Image Analysis Software Advances Biomedical Research Applications

Applications Author: Robert Finn SIDEBAR : COMPANIES SERVING THE IMAGE-ANALYSIS MARKET Improvements in computer-based image analysis are making the technique a routine tool of the biomedical laboratory. The hardware for obtaining digital images is getting more sophisticated (L. Krumenaker, The Scientist, Oct. 3, 1994, page 17), the software for processing and analyzing the images is getting easier to use, and scientists are finding ever more creative ways to use them. Surveys indicate that fully 30 percent of scientists funded by the National Institutes of Health use some version of the technique in their work, according to "Fundamentals of Image Analysis," a useful primer written by Peter Ramm. Ramm is president of Imaging Research Inc., an image-analysis company based at Brock University in Ontario, Canada, where he is an associate professor in the department of psychology. (This 1995 booklet is available for free by contacting the company at Brock University, 500 Glenridge Ave., St. Catharines, Ontario, Canada L2S 3A1. E-mail: imaging@spartan.ac.brocku.ca. World Wide Web: http://imaging.brocku.ca.) "On the hardware side it's been incredible," says Paul C. Goodwin, manager of the image-analysis laboratory at the Fred Hutchinson Cancer Research Institute (FHCRI) in Seattle. "And on the application side-finding something to do with that information-it's also been phenomenal. We're doing things with microscopy now that we never dreamed of doing. We're getting resolutions that challenge some of the fundamental knowledge of what magnitude of resolution is possible with light. "I'm not a computer scientist. I'm a biologist," Goodwin notes. "And to me, you can have all the tools in the world, [but] if it doesn't advance the science-if it doesn't allow you to ask better questions in biology-then it's just a neat toy. And I think what we're seeing is that the tools becoming available are allowing the scientist to ask a lot more intelligent questions of the data." Image-analysis tools range in complexity from those that merely present a publishable picture of an electrophoresis gel to those that use sophisticated data-processing techniques to build up three-dimensional images of living cells. The programs also vary in how much "massaging" they do of the data. With several hundred companies selling image analysis hardware, software, or both, selecting a system can be a daunting task. "[Scientists] should realize that [an image-analysis system] is going to be obsolete from the moment they buy it," advises James L. Olds, executive director of the Bethesda, Md.-based American Association of Anatomists, and an avid user-and inventor-of image-analysis tools. "So they should concentrate less on technology and more on what they want to achieve with the technology. "How close is the output of the machinery to the published manuscript? One thing that's great about Peter Ramm's software [Imaging Research's MCID system] is that the output of the image-processing software is in a tabular format which can be readily put into a scientific paper, and is meaningful without a lot more mathematical processing." Another important consideration is the extent to which an image-analysis system is tailored to a specific task (these are called "turnkey" systems) or broadly functional (these are called "toolbox," or "macro-based," systems). "What you find is that the more sophisticated users want a toolbox," notes Theodore D. Inoue, president and founder of Universal Imaging Corp., an image-analysis company based in West Chester, Pa. "They want the flexibility to put things together in any way that may be appropriate for their research. That might be the top 5 percent of users. Then you have the other mass of users who are more applications- oriented. They have a specific problem, and they really want one pushbutton solution to the application." And Ramm says that there's another reason to prefer a turnkey system. "If you're in a setting in which replicable data is the most critical aspect of what you do, then it might be preferable to get a system that functions in a standard way from laboratory to laboratory." TOOLBOX Visilog, from Noesis Vision Inc. of Quebec, Canada, is a full-featured image-analysis package. Scientists investigating image-analysis vendors will discover that some concentrate on toolbox systems, some concentrate on turnkey systems, and some produce both. A good example of a full-featured image analysis toolbox is Visilog, produced by Noesis Vision Inc. of Quebec, Canada. According to Luc Nocente, Noesis' general manager for North America, "We have the widest selection of image-processing algorithms of any software on the market. In particular, we have a very large selection of morphological operators," for teasing out information on the shape of objects from digital images. MACRO-BASED: The OPTIMAS package originatd as a Windows program. Another popular macro-based package is OPTIMAS, from Optimas Corp. of Bothell, Wash. "One of the things I've always appreciated about OPTIMAS is that it was a program that was written for Windows from the very beginning, unlike some other programs that were written for DOS and transferred to Windows," says Paul Goodwin. "This means it does the best job of integrating all that Windows has to offer. It's relatively easy to link an analysis module to a spreadsheet [for further numeric analysis]. It's easy to move data from application to application, to streamline the process of getting the data to the final form you want." Jandel Scientific Software of San Rafael, Calif., sells two versions of its macro-based image analysis package: SigmaScan and SigmaScan Pro. The packages are similar, although SigmaScan Pro permits a greater degree of automation. "SigmaScan Pro allows you to measure up to 64,000 objects at a time," explains Norman Link, a senior project manager at Jandel. "Looking at large sheets of tissue, for instance, you can take an image and threshold it such that all the pycnotic levels are identified." Cancer cells tend to be more pycnotic-denser and darker-than normal cells. "You then count and size all of those objects," Link notes. "And by looking at a histogram of the relative size and number of these objects, you can get an idea of whether a tissue would be cancerous or not." With the advent of the Windows 95 operating system, Jandel will be releasing a 32-bit version of SigmaScan Pro in March. The most significant advantage of the new software will be its ability to work with images that are larger in file size. "In the 16-bit environment we could only go to an 8-megabyte image," explains Link. "We're not sure what the limit is now. The largest I've loaded myself is about 20 megabytes." Some companies sell software designed specifically to work with their hardware. One example is Data Translation Inc. of Marlboro, Mass. Its GLOBAL LAB Image software is a macro-based program designed to work with its extensive line of "frame grabbers"-circuit boards specialized for capturing and manipulating large image files with great speed. Robert H. Gregory, a senior product marketing manager at Data Translation, explains how a scientist works with the GLOBAL LAB Image scripting language, once he or she brings an image onto the screen. "He'll do different tweaks to it-adjust the contrast, do some sharpening and some other filtration operations. The package learns all that. So when he brings up the next image, he can say 'Do to this image just what I did to the previous image.'" Trenton, N.J.-based Princeton Instruments Inc., which specializes in high-performance digital charge-coupled device (CCD) cameras, is another example of an image-analysis hardware company that also sells software. "Princeton's approach is to have a very powerful acquisitions engine," explains Mark A. Christenson, a biological applications scientist at Princeton. "We've optimized our software for collecting data. It has the power for [further analysis], but it's hidden in that you usually have to bring it out with the macro language. We turn to our collaborators, like Universal Imaging, to make the specific solutions." Oncor Inc. of Gaithersburg, Md., is a good example of a company that produces both a general-purpose toolbox package-Oncor Image-as well as a line of highly specialized turnkey packages. Because Oncor Image runs on Apple Computer Inc.'s Macintosh line of computers, it competes with NIH-Image, a highly regarded general-purpose image-analysis package, developed with public funds at the National Institutes of Health, that's available for free ftp://zippy.nimh.nih.gov/pub/nih-image/). "My own feeling is that [NIH-Image allows the user] to get up and running very quickly, but for more detailed work or for a more precise type of work, the Oncor Image package allows a lot greater power," says James C. Mullikin, Oncor's director of R&D for imaging. Other limitations of NIH-Image are its inability to interface with certain devices, its relatively slow speed, and a lack of product support, aside from the informal assistance a scientist can obtain from fellow users. Oncor's turnkey products include software optimized for both FISH and CGH (fluorescence in situ hybridization and comparative genomic hybridization), a hardware/software package for three-dimensional fluorescence imaging (the Oncor 3-D Cytometer), and Oncor Karyotype, which is intended for chromosome analysis in the clinical laboratory. Fluorescence imaging is one of the hottest fields in microscopy today, but it has some inherent problems that can be solved by image analysis. Ramkumar Moorthy, senior applications scientist at the Scanalytics division of CSPI in Billerica, Mass., explains, "The main problem you encounter with fluorescence microscopy, especially when you look at something under high magnification, is that you end up having haze that comes in from objects that are in planes above and below the plane of focus. This haze doesn't allow you to see things clearly." Optical confocal microscopy is a popular solution to this problem (H. Ahern, The Scientist, Jan. 22, 1996, page 18). In optical confocal microscopy a small dot of light is moved across the sample, and this small aperture allows a fully focused, three-dimensional image to be built up. But optical confocal microscopy has problems of its own. "The biggest caveat [to optical confocal systems] is that the light levels we need to use to get good signal-to-noise ratios is very high," says James T. Russell, head of the National Institute of Child Health and Human Development's section on neuronal secretory systems. "You lose a tremendous amount of fluorescence due to photobleaching." A solution to this, contained in the Scanalytics/CSPI CELLscan system, uses computation-intensive deconvolution algorithms instead of optical confocal techniques to de-blur the image. (This is sometimes referred to as "digital confocal imaging.") The CELLscan software controls the microscope stage and takes a series of images at different focal planes. The computer uses the data in these images to produce a razor-sharp, three-dimensional representation of the image. "With digital confocal, your detector is never starving for light, so you have good signal-to-noise," explains Russell. "That means you can get away with very low amounts of light impinging on your specimen, so there's negligible photobleaching. I can go back and look at the same cell many times." And, says Russell, there's another advantage as well: "Depending on the optical qualities of your microscope, one can achieve resolutions on the order of 0.4 to 0.5 micron, which is 2 to 2.5 times better than with optical confocal." Image analysis is also used for fluorescence ratio imaging, a type of microscopy that can detect calcium-ion and hydrogen-ion concentrations in living cells. According to Mark Christenson of Princeton Instruments, "There's many things you just must do in the calcium imaging process: switch filters at the right time; collect the image pairs; ratio the image pairs after doing subtraction; and while you're showing that you also want to draw a graph of the results so you can be updated in real time, plus showing you each of the cells in the field of view individually." Several companies have developed software/hard-ware packages for this. For example, the InCa++ System from Cincinnati-based Intracellular Imaging Inc. includes a fluorescence microscope, a low-light-level CCD camera, a UV/visible light source, an automated filter changer, and a Pentium computer with all the necessary software. Similar packages are available from Universal Imaging (the MetaFluor System) and Imaging Research (as part of its MCID system), among others. The analysis of electrophoretic gels is another popular application for image analysis, and many companies make products for this market. Jandel's SigmaGel software is at the less complex end of the scale, according to Norman Link. One of its great advantages is that it will accept image input even from an inexpensive hand-held scanner. COMPLEX SYSTEM: The AMBIS hardware/software systemfrom Scanalytics/CSPI At the other end of the complexity scale is a series of products from Scanalytics/CSPI. The company sells DNAscan for automated DNA sequencing, RFLPscan for one-dimensional electrophoretic banding pattern analysis, and the AMBIS System-a hardware/software system for radioisotopic imaging and quantitation of gels, blots, thin-layer chromatography, and whole-body autoradiography. These products take advantage of complex image-analysis tools such as background correction and the use of artificial intelligence to detect and quantify lanes and bands even in badly distorted samples. Rapid progress has been the rule in image analysis since the late 1970s, and computing power is one area in which the most rapid advances are being made. The expensive MicroVax minicomputers of the early 1980s are a dim memory to many scientists, replaced by inexpensive and powerful Pentium-based computers and Power PCs. James Olds expects that the next development in computing for image analysis will involve the use of currently exotic massively parallel computer systems, which are particularly well suited to the rapid processing of large images. Another series of advances is occurring at the front end of image-analysis systems. "The tendency now is toward the development of detectors for ultra-low-light specimens-chemiluminescent and fluorescent specimens," notes Peter Ramm. "Part of that is driven by the need to replace isotopes for safety and disposal reasons." An example of this in the realm of gel electrophoresis is the System 200 series of imaging scanners from Bioscan Inc. of Washington, D.C.. The scanners which can directly detect all commonly used radioisotopes, including 3H, 14C, 32P, 125I, 131I, 18F, and 99mTc, are able to perform 2-D radioimaging up to 30 times faster than autoradiography. And James Olds owns a patent on a related apparatus. "I've invented devices that acquire data not through a camera, but through the chip itself. Let's say you have a slice [of tissue] that's radioactive. Instead of bringing a camera down on the slice, you can in fact bring a derivative of a CCD chip down, and the chip itself is sensitive to the radioactive disintegrations. So you can dump that data directly to the computer. It's direct detection of radioactive disintegration. We've worked at modifications of that for direct detection from gels, and actually looking at living brains." While image analysis systems seem to be growing every more complex, at least one company is bucking the trend. Boeckeler Instruments Inc. of Tucson, Ariz., sells video image-analysis systems that, in the words of a company news release, "take advantage of the operator's intellect to make the visual interpretations a computer would make. The operator does this by using a 'video micrometer,' . . . which overlays a video image." The company claims that its systems can be mastered in just 10 minutes, do not sacrifice accuracy, and in some cases cost only a tenth of comparable automated systems. With image analysis systems simultaneously growing simpler and more complex, life scientists are in an enviable-if dizzying-position. Paul Goodwin, who started at FHCRI seven years ago, remarks, "Nothing that's in my lab today existed seven years ago. It's difficult to imagine what's going to be here seven years from now. The one thing we can say is that it will probably be more usable, it will be cheaper, and it will be better than what we have today. And what we have today is incredibly better than what we had yesterday." Robert Finn, a freelance science writer based in Long Beach, Calif., is online at finn@earthlink.net SIDEBAR : COMPANIES SERVING THE IMAGE-ANALYSIS MARKET

Image Analysis Software Advances Biomedical Research Applications

Interested in reading more?

Become a Member of