JPH02232550A - Tissue assay by image analysis - Google Patents

Abstract

PURPOSE:To enable assay of a tissue quickly and accurately by measuring a section of a coupling substance having a plurality of tissues with an image analyzer to prepare a histogram of a distribution area corresponding to a number of density stages of an image. CONSTITUTION:A histogram of a density distribution area is prepared according to a preset density stage for an image of a section of a coupling substance having a plurality of tissues. A density range is determined corresponding to subject tissues for the hitogram to determine a composition of the subject tissues in terms of an area ratio of the density range. Here, estimate thresholds for the density range corresponding to the respective subject tissues are set as values having a specified allowable range. Then, a defined threshold is determined based on at least one or more of a geometry of the histogram centered on the allowable range thereof and a geometry of the density distribution of tissues to obtain an area ratio of the density ranges. The defined threshold may be, for example, a minimum value within the allowable range of the histogram subjected to a noise removal processing or a position of a center of gravity of an area within the allowable range of the histogram subjected to a noise removal processing.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a tissue quantification method using image analysis, and in particular, a method for determining the boundaries of a gray-scale image of a bonded substance having a plurality of tissues and adjusting the boundaries accordingly. This paper relates to a tissue quantification method for quantifying tissues by determining the area ratio of each tissue. [Prior Art] A technique for measuring the tissue composition ratio of a lumpy substance having a plurality of msa is disclosed in JP-A-58-153144. This is the distribution of reflectance of multiple tissues with different reflectances, with the strength of the reflectance as the x-axis and the frequency of each strength as the y-axis.
Obtain a reflectance histogram with the axis as the axis, obtain multiple Gaussian curves with peaks at multiple maximum points on the envelope of this histogram, and calculate the average reflectance and power of each of the multiple Gaussian curves obtained in this way. Based on the integral value and the reflectance distribution of a plurality of plain weaves obtained in advance for the sample to be measured, the composition ratio in each of the plurality of parts of the sample is determined. [Problems to be Solved by the Invention] The above-mentioned conventional technology calculates a histogram of reflectance of a sample, then calculates the values of multiple peaks of this envelope line, and then calculates the shape of the histogram around each peak value. is converted into a Gaussian curve (normal distribution curve) and used as the reflection distribution. Therefore, it is good if the tissue is distributed evenly on the left and right around the peak value, but the error will increase if the area ratio of tissues that are unevenly distributed on the left and right is identified. In addition, & [If there is a missing tissue in the tissue, an unreasonable deformed Gaussian curve may be applied to the missing part by mistake, giving an incorrect threshold level. The purpose of the present invention is to create a histogram of the density distribution area of a cross-sectional image of a connective substance having multiple tissues according to preset density levels, and to create a histogram of the density distribution area for each target &
The object of this invention is to provide a method for setting an estimated threshold for a concentration range corresponding to the 11 weave, and then determining a defined threshold based on the shape of this histogram, the shape of the tissue concentration distribution, etc. [Means for Solving the Problem] In order to achieve the above object, &[
The i-texture quantification method involves creating a histogram of the area of shading distribution according to preset shading levels for an image of a cross section of a binding substance that has multiple tissues, and then
f Image analysis that determines the concentration range corresponding to the tissue and determines the composition of each target tissue from the area ratio of the concentration range
In the tissue quantification method, the estimated threshold value of the concentration range corresponding to each target msa is set as a value having a predetermined tolerance range, and the shape of the histogram centered around the tolerance range and the shape of the concentration distribution of the tissue are determined. A special feature of this method is to determine the area ratio of the concentration range by determining a definite threshold value for at least one or more values. Further, the minimum value within the tolerance range of the histogram subjected to noise elimination processing may be used as the final threshold, or the position of the center of gravity of the area within the tolerance range of the histogram subjected to noise elimination processing may be used as the final threshold. good. Further, a position that bisects the area of the histogram within the permissible range may be set as the determined threshold. Further, after the histogram is subjected to noise elimination processing, the histogram is first changed from the center of the tolerance range to the left and right on the X-axis, with the gradation level of the histogram in the X-axis direction and the frequency of the gradation level in the y-axis direction. It is preferable to draw a tangent line at the curved point, and set the X coordinate of the intersection of the tangent line to the determined threshold value, and further define the position where each of the tangent lines intersects with the X axis as Xl+X! When lxzx+l is equal to or greater than a predetermined value, the final threshold may be set to two values, XI and x2. [Function] A cross-section of a binding substance that has multiple tissues is measured using an image analysis device, a histogram of the distribution area corresponding to the many gray levels of the image is created, and the area ratio of each target tissue is used to calculate the density of each tissue. The method for identifying this will be explained for the case of a binding substance consisting of tissues A and B. Even if two tissues have the same area, the basic concentration distribution of each tissue differs depending on the method of illuminating the target material (light type, angle, brightness, etc.), the magnification of the microscope, etc. Depending on the magnification of the microscope. In this case, spots, striped patterns, snowpack patterns, etc. appear and disappear.There are also various forms of concentration distribution within the same tissue.This will be explained with reference to FIG.

[a] The figure shows a case where the difference in density between fibers A and B is small, but the difference in density distribution is clear. (The bJ diagram shows a case where the concentration difference between the strings 1aA and B is relatively large. (The Cl diagram shows a case where the concentration distributions of tissues A and B have an overlapping part, but the distribution probability of the boundary part is small. (dl
The figure shows a case where the concentration distributions of tissues A and B overlap greatly and the boundary is unclear. (e) The figure shows the case where one of A and B has a narrow (sharp) fA degree range and the other has a wide (low trapezoidal) fA degree range. Although Fig. 5 shows a case where the area ratios of the weaves A and B are the same, there are also cases where the difference in area between the &Il weaves A and B is large. If we focus on the concentration boundary, this can be seen in Figure 5(e).
Close to. The method of identifying the amount of each tissue is the same as determining the boundary value of each tissue and calculating the area surrounded by this boundary value. Based on FIG. 5, the present invention assumes that the boundary between two adjacent tissues A and B exists at the bottom of the histogram curve of the distribution area corresponding to the density level. However, this bottom position is also relatively bright i1 in FIGS. 5(a), (C), and (e), but in FIG. 5(b). In case (d), it is unclear.
Therefore, a method for determining boundaries based on the shape of the histogram and the shape of the tissue concentration distribution will be explained using Figure 6. First, from the image of test sample f1, create a histogram as described above,
The threshold Ti serving as the boundary between each organization and its tolerance range ΔTi are estimated and set from existing data. Next, from this range Ti ±ΔTi, a definite shillenohold T1 is determined based on the shape of the histogram and the shape of the tissue concentration distribution. Note that the above Ti and ΔTi may be taken as absolute values of concentration, but it is better to use the absolute value divided by the entire concentration distribution range to obtain the concentration ratio. (In the al diagram, even if the density difference is small, the division between Il textures A and H is clear. After smoothing is performed to eliminate noise in the histogram, the bottom position is determined and this is used as the boundary (threshold). (bl diagram) !JlmWhen there is a large difference in the boundary characteristics of A and B, for example when the gradient in one direction is steep and the gradient in the other is gentle, Ti - ΔTi and Ti + ΔT
Let Ti' be the center of gravity of the area of the histogram surrounded by i. In this case, smoothing processing is not necessary. The center of gravity is calculated using the following formula. Σ y−x Ti ' = Σ y where X: abscissa of histogram (X coordinate) y; ordinate of histogram (y coordinate) a=Ti −ΔTi, b−Ti +ΔTi (C) The diagram shows the tissue concentration. The adoption is decided based on the shape of the distribution, and when this shape is simple, Ti - ΔTi and Ti
Let Ti' be the point that bisects the area of the histogram surrounded by +ΔTi. In this case, smoothing processing is not required, so the calculation speed is fast. The calculation formula is as follows. l/2 ×Σ y=Σ y −−−−−−−−−−
-(2) Find C that satisfies equation (2) and set it as c-Ti'. This can be done by finding the left side and finding C that satisfies ≦.

Note that X, y, a, and b are the same as in equation (1), (d
Figure l, (el. (r) shows a method in which after smoothing the histogram, lines are drawn at the first left and right inflection points on the χ axis with Ti as the center, and the intersection of these tangent lines is used. This method is used when the bottom position can be visually checked from the histogram, but the position is not clear.The way to draw a cutting line is to find the difference in the y direction between His 1 and Duram, and then find the reversal position of increase/decrease in the difference. (the position where the second-order differential becomes 0) and draw a regression line here.(d) Figure shows
This is a case where the y-coordinate of the intersection of the two tangent lines is positive, and the x-coordinate of this intersection is 71'.

(El diagram, (') figure shows that the y-coordinate of the intersection of two tangent lines is negative and the difference lxz- between the intersection xl+Xz of each tangent line with the
This is a case where there is a difference in the size of Xl1. (The difference in the EL diagram is when it is smaller than a constant value determined in advance by data analysis. In this case, the y-coordinate of the intersection of the tangent lines is negative, but the X-coordinate of this intersection is Ti'. ff) Figure If the difference is greater than the above constant value, in this case, as shown in figure (f), there is an area between the two tissues that it is unclear which one it belongs to, but its size is small, so this part is deleted. Therefore, it is not included in area calculations, etc. Therefore Ti' = x. , Ti'=x■. Based on the above six criteria, 2
The shape of the histogram and iJ1#s are the boundaries of the two Mi islands.
Since it can be determined from the shape of the concentration distribution, the amount of each tissue can be identified by calculating each area of the histogram surrounded by this boundary. (Example) An example of the present invention will be described below with reference to Figs. 1 to 4 and 6. A sample of sintered ore, for example, as a bonded substance having multiple structures is set in an image analysis device. For example, the image of the cross section of the sample is
Divide into 6 gray scales and create a histogram of the gray scale distribution area. This histogram can be created using software or hardware, but if it is done using hardware,
One of the hardware-based methods that can be created quickly is disclosed in Japanese Patent Application Laid-Open No. 100032/1983. Third
An example of a histogram is shown in the figure. This will be explained below using Figure 1. FIG. 1 is a flowchart showing the procedure for creating a calibration standard. The reason for creating a calibration standard is that, as mentioned above, the concentration distribution of each tissue differs depending on the illumination method, the magnification of the W1 microscope built into the image analysis device, etc., so first, set the field of view that will serve as the standard for the sample. , create a histogram that corresponds to as many gradation levels as possible by adjusting the illumination and magnification in 11 sections, and use the shape of this histogram and the shape of the density distribution of the tissue as parameters to calculate the shape of each tissue from the existing 'R$4. The boundary value (Slenshehold) and its permissible range are determined, and this is used as the standard for determining the tissue boundary value of each visual field thereafter. No. IUj! In J, first, in step 1l, a large number of threshold values (thresholds) that determine fA change stages are set. In this embodiment, the number of steps is 256, but it is preferable to have 500 or more steps for automatic adjustment and smoothing of the entire density range. However, the capabilities of imaging cameras currently cannot keep up with this. Each W of the target image with Stenbue 2! 1
Binarize the element according to the threshold, find the density, and create a gray histogram. An example of this is shown in Figure 3. The horizontal axis (X-axis) of Figure 3 shows the 4 degree threshold (divided into 256 pieces), and the vertical axis (y-axis) shows the area between each threshold as a percentage of the total area. In step 13, the entire range of gray histogram densities (
Find the iffi degree span). This is used to express the absolute value of concentration as a concentration ratio. Figure 4 is a smoothed version of Figure 3, and R shown in the figure represents the density span. In step l4, the shape of the histogram,
The boundary value Ti and allowable range ±ΔTi of each tissue are determined based on existing data using the shape of the concentration distribution of the tissue as a parameter. This tolerance range is based on the assumption that when determining the boundary value Ti' of the other field of view of the sample, the relative concentration ratio is not different from that of the calibration standard, and Ti' is within the range of Ti ±ΔTi. This is because it is used as a parameter for determining Ti'.

Next, the main routine will be explained using Figure 2. The main routine uses the standard threshold Ti ±△Ti created in the calibration standard creation routine shown in Figure 1 to quickly determine the threshold Ti' value for other samples with similar compositions. show.

Step 2l is the same as step 12 in FIG. 1, and a gray histogram is obtained for each target visual field, and step 22
Find the entire density range with and express the gray histogram relatively as a density ratio. By standardizing in this way,
Ti±8T1 determined in Figure 1 can be applied.
Set the total number of thresholds n in step 23. As a result, the value Ti′ for each thread
can be determined. In step 24, it is determined whether or not the shape of the gray histogram near Ti ±ΔT1 is clearly distinguishable between adjacent tissues based on the density distribution as shown in FIG. 6(a). If the distinction is clear, step 25 is performed. Performs smoothing of the gray histogram. There are several known methods for smoothing, but
One of them is the FFT method (Fast.
Transform). This performs a Fourier transform on the target curve, performs frequency analysis, cuts high frequency components to remove noise, and then performs an inverse Fourier transform to reproduce the curve. Figure 3 shows the shape of the gray histogram before smoothing, and Figure 4 shows the shape after smoothing. Next, find the bottom position using the stem 26 and set this position as Ti'. In step 24, adjacent &lI! in concentration distribution! If the distinction between a is not clear, in step 27, the characteristics of the boundary between adjacent tissues are examined based on the concentration distribution.
When this characteristic is shown in Figure 6 (C), where one slope is steep and the other is gentle, it is often difficult to determine the bottom position. In this case, in step 28, Ti ±Δ
Find the center of gravity position of the area surrounded by Ti using the formula ( ]) and let this center of gravity position be Ti゛. This method does not require smoothing. If the characteristic of the boundary is not as shown in FIG. 6(C) in the steno knob 27, in step 29,
! Examine the shape of the concentration distribution of l, and if the shape is knotty, use equation (2) to find the position where the area between Ti ±ΔTi is divided into two by the stem 30, and let that position be Ti'. This method does not require smoothing and is fast to calculate. In step 29, if the distribution shape is complex, in step 3l, the gray histogram is smoothed, and in step 32, T
Find the first inflection point on the left and right sides of i on the χ axis, and draw a tangent line at each point. Step 32 and subsequent steps are steps for determining the Ti' position when the bottom position on the gray histogram is visible, but it is not clear enough to determine the position. The steno knob 33 checks whether the y-coordinate of the intersection of the two tangent lines is positive or not. If it is positive, the steno knob 34 sets the χ coordinate of the intersection of the two tangent lines to Ti'. This is the case in Figure 6. If the y-coordinate of the intersection of the tangent lines is not positive in step 33, then in step 35, the point Xl+X! where each tangent line intersects with the X axis! In step 36, the length IXIX1| of X8 and In this case, the y-coordinate of the intersection point is negative 0, and this situation is shown in FIG. 6(e).In step 36, l x.
-x, is not smaller than the value D, in step 38 Ti
′”
As shown in
This is because the area between Xt and Xt is small, so removing this area will improve accuracy. Next, in step 39, prepare to examine the next boundary value by setting i to n-1. In step 40, it is checked whether all boundary values have been determined, and if they remain, the process returns to step 24 and all boundary values are determined. Once each boundary value is determined in this way, the structure of the tissue can be determined by calculating the area surrounded by this boundary value. In this way, the tissue composition of each field of the sample can be quickly determined. The above-mentioned embodiment is an example in which a plurality of boundary value determination methods of the present invention are combined,
Other combination methods are possible depending on the type of sample. [Effects of the Invention] According to the present invention, an image of a cross section of a binding substance having a plurality of tissues is darkened or darkened in accordance with preset grayscale levels. Create a histogram of the light distribution area, set the estimated threshold value of the concentration range corresponding to each target tissue with a tolerance range, and calculate mirror determination from the quasi-standard value based on the shape of the histogram and the shape of the concentration distribution in mm. Because the threshold value can be determined almost mechanically!
Quantification of Il texture can be performed quickly and with high accuracy.

[Brief explanation of the drawing]

Claims (6)

[Claims] (1) Create a histogram of the density distribution area of an image of a cross section of a binding substance having multiple tissues according to preset density levels, define a density range corresponding to each target tissue for the histogram, and In the tissue quantification method using image analysis, which determines the composition of each target tissue from the area ratio of the range,
The estimated threshold value of the concentration range corresponding to each target tissue is set as a value having a predetermined tolerance range, and at least one of the shape of the histogram centered around the tolerance range, and the shape of the concentration distribution of the tissue is set. A tissue quantification method using image analysis, characterized in that a definitive threshold is determined based on the density range and an area ratio of the concentration range is determined. (2) The tissue quantification method by image analysis according to claim 1, characterized in that the minimum value within the permissible range of the histogram subjected to noise elimination processing is set as the final threshold. (3) The tissue quantification method by image analysis according to claim 1, characterized in that the position of the center of gravity of the area within the allowable range of the histogram is set as the determined threshold. (4) Claim 1 characterized in that the determined threshold value is a position dividing the area of the histogram into two equal parts within the permissible range.
Tissue quantification method by image analysis described. (5) After the histogram has been subjected to noise elimination processing, the histogram's gradations are set in the x-axis direction, and the frequency of the gradation steps is set in the y-axis direction, and the first rows of the histogram are placed on the left and right sides on the x-axis from the center of the tolerance range. 2. The tissue quantification method using image analysis according to claim 1, wherein a cut line is drawn at an inflection point, and the x-coordinate of the intersection of the cut lines is set as the determined threshold value. (6) The position where each of the above-mentioned cutting lines intersects with the x-axis is x_1,
6. The tissue quantification method by image analysis according to claim 5, wherein x_2 is set, and when |x_2-x_1| is equal to or greater than a predetermined value, the determined threshold is set to two values, x_1 and x_2.