Hardware, Software Advances Brighten Image-Analysis Systems Picture

ADMAX Computer Inc. Nashua, NH Advanced Visual Systems Waltham, MA Aldus Consumer Division San Diego, CA Analogic Corp. Wakefield, MA Ambis Inc. San Diego, CA Amtec Engineering Inc. Bellevue, WA Applied Biosystems Foster City, CA BioCAD Corp. Mountain View, CA Bio-Rad Laboratories Hercules, CA BIOSYM Technologies Inc. San Diego, CA B/T Scientific Technologies La Costa, CA Cognex Corp. Needham, MA CoHort Software Minneapolis, MN Compix Inc. Mars, PA Data Translation Inc. Marlboro, MA DNA ProScan Inc. Nashville, TN DSP Development Corp. Cambridge, MA Dynacomp Inc. Webster, NY EG&G Instruments Princeton, NJ Epix Inc. Northbrook, IL Europa Scientific Software Corp. Hollis, MN Famous Engineer Brand Software Richmond, VA Golden Software Inc. Golden, CO Graftek Imaging Mystic, CT Hyperception Inc. Dallas, TX Imaging Automation Merrimack, NH Imagination Systems Inc. Virginia Beach, VA Imaging Technology Inc. Bedford, MA IMSL Inc. Sugarland, TX Infometrix Inc. Seattle, WA Innovative Imaging Inc. Wayne, PA Inovision Corp. Durham, NC Integrated Separation Systems Natick, MA Integrated Systems Inc. Santa Clara, CA IntelliGenetics Inc. Mountain View, CA Jandel Scientific San Rafael, CA Keithley Data Acquisition Division Taunton, MA Liant Software Corp. San Diego, CA The MathWorks Inc. Natick, MA Medica Cybernetics Silver Spring, MD Metric Systems Round Rock, TX Micro Data Base Systems Inc. Lafayette, IN MicroDisc Inc. Pennsauken, NJ Micromath Scientific Software Salt Lake City, UT Molecular Dynamics Inc. Sunnyvale, CA Molecular Simulations Sunnyvale, CA National Instruments Austin, TX Nicolet Instrument Corp. Madison, WI Nikon Electronic Imaging Melville, NY Noesis Vision Inc. Montreal, Quebec Optimas Edmonds, WA OTC Beaverton, OR Perkin-Elmer Corp. Norwalk, CT Pharmacia Biotech Inc. Piscataway, NJ Precision Visuals Boulder, CO Preston Scientific Inc. Anaheim, CA Quantitative Technology Corp. Beaverton, OR Research Systems Inc. Boulder, CO Resolution Technology Inc. Columbus, OH SAS Institute Inc. Cary, NC Scientific Endeavors Corp. Kingston, TN Signal Analytics Corp. Vienna, VA Signal Technology Inc. Goleta, CA Spectral Innovations San Jose, CA StatSci Inc. Seattle, WA StatSoft Inc. Tulsa, OK Tripos Associates Inc. St. Louis, MO United States Biochemical Corp. Cleveland, OH Universal Imaging Corp. West Chester, PA Univision Technologies Inc. Burlington, MA UVP Inc. Upland, CA Videologic Inc. Cambridge, MA Viewpoint Software Solutions Rochester, NY Visual Numerics Inc. Sugarland, TX Wolfram Research Champaign, IL The computerized conversion of images into data for analysis, quantification, and presentation has promised to become a widely used--even necessary--tool in the life sciences laboratory. But over the past decade, certain technical limitations have proved to be a barrier to market growth. These include deficiencies in computer processing speed and memory capacity on the hardware side, and ease of use and multiple applications on the software side. However, recent advances in computer hardware and software are lowering, if not removing, the barriers to wider acceptance of image-analysis systems by life scientists. The cutting edge of this technology is in the area of volumetric, three-dimensional (3-D) image analysis. These useful, highly detailed graphics currently come at the cost of millions of bytes of data and tens of thousands of dollars for state-of-the-art hardware and software. According to industry experts in image-analysis technology, the trend over the next few years appears to be that personal computers (PCs) and UNIX workstations will develop into more powerful, easier-to-use systems whose capabilities will be nearly indistinguishable from each other. Industry insiders point to improvements in the four basic components of this technology. The first is an image-acquisition device, often a charge- coupled device (CCD) camera or a video camera, which sends out analog signals like regular household camcorders. However, improvements have enabled these devices to now transmit digital signals instead. The range of cameras being employed in this technology can cost from $200 to more than $20,000. The price depends on a number of features: camera resolution; color capability; geometric correction; and the number of data bits per pixel of the image. Other imaging devices include hand-held scanners, desktop or flatbed scanners, proprietary interface devices for customized capabilities, and various specialized instruments. The second component of image-analysis systems is the device that transmits the image from the camera to the computer. This has been a frame grabber board. Prices for frame grabbers range from $200 to $8,000. But many new digital cameras now interface directly with the computer. Scanners also directly transmit their signals. The third component is the computer, and current imaging systems require at least a 386-level PC. But for optimal performance, the much-faster 486 or Pentium microprocessors or a UNIX-based workstation are essential for the next generation of image-analysis systems. The final component is software, because it's all meaningless numbers unless users have the software to make measurements of the biological sample or object being analyzed and export these data into a useful form, such as a report or a graphic. The software also controls the camera or scanner acquiring the image and does any image enhancement necessary to correct for problems in image acquisition. Often the user also wants to perform statistical analyses, and this capability must be built into the program. In many applications, the user will eventually want to automate the process so that it can be accomplished, more or less, at the touch of a button rather than by a laborious series of manual steps each time. Current image- analysis programs are still cumbersome to use, and few offer a wide selection of applications in a single package. The trend in software will be toward ease of use and multiple applications in an integrated package. Image-analysis software can cost from $500 to tens of thousands of dollars, though most fall in the range of $4,000 to $8,000. The Optimas software is Bob Hart's flagship product. The Edmonds, Wash.-based company, of which he is president and chief executive officer, recently changed its name to the Optimas Corp. from BioScan Inc. because customers kept calling them "the Optimas people." Like most PC-based image- analysis software, Optimas runs under Windows 3.1. Using a graphical user interface (GUI) such as Windows yields what its manufacturer, Redmond, Wash.-based Microsoft Corp., calls a "docucentric world," in which the user is focused on the work being done rather than the application he or she is using. "I don't have to worry about whether I am in Optimas or Excel, or how I'm going to do Word,S Hart explains. "They have become a single operating environment within my computer." "Before it was an application-centric situation--I'm working in my spreadsheet program. Now I'm working in my word- processing program. Well, today, I'm working in my image- analysis program and I'm going to export files to my other programs. This is a tremendous boost to productivity." Nice idea, Jandel Scientific's Craig Rappaport says, but most users arenUt in the docucentric world yet. "There is a real lack of understanding of the image-processing side of the application among the masses of scientists and engineering and technical specialists out there," the senior market manager at the San Rafael, Calif.-based company says. "Most scientists don't have a problem with quantifying their data, knowing what they want to measure. Their obstacle is the image processing, understanding what the digital image is. "I've been pushing for ease-of-use, image-processing `wizards,' to coin a phrase--some sort of artificial intelligence [that will make it easier for users to operate the systems]. That's what will make the image-analysis market grow. People have been saying this market would grow way back in the early '80s. And it never has. "The reason is, we have these basic obstacles. The technology is cheap enough now to afford it if you wanted it. So why are we still in the range of only 5,000 units sold per year by all the companies out there?" Rappaport quotes 1990 United States Bureau of the Census figures showing about 800,000 scientists and 1.5 million engineers in the U.S. "Either these numbers are a big puff of smoke or there is a real barrier to adoption," he states. He figures the barrier to more widespread adoption of image- analysis systems is due partly to the history of the industry and partly to the value-added resellers (VARs). These VARs, he feels, are doing a good job at keeping the industry going in terms of sales. But they are still following the lead of the early adopters of the technology-- the most sophisticated and knowledgeable users for whom ease of use is not necessarily a prerequisite for purchasing image-analysis systems. "The VARs have a vested interest in selling to those who have the most knowledge about imaging and want the most out of the applications," Rappaport explains. "The VARs then don't go out to the mainstream scientists and say, `Hey, I have an application that doesn't require any programming. You just sit down and do the work.' They don't want those people because it doesn't allow them to add value to the product." Rappaport concludes that this situation has led to very powerful but not very easy-to-use software packages. Also, these programs tend not to feature a comprehensive choice of applications in a single, integrated package. "Someone's work may require eight hours a week tracing graphs, but your application doesn't have a module in it for tracing graphs," he says. "It's too general. There's a lot of that, so customers are requesting specific applications within the program." Bill Strum, president of Media Cybernetics, Silver Spring, Md., agrees: "The greatest obstacle to the acceptance of electronic imaging is perceived ease of use. That's the key- -in bold italics." "Five years from now the typical system will look like a toaster. Every computer sold today already is image-enabled- -that's a fact. You will order the software you want, your PC will be plug-and-play compatible with devices that you use to capture the images. It will be easy to use. There will be suites of software--targeted for use in genetics, metallurgy, polymer science, forensics--and specialty suites within them. There will be an Internet full of custom modules to use." Ted Inoue, president of West Chester, Pa.-based Universal Imaging Corp., explains what those custom modules and application softwares are going to do: RThe field is evolving the way 2-D imagery did a decade ago. Initially people were interested in acquiring images, making pretty pictures, seeing details better, bringing out features that might not otherwise be seen. In 2-D, scientists were simply counting the number of cells in the field of view, and measuring their areas. "That's kind of the state you have now in 3-D imaging. But in 3-D, people are wanting to do this over the entire volume. They might look at a slice of rat brain and want to count the neurons, or they might want to examine the branching structure in these neurons. Now, the next step is making measurements and extracting quantitative data from these images." While the demand for 3-D and 4-D, or time-series, imaging systems will grow, manufacturers must address some problems, mainly that these graphic datasets are huge and require enormous amounts of computer memory. The rat brain specimen may involve taking 50 to 100 slices of image data, and the entire three-dimensional picture would require at least 25 megabytes of storage. The past state of the art for storing these images involved using laser disks, a video medium with limited resolution. The key for the future is to store data in a compressed format using mathematical algorithms that can achieve high resolution. "The data-compression field is really exploding," Inoue says, "and many of the methods used in multimedia are directly applicable to the scientific-imaging field, as well. You are probably getting from 10- to 20-to-1 data compression by using these algorithms. The problem still is the resolution quality of these images with high-compression techniques." Assistant anatomy professor David Dean of Case Western Reserve University, Cleveland, is so convinced that current PC hardware and software are a problem that he won't use them. His research in morphometrics, the biostatistical study of shapes--particularly human anatomy--is performed on a Silicon Graphics Inc. (SGI) workstation, using his own software routines. Much of his work revolves around neuro- and craniofacial surgery patient diagnosis, surgical simulation, and postoperative follow-up. But some recent research has included reconstructing the skulls of ancient Moroccans using fossil fragments, in collaboration with Alan Kalvin of the IBM Thomas J. Watson Research Center, Yorktown Heights, N.Y., and Jean-Jacques Hublin, Musee de l'Homme, Paris. Dean explains: "To move those volumetric images around in real time so you can see different parts requires an SGI. It involves matrix multiplication of tremendous amounts of data. PCs and Macs are just not punchy enough to do those kinds of things. SGI machines have graphics engines in them, powerful chips dedicated to graphics software." The software for the fossil reconstruction work is a Computer-Aided Anthropology system developed by Dean and Kalvin using the IBM Visualization Data Explorer (VDE). The VDE is a graphics visualization toolkit that runs on workstations from IBM Corp.; Hewlett-Packard Co., Palo Alto, Calif.; SUN Micro Inc., Palo Alto; Data General Corp.,Westboro, Mass., and Silicon Graphics, Mountain View, Calif. Redrawing volumetric images as they are rotated or otherwise moved on the screen is measured in polygons per second. A midrange SGI workstation redraws on the order of 100,000 polygons per second, which is an order of magnitude greater than that of PCs. The state of the art in the UNIX world is a minimum 32- megabyte memory and a 1-gigabyte hard disk, says Doug Benson of Inovision, a Durham, N.C.-based custom programming/hardware integration company. UNIX workstations use 64-bit buses. This means the computer throws 64 bits of data through the bus to the monitor from the hard disk, from a data-capture device all around the machine at each cycle of the main computer's central processing unit (CPU). PCs, until recently, have used 8- and 16-bit buses, but are becoming 32-bit bus systems. Thus, PCs are now approaching the power of the low-end workstations. Benson predicts, "In five years, my expectation is that Windows-based machines are going to run X Windows [the UNIX- based Windows style interface] and X-based machines will be able to run Windows. The users are going to have a switch to select which interface they want." One reason for the near equality of UNIX and PC systems are the upgrades of Windows--Windows 4.0 and Windows NT--from Microsoft. "Microsoft realizes that the things UNIX does well are good things to do, and they are going to do them as well," Benson observes. "The intention is for Windows NT to make the gap between PCs and workstations go away. I don't know if that is a fair statement. Until that is tried and proven I have no intention of being a guinea pig for Microsoft." To Jerry Fife, product manager at Photometrics Ltd., Tucson, Ariz., all the software and PC improvements are meaningless if users can't get a quality image. Photometrics specializes in high-resolution CCD cameras for digital image capture. High resolution not only can mean resolving the close separation of features, but also refers to the level of gray or number of colors that can be distinguished in each pixel. Though Photometrics was begun by a Kitt Peak National Observatory astronomer, much of its business comes from firms working on the Human Genome Project. "CCDs are being advanced in a number of different ways," Fife says. "A lot of things are happening in the areas of application specifics. For example, CCDs are taking on different formats as opposed to your standard 3 x 4 aspect ratio. A typical example would be spectroscopy chips, which are very rectangular, 4- or 8-to-1." Spatial resolution, the actual number of pixels, is very important because it determines the amount and detail of information captured in the image for display, analysis, and quantification. The highest available is up to 5,000 x 5,000 pixels. Typical video resolutions are only 700 x 500. Fife predicts, "I expect to see higher-resolution video cameras combined with some kind of frame grabber. The video connection will be built directly into computers so that frame grabbers will become more specialized. As the electronics shrink, more of the electronics will be involved in the camera head and less in separate controller boxes." He sees the greatest progress today in better-cooled CCD cameras, even though the price is keeping them out of the hands of many users. Cooling lowers the thermal "noise" in the signal, providing a cleaner and sharper image. The manufacturers' consensus is that the 1999 systems will be even faster than todayUs Pentium 90 Mhz models, with gigabytes of hard-disk space and at least a gigabyte of RAM memory. Image-capture de-vices will be tinier cameras, with higher resolution spatially, and in terms of digital data per pixel, chilled for greater thermal responsiveness. The digital output of the camera or other image-acquisition device will head directly into the PC by more powerful frame grabbers, some incorporated into the camera itself. All the hardware will be "plug and play," like an office stereo system. Data compression will ease some of the storage crunch. Software will come with a larger repertoire of application- specific modules, and the gap between PCs and workstations will disappear because of various windowing applications in both of these major operating systems. The gap will also close as a result of improvements in chip technologies, unless the workstation manufacturers can really boost performance much further than PCs. With all these potential changes, it remains to be seen who will be around to continue to provide products and services. Acquisitions and business failures within the image-analysis industry can make for a volatile marketplace. According to Optimas' Hart, "We're in what is still, surprisingly, a fairly emergent business. It's really just starting to consolidate. I think you're going to see an accelerating change over the next couple of years. I think there will be a lot of consolidation, a lot of shake-out. "We are entering that next phase of a maturing market where we start with hundreds of little players and end with a relative handful of dominant companies." Larry Krumenaker is a freelance science writer based in Hillsdale, N.J.

Hardware, Software Advances Brighten Image-Analysis Systems Picture

Interested in reading more?

Become a Member of