The Technology Of High Speed Slide Scanning
Metafer scanning systems excel by the
well-considered combination of effective and user-friendly software and hardware parts,
that are complementing each other in the best conceivable way. The result of this unique
concept is the worldwide acknowledged multipurpose scanning platform of MetaSystems,
prepared for automated scanning of slides originating from a nearly unlimited number of
preparational assays.
This document summarizes the strategies Metafer uses to reach its impressive and unique speed and efficiency.
1. General Scanning Strategies
1.1 Precise Slide Movement
In general automated slide scanning with Metafer is performed by moving the
slide with reference to the fixed objective lens in a regular meander-like pattern. Metafer
performs this movement using an 8-slide motorized microscope stage with two electronic
stepping motors (Maerzhaeuser, Germany). A proprietary 2-axes stepper motor
control board in the PC drives the scanning stage in micro-step operation to achieve a
positioning accuracy of 1 micron in x and y direction at a maximum speed of 70 mm/s. It is important to know that all motorized
microscope controls can be operated manually in the usual way. For manual stage movement a
trackball is provided.
1.2 Automatic Focussing
Precisely maintaining the plane of focus
during the scan is of utmost importance. Metafer uses a focussing strategy
which provides very high speed without sacrificing accuracy. At a number of grid 
positions that are
regularly distributed across the scan area the plane of best focus is determined by
automatically moving the stage in the z-direction, capturing images in different focus
planes, and analyzing the focus quality based on a local contrast criterion. Typically 11
z-positions are being analyzed within approximately 2 seconds. The number of grid
positions used for focus analysis is usually 60 for a complete slide (scan area 36 mm x 20
mm). Based on the grid focus measurements the software reconstructs the focus surface of
the scan area by performing a bilinear interpolation and displays it as a pseudo-3D wire
plot (see left image). During the subsequent scan, the slide is automatically kept within
the plane of best focus.
1.3 Image Acquisition
Each field of view is captured with a high
resolution CCD (charged coupled device) camera with long time
integration capability and fast digital read-out, for example the M1 camera from
JAI, Denmark with a resolution of 1280 x 1024 pixels and a capacity to capture 12
images per second at full resolution. Alternatively Metafer also supports the
new Axiocam MRm from Carl Zeiss, Germany, providing a 1300 x 1030 pixels
resolution. It is worth to mention that
cooled CCD cameras that provide lower electronic background at long exposure times are not
required. Automatic slide scanning always requires capturing large numbers of
images, often at multiple wavelengths and multiple focus planes. Therefore, reasonable
fluorescence intensities are prerequisite to achieve exposure times not exceeding a few
seconds per capture.
2. Image Analysis and Applications
2.1 Image Analysis
The digitized images are subsequently analyzed for the
presence of objects which match the search criteria (i.e., cells or metaphases),
predefined in a classifier. Depending on the result of the analysis individual objects
within a field will be stored in an image gallery along with their feature measurement
data (i.e., the number of FISH signals). Once the scan has been completed, the on-screen
image gallery can be used to review the detected cells and to reject unsuitable cells or
to do corrections if necessary. If the gallery image is not sufficiently informative, any
cell can automatically be relocated under the microscope for direct visual inspection.
2.1.1 Metaphase Finding with MSearch
Speed and
specificity are the demands for automated metaphase finding. MetaSystems
provides a very fast metaphase finder since 1986, and up to today it is still the fastest
system on the market. Metaphases can be found both in bright field illumination and in
epi-fluorescence. Each detected metaphase is stored as gallery image together with its
position on the slide. One mouse click relocates desired objects and allows to revise it
in the microscope.
Being a typical routine application, MSearch provides many helpful tools for the analysis of metaphases, including a customisable data entry form. The data entered into this form can be used to generate reports which summarize the data. Using the karyotyping software Ikaros, the metaphases can be captured with high magnification for karyotyping. Since metaphases can be of highly different shape, flexible adaptation to various conditions is necessary. MSearch provides a training module which allows to adapt the system to any type of metaphase. The user can select the positive objects in previously captured images and then let the system calculate an appropriate classifier.
2.1.2 Multiparametric cell feature
analysis with MetaCyte
In automatic interphase FISH analysis the thickness of the nuclei and the small size of the FISH signals pose additional problems. As a consequence, the optical magnification has to be raised to 20 x (for relatively large signals, e.g. centromeric probes) or even 40 x. Additionally image capture at different focal planes is necessary to avoid losing individual spots. And, if spot counts per cell are required, isolated cells need to be identified while cell clusters that cannot be automatically segmented need to be skipped.
2.1.2.1 Cell Detection
MetaCyte determines the focus surface using the counterstain information, before the counterstain information of each field of view is analyzed. The objects within the image field are automatically segmented and suitable cells are identified using appropriate shape criteria. Cell clusters are rejected by detecting the concavity depth of the object contour. If suitable cells have been identified the system selects the filter combination for the FISH signals, otherwise it proceeds with the next field of view.
2.1.2.2 Signal Detection
FISH
signals are captured in several focal planes. When using the 40x lens typically 5 focus
planes separated by approx. 0.7 micron are captured. The individual images are combined to
a projection image that includes only in-focus information of the
individual
focus planes. This projection (or extended focus-) image is used for detecting the FISH
signals. This procedure is repeated for additional color channels if more than one FISH
label is present. Currently, up to 6 simultaneous color channels (including the
counterstain), simultaneously displayable in pseudo colors, are supported. The detected
nuclei are displayed in an image gallery for on-screen review and interactive relocation.
2.1.2.3 Signal Analysis, Data Presentation and Gating
Due to its flexible and powerful parameter setup possibilities MetaCyte can measure numerous features per color channel during the scan. These parameters include shape and texture-related features for cell classification as well as intensity features which yield e.g. ploidy information. Three-dimensional spot distances are also measurable, e.g. for automatic detection of signal fusions in translocation analysis. After the scan, any of the features can be displayed either cell by cell in the galllery image or as one of two histograms available. Alternatively, two features can be compared with two-dimensional scatter plots. Selecting sub-populations based on any one or two feature values (known as "gating" in flow cytometry) allows detailed analysis of the scan data.
2.1.2.4 Scanning in Tissue Sections: Area Measurement and Tile Sampling
While the
single cell analysis approach described above is essential if signal counts per
cell are required, there are situations where sufficient individual cells are not
available or where the cells of interest appear in cell clusters. Smear preparations and
tissue sections are specimen types where the analysis has to rely on cell clusters as
well. In such situations ratio approaches can be 
helpful. A special sampling mode of MetaCyte
allows measuring spot count ratios within automatically positioned sampling windows in the
field of view. For example, gene amplifications in tissue sections can be automatically
assessed by measuring the signal area of the probe for the amplified gene, measuring the
signal area of a reference signal and calculating the ratio. Areas rather than counts are
used if high level amplifications yields clustering of FISH spots which makes the
identification of individual spots impossible. As problems like cut or incompletely
sampled nuclei will affect both signals similarly, they will average out statistically if
a sufficiently large area is analyzed.
On samples with a very high cellular density (e.g. tissue sections) it may be impossible to segment a sufficient number of single nuclei for the analysis in the available slide area. Therefore MetaSystems developed a simple alternative strategy, which was called "Tile Sampling". Non-overlapping equisized square tiles are placed in the counterstain image, maximizing the total fluorescence intensity inside the tiles. Multiple channel tile sub-images are then used for the quantitative analysis and gallery display. The tile placing algorithm tries to include as much cellular material and as little empty space as possible in the tiles.
2.1.2.5 Speed Considerations for Interphase FISH Signal Analysis
The limiting factor in interphase FISH signal analysis is the image acquisition time. The relatively small signals normally require integration times of up to one second for each image of the focus series. This time has to be multiplied by the number of color channels. To keep acquisition times as short as possible, single band filter cubes are preferred, despite the longer filter changing time of approximately one second (compared to the 60 milliseconds of the custom-modified excitation filter wheel). Another essential time factor in spot counting is the cell density on one hand, and the necessary optical magnification on the other hand. With optimized cell spreads scanning speeds of up to 1,000 nuclei per minute (large, eg. centromeric FISH signals) are possible at 20 x, while scanning with a 40x yields up to approx. 200 analyzed nuclei per minute. This scanning rate corresponds to that of a laser scanning cytometer which provides about 10 times lower spatial resolution and no z-information.
2.1.3 Rare Cell Detection with RCDetect
On typical slides for rare cell detection in fluorescent light mode a large number of counterstained cells and a small number of specifically labeled cells of interest are present. The label can consist of a cytoplasmic or a surface marker with a fluorescence tag attached to it and will cover a significant part of the labeled cell. If a immunohistochemically stained slide is used, the conditions are rather similar, with the difference that no single color channels are existing. For automatic scanning in the RCDetect mode a low optical magnification can be used. For high light efficiency Metafer uses a high aperture Fluar 10 x/ 0.5 objective lens.
2.1.3.1 Cell Detection
After the initial focus analysis based on the counterstain (usually DAPI) image the system selects the appropriate filter cube to capture the fluorescence signal of the specific label. The contrast of the automatically captured image is analyzed, and images with insufficient contrast are rejected. Next, in case of a sufficient image contrast a locally adaptive detection algorithm identifies objects that are brighter than their surroundings and at the same time meet predefined size criteria. In contrast to simple thresholding techniques the employed algorithm assures that labeled cells are reliably found despite variable labelling efficiency across the slide, without increasing the likelihood that elevated background is misinterpreted as a labeled cell. If labeled objects have been detected a DAPI (counterstain) image is captured and the software determines if a nucleus of appropriate size is present. If this is not the case the object is considered a staining artifact and is rejected.
2.1.3.2 Increasing the Specificity: Cell Selection
After passing this test, the position of the object is stored together with a gallery image of the cell and its surroundings. This procedure is repeated until the desired scan area has been completely covered. During the scan, the average counterstain intensity of isolated nuclei is automatically determined which allows the software to estimate the total cell count by dividing the total counterstain intensity by the average nucleus intensity. The system can thus provide a quantitative relative occurrence rate of labeled cells. A simultaneous second cell label tagged with a distinguishable second fluorochrome can additionally be detected during the scan. Logical conditions can be used to increase the specificity of the result. For example the coincidence of two labels may be used to select cells of interest.
2.1.3.3 Increasing the Specificity: Sequential Assays
In many if not most applications the rare cell detection
is only the first step that will be followed by a second assay. To this aim, the
slide may be removed from
the scanning stage after detecting the rare cells of interest, and a sequential assay
(e.g. a fluorescence in-situ hybridization) may be performed. After
re-inserting the slide in the scanning system, any rare cell can be relocated with a mouse
click, and the FISH result can be checked. The sequential correlation of different assays
provides a maximum of information in characterizing rare cells.
2.1.3.4 Speed Considerations for Rare Cell Detection
It is obvious that speed is essential in rare cell detection. For optimizing speed the main time-consuming steps in the scanning process need to be identified. Counterstain images as well as cell labels are usually pretty bright, so that exposure times will be fractions of a second per image. Stage displacement time (including the settling time necessary to wait for vibrations to fade out before capturing the next image) is typically 0.1 seconds. Image analysis does also not take longer than a fraction of a second and can be performed while the camera is already capturing the next field of view. Changing the filter block position takes approxi- mately 0.4 seconds. Obviously, the key to speed improvement lies in accelerating the filter change. With Metafer, this is achieved by using a dual band filter in combination with individual excitation filters mounted in a modified excitation filter wheel (Carl Zeiss). The original stepper motor of the filter wheel was replaced with a more powerful model which is driven by a proprietary controller board. This solution improved the switching time between DAPI and FITC excitation from about 0.4 seconds to 60 milliseconds, resulting in an overall scanning speed of two images per second (for DAPI plus an additional cell marker). The employed high resolution camera (1280 x 1024 ) with a pixel size of 6.7 micron results in a threefold increase of information per field of view as compared to a regular video type CCD with typically 752 x 572 pixels. Finally, it turned out that an optical magnification of 6.3 x is sufficient for rare cell detection. This can be achieved by using a 0.63 x camera adapter instead of the 1 x adapter in combination with the 10 x objective lens. The resulting scanning time per slide is 6 to 8 minutes for a complete slide. In monolayer Cytospin preparations with approximately 3 million cells per slide (scanned area 36 mm by 20 mm) this corresponds to approximately 7,000 scanned cells per second, which is one order of magnitude above scanning rates of previously reported image cytometers and comparable to flow cytometry.
2.2 Training
Easy adaptation and parameter optimization to different cell and label types is of crucial importance for making an automated system useful for routine analysis. To facilitate this task Metafer provides sophisticated training and optimization routines. The principle of optimization by training relies on a set of training images which have been preclassified by the operator.
In a first step, the training images are automatically
captured using slides that are representative for the type of specimens to be analyzed.
Next, the operator can display image by image and can mark positive cells by using the
mouse. Based on this a-priori information the training routine then
performs the optimization by systematically varying the parameters that are used for cell
classification within predefinable limits. Each set of parameters can be applied to the
pre-classified training images, pretending a normal search. The result of this test scan
is then compared with the manual pre-classification of the user. The parameter combination
is adapted iteration by iteration, until the best match between automatic and human
classification has been found.
The optimization can be applied to the metaphase detection in MSearch, the cell selection criteria as well as to the spot counting parameters in MetaCyte and to the cell detection and cell selection parameters in RCDetect. In MetaCyte, the operator will not only preclassify suitable cells to train the cell selection but will also interactively score for each hybridization channel the number of spots. Again, by automatically varying the spot count parameters like minimum/maximum spot size, minimum spot intensity, minimum distance required between two adjacent spots of the same color to interpret them as separate signals, the parameter set resulting in the best match between human and machine interpretation is determined.
