Nigel De Wit

Millions of dollars may have been spent on the latest CT, PET, MRI, or other digital modality in your imaging center. The purchase of a PACS, RIS, or electronic medical records system may have been made, too. The need to move away from film is as high on the agenda as it is expensive to produce, store, and distribute.

Ultimately, all this investment supports a process that includes scheduling a patient, conducting an imaging scan, performing a radiological interpretation, and providing a report. If anything in this chain of events goes wrong, it can cause frustration and cost money.

One of the problems with patient scans in the digital world is large file size, particularly files that derive from 64-slice CT scanners. It is not uncommon to see files over 500 MB for a single study. There is no sign that these files are getting smaller, only bigger.

The larger the file, the more resources needed to move data around a local area network or wide area network. The problem can be greater over a WAN since the amount of bandwidth is substantially lower.

Despite so much money spent on hardware and software, why do radiologists report that the "system is slow" on a regular basis? In many cases, the reason is that the care and attention that were invested when buying the modalities were not present when the network infrastructure was purchased. The network infrastructure investment is often the smallest part of the budget. Yet everything is dependent upon it working well, as it provides the platform for the entire operation.

In many cases, sites must put their trust in the local computer network company to help with networking needs. Unfortunately, most computer network technicians are not familiar with terms used in imaging centers, such as modality, RIS, and PACS. As a result, the network is not optimized or designed in a way that promotes efficient networking.

Was the network cabling in the practice professionally installed? All that money invested in hardware and software could be tainted by a handful of $60 cable runs. If Category 5 unshielded twisted pair cabling is not installed correctly, it can significantly downgrade network performance.

Cable testing is another important aspect of professional installation. Cable testers can cost thousands of dollars, but they can provide vital information. Most professional cable businesses should have one.

Have your cabling tested to make sure that it meets the minimum standards. The tests should include details on the length of each cable run, the amount of attenuation, and the near-end-crosstalk test.

Beyond cabling, another important determination is network structure. In imaging centers where it is not necessary to integrate the PACS and the RIS/EMR, physically separating the network can be a good choice.

A typical medical imaging business can be broken down into two sections. The modality side covers equipment and software for reading, postprocessing, printing, and routing image data. For the sake of simplicity, this is the PACS section. The PACS section could have its own IP subnet, such as 192.168.100.x.

The other side of the network supports the back-office functions that include patient demographic information software and hardware, e-mail, billing, and documents. This can be called the RIS or EMR section. This section could have a different IP subnet, such as 192.168.200.x (Table 1).

The result is like having two LANs in one building. It does add some cost in duplicate hardware and Internet fees, but it also offers several advantages. For example, if the imaging center has a single Internet connection and is sending or receiving medical images to and from the network, the Internet pipe is reaching its maximum capacity throughout the day due to the large file sizes. E-mail, Web, and other Internet traffic must compete for bandwidth on this Internet pipe as well, creating slowdowns. Two separate Internet pipes prevent this by allowing both sections to run better.

Separation of the network subnets reduces the risk of viruses reaching the PACS network or unwanted broadcasts moving from one network to the other. It can also prevent another issue that arises when software fails or does not respond correctly. Software vendors can easily deflect blame or criticism by claiming another process is interfering or degrading their software. With this separated network design, finger-pointing is far less likely. The separation of the networks can be physically achieved on two separate switches (as shown in Table 1) or from a single virtual LAN (VLAN), which has the capability of separating the networks by ports.

As imaging centers evolve, the need to integrate PACS and RIS/EMR becomes more important. But doing so can be quite a challenge. The most common mode of communication between PACS software and RIS/EMR is with an HL7 interface.

Placing a suitable router at the heart of the networking infrastructure can make the connectivity between two networks. The only time the two subnetworks communicate is when a specific request is made from one to the other. The router allows this communication to occur.

The router is probably the most flexible device at the core of the network. It can handle T1 (serial) connections that can be used for direct connectivity to another office using point-to-point circuits, MPLS, or a simple T1 to the Internet. These routers usually have 2-gigabit Ethernet ports (one for PACS and the other for RIS/EMR) and have the capability of adding card modules such as a switch that can be set to provide VLANs. One of the VLAN ports can connect to a firewall or to a router in another office (useful when resources need to be shared between different businesses in the same building while keeping control of internal security).

By installing a router or VLAN switch, the networks can talk to each other when needed but still keep their separate identities and efficiencies.

Opening images can be a slow process if they are read directly on modality workstations, because the image data are not compressed. One of the big advantages of PACS software is that it provides lossless compression of 2:1 or more, so the time to open a study-and archive it-can be twice as fast. This allows the radiologist to read more studies per day. PACS software offers a host of enhancements for image manipulation and control, but it can be a substantial investment.

PACS software packages can also be configured to compress at lossy rates of 3:1 and more. When using lossy compression, radiologists will notice the speed at which the study opens, even from remote locations. Deciding to read lossy images requires careful consideration. The decision to read lossy images should consider type of modality, file type, anatomy, and pathology.

Radiologists are not always able to read a study in the same location where it was performed. This presents challenges for PACS administrators. It means that the image set has to be sent outside of the network to the location where the radiologist will perform the interpretation. In many cases, this is at another branch that is connected via a WAN circuit (Table 2).

The radiologist still wants to receive the data quickly but not at the cost of starving bandwidth for RIS/EMR data that may also be needed. This is a common problem where Quality of Service (QoS) techniques are not implemented, and slowdowns, lockups, or time-outs are common.

In private WANs, the flow of traffic between PACS and other software can be managed with QoS techniques. QoS tools provide more control over data than is possible with an Internet connection such as with a virtual private network. QoS can be controlled by the routers at either end of the WAN.

Following network analysis, you may find that a network that is split 75% PACS and 25% RIS/EMR over the WAN is about right. In larger operations, where there are several WAN connections, it is vital that QoS be implemented correctly, as it is possible to starve some WAN pipes of bandwidth in favor of other WAN pipes. An experienced network specialist who can set this correctly is worth his or her weight in gold.

As file sizes increase, the LAN is the best place to read studies, since gigabit bandwidth provides enough speed to move the images quickly. PACS administrators should try to manage workflow so that studies can be read by radiologists in the location where they were acquired.

But images may need to be sent to other locations, and they must arrive in a reasonable time frame. Ensuring enough wide area bandwidth involves balancing the needs of the radiologist and the costs of bandwidth.

For WANs, available offerings can be expensive. Point-to-point circuits are widely used but expensive, particularly over long distances. Other options include T1s, which provide 1.54 Mbps of bandwidth. T1s can be bundled together to form even larger pipes. A T3 is a high-bandwidth option and usually provided over fiber-optic cable.

T1s cost from $400 to $800 per month, but the upstream speed is the same as the downstream speed; that is, it is symmetrical. This is important if studies are pushed across the Internet, which involves the use of upstream bandwidth.

Additional options include cable or digital subscriber lines (DSL). These technologies are faster downstream than upstream, making them much slower when pushing images. DSL and cable are adequate if users are relying on a Web client for downloading and viewing lossy images. Otherwise, these two options can be painfully slow if full lossless studies are pushed.

Consider the math: 1 GB of data traveling on a 1.54 Mbps pipe should take one hour and 28 minutes to move from one location to another, assuming no compression. But the reality is that the times are usually much longer.

Circuit latency and competing traffic can add time despite compression options. At 43 Mbps (T3), 1 GB takes just over three minutes under ideal network conditions. The problem is that T3s are expensive and not always available. The same data over a gigabit LAN takes seconds rather than minutes (Table 3).

New technologies are now available in the form of high-speed wireless WANs that can operate up to 100 Mbps (or more). The idea is that the last five miles to the imaging center are covered by a wireless connection rather than a wired connection.

This is achieved using an antenna erected on the imaging center roof and connected to the nearest wireless tower by line of sight. A fiber connection links the wireless tower to a data center or group of powerful wireless towers.

Data centers that offer colocation facilities and local wireless companies are the organizations that appear to be driving this offering. They are offering service level agreements and rates varying from $1000 to $1500 per month for 10 to 12 Mbps to the Internet or data center. The price per megabit reduces as more bandwidth is ordered. At last, a light exists at the end of the tunnel for large interoffice files sent over WANs.

A colocation data center is attractive to imaging centers because it also provides a place to rent rack space. A PACS server and a large storage device would fit nicely in such a space. The data center could also be the place to store studies, providing Health Insurance Portability and Accountability Act compliance (the need to retain studies for up to seven years offsite). It also creates a single point for referring physicians to access patient studies (data centers can offer Internet bandwidth cheaply). Using the data center bandwidth saves the imaging center's bandwidth for other needs. There are companies that also provide offsite DICOM archiving.

Moving large amounts of data efficiently across networks is one of the biggest challenges that face the digital world of imaging centers. Implementing efficient network infrastructure can provide the speed of service that radiologists need.

Mr. de Wit is an analyst and network specialist with Discount Telecom in Sarasota, FL.

Articles

Millions of dollars may have been spent on the latest CT, PET, MRI, or other digital modality in your imaging center. The purchase of a PACS, RIS, or electronic medical records system may have been made, too.

Nigel De Wit

Articles

Digital workflow depends on well-designed networks