Scanning fluorescent microscopy is an alternative for quantitative fluorescent cell analysis

Authors


  • Presented at the ISAC XXI International Congress in San Diego as a paper entitled “Tutorial VI. Slide-Based Cytometry Systems,” and as a poster entitled “Development and Standardization of a Scanning Fluorescent Microscope System for Cytometric Measurements Using Digital Slides.”

Abstract

Background

Fluorescent measurements on cells are performed today with FCM and laser scanning cytometry. The scientific community dealing with quantitative cell analysis would benefit from the development of a new digital multichannel and virtual microscopy based scanning fluorescent microscopy technology and from its evaluation on routine standardized fluorescent beads and clinical specimens.

Methods

We applied a commercial motorized fluorescent microscope system. The scanning was done at 20× (0.5 NA) magnification, on three channels (Rhodamine, FITC, Hoechst). The SFM (scanning fluorescent microscopy) software included the following features: scanning area, exposure time, and channel definition, autofocused scanning, densitometric and morphometric cellular feature determination, gating on scatterplots and frequency histograms, and preparation of galleries of the gated cells. For the calibration and standardization Immuno-Brite beads were used.

Results

With application of shading compensation, the CV of fluorescence of the beads decreased from 24.3% to 3.9%. Standard JPEG image compression until 1:150 resulted in no significant change. The change of focus influenced the CV significantly only after ±5 μm error.

Conclusions

SFM is a valuable method for the evaluation of fluorescently labeled cells. © 2004 Wiley-Liss, Inc.

For quantitative and statistical analysis of large cell populations, FCM (1,2) and laser scanning cytometry (LSC) (3, 4) are two widespread tools with partially overlapping routine and research analytical cytology applications (5, 6, 7). Both modalities can excite cells labeled with fluorescent dyes using a laser, and both can measure the emitted light intensity with photomultipliers. FCM can use arc lamps as well, but most commercial instruments use one or more lasers.

In a flow cytometer, the cells flow through the laser beam in suspension; in the laser scanning cytometer, the laser scans the cells on a microscopic slide. LSC can record the cells' coordinates on the slide along with the other parameters; thus, individual cells can be re-located on the slide for visual examination of the cells' morphology. Fluorescently labeled cells can also be examined by a conventional fluorescent microscope; in this case, the dye is most frequently excited by a mercury arc lamp and the emitted light is detected by a charge-coupled device (CCD) camera. Other light sources can be used with fluorescent microscopes, as the laser scanning cytometer is also built around one, but xenon and mercury arc lamps are used most often.

The photomultiplier used in the flow cytometer, after amplification, pulse processing, and analog-to-digital conversion (ADC), immediately provides the measured data to a computer, so that it can be immediately analyzed and stored. Data extraction from CCD camera–generated images by image analysis requires extensive computation. Inexpensive personal computers and reasonably priced, low noise, high resolution CCD cameras are now available for this application.

Attempts were made in the late 1980s and in the 1990s to convert the fluorescent microscope into a scanner for single- (8, 9) and multifluorescence measurements. One of the first studies was published by Galbraith et al (10, 11). In their study, they showed that multicolor fluorescence analysis can be performed by motorized fluorescent microscopes. A comparison between FCM and imaging cytometry showed similar distribution of monocyte and NK cells from peripheral blood.

Kozubek et al. (12) developed high-resolution cytometry, which enables automated acquisition and analysis of FISH-stained nuclei using wide-field and confocal imaging techniques. Automated FISH analysis was also developed by others (13, 14, 15). Mehes et al. (16) developed systems dedicated to rare-cell detection.

All of these techniques required substantial time and effort to determine the focus level. A high-speed automated approach was presented by Bajaj et al. (17), using an upright microscope and phase contrast illumination strobed under computer control. This system has been successfully applied to high-speed rare-cell detection. However, in all of these techniques, measurements are performed directly after image capture. Only measured parameters and selected images are stored for later evaluation.

We recently reported the development of digital slide and virtual microscopy technology for histological sections for fluorescent microscopy applications (18), with the advantage of separated slide digitization and time and location independent “after scanning” evaluation.

In this work, attempts were made to standardize and calibrate this system. The opportunity for routine cytological and histological application was also evaluated. The effects of focusing accuracy and image compression were examined. Our system is called the “scanning fluorescent microscope” (SFM). The word scan is used in a sense that the images of multiple fields of view are grabbed and recorded as they are. During evaluation, the complete slide is digitally reconstructed from the stored images by the software, hence the name “digital slide.” The software part of the system that simulates the functionality of a real microscope on digital slides is called the “virtual microscope.”

MATERIALS AND METHODS

Sample Preparation

For testing and calibration of the system, 10 μm diameter cytometric calibration beads were used (Immuno-Brite Fluorospheres, Part No. 6603473; Beckman Coulter, Fullerton, CA). These beads had the advantage of being a mixture of populations, each with an exponentially increasing number of fluorescent molecules.

For evaluation of clinical samples, residual samples from young, cancer-free patients were used. Mononuclear cells were isolated from EDTA anti-coagulated blood with standard density gradient centrifugation (Histopaque-1077; Sigma-Aldrich, St. Louis, MO). After removal of mononuclear cells from the layer floating on the Histopaque, the cells were washed three times by PBS and counted in a Buerker chamber. A total of 10 μl of the cell suspension was smeared on a microscope slide and dried. Cells were stained in 100 nM Hoechst 33258 for 20 min, and washed in PBS. As recommended by the manufacturer, ProLong Antifade (Molecular Probes, Eugene, OR) was used on dry smears to minimize fluorescent fading and permit multiple scanning.

For the correction of mercury arc lamp uniformity errors, a slide with evenly distributed FITC stain was prepared. A 1 mg/ml FITC solution was mixed with ProLong in a 1:10 dilution, and 30 μl was applied to a glass slide and covered with a 22 × 22 mm coverslip.

SFM HARDWARE

The SFM includes hardware and software components. In this study, an Axioplan 2 imaging MOT (motorized) microscope (Carl Zeiss Gmbh, Oberkochen, Germany) was used, with motorized objective and filter changer. A Plan-Neofluar 20×, 0.5 NA dry objective was used for every scan in this study. The system was equipped with a high-resolution AxioCam HR color camera (Carl Zeiss Gmbh). Though the highest resolution of the camera is 3,900 × 3,000 pixels with 14-bit depth, only 650 × 515 pixel resolution was used, with 8-bit depth in black and white mode. The relatively low resolution and bit depth was used to lower the exposure time as much as possible, in order to speed up scanning.

The software ran on a 700 MHz PIII PC with 256 MB of RAM and a 40 GB hard drive (HDD). The operating system was Microsoft Windows NT Workstation 4.0 with Service Pack 6 (Microsoft Corp., Redmond, WA). Only the PC interface card provided with the AxioCam was required for image acquisition. The microscope was controlled through the RS-232 port of the computer. The microscope and camera were controlled using the software libraries provided by the AxioCam manufacturer. SFM will work with any PC with the above or better characteristics.

SFM Software Components

The SFM software components were written by the authors in C++, using Borland C++ Builder (Borland Software Corp., Scotts Valley, CA). Figure 1 is a flowchart that describes the interactions of the software components.

Figure 1.

Flow chart of the working logistics with the SFM program. Following the arrows, one can see the steps of scanning or evaluating a slide.

Hardware handling modules.

The hardware handling modules are dynamic linked libraries (DLLs), as are all the other modules, and contain a low but abstract level of functions to control the specific hardware to which they belong.

Scanning module.

This is a complex module that unites other smaller modules. This module covers the complete functionality of scanning. It contains the following modules: settings, image-processing, scanning strategy, and scanning area designation. The scanning strategy module can be changed to realize different scanning approaches.

Autofocus module.

This module finds the best focus level for a field of view during scanning. The standard overall image sharpness methods did not work properly for autofocusing, because, in most cases, the cells fill out less than 2% of the field of view. In this case, calculating the image sharpness for every pixel will result in measuring the background noise, instead of measuring the proper focus level of the cells (19). The basic characteristics of the fluorescent images are that on a dark background there are some bright cells. Focusing on the image of the cell is also difficult, because it is often small and has no distinct pattern. Because of the properties of the point spread function of the optical system, the cell's image becomes increasingly blurred as it is removed from the right focal plane. The focus measurement method utilizes the relatively great intensity difference between the background and the cells. It thresholds the image by a predefined threshold and measures the area of the remaining objects. If the cells are in focus, their images are sharp, and their areas are small. If the cells are out of focus, their images are blurred, and after thresholding, the value of the remaining area, which includes the cell and its close surroundings, is larger. First, a coarse focusing is performed for every field of view by calculating the focus value in a predefined range with a step size of 10 μm. In the second phase, the focus is refined using successive approximation around the best focus level determined. The steps of the focus refinement used are 5, 3, 2 and 1 μm.

Slide viewer module.

Provides the functionality of a virtual microscope. Displays the data recorded from the real slide. It preserves the 2D structure of the images and provides slide zooming and moving, plus visually displays the evaluated data.

Image processing module.

This module evaluates the scanned digital slide offline. Since all measurement data is saved into the digital slide, the results are not lost when the digital slide is moved or copied. Only the latest measurement data is stored.

The first step of image processing is illumination nonuniformity compensation or, in other words, shading compensation. An illuminated reference (white) image and a CCD dark current noise reference (black) image are made at the beginning of every scan. The noise present in both images is decreased by separately averaging 16 images of each type of reference image. The images in both groups were grabbed consecutively, and one image was made by averaging the same pixel in every image. The white reference image is made using the evenly spread FITC slide described in the sample preparation section. The black reference image is made by closing the camera shutter on the microscope. The black reference image is for compensating for the dark current error of the camera. The black reference image is subtracted from every image, because it represents the standard error of the camera. Every field of view is compensated according to the following equation (Eq. 1.a). I′ denotes the compensated image, I denotes the original image, B denotes the black reference image, and W denotes the white reference image. The x and y indices denote an image pixel, and u and v denote the coordinates of the brightest pixel in the white reference image.

equation image(1.a.)

The following example explains how compensation works. In equation 1.a, the corresponding pixel value of the black compensation image is subtracted from every pixel value. In our example, we suppose that the camera is ideal and has no dark current error. In the case of modern Peltier cooled CCD cameras, this is almost true, and the black reference image has gray-level pixel values less than 3. If we eliminate the black reference subtraction from equation 1.a, we get equation 1.b.

equation image(1.b.)

Figure 2A represents a white compensation image; the nine regions of the image are illuminated with different intensities. The gray-level values of the pixels in a region are written in the region. Figure 2B represents one field of view with beads of equal numbers of fluorescent molecules; each bead's gray-level value is indicated next to it. For the ideal case of even illumination, the gray-level values should be equal, but the values are linearly proportional to illumination intensity.

Figure 2.

Artificial images demonstrating shading correction on one field of view. A: Demonstration of white compensating image. The nine regions represent the different illuminations. Numbers in regions show the gray level values. B: Demonstration of uncompensated field of view. The beads should have the same brightness, but they are different because of the uneven illumination. Beads' gray level values are shown beside them. C: Same demonstration field of view as in (B). After compensation, beads have the same intensity. Beads' gray level values are shown beside them.

Compensation works in the following way. First, a search is performed for the gray-level value of the brightest region in the white reference image. This will be Wmax, which is 210 in the example. Then, we modify the gray-level values of the beads. The gray-level value of the bead in the lower right corner of Figure 2B is 77; this will be Ix,y in equation 1.b. The gray-level value of this bead's region in the white reference image is 70; this will be Wx,y in equation 1.b. If these values are substituted into equation 1.b, the result, I′x,y, will be 231. Equation 1.c demonstrates this calculation.

equation image(1.c.)

Figure 2C represents the compensated field of view. Unlike in the example, in the SFM the compensation is done per pixel and not per bead.

In the second step of image processing, images are thresholded. Every channel has a separate threshold value. If the blobs remaining after thresholding have overlapping pixels in the different channels, their union is treated as a cell.

The following morphometric parameters are calculated for every cell: maximum diameter, minimum diameter, average diameter, area, and perimeter. The following fluorescence parameters are calculated for every cell: integrated fluorescence, minimum fluorescence, maximum fluorescence, average fluorescence, and fluorescence range.

In our terminology, fluorescence means the intensity of a pixel in the cell's image. Minimum fluorescence is the value of the darkest pixel, maximum fluorescence is the value of the brightest pixel, average is the average of all the pixels, and range is the difference between the maximum and minimum.

Integrated fluorescence (IF) is the same as fluorescence in FCM. In equation 2, the variable IF is the integrated fluorescence of one cell in one channel. Ci is a pixel of the cell image and B is the average value of the neighboring background pixels around the cell. Pixels considered as neighboring background pixels, are those whose gray-level value is below the channel's threshold and are within a rectangle. This rectangle's edges are five pixels farther from the cell's leftmost, rightmost, topmost, and bottommost pixels. In other words, to calculate the integrated fluorescence, the average neighboring background value is subtracted from every pixel of the cell, and these results are summed.

equation image(2)

Cytometry evaluation module.

Scatterplots, histograms, and galleries can be used for evaluating the image processing data. The SFM software can export data to an Flow Cytometry Standard (FCS) file format; however for purposes of efficiency, we have implemented the functions of a standard FCS analysis program to integrate with the digital slide galleries and export to Microsoft® Word.

The software handles a maximum of six fluorescent channels. For every channel, a separate pseudocolor and image processing threshold can be defined manually. Any parameter of each channel can be displayed on the x- and y-axis of a scatterplot or on the x-axis of a histogram. The gates can be linked, i.e., the cells defined by a gate on a scatterplot can be the source data for a histogram and vice versa. An arbitrary number of gates can be linked and any gate's data can be displayed in a gallery. If a cell is clicked in the gallery, the dot representing the cell will turn to red from black in every displayed scatterplot and the digital slide will be centered around the clicked cell, which is now highlighted. The scatterplots, histograms, and galleries created can be exported into a Microsoft Word document as an image and the cells' data can be exported as a table.

The fields of view in the digital slide were stored in standard JPEG format using Intel JPEG Library 1.5 (Intel, Santa Clara, CA).

The statistical analysis was performed by the Statistica program package (StatSoft, Tulsa, OK).

RESULTS

The scanning speed of the system is dependent on the emission intensity of the sample, the camera speed, and the requirements for autofocusing. In the SFM concept, the fluorescent channel with the highest contrast is the one that is used for focusing and object segmentation. In general, the most prominent dye applied for this purpose is nuclear staining. If the system is intended to quantify cytoplasmic and/or cell surface staining, there is no need for image recording in additional channels. This way the usual time for movement, autofocusing, and image capture for a field of view was 5 ± 0.5 s; and 975 KB were required for storage of an area 430 × 320 μm. A 3.7 × 3.7 mm cytospin, digitized in three fluorescent channels, required 100 frames and, without image compression, 285 MB of storage.

Standardization beads were used to evaluate the system's performance. The Immuno-Brite bead set's brightest fluorescent peak was taken as reference value.

Without shading compensation, the CV of the specimen was 24.3% (Fig. 3A). After the use of the white and black reference images for compensation without image compression, the CV decreased to 3.9% (Fig. 3B).

Figure 3.

CV of the IF values of standardization beads with and without compensation. The dark gray columns represent the measured single beads and the light gray columns represent the doublets, clusters, and other artifacts that were gated out based on area and perimeter. A: IF histogram of single-gated beads without compensation, 535 beads are measured. CV = 24.3%. B: IF histogram of the same beads with compensation, 529 beads are measured. Some beads fell out of the original gate after compensation. CV = 3.9%. C: On the left, six beads without compensation and on the right the same beads after compensation. The beads are in focus, the difference in intensity is only because of the uneven illumination. D: The white compensating image used for shading compensation in this example. In the case of ideal illumination, the whole image should be uniform and bright. In reality, the upper left part is 2.6 times brighter than the lower right. The bit depth of the image is 8 bits, the brightest area's gray level value is 205, and the dimmest gray level value is 78.

Two methods were applied to test the system's linearity. The first method used was to scan a homogenous fluorescence bead sample multiple times using different exposure times, and to calculate the correlation of the mean value of the integrated fluorescence. With exposure times between 1,000 and 4,000 msec in 500 msec steps, the correlation of the integrated fluorescence and exposure times was 0.999963 (P < 0.0001).

The second method used was to compare the fluorescence ratio of different beads measured using FCM and SFM. The beads were measured on a FACScan (Becton, Dickinson and Co., Franklin Lakes, NJ) (Fig. 4A). The ratio of the two brightest peaks using FCM was 4.11, and using SFM it was 4.00. The ratio of the second and third peak was 3.84 using FCM and 10.67 using SFM. (Fig. 4A and B)

Figure 4.

Linearity of the integrated fluorescence measurement. A: Fluorescence histogram of four different bead intensities measured on flow. R3 mean = 60.3; R4 mean = 231.5; and R5 mean = 952.6. The ratio of the R5 mean and R4 mean is 4.11; the ratio of the R4 mean and R3 mean is 3.84. Doublets were gated out. B: IF histogram of the same beads measured with SFM. R3 mean = 624; R4 mean = 6,649; and R5 mean = 26,605. The ratio of the R5 mean and R4 mean is 4.001; the ratio of the R4 mean and R3 mean is 10.67. All ratios were manufacturer-specified at 4.0. The dark gray columns represent the measured single beads and the light gray columns represent the doublets, clusters, and other artifacts that were gated out based on area and perimeter.

The effect of image compression can be evaluated by the results in Figure 5.

Figure 5.

Effect of JPEG compression on the CV value of integrated fluorescence (rhombus symbol), digital slide file size (square symbol), and image quality. JPEG compression defines the level of compression by the quality of the resulting image. Quality of 100% means that it uses only a slight compression and the image will be very good; 10% quality means a very small file size and poor image quality. UC, uncompressed images. Bitmap images were used as uncompressed images. A: Lowering JPEG quality increases the CV value of the compensated single-fluorescence bead population used in Figure 3B, and decreases the digital slide file size. The CV curve is marked with rhombuses and the file size curve is marked with squares. B: Images of a bead (top) and lymphocyte (bottom) from the clinical sample at different compressions. There is no significant difference between the uncompressed and the 80% quality image. At 20% quality, the cell lost its texture, and the blocking effect of JPEG can be observed. At 0% quality, the images hardly resemble the originals.

We found that when using standard JPEG images up to a compression ratio of 1:150, the CV does not change significantly (Fig. 5A). A general property of standard image compression techniques that reduce the amount of information in the image (known as “lossy” compression), such as the JPEG standard, is that the loss is in the resolution domain and not in the intensity domain. Since the quality measurement is based on integrated fluorescence or intensity, the CV is good at high rates of compression, but the image quality is not acceptable (Fig. 5B). The best compromise in compression is between 1:50 and 1:100; in this range the CV and image quality are still very good, but the size is reduced dramatically. Standard JPEG defines the compression in image quality instead of compression ratio; a compression quality between 90% (1:61) and 80% (1:116) provided images suitable for future research. The quality percentage does not relate to exact compression ratios; other samples would have different ratios for the same quality settings. The newer version of JPEG, JPEG 2000, uses compression ratios directly, thus the final file size can be better controlled. JPEG 2000 uses new image compression techniques too, and will probably provide better quality at the same compression rate (20).

The effect of the focusing accuracy is shown in Figure 6. The manufacturer does not specify it, but the calculated depth of focus is 1.914 μm for a 20× (0.5 NA) objective (21). The CV of single beads is 2.8 in the ideal focal plane and is the same in the 2 μm range, which is in correlation with the calculated depth of focus. As the focusing range is extended, the CV increases. Up to a focal range of 10 μm, the CV is below 4 and is acceptable for measurements.

Figure 6.

Effect of focusing accuracy on the CV value of the IF of the standardization beads. One field of view of beads was recorded in 1 μm steps from –10 μm to +10 μm from the ideal focal plane. From the single images, several digital slides were made, corresponding to a certain focus range. On these created slides, the CV of the populations was measured. There were 48 single beads and three clusters in the field of view. In our measurements, the focus range is always an even number, because it includes the same number of focal planes in plus and minus directions. In this case, we defined the focus range of 10 μm (± 5 μm) acceptable because the CV of single beads remained below 4.

The system was also tested with a slide of Hoechst stained lymphocytes. The scatterplot of area and perimeter, a histogram of integrated fluorescence, and a portion of the cell's gallery are presented in Figure 7. The CV of the gated population shown in Figure 7A was 5.6.

Figure 7.

Application of the technique on a clinical specimen (Hoechst stained Ficoll separated lymphocytes) and the use of the scatterplot and histogram linking. A: Ungated scatterplot of the sample. Since SFM can not measure forward or side scatter, usually the cell area and perimeter values are used to gate debris and clumps. B: Histogram linked from the gated scatterplot. The dark gray columns represent the cells inside the gate and the light gray columns represent the cells outside the gate. C: Portion of cell gallery. Gallery is of cells contained within the gate shown in scatterplot A; the cells are shading-compensated.

DISCUSSION

Fluorescent cell measurements are superior to transmitted light determinations due to higher sensitivity, absence of the distribution error, and a greater number of simultaneously measurable molecular species. A major advantage of slide-based microscopic measurements is that they can be performed in parallel or sequentially.

The earlier automation of FCM systems in sample handling and their higher measurement speed (1) are the cause of the overall dominance of FCM systems over microscope-based systems in cell analytics. However, the lack of opportunity for the morphological re-location of selected cells proved to be one of the most limiting factors of this technology.

The work of Kamentsky et al. (3) on developing a LSC system aimed to fulfill these requirements. The broad range of applications using the LSC system and papers reporting on the LSC system shows that this technology has real acceptance by cytometrists (5, 6, 7).

Attempts at automated fluorescent microscopy have shown the close correlation of the measured fluorescence parameters to FCM (10, 11).

However, the principle of online measurements of the multifluorescent-labeled cells and not storing the individual fields of view caused the loss of one of the most important features that microscopy can yield: the correlation of the measured parameters to morphological data.

SFM has several advantages over FCM and LSC. To add a new stain with new excitation and emission wavelengths requires only a new filter block; while with the other two modalities, it requires the installation of a new laser and photomultiplier. Making a digital slide from the original is ideal for archiving and documentation. The digital slide can be sent via the Internet to fellow researchers, or students can take it home on a compact disk (CD). Because fluorescent samples fade, in some days or weeks the sample itself might become useless. Arbitrary selection of cells can be displayed in galleries in seconds, because every cell's image can be accessed in a random manner from digital storage. This feature gives real freedom for morphological analysis, which is missing in FCM and is only partially supported using LSC.

For rare-cell detection, scanning the complete sample provides the safety of not losing any cells. Offline evaluation permits the measurements to be repeated using different parameters as long as rare cells are found. The digital slide also permits the evaluation of fluorescent histological samples, because the process is not limited to a single field of view.

The main problem of quantitative measurement using mercury arc lamps is nonuniform illumination. Scrambling the light with a fiber optic or liquid waveguide can improve illumination uniformity (22), but these are not yet widely accepted and are not available through standard commercial channels for all microscopes.

In LSC, the scanning laser beam provides a constant intensity in every part of the field of view. Since this is not the case for arc-illuminated microscopes, the illuminated reference (white) image and CCD dark current noise reference (black) image were used for compensation. The first image is called white because in the case of an ideal illumination, all of its pixels should be white, but not saturated. The white image is for correcting the error of the lamp; the black image corrects the error of the camera. The pixels values in a CCD camera's picture are higher than zero, even if there is no light at all. This error means that there is a constant positive shift in the pixel values. On the black reference image, in the ideal case, every pixel should be zero. The use of these two images in equations 1.a through 1.c compensates for the raw images so effectively that the system provides comparable data to FCM or LSC.

In the compensation example in Figure 2B, the mean of the uncompensated beads is 156.2, the standard deviation is 60.9, and the CV is 39%. The mean of the compensated beads in Figure 2C is 231, the standard deviation is 0, and the CV is 0. Compensation scales up gray-level values to the brightest level, thus the new mean is always higher than the original. The same effect can be observed in Figure 3A and B; in the uncompensated histogram, just few beads are above 30,000, and in the compensated histogram, all of them are around this value. In the example, beads are brighter than the white reference image. This is not a contradiction; the white reference does not have to be the brightest value; it has to represent the illumination pattern. We chose upscaling over downscaling, because downscaling would lose information.

SFM stores the original image for every field of view in every channel, plus one common black reference image and one white reference image for every channel. Compensation is always done on the fly during slide viewing or image processing. Compensation is applied to every fluorescent channel of a field of view and, of course, must be repeated for every field of view. This approach yields slower viewing and image processing compared to stored compensated images, but the goal was to maintain maximum confidence and verifiability. What the original image looked like can be always checked in the digital slide viewer: Was it saturated? Did the compensation introduce any artifacts?

Our data showed that using compensation slides, image compression techniques, and a sophisticated autofocusing strategy with the implementation of standard cytometry techniques, an applicable working environment was developed.

Model (23) found that concentrated fluorophores between standard slides and coverslip are better for shading correction than commercial fluorescent slides or fluorophores in chamber. They used a method similar to ours for shading correction, but did not measure large cell or bead populations. Lockett et al. (24) used beads for shading compensation. Their results are better, but they did not measure large cell or bead populations either. Instead of grabbing a single field of view of fluorophores, they position one bead or cell to multiple locations in the field of view. They calculate a compensation image based on the intensity of the same object in these different locations. The consequence of this method is that they also have to compensate for the photobleaching of the object.

With regard to the linearity of the system, in the second ratio of the fluorescence of the standardization beads, the great difference between FCM and SFM is due to the dynamic range. The flow cytometer used for the FCM study had a 10-bit range, and in the SFM system the camera's bit depth was set to 8 bits. The ratio of the first peak and the third peak according to the FCM measurement was around 16. On the 8-bit image, the brightest pixels of a bead had a value of 240; 240 divided by 16 results in 15. If the brightest pixel of the spherical bead is 15, then the edge of the bead fades into the background and is cut off by the threshold and the integrated fluorescence is calculated on a smaller area. The fluorescence dynamic range is around 1:6 using 8-bit depth. FCM and SFM dynamic range can not be compared directly, because in real life the integrated fluorescence value also depends on the size of the cells. Due to its measurement method, SFM's dynamic range is comparable to FCM because it can measure objects up to the size of the field of view. The AxioCam has 14-bit depth and other commercially available, Peltier cooled, monochrome cameras also have 12-bit ADCs and a residual noise of often less than five photoelectrons. In this study, we used the lower 8 bits of AxioCam's 14-bit depth to decrease acquisition time. Fluorescence values are different in Figure 3A and B, because both systems use arbitrary units for quantification. In this measurement, the flow cytometer had a 10-bit ADC, thus the highest measurable value is 1023. In the SFM measurement, one bead's image had approximately 200 pixels; 200 multiplied by 255 (the maximal gray level of an 8-bit image), results in 51,000, which is the highest theoretical integrated fluorescence value for a bead with this bit depth and magnification.

SFM stores every channel in every field of view; therefore the final file sizes would be very large without image compression. Using standard JPEG lossy image compression, acceptable file sizes and image quality were achieved. Standard JPEG supports only 8-bit depth, but the digital camera used provided 14-bit images. Since we had to sacrifice 6 bits anyway, we chose the lowest 8 bits of the camera, to have as short exposure times as possible to increase scanning speed.

Our technology still uses a commercially available, motorized fluorescent microscope. The advantage of this technique is that these microscopes can also be used as a manual tool for standard microscopy. Also, with our technique, they can be converted into a multichannel fluorescent scanner. This technique could be enhanced by sophisticated biotechnology methods to evaluate large number of antigens (25).

The disadvantages of our system are also clear: Manual slide loading and scanning area definition are still necessary, which costs time and manpower. Some companies have developed slide loaders; an example is in the field of cervical cytology analysis. These loaders are only available as a part of a complete and closed system and currently they are not open to the use of other applications like SFM.

The development of Bajaj et al. (17) decreases the autofocusing time by at least an order of magnitude. However, their work was also done around a commercial inverted microscope. A new attempt for automation and enhanced detection was recently described by Tibbe et al. (26) using magnetic cell isolation and compact disc technology.

With the enhanced focus of medical interest on rare cells (stem cells, malignant residual disease, and organ-specific stem cells in histological sections) the mainstream of cell analytical research will need novel microscopy-based techniques. The work presented demonstrates one solution for this requirement.

Acknowledgements

We wish to thank Ferenc Szipőcs for correcting our text.

Ancillary