JPH0443311B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a method for dividing an image into regions using color, and is particularly suitable for dividing a blood cell image into regions of white blood cell nuclei, cytoplasm, red blood cells, and background. This paper relates to a region segmentation method.

"Prior art" The wavelengths of color separation for blood cell image input are as follows:
r: 560-600nm, g: 510-550nm, b: 400-
Although 430 nm is suitable, the conventional area division method (see Japanese Patent Application No. 185413/1987 for details) does not necessarily give satisfactory results for filter combinations in the above three wavelength ranges.

For example, when g': 550 nm and b': 430 nm are selected, even with the conventional region segmentation method, the white blood cell nucleus N, cytoplasm C, red blood cell RBC, and background BG are displayed as shown in Figure 1a.
It is possible to divide g: 525nm, b: 420nm.
In this case, as shown in Figure 1b, the white blood cell nucleus N,
This results in drawbacks such as the distribution of cytoplasmic C and red blood cell RBC becoming close to each other, making it difficult to divide them. Also r:
When it is 590 nm, g′, b′ two-dimensional plane and r, b2
Although the distributions on the dimensional plane are similar and the above-mentioned drawbacks are eliminated, there is a drawback that red blood cells RBCs and background BG cannot necessarily be separated with a threshold value T ^BG _g ′ corresponding to the threshold value T ^BG _g ′.

``Object of the Invention'' The object of the present invention is to provide a region segmentation method that eliminates the above-mentioned drawbacks and divides regions more accurately than conventional region segmentation methods based on r and g images that emphasize the color tone of the cytoplasm. There is a particular thing.

“Summary of the Invention” Concerning the contamination of red blood cells RBC into the background BG, T ^BG _g or
This can be solved by finding T ^BG _b (Figure 1b). By using this threshold together with the two thresholds T ^C _rb (Fig. 1b) obtained from the difference image between the r image and the b image, the white blood cell nucleus N, the white blood cell cytoplasm C, and the background BG It is possible to divide the red blood cells into three types and red blood cells RBC.

On the other hand, r after removing red blood cells RBC in this way,
In the density histogram of the g image, as shown in FIG. 2, the state of distribution corresponding to background BG, cytoplasm C, and nucleus N differs depending on the color tone and density of the cytoplasm. For example, in the case of eosinophils, in the concentration histogram of the g image, cytoplasm C is greatly different from the distribution of nucleus N, resulting in a bimodal histogram, but in the concentration histogram of r image, background BG, cytoplasm C, nucleus N is separated into three peaks, and the threshold that divides the three peaks can be easily determined. In this way, it is easier to find the threshold for dividing the above three groups from the density histogram of the r image, and it is easier and more accurate to find the threshold for white blood cells, which allows for more accurate region segmentation. There are different types of white blood cells that can be divided into regions with high precision.

Therefore, based on the above two facts, we devised a new area division method as shown below.

"Embodiment of the Invention" An embodiment of the present invention will be described below with reference to FIG. In this example, the smeared blood specimens are passed through a green filter (main wavelength around 525 nm), a blue filter (around 420 nm), and a red filter (around 590 nm) to a photoelectric conversion device (image pickup tube, etc.). The converted density signals, g image, b image, and r image, are handled. The blood image of the smeared blood sample is represented in a three-dimensional density space with the densities of the g, b, and r images as axes.
The density signals of each pixel of the g image, b image, and r image are stored in an image memory described later, and based on these density signals, the following method is used to
The area is divided into four areas: nucleus N, cytoplasm C, background BG, and red blood cell RBG of white blood cells, which are components of a blood cell image.

1. Find the minimum value min (g) in the g image, and calculate the minimum value min (g) and the preset upper limit value UL of the image as "1. From the signal, consider the density space for the b image and the g image as shown in Figure 3a, that is, consider the secondary space of the three-dimensional space mentioned above.In Figure 3a, the axis b and The axis g indicates the density of the b image and the g image, respectively, and in FIGS. 3b to 3g, the vertical and horizontal axes similarly mean the density of each image.
Find the minimum density min(g) in the image, and set the fixed ratio value of the minimum density min(g) and the preset maximum density UL of the g image (for example, 17
%) _is ^calculated . (3rd
Figure a) However, in the case of white blood cells with a low cytoplasmic concentration in the g image, the cytoplasmic C is in the area below the threshold T ^BG _g .
Some of them may be mixed in. (Fig. 1 b) 2. Find the density histogram H ^BG _b in the b image of the pixels existing in the area below the threshold value T ^BG _g in the g image, and calculate the center value of the histogram by the half width of this histogram H ^BG _b . The more concentrated value, the background
Set the threshold value T ^BG _g to divide BG and red blood cell RBC. (Figure 3b) In this case, cytoplasmic C
Even if a low-density part of BG is mixed into the background BG, the density of both images is almost the same even in the b image, so it will not cause _a change in ^TBGg .

3 r-b image of pixels existing in the area above the threshold T ^BG _g in the g image (difference image between the r image and the b image)
The density histogram H ^BG _rb is obtained on the axis marked as r-b, which is at 45° with the b-axis and the r-axis, as shown in FIG. From this histogram H ^BG _rb , a threshold value T ^C _rb for dividing white blood cell nucleus N, cytoplasm C and red blood cell RBC, and a threshold value T ^N _rb for dividing white blood cell nucleus N and red blood cell RBC are set. In addition, Fig. 3c shows the three-dimensional density space described earlier in a plane (two-dimensional )
It is expressed in the concentration space of . The same applies to FIGS. 3d to 3g.

4. After binarizing the b image and the r-b image described in 3., the area of red blood cells is determined by performing the logical OR of these images. This can be achieved, for example, as follows. For each pixel of image b, if the density signal is less than or equal to the threshold T ^BG _b , that pixel is set to binary 1, and if the density signal exceeds the threshold T ^BG _b , that pixel is set to 2. The b image is binarized by setting the base number to 0, and for each pixel of the r-b image described in 3., if the density signal is equal to or higher than the threshold value T ^C _rb , that pixel is set to the binary number. 1, and when the density signal is less than the threshold value T ^C _rb , the pixel is set to binary 0 and the r-b image is binarized, and all of the two binarized images are When the values set for the pixels are logically ORed between the corresponding pixels, the pixel that becomes 0 in binary becomes the pixel in the red blood cell RBG area, and r
The red blood cell RBC region can be extracted and removed from the image and g-image. In this way, red blood cells
Obtain the density histograms H ^RBC _r and F ^RBC _g in the r image and g image of the image from which RBCs have been removed. Thresholds T ^N _r , T ^N _g that divide the nucleus N and cytoplasm C of , cytoplasm C and background
Find thresholds T ^C _r and T ^C _g for dividing BG. Third
Figures d, f show different concentration histograms, d
shows an example in which region segmentation is more stable when the threshold value is determined using the r image, and f shows an example in which region segmentation is more stable when the threshold value is determined using the g image.

Here, as a method for selecting r, g images, the density histogram H ^RBG _i (i=r, g) is
It is classified into six types f, and the six types are prioritized in order from a to f. That is, if the density histogram of the r image is 30 type and the density histogram of the g image is 20 type, the r image is selected.
The eosinophil whose cytoplasm is dark orange in FIG. 2 corresponds to this example. Further, if the density histogram type of the r image and the density histogram type of the g image are the same, selection conditions are provided for each type. For example, in the case of type 30, the positional relationship between the peak position of the central distribution (cytoplasm C) and the peak position of the left and right distributions (background BG/nucleus N), and the peak value of the central distribution and the nuclei at both ends. Selection is made based on two evaluation variables: the ratio of the sum of values at LV and UV. For example, in the case of a neutrophil whose cytoplasm is light orange in FIG. 2, the g image is selected based on the above two evaluation variables.

Here, type classification of density histogram and
For details on how to determine T ^N _rg , T ^C _rg , T ^N _r , T ^C _r , T ^N _g and T ^C _g , please refer to the specification of Japanese Patent Application No. 1983-9438.

According to this embodiment, a certain leukocyte can be divided into four regions on the r, b two-dimensional plane as shown in FIG. As shown in Figure 3g, the threshold values T ^N _rb and T ^C _rb of the r-b image are set to g, b2
The division into four regions is accomplished by mapping onto a dimensional plane. (FIGS. 3e and 3g) The present invention can be realized using a computer program, but it is more practical to process images using a dedicated circuit and process histograms using a computer. FIG. 5 is a diagram showing an example of the configuration of an apparatus that implements the method of the present invention.

In the figure, 1 is a g-image image memory that stores a density signal converted from a smear-stained blood specimen by a photoelectric conversion device (image pickup tube, etc.) via a green filter (main wavelength around 525 nm); 3 is an image memory of the b image that stores the concentration signal obtained by converting the above blood sample through a blue filter (main wavelength around 420 nm), and 3 stores the concentration signal obtained by converting the above blood sample through a red filter (main wavelength around 590 nm). This is an image memory for r images that stores density signals.

Read the g image from the image memory 1, find the minimum density min(g) of the g image using the minimum value detection circuit 4,
Input this into the computer (for example, μ computer) 20, and use the above minimum density min(g) and the predetermined maximum density UL of the g image to calculate T ^BG _g =
The threshold value ^TBG _g is determined by calculating (UL-min(g))×ω+min(g). However, ω is a constant,
For example, it is 0.17. Next, the computer sets the threshold T ₁ of the threshold circuit 9 to the threshold T ^BG/g obtained by the above processing using the signal from the logic circuit 11.
The logic of T ₁ is determined by the control signal d from the computer 20.
Set as below. The switching circuit 12 is set to use the output of the logic circuit 11 as a control signal for the histogram creation circuit 13 by the control signal E from the computer 20, and the selection circuit 6 is set to use the b image of the image memory 2 by the control signal A from the computer 20. In addition, the selection circuit 7 is set to select the g image in the image memory 1 in accordance with the control signal RO from the computer 20. In the following processing as well, the settings of each circuit are performed using control signals from the computer 20. When g and b images are simultaneously read out from the image memories 1 and 2, a density histogram H ^BG _b of the b image in a region including the background BG and part of the cytoplasm C is determined as the output H of the histogram creation circuit 13. This density histogram H ^BG _b is input to the calculator 20 and a half-width calculation is performed to obtain a threshold value T ^BG _b .

Next, set the theory of logic circuit 11 to T ₁ or more,
The selection circuit 6 is set to select the output of the arithmetic circuit 5. Do not change the settings of other circuits. Then, when g, b, and r images are simultaneously read out from the image memories 1, 2, and 3, a density histogram H ^BG _rb of the r-b image in the area excluding the background BG is obtained as the output H of the histogram creation circuit 13. This density histogram H ^BG _rb is input into the calculator 20, and the density values of valleys with low density are determined as threshold T ^C _rb , and the value obtained by subtracting a constant value (for example, 10) from the density value of valleys with high density is T ^N _{rb .} shall be.

Next, set the threshold T ₁ of the threshold circuit 9 to the threshold
T ^C _rb , the threshold value T ₂ of the threshold circuit 10 is the threshold value
T ^BG _b is set, and the logic of the logic circuit 11 is set to the logical sum of T ₁ or more and T _{2 or} more. The selection circuit 6 selects the r image in the image memory 3, the selection circuit 7 selects the output of the subtraction circuit 5, and the selection circuit 8 selects the b image in the image memory 2 according to the control signal c from the computer 20. Set to select. Settings of other circuits cannot be changed. Then, when the b image and the r image are simultaneously read out from the image memories 2 and 3, a density histogram H ^RBC _r in the area excluding the red blood cells RBC is obtained as the output H of the histogram creation circuit. This density histogram H ^RBC _r is input into the computer 20 and stored.

Next, the selection circuit 6 is set to select the g image in the image memory 1, and the settings of other circuits are not changed. Then, from image memory 1, 2, 3, g image, b image, r
When reading images simultaneously, histogram creation circuit 1
As the output H of 3, the concentration histogram H ^RBC _g in the area excluding red blood cells RBC is determined. This density histogram H ^RBC _g is input into the calculator 20, and the type determination of the memorized H ^RBC _r and the above ^{H RBC} _g is performed _.
H ^RBC _g Determine the threshold value for dividing the nucleus N, cytoplasm C, and background BG, and determine the concentration value of the valley.
Find T ^N _r , T ^C _r or T ^N _g , T ^C _g .

For example, to divide the region of the nucleus N based on the above thresholds, if H ^RBC _r is selected as the concentration histogram for threshold calculation, the threshold value T of the threshold circuit 9 is Set the threshold T ^N _r as ₁ ,
Threshold value as threshold value _T2 of threshold value circuit 10
Set T ^N _rb and set the logic of logic circuit 11 to T ₁ or more, T ₂
Set with the logical product of the above. The selection circuit 7 selects the r image in the image memory 3, and the selection circuit 8 selects the r image in the image memory 3, and the selection circuit 8 selects the r image in the image memory 3.
Set to select the output of Then, the area of the nucleus N can be obtained by simultaneously reading out the b image and the r image from the image memories 2 and 3 and counting the output of the logic circuit 11 through the switching circuit 12.

The same applies to the region including the nucleus N and cytoplasm C, and the background BG region.

In the above embodiment, the area division method is implemented using two systems of threshold circuits, a histogram creation circuit, and a computer, but partially parallel processing is possible by providing two systems of histogram creation circuits, for example.

"Effects of the Invention" According to the present invention, the threshold value that separates the components of a blood image, the nucleus N of white blood cells, the cytoplasm C, the background BG, and the red blood cell RBC, is set according to the distribution in the r, g, and b three-dimensional color space. Therefore, the region division of nuclear components can be performed accurately.

[Brief explanation of the drawing]

Figures 1a and b are diagrams for explaining the principle of the conventional method of dividing blood cell images into regions, and Figure 2
The figures are diagrams to explain that the concentration histogram differs depending on the color tone and density of the cytoplasm, Figures 3 a to g.
4 is a diagram showing the procedure of the area dividing method according to the present invention, FIG. 4 is a diagram for explaining the types of histograms, and FIG. . 1, 2, 3... image memory, 4... minimum detection circuit, 5... subtraction circuit, 6, 7, 8... selection circuit,
9, 10...threshold circuit, 11...logic circuit,
12...Switching circuit, 13...Histogram creation circuit, 20...Calculator.

Claims (1)

[Claims] 1 Input a blood image containing a mixture of white blood cells and red blood cells by wavelength of red component, green component, and blue component to obtain r images, g pixels, and b pixels each divided into pixels,
In a region dividing method of dividing each pixel into four regions, ie, the nucleus and cytoplasm of the white blood cell, the red blood cell, and the background, in the density space regarding the b image and the g image, the g image determining a first threshold value for dividing the background and the red blood cell region on the density axis of the b image from a density histogram of blue components for pixels having a certain density or less obtained from the minimum value of the pixels of the g image; For pixels whose density is higher than a certain density determined from the minimum value of the g image, a density histogram of the difference image between the r image and the b image is calculated, and the nucleus and cytoplasm of the white blood cell are calculated on the density axis of the difference image. and determining a second threshold for dividing the red blood cells, binarizing the density of each pixel of the b image based on the first threshold, and dividing each pixel of the difference image into a binary value. The difference image is binarized based on the density of the second threshold value, and the logical OR is performed between the corresponding pixels of these two binarized images with respect to the values set for those pixels. determining the area of the red blood cells, and obtaining an image in which the red blood cell areas are removed from the r image and the g image, a concentration histogram of red components in the r image with the red blood cells removed, and a g image with the red blood cells removed. determining the density histograms of the green components respectively; comparing the density histograms of the red component and the density histogram of the green component and selecting one of the density histograms in which the three peaks appear more clearly; and determining the density histogram of the selected density histogram. A threshold value that separates the nucleus and cytoplasm of the white blood cell, and a threshold value that separates the cytoplasm of the white blood cell and the background are determined from
The r-image or g-image from which the red blood cell region has been removed is converted into a two-dimensional density space of one of the red component density or the green component density and the blue component density corresponding to the selected density histogram. A region dividing method comprising dividing the region into each region of the nucleus of the white blood cell, the cytoplasm of the white blood cell, and the background.