Deep Wells and Dark Currents: CCD Cameras Offer Microscopists an Immediate, Distortion-free, Quantifiable Image

Date: July 20, 1998CCD Cameras, Image Processing Software For the biologist interested in cell structure-function relationships or cell and tissue dynamics, the quest has long been for a magnifying detector that is stable, quantitative, and sufficiently sensitive to eliminate phototoxicity or photobleaching and, in the limit, allow continuous observation without perturbation. If the data from one's detector can be immediately analyzed and displayed in a favorite software application, so much the better. Charge coupled device (CCD) cameras now offer the flexibility to acquire distortion-free, quantitative, high resolution images from a microscope or other instrument in an easily manipulated format. Image data can be efficiently collected, processed, and displayed in a number of ways. Rapid developments in other digital device designs such as charge injection device (CID) and complementary metal oxide surface (CMOS) promise to press the manufacturers of CCD cameras to even greater improvements and to put these technologies in the microscopist's toolbox. The number of different producers and the variety of CCD cameras is truly staggering. There is a camera for every reason and a sampling time for every purpose under microscopy. Not so long ago a low-light CCD would necessarily be liquid nitrogen cooled and it would cost more than $30,000. At the present time, a reasonably sensitive camera, lacking, perhaps, some of the sampling flexibility of the more expensive products, can be purchased for as little as $2,500. The major improvement in recent years has been the reduction in dark current, which translates into a greater sensitivity. Virtually every major microscope manufacturer offers a complete system for a variety of digital microscopy applications. For anyone who doesn't want to plow through the literature of the 100 or more camera-frame grabber-image processing software products available, these systems represent an excellent solution. Typically, the systems are modular, and the manufacturer can assist in matching a system to the job. Cells loaded with fura-2 and calcium signal induced by NMDA. Image taken with The Cooke Corporation's SensiCam SVGA camera and Axon AIW software. Image courtesy of D. Babcock, Washington University, St. Louis. While fully realized systems are convenient for well defined research projects, for more open-ended research projects, there is motivation toward modular systems in which the microscope, illumination device, digital camera, frame grabber-image processing board, software, and computer are selected to optimally perform the intended project. As a practical matter, most labs will have some of these components and will want to build around them. This having been said, it is important to understand that many CCD cameras are sold with software and will perform optimally only with the software provided. This article deals primarily with CCD cameras and does so in a non-exhaustive manner. The total number of cameras, frame grabbers, and imaging software packages is beyond the scope of the review. As will become clear, the price range for CCD cameras is considerable. As mentioned, $2,500 will get you good performance with respect to one of the major variables of performance, dark current. For this price, the camera will have a CCD array of about 750 x 500 pixels and a small pixel size but will not have all the bells and whistles. At the other end of the price range, for about $70,000, you can have a back illuminated, high quantum efficiency, liquid nitrogen-cooled CCD with possibly a very large array (4,000 x 4,000) made up of large 27 µm2 pixels. A CCD is an array of light-sensitive picture elements, or pixels, each measuring 5 to almost 30 µm across. The camera's lens focuses the scene onto this pixel array. Just as the resolution of conventional photographic film is related to the size of the grain, the resolution of a CCD chip is measured by the number of pixels in the array and the size of the individual pixels. Because the technology that the CCD is based upon is an old one now used only to produce CCDs, the economies of scale that exist in the CMOS chip manufactory do not hold for CCD production. Consequently the price of these cameras has stayed high. Apart from a few manufacturers specializing in the top-end CCDs, most chips are made by some familiar names--Kodak and Sony, for example. So while there are many CCD cameras, they contain a very limited set of chips. This should be helpful in the sense that one can see what the camera maker has done to bring the best performance out of the chip. Fluorescent image taken with Optronics DEI-750D three-chip color camera The choice of chips for use in CCD cameras is limited. Obviously, large arrays--2000 or 4000 cell square arrays--will cost much more than a 400 x 600 cell array. For a given chip, price is determined by the ability to control the camera settings--for example, sampling intervals and binning--by the read-out system and by the level of sensitivity achieved through cooling. There are four major factors in determining camera sensitivity: quantum efficiency, pixel size, read-out and switching noise, and thermal noise. Other factors to consider in purchasing a CCD camera would include digital to analog resolution and signal-to-noise ratios. As we will see below, these are not all independent properties of CCDs. Thermal noise reduction is the component over which the buyer has the greatest control. One can choose among uncooled, Peltier-cooled, and liquid nitrogen-cooled cameras. If you know that your sample is sufficiently bright and your excitation is not excessively destructive, then a modern, low dark current, uncooled CCD may serve you well. Experts would say that exposure times in the range of a few seconds could be performed with an uncooled camera. If the exposure time is not much more than a few seconds to 30 seconds, water-cooled cameras should be explored as a possible solution. For exposures in the range of a few minutes to 60 minutes, a Peltier-cooled camera may be required. For applications requiring the highest sensitivity, liquid nitrogen-cooled cameras are the necessary choice, necessary because liquid nitrogen introduces potentially annoying limitations--the principle one among them being that the nitrogen needs to be replenished every few hours in cameras using a cold finger to cool the chip. For long duration observations, cameras in which the chip is seated within the Dewar should be considered; if this is not done, the Dewar nitrogen level will fall below the cold finger. Moreover, since the temperature of the chip will increase slightly as the nitrogen level in the Dewar decreases, standards must be run to account for increased background, at least for the most sensitive quantitative measurements. For most applications, Peltier cooling is adequate and considerably less trouble. Manufacturers should be queried on the required frequency for degassing CCD chips. Trace amounts of volatile materials used in the production of the chips and supporting electronics are released over time, which reduces the vacuum surrounding the chip and can result in significant reductions in performance. This means that time-consuming and costly vacuum pumping of the chip is sometimes required. The individual pixels of the CCD chip should be matched if possible to the characteristics of the imaged sample. Currently pixel sizes range from about 6 to 27µm. Large pixels, because of their greater extent, will have a greater individual sensitivity and greater full well capacity (maximum number of electrons that can be stored between read-out cycles). Point sources of emitted light such as those produced by FISH "headlights" for example, will be better detected if they are of roughly equal intensity, using smaller pixels, while very dim diffuse light will be better detected with larger pixels. Note though that if the full well capacity of the smaller pixel is exceeded, the charge will bleed into the next bin and corrupt the data. On the other hand, because it is usually possible to bin arrays of pixels into a single value during read-out and thus gain most of the sensitivity of large pixels, smaller pixel arrays have additional flexibility in their applications. For the serious student of microscopy, it is recommended that the pixel size be matched to the object type and magnification. It is possible to calculate the ideal pixel size to collect an image at a given magnification with the following equation (M. Christenson, Princeton Instruments, personal communication): Pixel Size = 0.305 x lambda x Mobj x Mtl / NA where lambda = wavelength Mobj = magnification of the objective Mtl = magnification of the tube lens NA = numerical aperture Full well capacity refers to the maximum number of electrons that can be stored in a pixel without saturation. For example, the popular Kodak KAF series, which has a nine micron square pixel, is capable of storing approximately 85,000 electrons before becoming saturated. Chips such as the SITe 512 x 512 chip series have larger 24 µm pixels and can store 350,000 electrons before saturating. Quantum efficiency (QE) is a measure of how well a CCD detector converts incident photons into electron pairs and thence to a detectable signal. Most front-illuminated CCDs have quantum efficiencies of 30% to 50%. Back-illuminated detectors have the best quantum efficiencies. In the blue wavelengths, these chips will have QEs of 70% to 80%, while in the green and red wavelengths they may reach 90%. For comparison purposes again, the front-illuminated detectors, such as the Kodak KAF series, have a peak efficiency around 40%, dropping to 4% at the blue end of the spectrum. Human striated muscle taken with SPOT camera from Diagnostic Instruments In all but the most expensive CCD cameras, light is detected through the top surface of the chip. Since the charge transfer surface and, if present, the antiblooming gates are necessarily there as well, a significant fraction of the incident light is blocked from reaching the detector. However, the mass production of back-illuminated chips has not become a reality, and consequently their cost is high compared with the front-illuminated devices. A familiar experience for microscopists working with samples where the objects of interest are punctate and of highly variable intensity is a bloom or streak that destroys the image. This problem occurs when a pixel of a CCD becomes saturated. In a CCD with small pixels and therefore a small full well capacity, this can happen rather quickly. The charge will overflow into the adjacent pixel, which may also overflow, and the effect will propagate along the entire line of an array. Although the problem of resolving punctate spots may have led to the purchase of a small pixel size camera, larger pixels actually would have given a better result. The overflow can be countered by adding to the surface of the CCD an antiblooming gate that bleeds off the excess charge. The presence of an anti-blooming gate can protect against overexposures that would otherwise ruin the data set, but it reduces the sensitivity of the CCD by as much as 30%. These gates also cause a reduction in full well capacity and a slight reduction in resolution of the array. For quantitative imaging, these are not the CCD detectors of choice since chips with anti-blooming gates are not linear over the full well capacity. In the discussion of CID cameras that follows, note that these devices have a superior mechanism for eliminating blooming without the unwanted side effects. Fluorescence Image of Molecular Probes Sample #2 taken with REAL-14 precision digital camera from CRI Some CCDs are designed to be illuminated through their substrate layers. The substrate needs to be "thinned," which is a major cause of chip failure during manufacture, hence these devices are costly. Such "back-illuminated" CCDs, however, have excellent quantum efficiencies, in the order of 80%-90%, and perform extremely well at the blue end of the spectrum. For quantitative multiwavelength fluorescence, these chips should be the first choice. A frame transfer CCD consists of two arrays, one of which is masked. Output from the unmasked array can be transferred rapidly to the masked area, which is then read at a slower rate. The ability to rapidly transfer data from the unmasked array allows for a very short exposure time. Described by some as the latest in CCD technology, these so-called "lens on a chip" devices provide an alternative to back-illuminated CCDs, particularly when working in the blue and green, where QE's are quite high. These instruments are shutterless, thereby providing shorter exposure times and faster readout rates. Signal-to-noise (S/N) ratio is perhaps the best single measure of a camera's performance since it will be influenced by many, if not all, of the factors discussed thus far. What we are trying to separate is the incident light from the sample material from thermal electrons, switching and readout noise. S/N, which must be, as a minimum, two to five, is defined as the number of electrons generated by the signal divided by the unit of noise. For example, a pixel with 50,000 electrons per well with background of 10 electrons will have a S/N of 5,000, slightly better than 12 bits. For almost any application in microscopy, CCD cameras are superior to tube cameras or their intensified counterparts and should be the choice for now. However, before laying down your valuable grant dollars for a camera, talk to as many of the camera makers as you can. You will get a variety of viewpoints and probably some conflicting information but, if you talk to enough of these generally knowledgeable people, you'll be able to make a consensus choice with confidence. Mouse Cochlea Nerve Fibers taken with Photometrics HCCD camera system. Image courtesy of Bruce Tempel and Hao Wang, University of Washington and Paul Goodwin, Fred Hutchinson Cancer Center Twenty years ago, image processing could be considered only if one possessed a specialized computer or access to a supercomputer. Though even by that time essentially all the imaging tricks were well understood and the algorithms encoded in specialized image-processing boards, it was something of an amusement to talk with the military and space agency imaging folks who liked to believe that there were imaging secrets--but a bit plane is a bit plane and a gray scale a gray scale whether it is generated by an ICBM or a tumor cell. What was different was the amount of money that biologists versus Defense Advanced Research Projects Agency could spend on programming and computers. In the decade that followed, imaging applications moved from specialized computers to specialized boards. Now it seems that image processing is a software function on a fast personal computer. However, it may still be the case that if one wants to do real-time, machine recognition of some cellular event, then a specialized board or computer with multiple arithmetic logic units and some ability to do pipeline processing would be required. To follow dynamic changes over time, where no processing of the image beyond frame grabbing is required, or if digital time-lapse recording is of interest, the study may still be facilitated by specialized hardware. But for most applications where computation time is mostly a matter of the investigator's patience, fast personal computers, a simple frame grabber, and imaging software will serve nicely. By avoiding specialized boards, one avoids the experience of having obsolete hardware make obsolete thousands of programming hours' worth of software. If it can be done in software alone, do it. It seems likely that sometime in the near future, research-quality CMOS-based cameras will be available for applications that are not trying to push the sensitivity envelope. CMOS cameras use the same complementary metal oxide semiconductor technology used to make 95% of all the electronic circuits produced today. This means that the manufacturers can use the enormous productive capacity of the integrated circuit industry to mass produce these camera chips in the same way that microprocessors and dynamic random access memory components are produced. In the consumer camera market it is clear that CMOS will soon be the dominant product. This is largely due to the investments of industry giants such as Eastman Kodak, Motorola, Toshiba, Intel, Rockwell, and Sarnoff. What CMOS sensors share with all CMOS chips is their low power consumption. CMOS sensors can be directly integrated with other circuitry--for example, for analog-to-digital conversion or image processing--resulting in further savings in cost, power consumption, and size. However, these features are not particularly important in the digital microscopy environment. CMOS cameras currently in the market are far too noisy for many research applications, and the problem is due to switching noise that cooling will not eliminate. In an attempt to solve this problem, Motorola and Kodak in a CMOS sensor venture, are modifying their production lines to produce better detector quality products. But as has been pointed out on a number of occasions, if you modify the production line you may lose the major advantage of CMOS: low cost Fibroblast stained with anti/tubulin antibody imaged with MicroMAX Interline Camera System from Princeton Instruments CID cameras have been around since the early 1970s but have only recently been recognized for their particular strengths. Unlike CCDs, CIDs can be sampled nondestructively for the accumulated charge in each well. Moreover, if the charge saturates, it does not spill over into adjacent wells and cause a bloom or smear as it does in a CCD. Rather, it can be bled off nondestructively. This has certain advantages; if the object of interest is faint and located in the vicinity of a bright object, it is possible to allow the charge to accumulate in the dim object well without losing it in the bloom of the nearby bright object. Exposures can be continuously monitored in a time-lapse manner and sampled at the time of optimum exposure. Again, as with CMOS, these cameras suffer from their inherently noisy electronic switching in the read-out. Applied Scientific Instrumentation Applied Scientific Instrumentation offers complete turn-key imaging and photometric systems with closed-loop microscope focus controllers, automated stages, shutter controllers, custom shutter housing, high speed monochromators and microinjection devises, bundled with a number of software packages. A broad based imaging system based around Cytos (Mac based program) is available for time-lapse, calcium ratio and other low light image acquisition options. A high speed, calcium ratio system is also available. TillvisION operates on an NT4.0 platform and can acquire over 100 ratios per second. Camera:Choice of many, both analog and digital Software:TILLvisION, CYTOS Platform:Windows 95 Axon Instruments Axon Instruments, Inc., offers both modular components and turnkey systems for cellular imaging, and for electrophysiological stimulation and recording coupled with imaging. The Integrated Imaging System offers an array of components, among them the Imaging Workbench image acquisition and processing software, a Dell Dimension computer, and a choice of digital or analog cameras and accessories. Imaging Workbench provides ratio-based ion imaging and dynamic fluorescence measurements, and controls most digital cameras and most wavelength switching devices. Company scientists will assist in the choosing of components and evaluate the appropriateness of system for the task on site. Camera:Digital (Photometrics, Princeton) or analog (Photonic Science) Software:Imaging Workbench 2.1, Image-Pro Plus 3.0, pAlamp7, AxoScope 7 Platform:Windows 95 or 98 Price Range:$20,000 to $47,000. Components available. Imaging Research Imaging Research Inc. makes advanced image analyzers (including specialized cameras and optics) for ultra-low light imaging (e.g. bioluminescence, chemiluminescence), microscopy, quantitative densitometry (including all forms of autoradiography), genomics and pharmaceutical lead discovery. The company offers MCID and AIS image analyzers, and ArrayVision software for analysis of high density hybridization arrays. Camera:Choice of many, both analog and digital Software:MCID, AIS, ArrayVision, optional 3d reconstruction modules Platform:Windows NT Price Range:AIS software from $2,500, MCID systems from $20,000. Inovision Corporation Inovision Corporation offers ISee Analytical Imaging Software and fully integrated ISee Imaging Systems with applications that include cell tracking, multi-camera imaging, real time image acquisition and storage, ratio imaging, probe co-localization, and 2D/3D deconvolution. ISee's modular design enables Inovision to incorporate new equipment as it becomes available and adapt its imaging applications to new methods. Camera:Dage MTI, Hamamatsu, Photometrics, and Princeton Instruments Software:ISee FL and ISee APA Analytical Imaging Software Platform:Silicon Graphics Inc., Sun Systems, Hewlitt Packard Price Range:starting at $4,000 (software), systems starting at $11,00. Intracellular Imaging Intracellular Imaging offers several complete, turnkey fluorescence imaging systems and microphotometry systems. While specializing in ion imaging, general purpose and morphometric image packages are also available utilizing Media Cybernetics's ImagePro and Fluoropro SW packages. All Intracellular systems include an epi-illuminationg source, filter changer, integrating CCD camera, Pentium Pro 233 Mhz Windows NT based image acquisition and analysis workstation Camera:8 bit cooled CCD, Princeton Instruments 12 bit cooled CCD, ICCD Software:Intracellular developed acquisition and analysis package; ASCII format (graphical data), TIFF (images) Platform:Windows NT 4.0 Price Range:$12,000- $60,000 Leica Microsystems Leica Microsystems offers the new DC 100 digital imaging system. This integrated system has optical components and software designed specially for microscopy that enable digitized images to be created, manipulated, and stored with both incident or transmitted light or fluorescence. Image quality results from digitization in the camera, pixel duplication, and direct transfer of the original pixel information. Measurement and analysis are facilitated by synchronization of pixels between the camera and monitor. Progressive image scanning prevents flicker and provides fast transfer in real time: live image is immediately available for focusing and framing. Camera:Leica DC 100 Software:TWAIN, Leica QWin software Platform:Wind95, NT 4.0, TIF, BMP, and JPG formats Price:$43,000 includes camera, software and PCI card Olympus America Olympus America Inc. offers Merlin, an integrated package consisting of a high speed, low light level imaging camera (analog or digital); SpectraMASTER, an ultrafast monochromator wavelength selection device; and a workstation with mass optical storage and components that interface to most optical microscopes-fluorescent, brightfield, or interference contrast instruments. The Merlin workstation provides experimental control throughout the experiment with event marking, variable capture intervals, experimental pausing, and synchronization with external events. Cooled CCD cameras are available that have onchip integrating capabilities with high signal to noise and large dynamic range. Platform:Windows 95 Price Range:competitive Carl Zeiss The Carl Zeiss Vision Division offers its modular designed Axiovision and KS series of image analysis software for low light level and bright field applications. The series covers software for image acquisition for a wide range of state of the art digital and analog cameras, digital photography, archiving and image processing/analysis software for two- dimensional and three-dimensional applications. In connection with its line of microscopes, which are fully controlled by the software, Carl Zeiss offers a total imaging solution. Price Range:$2,000-$11,000 The first critical step in three-dimensional (3D) digital microscopy, whether laser-based or CCD-based, is capturing raw images. This requires an image detector, a computer, and specialized software that can communicate with the image detector and read off images as a series of pixel values (2D image elements). It is critical for most applications that the image detector is linear. That is, a spot that is twice as bright in an image has a pixel value that is twice as great. Data linearity is a basic assumption for most types of digital microscopic processing and analysis. Modern detectors of choice include calibrated photomultiplier tubes (PMTs) for laser confocal systems and charged coupled devices (CCDs) for digital detector based systems. Bright field, dark field, phase contrast, and Nomarski digital imaging require CCD based detectors. Epifluorescence imaging, however, can be performed with either a laser-based or a CCD-based system. Different types of digital microscopy require additional data collection hardware and software. Some of the popular types of digital microscopy are three-dimensional (3D), ratio (for Ca++ and pH imaging), and live cell. This article will focus on 3D digital imaging. 3D imaging systems require control of the Z-axis, usually in the form of a stepper motor or piezoelectric focusing collar. An excitation light source shutter is also usually required. 3D imaging also uses automated optical filter control. This can achieved via automated filter wheels, filter cubes or monochromators. An automated X-Y stage may also be beneficial for collecting large 3D montages. Other elements of the optical path may also be automated such as lens selection, illumination intensity, stage temperature, etc. The exact automation required depends on the specific application. In addition to automation, digital encoding allows imaging systems to be aware of the state of optics such as filter wheels, optovars, prism orientations, etc. As a general rule, the most convenient systems automate and encode all the relevant optics for a certain application. This insures that the data capture parameters are logged correctly for later processing. For correct system integration, the image capture software should be able to communicate with all the hardware automation and encoding. After images have been collected as a two-dimensional array of pixel values, they can be de-blurred using a mathematical algorithm called deconvolution. Deconvolution is the process of computationally removing out-of-focus blur from images by employing knowledge of how a particular set of optics blurs a point source of light. Currently there are four basic types of deconvolution: No Neighbors: An approximation that allows de-blurring of 2D images. Nearest Neighbors: An approximation that requires capturing planes above and below the plane of interest to remove haze. Constrained Iterative: A rigorous iterative approach that uses a measured point light source or point spread function (PSF) and yields a quantitatively true 3D model of the sample. Blind: An extremely computationally intensive approach that calculates both the PSF and 3D model from the raw data. No Neighbors, Nearest Neighbors, and Constrained Iterative deconvolution can be performed on a high-end desktop computer. Blind deconvolution is too computationally intensive for the computer facilities available to most microscopy labs. Once the raw data have been deconvolved, they can be analyzed, displayed or rendered. One of the most significant advantages that digital microscopy has over conventional microscopy is its versatility in exploring and presenting the 3D structure of cells and subcellular features. Often hidden from view in a conventional microscope, many features can be seen by rotating data to its best vantage point. Displaying 3D digital microscopy data poses several key challenges, however. Data sets are typically extremely large and rarely lend themselves to simplification. Investigators are often interested in features that approach the scale of the resolving power of the light microscope itself, so it is important to faithfully preserve and present every single data point. Further, fluorescence data rarely gives sharp boundaries but rather varying intensity gradients. Finally, the physical nature of fluorescence data is unlike what most 3D graphics systems are used to rendering: instead of opaque objects that reflect light, data consist of numerous light sources of varying intensity. Direct volume rendering has been the most common approach, where every data point contributes to what is displayed on the screen. One algorithm is maximum intensity projection, where each pixel of the display is determined by the brightest point along a ray that travels through the data. Although fairly fast, individual images generated from direct volume rendering do not reveal 3D structure nearly as well as movies created from a number of viewing angles. Model-based rendering algorithms try to represent clusters of data points as complex geometric solids. While they can lead to attractive 3D models similar to those produced in CAD programs, they almost invariably compromise some of the data's fine structure. Finding good rendering approaches for fluorescence data is an active research area, where the goal is to reduce the tradeoff between how easily 3D structure can be surmised from a single view and to what extent artifacts are introduced by the rendering algorithm. Initially, the development of digital microscopy concentrated on acquiring and deblurring data. Now that most of the obstacles to rapidly collecting accurate 3D data have been eliminated, many in the digital microscopy community are shifting their focus to analysis. While digital microscopy dramatically extends the capability of the light microscope as a qualitative tool, its greatest promise is to advance light microscopy to the point of answering meaningful quantitative questions. Most image-processing systems now have facilities for basic morphometry and densitometry: computing the volume, orientation, and distribution of features within a data set. Some systems have extended capabilities, such as template-based shape classification, multiple fluorescent-dye colocalization measurement, and even fourier descriptor analysis to quantify molecule translocation. There are two areas of great potential improvement, however. The first is in standardization. While more and more research papers include analysis of digital microscopic data, almost every approach to answering difficult morphometric and densitometric questions is unique. One reason for this is that most commercial image-processing systems either offer only the basic tools for performing quantitative analysis (leading each researcher to construct his or her own solution) or have their own unique approaches to more advanced analyses. Clearly there should always be room for new approaches, but at the same time any approaches that can be standardized should be. More cooperation between academic researchers attempting quantitative digital microscopy and commercial software providers will help forge image-analysis standardization. The second is in increasing analytical throughput. Currently, most analyses require significant user interaction with the image-rocessing system. This means that in most labs, far more time is spent analyzing even modest amounts of data than in collecting new data. A critical factor in increasing the throughput of quantitative microscopy systems is the ability of the computer to automatically detect in a large number of samples the kinds of features that an investigator is interested in. Most promising are approaches that learn "rules" from the investigator and can apply them to orders of magnitude more data than the investigator could on his or her own. Neural networks and other statistical pattern-recognition approaches can perform much of the feature recognition and delineation that once had to be done manually. Colin Monks is vice president of Intelligent Imaging Innovations. He can be reached at colin@intelligent-imaging.com CCD Cameras, Image Processing Software Robert R. Klevecz is a Senior Scientist and Director of Cellular Dynamics at the Beckman Research Institute of the City of Hope National Medical Center He can be reached at neumann@earthlink.net.

Deep Wells and Dark Currents: CCD Cameras Offer Microscopists an Immediate, Distortion-free, Quantifiable Image

Interested in reading more?

Become a Member of