An MD&DI November 1998 Column
Older technologies still have a place in medical devices, but the upcoming models of high-performance chips may well pay for themselves in speed and flexibility.
Proliferation of the personal computer has created fertile ground for the growth of new medical technologies whose control, function, and continued development are dependent on embedded computer chips. These embedded systems may guide the actions of mechanical or electronic components, monitoring the gear speed or pneumatic pressures that drive microsurgical instruments. They may govern the collection and processing of data by analytical devices reporting the chemical makeup of body fluids or control the operating parameters of medical lasers. They might even monitor patient vital signs or gather and reconstruct imaging data obtained with medical scanners. In short, "virtually all electronic medical devices these days are run by embedded controllers," says Jerry Epplin, senior software engineer for Sterotaxis, Inc. (St. Louis).
Renal 3-D volume acquisition shows simultaneous display of incremental volume reconstruction (Siemens Ultrasound, Issaquah, WA).
As varied as these applications are, so are the choices of computing technologies to embed, ranging "from a one-dollar 8-bit microcontroller to the latest 64-bit microprocessors," says Epplin. Very often, engineers choose the low-cost optionsthe Intel 386 and 486 chips. But although these products conjure images of obsolescence, they're still quite valuable to many medical applications.
"The embedded world revolves around two ideassize and functionality," says Donald Springfield III, an embedded systems technician at EMJ Embedded Systems Tech (Apex, NC). "Sometimes a second-generation Pentium system, with its power-hungry chip set and immense fan or heat sink, is just not usable."
The "old" chips386 and 486have an additional advantage in that they are proven technology. Minor enhancements make them suitable for medical and other critical devices. Moreover, semiconductor manufacturers have a big stake in making these enhancements. "They keep producing these old chips because they have the manufacturing facilities to do so," says Epplin. "These old facilities cannot be used for manufacturing the new chips."
As the sophistication of the medical task rises, however, more-powerful chips become necessary. Like their simpler cousins, the newest breeds of PC offspringPentiums I and IIare available and are being used when operators demand certain features, such as graphical user interfaces similar to those found on the latest PCs. "One can develop slicker products with the newer chips because it is possible to provide more-intuitive user interfaces, high-speed networking features, internationalization, and other features with a higher-speed processor," Epplin says.
Then there are the elitethe digital signal processors (DSPs)whose performance can virtually transform the machines bearing them into miniature supercomputers. Among such applications are the most sophisticated medical devices available today, including computed tomography, magnetic resonance, and ultrasound scanning equipment.
TOO MANY OPTIONS?
Such a wide range of options is both advantageous and problematic. The expanding power of embedded chips encourages designers to expand the capability of their medical devices. Users want the most powerful technology available, says Anthony Ferguson, customer support technician for EMJ Embedded Systems Tech. Suppliers, however, have to match the capability of the embedded computer with the task.
"A simple controller used to monitor vital signs does not need to have the power of today's home or office solutions," says Ferguson, whose company distributes single-board computers and accessories for use by equipment vendors in several industries, including medical devices. "Oftentimes, a 386 will do the job."
For blood analyzers and the like, overengineering is usually not a temptation. There are only so many functions to perform, only so much computing to be done. The chip choice is obviousat least to the engineers building the embedded system.
"If we don't stray from Intel architecture, we can take advantage of what we call our technology briefcase, most of which is PC compatible," says Roy Kravitz, director of engineering at RadiSys (Hillsboro, OR). RadiSys develops computer systems designed to be embedded in medical devices and other equipment. Kravitz and his staff have developed a product line of circuitry blocks that can be optimized to fit customer needs.
"Each customer thinks he is getting a custom computer from usand [the computers] are, but they have common parts," explains Mark Budzinski, director of marketing for commercial equipment at RadiSys.
An Intel 386 chip, tweaked by RadiSys, is the core computing technology in the Accurus 400 VS, an integrated vitreoretinal surgical system manufactured by Alcon Surgical (Irvine, CA). This device repairs detached retinas or macular degeneration by drilling three holes in the eye: one to infuse liquid to keep internal pressure constant, a second for a fiber-optic light pipe to illuminate the surgical field, and a third for a surgical probe that cuts and vacuums away excess tissue.
"The 386 chip runs the system," says Douglas Downing, operations product manager at Alcon. Downing notes, for example, that the 386 chip regulates the pneumatic pulses that drive the surgical cutter, while monitoring the vacuuming of debris. "It's the brain of the unit," he says.
But chips from the 386 and 486 era of personal computing bring software limitations inherent in their circuitry. For example, 386 chips run best on MS/DOS, yet this operating system is anything but ideal. DOS is a programming nightmare, vulnerable to the year 2000 problem because years are limited to two digits. But there are alternatives.
As interest has grown in the use of these chips as part of embedded systems, developers have come up with numerous specialty operating systems. These include RTOS (real-time operating system), QNX (a real-time implementation of UNIX), as well as two other real-time operating systemsvxWorks and pSOS+. Others have developed DOS clones that are compatible with MS/PC-DOS and that offer extended capability. Examples are ROM-DOS and DR-DOS. Each of these is year 2000 compliant.
Roy Kravitz (left), director of engineering for the automation and controls division, RadiSys Corp. (Hillsboro, OR), works with a hardware engineer to bring up a new embedded computing systemn.
These options provide designers with alternatives to MS/DOS, yet simpler chips still are not always the answer. Hospital networking that requires a medical device to interact with Windows NT, for example, demands the Pentium breed of chips. Such sophisticated applications commonly require more computer power anyway, as in the case of imaging systems.
In late summer, ATL Ultrasound (Bothell, WA) released its newest scanner, the HDI 1500. The device, which still awaits FDA clearance, is a combined engineering effort of ATL and Medison, a Korean company. Especially interesting is the machine's ability to generate 3-D images, a capability whose clinical value is only beginning to emerge. On the HDI 1500, data accumulated by sweeping the patient are rendered in multiple planes or volumes on-screen.
The ability to capture and rapidly assemble these data is largely the result of ASICs (application-specific integrated circuits), which are designed for defined purposes. Because of their tight focus, ASICs can perform certain calculations at blinding speed. "ASICs essentially take what used to be on circuit boards and miniaturize that into single chips," says Jim Brown, ATL senior director, product management. "They are very specialized microcircuitry built to do specific tasks."
Whether ASICs are truly embedded computers, however, is debatable. Epplin notes that they meet most definitions of the term, but many in the industry stop short of conferring such prestige on these chips because of the difficulty in programming them. "You need sophisticated and expensive development tools [to program ASICs], and this development is typically done by people with hardware rather than software backgrounds," he says. ASICs can be used, however, to control the function of other embedded systems, such as DSPs, which are driven by software. The more advanced scanners built by ATLthe HDI 5000 and HDI 3000use this combination to provide speed and flexibility.
"With ASICs we can tell two signal processors to complete a certain function, two others to do something else, or we can apply all the DSPs to the same function, under the direction of an ASIC," Brown says. "So you can really apply a lot of power to what you want to get accomplished."
PROGRAMMABILITY EQUALS FLEXIBILITY
Ideally, manufacturers would like the flexibility of a totally programmable device. In such a case, software defines function. New clinical applications can be written and simply downloaded into installed machines. Here, Siemens Ultrasound (Issaquah, WA) stands alone among its peers in the ultrasound industry.
In developing the Sonoline Elegra, its flagship ultrasound scanner released about three years ago, Siemens partnered with nearby University of Washington (Seattle) to develop an advanced computing chip, a DSP called the multimedia video processor (MVP). The two groups also collaborated on the development of a programmable ultrasound image processor (PUIP), which is composed of electronics built around two parallel processing MVPs. Together, the two chips can handle up to 4 billion operations per second.
The MVP is programmable, which means engineers can create new functionality by adding software. The programmable chip itself is a critical component, but not sufficient to achieve the true potential of embedded systems. When using such chips, engineers must design the architecture of the medical device not only to work with the chips but to take full advantage of them.
Siemens has created impressive new applications in a relatively short time, because development focuses primarily on software implemented on the PUIP, which Siemens has dubbed Crescendo. First was SieScape, a novel capability that strings ultrasound images into a panoramic view of the bodyfor example, a long view of the arm or leg. The first iteration allowed the visualization of gray-scale images only. A software upgrade last year added color, visualizing panoramic blood flow of the carotid, for example. Before the end of this year, Siemens plans to release a 3-D version, which accumulates data representing individual planes in the body, assembles these data into a volume, and then displays the volume as a cube that can be interactively sliced in virtual space to provide literally any view desired. Again the upgrade will primarily entail software.
"It is only because we have an image processor built into the computer that we can do this without needing a stand-alone workstation," says Rich Fisler, a senior product manager at Siemens Ultrasound.
And this is just the beginning. The successor to the current MVP, due out in 1999, will operate five times faster than the current model. The cost is expected to be much less than the cost of designing, developing, and manufacturing proprietary ASICs. And like the current MVP product, it will be designed to work in parallel.
"In 1999, we will be able to perform 10 billion operations on a single computing chip, which will cost maybe $99," says Yongmin Kim, PhD, professor and director of the University of Washington Image Computing Systems Laboratory. "How about using four of these chips to bring up the processing speed to 40 billion operations per secondor even 10 of them for 100 billion?"
MORE THAN CHIPS
But the challenges to reaching 100 billion operations per second are as formidable as the benefits that might be achieved. "Our biggest challenge is input and output," Kim says. "We have to divide and route the data to the appropriate processor and get the result." That process is complicated when data must be shared by several processors. Kim explains that in such cases, the data must be accessed simultaneously. The coordination of this simultaneous access and the need to synchronize multiple processors will demand approaches beyond current knowledge.
An Intel Pentium chip customized by RadiSys.
That knowledge might come from other industries, particularly those related to defense, where superfast processing is already a requirement. The rapid identification of targets such as aircraft, missiles, and tanks has driven the development of processors capable of handling huge quantities of data, as in the case of airborne and ground-based radar systems. Nowhere is the potential of transitioning these processing technologies greateror more clearly relevantthan in medical imaging. Pressed by increasing cost concerns, manufacturers are trying to build equipment that will streamline the diagnostic process, increase patient throughput, and support the trend toward minimally invasive diagnosis.
Processors now serving as the cornerstone for massive phased-array radar systems and in-flight weapons targeting systems are being optimized to address those goals. Mercury Computer Systems (Chelmsford, MA) has launched a five-year, $100 million R&D effort aimed at developing new technologies for digital signal and image processing.
"We anticipate that our next-generation systems for diagnostic medical imaging will enable the physicians to see a full three-dimensional picture of various parts of the body, instead of simply two-dimensional slices from a scan," says Jay Bertelli, president and CEO of Mercury Computer, which makes the computer boards that power advanced phased-array radar systems for the U.S. military.
As this technology trickles down to commercial applications in medicine, diagnostic images might not only be instantaneous but widespread, appearing in ultrasound, computed tomography, magnetic resonance imaging, and even some forms of x-raybased imaging, such as mammography. In the next decade, there is a very real possibility that computers embedded in advanced radiology devices will deliver not only images but an understanding of those images, where algorithms designed to assist in the interpretation of data highlight suspicious lesions, quantify the dimensions of those lesions, and compute the volume. Under a physician's control, such data processing might be used to interactively target surgical probes or beams of therapeutic radiation that precisely excise or destroy tumors.
Such capabilities will have a profound effect on the practice of medicine. But given the speed of development and technological advances in embedded systems, the real issue isn't whether such devices will be built, but when.