JP4944641B2 - Automatic detection of positive cells in stained tissue specimens - Google Patents

Description

  The present invention relates to a method for objectively quantifying positive cells in stained tissue specimens in the fields of histology and pathology.

In the fields of histology and pathology, staining tissue specimens by immunological means or hybridization, etc. and quantitatively analyzing the positive cells are a means of research in these areas and a means of pathological diagnosis. Very important and widely implemented. For the quantification of the number of positive cells, a method of counting the area set by the experimenter himself from the image is used (Non-Patent Documents 1 and 2).
Perrotti et al, J. MoI. Neuroscience, 24 (47): 10594-10602 (2005). Lai et al, J.A. Neuroscience, 25 (49): 11239-11247 (2005).

However, in the conventional quantification method, only the area set by the experimenter himself is quantified, so it is not possible to obtain the data of the place where the experimenter is not paying attention. The obtained data is only the count data and is within the set area. There was a problem that the information of the space was lost.
Therefore, the present inventor, in quantitative analysis of the number of positive cells in a stained tissue sample, combines detection based on cell size in addition to detection of a region stained above a certain threshold in the image processing step. Region detection is performed by detecting positive staining cells, pseudo-coloring positive cell images and detecting them as positive cell density, standardizing the colored images, and creating a map with average values using multiple specimens Has been found that it is possible to visualize quantitative data while maintaining spatial information by standardization and creation of an average value map, and filed earlier (Japanese Patent Application No. 2006-13465). By this method, the sample area setting and quantification can be automatically performed by a computer, and the data of a plurality of samples can be analyzed simultaneously, so that it is possible to visualize not only single-point data but also spatial data.
However, in the region where the cell density is high (> 800 / mm ² ), the overlap of the cells causes a measurement error, and the staining positive cells cannot be detected accurately. Counts were unreliable (Wada et al., Neuroscience Research, 56: 96-102 2006).

  Therefore, an object of the present invention is to provide a new method for quantitatively analyzing the number of staining positive cells even in an area where cells are densely present.

  The present inventor has examined various methods for detecting staining positive cells in an image obtained by imaging a stained tissue specimen, and has set a threshold value on the lower concentration (high luminance) side than the highest concentration (low luminance) of the captured image. Then, by gradually changing the detection threshold from the high concentration side to the low concentration side up to the certain threshold value, only the positive cell image detected for the first time at each threshold is detected as a positive cell image, thereby overlapping the cells. It was found that positive cells can be detected with extremely high sensitivity as compared with the conventional method even in a high-density region such as that shown in FIG.

That is, the present invention is a method of detecting a stained positive cell by imaging a stained tissue specimen and processing the obtained image by a computer, and (1) a density having the highest luminance distribution in the image is determined as a background density. (2) A threshold value is set on the lower density side than the highest density of the image, and the detection threshold value is gradually changed from the higher density side to the lower density side until the predetermined threshold value. Selecting only positive cell images detected for the first time at each threshold as positive cell images, the step (2) comprising: (a) detecting a region stained above the detection threshold; and (b) detecting A step of selecting only a portion where cells of a certain size are stained as positive cell images, and (c) a step of recording the number of detected positive cell images and the coordinates of the center of gravity. Of stained positive tissue cells characterized by It is to provide a method out.
In addition, the present invention adds the counts of positive cell images at each threshold selected in the step (2), and adopts the positive value of the stained tissue sample to be used as the count value of positive cells in the imaged tissue sample. A method for counting cells is provided.

  According to the present invention, staining-positive cells can be automatically detected with high accuracy even in a region where cells are densely present. In addition, according to the present invention, it is possible to count the number of positive cells in the whole tissue and to easily perform statistical comparison between a plurality of groups, so that tissue changes in regions not intended by the experimenter can be clearly and objectively performed. I can observe. Furthermore, objective diagnosis is possible by applying this method also in the pathological diagnosis site. In addition, the method of the present invention can be applied to quantification in particle analysis in cell counting, nuclear medicine or the like using stereology.

  In the present invention, first, a stained tissue specimen is imaged, and the obtained image is input to a computer. Here, as the tissue specimen, a tissue specimen collected from biological tissues such as animals including humans and plants is used. For example, a frozen section of a whole organ image, a section of tissue removed by surgery, and the like can be mentioned. Examples of staining means include immunostaining, in situ hybridization, nuclear staining, and the like for a target in which only cell nuclei and cell bodies are stained. For imaging, a microscope and an imaging device are preferably used. As the imaging device, for example, a camera capable of taking a digital image such as a color CCD camera is used. An image obtained by the imaging device is sent to a computer converted into image data of a digital signal.

  Image processing by a computer is performed by an input device, a display device, and a computer.

  The input device is a device for receiving an instruction input related to the implementation of the method of the present invention, inputting data including various characters and symbols, and the like. Specifically, it is configured by a combination of devices such as a keyboard, a mouse, a touch panel, and a voice input device that can be used for inputting the above-described instructions and data.

  The display device is for displaying an acquired image, a measurement result, a colored image, and the like in addition to a menu screen, an operation screen, an instruction screen, and the like. Specifically, a device capable of displaying an image using a flat panel display such as liquid crystal or plasma, or a display tube such as a CRT is used. In addition, a slide projector or the like for displaying an enlarged projection can be connected.

  The computer has a central processing unit (CPU), a memory, and an auxiliary storage device. The auxiliary storage device stores a program group executed by the CPU, various data, and the like.

  The following image processing is all performed on a computer.

First, a luminance distribution is examined from an image obtained by imaging a stained tissue specimen (FIG. 1). (1) A density having the highest luminance distribution is detected as a background density, and this is detected as a maximum value (for example, 8 bits). In this case, standardization of 255) is performed (FIG. 2). This operation can be performed, for example, by expressing an image as matrix data on the MATLAB software and adding arithmetic processing, thereby making it possible to match cell detection conditions.
At this time, when noise such as background dyeing unevenness is recognized in the image, it is preferably removed in advance using a known image processing software (NIH-image) or the like.

Next, (2) a threshold value is set on the lower density side (high luminance side) than the highest density (luminance 0/255) of the image, and the detection threshold value is gradually increased from the higher density side to the lower density side until the predetermined threshold value. And positive cells are detected at each threshold. The detection threshold is set to an interval of about 1 to 20 before reaching a certain threshold, that is, “0, 20, 40, 60, 80, 100, 120” (20 intervals). The set number can be determined as appropriate depending on the condition of the intercept, the balance between the processing speed of the computer and the detection accuracy.
Here, the constant threshold value (final detection threshold value) may be a threshold value that is normally determined based on the experience of the experimenter, but about half to about 2/3 based on the background value (120 to 160 when the background value is 255). It is preferable to set the concentration to the extent of approximately.
Detection of positive cells at each threshold is performed as follows.

  From the image, (a) a region stained above the detection threshold is detected. This operation can be realized, for example, by performing an operation of selecting data of a certain numerical value or more from an image expressed as matrix data on the MATLAB software.

  At this time, in order to prevent contamination of the sample with noise such as dust in the specimen and scratches on the section, select only the portion that is suitable for positive cells (b) stained with a certain size of cells as positive cells. . Only this part is determined as a positive image. Here, selection of cells having a certain size can be determined by the size of the cell size or the size of the cell nucleus. This operation can be realized, for example, by using a library function that calculates the area of the selected region included in the image processing toolbox that can be added to the MATLAB software. Only areas within a certain range among the areas of the respective areas calculated in this way can be selected as positive cell images.

  Next, (c) the number of detected positive cell images and the coordinates of the center of gravity are recorded. This operation can be realized, for example, by preparing an empty matrix having the same size as the image on the MATLAB software, and substituting numerical values only for locations that coincide with the coordinates of the center of gravity and recording them as variables.

  The same operation is performed at the next detection threshold value, and positive cells at the threshold value are detected. At that time, the positive region including the barycentric coordinates of the positive cell image already detected and recorded is omitted, and only the positive cell image detected for the first time at the threshold is selected as the positive cell image. Thereby, even if it is a high-density area | region where cells overlap, a positive cell can be detected correctly.

As described above, the sum of the count values of the positive cell images detected at the respective threshold values is calculated and adopted as the count value in that region.
By performing analysis based on multiple concentration thresholds, positive cells can be automatically detected with high accuracy even in thin images and areas with high cell density compared to conventional detection methods based on binarization. A positive cell count can be performed for the entire tissue, or a group comparison can be performed by adding data of a plurality of individuals, and subtle changes in the number of cells can be detected. In addition, the positive cells detected by the detection method of the present invention can be visualized and analyzed by the method described in Japanese Patent Application No. 2006-13465, or the method of the present invention can be used for cell counting using stereology, particle analysis in nuclear medicine, etc. It can also be applied to quantification and the like.

When the staining positive cells detected using the method of the present invention are visualized and analyzed by the method described in Japanese Patent Application No. 2006-13465, the following is performed.
That is, (3) the image is divided into grid-like pixels, and the positive cell density in each pixel is measured (FIG. 7A). This counting of positive cell density is performed automatically by counting software. For example, in the case of the illustrated example, the size of each pixel can be divided into nine equal parts by a 200 μm block. This operation can be realized by, for example, a library function group included in an image processing toolbox that can be added to the MATLAB software. Convert each area to be represented by one point using a function that calculates the center of gravity of the selected area, and obtain an average value in the block using a function that can perform arbitrary numerical operations on this converted data for each block. To calculate the positive cell density.

  (4) Each pixel is given a pseudo color corresponding to the positive cell density (FIG. 7B) to obtain a colored image of the entire tissue specimen (FIG. 8). Thereby, the area | region with a high positive cell density can be screened in the whole tissue specimen, and the image which visualized it by coloring is obtained. This operation can be realized by, for example, a function of displaying image data included in the MATLAB software with an arbitrary color map.

  However, the image obtained in the step (4) is the result of one slice per individual. To obtain the overall trend, it is necessary to average data obtained from a plurality of sections and a plurality of individuals. However, since the shape of the tissue sample is slightly different depending on the individual and conditions, it cannot be overlaid with the images of a plurality of samples as it is. Therefore, (5) a step of standardizing the shape of the entire tissue specimen in accordance with the known tissue specimen shape is required. This standardization is based on known tissue specimen shape data, for example, known brain map data (Non-Patent Document 2), by rotating, enlarging, reducing, etc. the colored image of (4) above. Standardization is performed (FIG. 9). By this standardization, the colored image obtained in (4) can be overlaid with the image of another sample to create an average value map. This operation can be realized by a library function group included in an image processing toolbox that can be added to, for example, MATLAB software. An entire section image is selected by a function for selecting an arbitrary region, and a major axis, a short axis, and a rotation angle are calculated by a function for extracting features by approximating the slice. Standardized data can be obtained by conversion using a function that performs image manipulation based on this value. This information is expressed as matrix data on the MATLAB software, and an average value map can be calculated from each map in the form of matrix operation.

  (6) For the tissue specimens derived from a plurality of individuals, the operations (1) to (5) are performed, and an average value map of colored images for the obtained plurality of tissue specimens is obtained. This can be done by superimposing colored images of tissue specimens derived from a plurality of individuals, obtaining an average value for each point, and mapping it (FIGS. 10A and 10B).

  Further, (7) comparison between groups is possible by performing the operations (1) to (6) and comparing the groups of individuals with a plurality of conditions. FIGS. 10A and 10B are colored images for groups in which different experimental operations were performed. 10 (A) and 10 (B), it can be seen that there is a region on the ventral side where the positive cell density is greatly different. The positive cell density is pseudo-colored, but is quantified and can be quantitatively analyzed. Here, the number of tissue samples is preferably a number that can be statistically significant, for example, 5 or more.

  Furthermore, if the result of comparison between groups is digitized as a statistical parameter and a pseudo color is assigned to each pixel, it is possible to visualize only a portion where there is a significant difference in the significant difference test. For example, FIG. 11 is a diagram in which a portion having a significant difference (t-test) is pseudo-colored based on the data in FIG. 10B. According to FIG. 11, it can be seen that only the lower right portion is a portion where positive cells are significantly present. This operation can be realized by, for example, a function for performing a t-test included in the MATLAB software. Using a function for converting the calculated statistic into a logarithmic value, it is possible to display the calculation result visually by the method described in the step (4).

  EXAMPLES Next, an Example is given and this invention is demonstrated still in detail. Note that MATLAB software was used as a means for expressing image analysis in programming.

Example 1
(1) As shown in FIG. 3A, four positive cells are visible under the straight line (yellow) in the figure. However, it is extremely difficult to separate and detect these four cells by the conventional detection method using binarization. If the threshold value is set to the high density side (luminance 0 side), the cell image stained at a relatively low density shown on the left of the graph cannot be detected. On the other hand, if the threshold value is set to the low density side, the graph right The dense cell images shown in Fig. 1 are fused into one and cannot be detected individually. In the present invention, detection accuracy is improved by detecting positive cells at a plurality of threshold values set from the high concentration side.
(2) The present invention was applied to a tissue specimen (mouse brain) that had been immunostained with a neuron-specific labeled protein NeuN, and automatic detection of nerve cell nuclei was performed. Specifically, the background value detected from the captured image is standardized as the maximum luminance (255), the final detection threshold is set to 140, and positive cells are detected at the thresholds 20, 40, 120, and 140, respectively (FIG. 3). -B). As Comparative Example 1, positive cells were detected only at the threshold 120.

  When the detected positive cell images were counted, it was 41 in Comparative Example 1, which was about 50% of the counted value by visual observation (left of FIG. 4), but 66 according to the detection method of the present invention. Yes, it was possible to detect about 85% when visually observed (right side of FIG. 4).

Example 2
(1) The present invention was applied to a tissue specimen (mouse brain) similar to that in Example 1, and automatic detection of nerve cell nuclei was performed. Specifically, with respect to the standardized background value 255, the final detection threshold is set to 160, and positive cell images were detected with the thresholds 0, 20, 40, 60, 80, 100, 120, 140, and 160, respectively. . For each threshold, an image having an area of 2 pixels or more was determined as a positive cell image. Further, as Comparative Example 2, positive cell images were detected at only one point of the threshold 120 or the threshold 160.
The results are shown in FIGS. In the figure, red circles indicate positive cell images detected by the detection method of the present invention, and blue circles indicate positive cell images detected by conventional binarization (Comparative Example 2). As is clear from FIG. 5, the cell image can be separated by the detection method of the present invention even in a thin image or a region where cells are fused and become one, and accurate positive cell image count is possible. It was.

(2) Further, positive cell images are counted by visual observation (counting at a resolution conceivable by a × 20 objective lens) in a 100 × 100 μm region arbitrarily extracted from the section image, and the automatic detection method (threshold value 0 to 160) and The detection method was compared with conventional binarization (threshold 120 or 160). As a result, as shown in FIG. 5C, in the conventional detection method based on binarization, when the density of the cells exceeds 20 cells, the accuracy is greatly reduced as compared with visual observation. In this case, there is a strong correlation with the visual result, and as shown by the regression line, an average of about 90% of the visual value was secured even in a densely packed region. In the embodiment, analysis was performed with a relatively low resolution (1 pixel = 2.5 μm corresponding to objective × 4), but the resolution of the analysis image can be arbitrarily set.

(3) An image of a region where cells are densely packed in the NeuN-stained section (region indicated by black triangles in FIG. 6A; cortical layer IV) (1 pixel = 2.5 μm) is taken, and the automatic detection method ( Staining positive cells were detected by the threshold value 0-160) and the conventional binarization detection method (threshold value 120).
By applying the method described in Japanese Patent Application No. 2006-13465, the captured image was divided into 100 × 100 μm regions (40 × 40 pixels), and the positive cell density in each region was measured. The measurement result was displayed as a positive cell density map as a pseudo color in which a region having a high positive cell density was red to yellow.
FIG. 6B shows a positive cell density map detected and created by the conventional binarization detection method, and FIG. 6C shows a positive cell density map detected and created by the detection method of the present invention. . In the conventional binarization detection method, the region indicated by the black triangle could not be correctly recognized although the cell density was high (FIG. 6B). On the other hand, according to the detection method of the present invention, individual cells could be detected even in a high-density region that could not be detected well by the conventional method (FIG. 6C). As a result, it became possible to show the difference in nerve cell density due to the layer structure of the cerebral cortex.

It is a figure which shows the dyeing | staining positive cell image of the tissue specimen (rat brain) which performed the immuno-staining with respect to the labeled protein NeuN specific to a nerve cell. Arrows in the graph indicate measurement errors in a region with a high cell density. FIG. 4 is a diagram showing an image (left) obtained by imaging a stained tissue specimen and an image (right) that is automatically detected and standardized as a background density at a density having the highest luminance distribution among the images. It is a figure which shows the conceptual diagram of the positive cell detection method of this invention. It is the figure which detected the neuron nucleus (NeuN) automatically in each detection threshold value. It is a figure which shows the result of having counted the positive cell image (left) detected by the detection method by the conventional binarization, and the positive cell image (right) detected by the detection method of this invention. (A) It is a figure which shows the result (red circle) which counted the positive cell image detected by the threshold value 120 (blue circle), and the positive cell image detected by the detection method of this invention. (B) It is a figure which shows the result (red circle) which counted the result (red circle) which counted the positive cell image detected with the threshold value 160, and the positive cell image detected by the detection method of this invention. (C) A graph in which the count value of the positive cell image detected by the conventional binarization detection method and the count value of the positive cell image detected by the detection method of the present invention are respectively compared with the count value by visual observation. It is. (A) It is a figure which shows the image of a NeuN dyeing | staining section | slice. (B) It is a figure which shows the positive cell density map detected and produced by the detection method by the conventional binarization. (C) It is a figure which shows the positive cell density map detected and produced by the detection method of this invention. (A) It is a figure which shows an example of the result of having divided the image into pixels and measuring the positive cell density in the pixels. (B) It is a figure which shows an example of the state which attached | subjected the pseudo color according to the iba-1 positive cell density. It is a figure which shows an example of the state by which the area | region with a high iba-1 positive cell density was visualized about the whole section. It is a figure which shows an example of the image which performed standardization of data. (A) It is a figure which shows an example of the average value map of the coloring image of the tissue sample derived from several individuals (control group). (B) It is a figure which shows an example of the average value map of the coloring image of the tissue sample derived from several individuals (drug administration group). It is a figure which shows an example of the image pseudo-colored based on P value by t-test (P <0.01).

Claims (5)

A method for detecting stained positive cells by imaging a stained tissue specimen and processing the obtained image by a computer. (1) Standardization is performed by detecting a density having the highest luminance distribution as a background density. (2) A final detection threshold is set on the lower density side than the highest density of the standardized image, and the detection threshold is gradually changed from the higher density side to the lower density side until the final detection threshold, Selecting only positive cell images detected for the first time at each threshold as positive cell images, the step (2) comprising: (a) detecting a region stained above the detection threshold; and (b) detecting A step of selecting only a portion where cells of a certain size are stained as positive cell images, and (c) a step of recording the number of detected positive cell images and the coordinates of the center of gravity. Stained tissue specimen characterized by The method of detection.   A method of counting positive cells in a stained tissue sample, which is used as a count value of positive cells in an imaged tissue sample by summing the counts of positive cell images at each threshold selected in the step (2).   A method for visualizing and analyzing stained positive cells detected by the method according to claim 1, wherein (3) the obtained image is divided into grid-like pixels, and the density of positive cells in each pixel is measured. A step, (4) attaching a pseudo color according to the positive cell density, obtaining a colored image of the entire tissue specimen, (5) performing a standardization according to a known tissue specimen shape, and (6) A positive cell of a stained tissue sample, comprising the steps of (1) to (5) for a tissue sample derived from a plurality of individuals to obtain an average value map of colored images for the plurality of tissue samples. Visualization analysis method.   Further, (7) performing the operations (1) to (6) on a group of individuals under a plurality of conditions, and comparing the groups, the positive cells of the stained tissue sample according to claim 3 Visualization analysis method.   5. The method for visualizing and analyzing positive cells of a stained tissue sample according to claim 4, further comprising the step of digitizing a result of comparison between groups as a statistical parameter and adding a pseudo color to each pixel.