Implementing an ATM-Based Teleradiology Network

There are numerous applications in the medical field that can benefit from deployment of high-speed switching technologies. In this author's opinion, three major factors will drive consideration of a high-speed, broadband network for delivery of cost-effective quality healthcare in the future:

High bandwidth imaging applications such as teleradiology.
The growing volume of medical and patient data.
Emerging decision support systems using artificial intelligence database queries that require enormous bandwidth resources.

One of the emerging high-speed switching technologies which will support a wide variety of medical imaging and data applications is ATM (Asynchronous Transfer Mode). It offers the performance advantages of switched connection-oriented channels, and allows users to avoid the bandwidth problems inherent in shared-media networking systems. Key benefits of ATM include:

The ability to intermix traffic with different latency and delay requirements such as voice, video, and data on the same communication circuit, providing a certain quality of service.
A fixed cell-size that allows it to support bursty traffic with stringent performance requirements.We envision ATM enterprise switches to initially be deployed within corporate backbone networks.

OSF HEALTHCARE SYSTEMS

OSF Healthcare Systems, based in Peoria, Illinois, is an integrated healthcare provider which includes seven hospitals, two long-term care facilities, and over 50 independent physician practices. Spread over three states -- Illinois, Iowa, and Michigan -- OSF decided to take advantage of high-speed, broadband network technologies, such as frame relay, ATM, and SONET, to tie its diverse pieces together. At issue were two key problems:

Medical specialists, such as radiologists, traveled extensively between geographically dispersed sites to meet critical healthcare needs. OSF hoped to reduce -- even eliminate -- travel time by using high-speed switched networks to transfer patient data and images.
OSF staffers had determined that the lag between time of treatment and billing was the primary cause of billing disputes. OSF hoped a patient-centered network would reduce the lag time which, in turn, would translate into cost savings.

The goal at OSF Healthcare Systems was to design a network where the patients' clinical and financial data could follow them throughout the system whether at the hospital, a patient-care facility, or the doctor's office. This patient-centered data network would:

Seamlessly integrate local area networks with wide area networks.
Support bandwidth intensive applications such as imaging.
Support workgroup applications.
Support software-defined changes in network topology wherever possible.
Offer ubiquitous access to high-speed data.

In the early phases, OSF plans to use ATM within Peoria over local and metropolitan area networks, while using frame relay to tie-in remote sites. However, as the cost of ATM falls, it is hoped that all the sites within the OSF Healthcare system will be connected via ATM-based networks. To demonstrate the switching technology in a patient-centered data network, it was decided to implement a test ATM-based local teleradiology network. The remainder of this article details the ATM test project and reviews the overall project status to date.

ATM PILOT OBJECTIVES

The primary goal was to test the viability of ATM as an efficient network backbone that would support high-speed, bursty applications such as data and image transfers. Using a two-node ATM network, the OSF Healthcare Systems corporate MIS data center was connected to the Saint Francis Medical Center (SFMC) via a 155 Mb/s fiber optic link. Located in downtown Peoria, these sites are two blocks apart. The objectives were to demonstrate:

The effectiveness of ATM as a network technology.
Seamless connectivity between Ethernet and ATM-attached clients. This means a client on an Ethernet network would be able to transparently communicate with an ATM-attached client without having to manually set up the connection within the ATM switch first.
Switched Virtual Circuits (SVC) and Permanent Virtual Circuits (PVC) connections.
The viability of image transfers from an MRI machine to a radiology workstation via ATM.

THE SFMC ATM NODE
Figure 1 illustrates the ATM node deployed at the SFMC. It consists of an ATM switch, an EtherCell, SUN workstations, and an MRI machine. The principal goal is to demonstrate the ability to transfer radiology images from the MRI machine to a UNIX-based workstation via the ATM switch. Note that the MRI machine is on an Ethernet network attached to the EtherCell which, in turn, is connected to the ATM switch via a fiber-optic connection at 155 Mb/s. The patient study (a collection of patient images) can be shared between two radiologists (workstations) within the same physical facility or between different facilities. Currently, the images (in hard copy film) are sent between facilities via courier which can take anywhere from several hours to several days depending on the location.

ETHERCELL FEATURES
The EtherCell shown in Figure 1 has 12 Ethernet ports and one ATM port. Each of the 12 ports provides a 10 Mb/s dedicated Ethernet segment to which individual servers/PCs or conventional Ethernet HUBs (with several clients attached to these HUBs) can be attached. The ATM port (155 Mb/s) connects to an ATM switch. It is perfectly suited for building hybrid Ethernet/ATM networks. The main features of the EtherCell are highlighted below:

Allows users to build hybrid Ethernet/ATM networks.
Multiplexes Ethernet traffic into the ATM link.
Functions as an ATM proxy for each Ethernet attached client.
Supports up to 1,024 Ethernet clients.
Segments and reassembles Ethernet frames into ATM cells and vice-versa.

THE ATM SWITCH
The LattisCell™ ATM switch facilitates call set-up between network end points with the flexibility of allocating precise bandwidth, quality of service, and performance guarantees. Each of the switches deployed in the OSF network have 16 individually configured user-defined ports, each operating at 155 Mb/s. The bandwidth requirements of the individual ports can be configured via software at connection set-up. For example, a videoconferencing session between ports three and four could be allocated 32.45 Mb/s if desired. When a connection is requested, the switch determines if there are available resources; if there are not, the request is denied. This type of performance guarantee is not available with conventional shared bandwidth media or switched media. Some of the main features of the switch are that it:

Supports a maximum of 16 ports.
Supports 1,024 cell output buffers per port.
Is a high-performance internal non-blocking switch fabric.
Has dual output queues to support high priority and normal traffic.
Provides permanent Virtual Circuit (PVC) and Switched Virtual Circuit (SVC) support.

TYPES OF CIRCUITS
The concept of virtual connections is similar to traditional switching networks, but the actual mechanics are different. There are two basic types of virtual circuits: switched and permanent.

A switched virtual circuit is dynamically established on an as-needed basis. A Virtual Circuit Identifier (VCI) and Virtual Path Identifier (VPI) are established for each direction of communication. Therefore, for every full-duplex SVC connection, there will be two pairs of VPI/VCI assigned by the switch. To benefit fully from ATM switching technology, SVC implementation is a must because this allows users to add and delete connections dynamically.

A permanent virtual circuit is established from a source end-point to a destination end-point without the ability to dynamically release the connection. PVCs are manually established and must be manually released.¹

THE RADIOLOGY WORKSTATION
The radiology workstation is UNIX-based, with an ATM network interface card that enables it to connect to the ATM switch at 155 Mb/s. Figure 2 shows the ATM protocol architecture. SONET and ATM form layers one and two respectively. The ATM Adaptation Layer (AAL5) forms the periphery of the ATM network, which converts conventional LAN packets into ATM cells and vice versa. Upper layer protocols, such as Internet Protocol (IP) and Internet Packet Exchange Protocol (IPX), are encapsulated within LAN emulation headers before being segmented into ATM cells.

The ATM interface cards have several software components, including:

LAN Emulation² (LANE) allows ATM attached workstations to communicate with Ethernet attached workstations. The LANE software running on the ATM clients emulates the client as if it were on an Ethernet network. The advantages of LANE are that it:

Controls the flow of multicast and broadcast traffic.
Provides address resolution.
Maintains the database of virtual LAN domains.

The ATM forum specifies that the key to handling multicast/broadcast traffic from an Ethernet LAN is a Broadcast Unknown Server (BUS). To function as today's shared LANs, the LAN Emulation Configuration Server (LECS) enables switched LANs to be configured in broadcast domains or virtual LANs. Each emulated LAN receives and broadcasts traffic over its own specific BUS call tree. The LAN Emulation Server (LES) uses its own call tree to perform various tasks such as address resolution. LANE deliberately hides the ATM, allowing virtually any network layer to operate over ATM.

User-Network Interface (UNI) signaling, as defined by the ATM Forum, specifies how a client will communicate with an ATM switch.

SVC operation allows ATM-attached clients to dynamically set up connections as and when required.

The Interim Link Management Interface (ILMI) protocol uses Simple Network Management Protocol formatted packets across either a User-Network Interface or Network-to-Network Interface to access the ILMI Management Information Base for that particular link. The most important feature is address registration that facilitates the registration of ATM addresses.

Figure 3 illustrates a typical radiology workstation with ATM connectivity. We have attached an external high-resolution monitor for viewing x-rays. The features of this workstation include:

High resolution monitor (2,000 X 2,000 pixels).
ATM-ready UNIX workstation.
Transmit and receive NTSC-quality video at 30 frames per second.
Built-in JPEG compression.
CD-quality sound.

Using videoconferencing software, we enable doctors to set up a full-motion videoconferencing session with CD-quality audio. It is important to stress the audio quality, because this high-quality sound allows doctors to transmit and receive audio clips of a patient's heartbeat, lungs, etc.

CORPORATE MIS ATM NODE

The corporate node is the heart of the ATM domain because it controls the ATM end-points at both the corporate and SFMC sites. The Connection Management System (CMS), the Network Management Application (NMA), and the Multicast Communication Server (MCS) are all located at the corporate facility. ATM connection management involves resource reservation, maintaining status of connections, bandwidth allocation, and congestion control. This is a far cry from traditional management systems deployed in today's legacy networks. Figure 4 illustrates the different components of the network. The System 3000™ is a switching HUB that emulates a SUN workstation. Each built-in SUN card implements the CMS and the NMA, respectively, that will be described in more detail later. The EtherCell clients shown are essentially DOS-based clients with IPX and IP stacks that will communicate with Novell servers attached to the EtherCell.³

CONNECTION MANAGEMENT SYSTEM FEATURES
The CMS which is responsible for setting up and managing connections within a switch or multiple switches is the heart of the ATM network. Important features of CMS are:

Provides call management, i.e., call set-up, main-tenance, and tear-down.
Manages on-demand SVC and PVC connections.
Ensures that there are adequate resources available to support traffic.
Automatically re-routes calls in the event of a network failure.
Collects network management information.

MULTICAST COMMUNICATION SERVER FEATURES
The MCS and the CMS work together to manage the ATM network. There are several important features of the MCS:

It provides broadcast and multicast data distribution.
It administers the multicast call trees between clients in a virtual local area network (VLAN).
It selectively routes broadcast and multicast data to clients that are members of the same VLAN.
Briefly, a VLAN can be defined as a broadcast domain that is not geographically contained and can be comprised of arbitrary groups.

NETWORK MANAGEMENT APPLICATION FEATURES
The NMA uses three protocols to communicate with all management agents.

Simple network management protocol.
User datagram protocol.
Internet protocol.

The NMA works with the CMS software to provide a graphical user interface that displays the ATM network and its status in real time.

LattisCell Domain View™ displays the complete network topology by providing a color-coded view, displaying the status of the switches and the associated links. From this domain view, the user can profile a network or clients within that network. For example, you can obtain information on a switch, reset the width, or view bandwidth usage on any particular link.

LattisCell Expanded View™ displays a picture of the front panel of the LattisCell and EtherCell switch. Included in this display are ports and their associated LEDs that turn on and off in real time based on the status of the network. This application also provides pull-down menus that allow the user to enable/disable ports on the switches.

ATM PILOT DEMONSTRATIONS

OSF has identified several applications or uses it hopes to demonstrate over the ATM network described above. IP is currently the only legacy protocol that is widely supported and that will run over ATM. However, Novell has recently announced the Connection-Oriented IPX or CO-IPX, a modification of the traditional IPX protocol designed specifically for ATM networks that will include Quality of Service (QOS) support.

NetWare™ Connectivity is included to demonstrate transparent communication between a Novell file server and its associated clients through the EtherCell. The EtherCell supports up to 1,024 Ethernet clients on the ATM backbone. The primary benefits of using the EtherCell are that it:

Allows each EtherCell client to have a dedicated 10 Mb/s segment. Note that this bandwidth can be shared by attaching a conventional Ethernet HUB with multiple Ethernet clients.
Allows for transparent operation between Ethernet clients. Communication between the Ethernet clients actually takes place via the ATM switch, but this complete process is transparent to the Ethernet clients.

Ethernet-ATM Connectivity via IP demonstrates transparent communication between Ethernet and ATM clients, i.e., between PCs (loaded with TCP/IP stacks) and ATM-attached SUN workstations running the LANE specification. The primary benefit of LANE is that the user does not have to manually set up the connections within the ATM switch every time an Ethernet client wants to communicate with an ATM client.

Native ATM Connectivity demonstrates some of the applications that will run over an ATM-based network linking ATM-attached workstations at different physical sites. Demonstrated applications will include:

Radiology image sharing and integrated editing capabilities to allow radiologists to view and mark the same image.
Full-motion videoconferencing for case discussions.

Videoconferencing will demonstrate conferencing capabilities between SUN workstations at a full 30 frames per second using commercially available frame grabbers and small cameras mounted on the workstation itself. This will allow doctors to visually communicate with colleagues and share data (text and images) simultaneously. To connect to another user, one has only to select the user name from a given list. The connection through the switch/network is automatically set up.

FUTURE NETWORK INTEGRATION

OSF is currently in Phase 1 of its implementation of a patient-centered network (Figure 5). A 68-node frame relay network is being deployed that will connect 12 facilities and approximately 56 physician offices over the three state area. It is important to note that, in the future, OSF plans to integrate the legacy networks on frame relay via ATM. We expect the present frame relay network to grow about 20% every year (in terms of additional sites and bandwidth). As the network grows, so will the bandwidth at the corporate head-end. Presently, four T-1s are adequate to support the SNA, IP, and IPX traffic between these facilities. Based on our projected growth rates, we expect we will need over 10 T-1s within two years to support existing legacy applications. If you factor in imaging applications, it is quite easy to see that the demand for WAN bandwidth will explode. Presently, we have justified the need for frame relay as a WAN technology that will connect our remote sites to the corporate data center. We have also built a strong business case for the use of SONET/ATM technology in the construction of a metropolitan area network in Peoria, which will connect several local hospitals.

In Phase 2, we plan to have direct connectivity between frame relay and ATM via the service provider's switch (Figure 6). We are currently working with several service providers, including independents and competitive access providers. The prices being quoted on 155 Mb/s SONET services are attractive enough to warrant deployment projections of ATM across the metropolitan area and beyond.

CONCLUSION

There are several other applications that we intend to implement in the coming months. Our plan for network migration is in Phase 1 with an emphasis on cost-justification, application availability, etc. In the coming months, we will be testing the following as an extension to our ATM network:

Ethernet and token-ring connectivity via ATM.
Routing between virtual LANs.
Interconnect the ATM domain at Glen Park with the corporate ATM domain.
Attach ATM-based servers.
Attach a token-ring switch to demonstrate token-ring connectivity.

The focus of this article was to demonstrate the feasibility of ATM as a viable backbone technology for medical applications such as teleradiology. I strongly believe that ATM is a cost-effective and technically viable technology that will enable us to share valuable resources such as medical expertise, images, and data among facilities and negate geographic distances.

¹ PVCs were the only option in early implementations of ATM. Currently, PVCs are used to exchange signaling information between clients and the ATM switch.
² LANE specifications, as defined by the ATM Forum, are used to build hybrid ATM networks. These specifications allow for both IEEE 802.3 (Ethernet) and IEEE 802.5 (Token Ring) networks to communicate with ATM clients. All LANE clients must use the same ATM Adaptation Layer (AAL5) packet size. What LANE means is that it defines a service interface for the upper layer protocols and sends data in the appropriate LAN Media Access Control (MAC) packet, i.e., Ethernet or token-ring. LANE in no way implies the emulation of the legacy access mechanism. In other words, emulating Ethernet within an ATM client does not mean emulating the Carrier Sense Multiple Access/Collision Detect (CSMA/CD) access protocol on the ATM-attached client. LANE allows users to build hybrid Ethernet/ATM networks. For example, Ethernet clients could access a high-speed ATM attached server transparently. Without LANE, we would have to attach all the Ethernet clients on the ATM network before they could communicate with the ATM-attached server.
³ I expect ATM-attached Novell servers to become a reality this year because several vendors will support Connection Oriented IPX (CO-IPX). Novell recently announced CO-IPX, a modified version of their IPX protocol for ATM.