day could be achieved from the very start. Down time of the PACS is very low and mainly due to scheduled maintenance work (a few hours per month). Unscheduled down times were rare and mainly due to networking problems. The transition from the initially adopted LAN emulation to classical IP in the ATM network helped to remedy the occasionally experienced networking problems and to dramatically boost network throughput . After tuning of the PACS database, image query and loading speeds are very fast (approximately 2 seconds per CR image). Prefetching information not only from the RIS, but also from the traumatology information system proved to be essential for efficient operation in the traumatology outpatient controls with up to 200 patients a day. For smooth operation, a change of the ”routing policy” for image distribution to the workstations and an upgrade of the RAID-system for keeping images online for at least 4 weeks became necessary. With a daily data production of 3.6GB in average and the prospect for an increase to about 6GB in the immediate future, the CD-R jukeboxes used are close to the limit of their writing performance. After one year of operation, certain desired but not indispensable features of the DICOM interface between RIS and PACS are still to be realized. With the expansion of PACS to other radiology department on campus and to the wards, establishing an elaborate  concept for access restriction is currently of key importance. Demand on personal resources especially in the initial phase of establishing and consolidating the PACS turned out to higher as initially expected.

Discussion

An ongoing review of system performance, continuous feedback from the users and close interaction between the hospital PACS project team and the software solution provider is essential for establishing a refined and efficient working environment utilizing all the potential benefits of  information technology. PACS, especially when it has to be imbedded into the complex context of a large university hospital with various information systems and imaging modalities from diverse vendors to be interfaced, is far from being a plug ‘n’ play system but rather a tailor made solution requiring long and strenuous fine tuning, especially in the configuration of the ”industry standard” DICOM and HL7 interfaces. For these tasks, a well trained team of IT technicians is indispensable for the hospital and the fact that about half a year of onsite working experience is required for a technician to provide valuable support has to be considered. The decision for ATM as networking technology proved to be of advantage in the long term because of its advanced features for bandwith scalability through multiple parallel links and network link failover, even though additional training for the network technicians was required and the now widely available 100Mbit/s Ethernet delivers similar network performance at a lower price.


Oral presentation at EuroPACS'98, Barcelona, Spain