JP2001264317A - Method for confirming tissue of organism using image processing - Google Patents

Abstract

PROBLEM TO BE SOLVED: To automatically confirm specific tissue by applying image processing to a two-dimensional image of tissue of an organism. SOLUTION: A sample environmental factor showing a feature of texture is extracted from an image of sample tissue (S1, S2). An image of specimen tissue to be processed is inputted (S3). A large number of imaginary organisms having the extracted sample environmental factor and random moving characteristic factors are generated on this image (S4) and moved on the basis of the moving characteristic factors (S5). Intake energy depending on the similarity of the environment after movement with sample environment is applied (S6), and the imaginary organisms are killed or propagated on the basis of an increase/decrease in consumption energy and intake energy (S7). If this processing is repeated (S8), a discrimination image showing a portion apparoximate to the sample tissue is formed on the basis of the basis of the final distribution of the imaginary organisms (S9).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for recognizing a biological tissue using image processing, and more particularly to an image processing technique for facilitating identification of a malignant tumor image such as a cancer tissue using a computer.

[0002]

2. Description of the Related Art Various apparatuses using computers have been used as medical diagnosis support apparatuses. For example, CT devices, MRI devices, and the like are widely used as diagnostic support devices that are indispensable for current medical diagnosis, and three-dimensional scanning of a specimen can obtain three-dimensional image data of a body tissue. It is. On the other hand, histological examination is indispensable for accurate determination of disease, and more than 5 million tissue diagnoses are performed annually.
In such a tissue diagnosis, a two-dimensional image of a biological tissue serving as a specimen is usually used. As a method for obtaining such a two-dimensional image of a biological tissue, a method using hematoxylin / eosin staining is generally used, and a microscope image of the stained biological tissue is captured as digital data via a CCD camera or the like.

[0003]

Even if a two-dimensional image of a biological tissue is obtained by the above-described method using hematoxylin and eosin staining, the individual portions on the two-dimensional image are determined by what kind of tissue If it cannot be recognized, it is useless for diagnosis. In general, for red blood cells, white blood cells, and the like, an identification device that focuses on morphological characteristics such as circularity, perimeter, and cell nucleus / cytoplasm ratio is used in practice, and automatically recognizes whether or not the blood cell is normal. Can be displayed. However, with respect to general biological tissues, an identification device for recognizing a specific tissue is in a research stage, and has not been put to practical use at present. For example, when determining whether or not a biological tissue on a two-dimensional image obtained using hematoxylin and eosin staining is a malignant tumor, the stained image is divided into a small number of regions based on a color difference threshold. Although there is a proposal of a method for performing feature extraction, it has not reached the level of practical use. One of the reasons is that, when a biological tissue is stained, cell nuclei and lymphocytes are stained in the same color, and it is difficult to distinguish a cancer cell region or the like only by characteristic evaluation based on color. Until now, the discovery of lesions such as cancer cells has been performed by skilled pathologists based on experience and comprehensive knowledge, so for example, an objective index is provided to evaluate the morphological characteristics of the cancer cell area At present, sufficient research has not been conducted on such technologies.

Accordingly, an object of the present invention is to provide a biological tissue recognition method capable of automatically recognizing a specific tissue by performing image processing using a computer on a two-dimensional image of the biological tissue. And

[0005]

Means for Solving the Problems (1) According to a first aspect of the present invention, a predetermined image processing is performed on an image of a biological tissue to be a specimen using a computer, so that a specific specimen tissue can be processed. In a biological tissue recognition method using image processing to obtain an identification image that can be recognized by distinguishing the approximated part, a step of inputting a sample tissue image from a specific sample tissue, based on the input sample tissue image, Extracting a sample environment factor indicating a texture characteristic of the sample tissue; inputting a sample tissue image from the sample tissue; having a movement characteristic factor indicating a predetermined movement characteristic and a sample environment factor; Generating a number of virtual creatures having defined values so as to be distributed on the sample tissue image; and (a) moving the virtual creature on the sample tissue image based on the movement characteristic factor, and (B) Similarity between the sample environment factor indicating the texture characteristic of the sample tissue image around the position after the movement of the virtual creature and the sample environment factor of the virtual creature (C) reducing the consumed energy value and adding the intake energy value to the energy value of the virtual creature, thereby updating the energy value. Do
When the updated energy value reaches a predetermined lower limit, the virtual creature is removed as being dead, and when the updated energy value reaches a predetermined upper limit, the virtual creature grows. A new virtual creature with the same sample environmental factor
(a), (b), and (c) are repeatedly executed a predetermined number of times, and a target identification image is created based on the distribution of virtual living organisms finally living on the specimen tissue image. And the steps.

(2) According to a second aspect of the present invention, a predetermined image processing is performed on an image of a biological tissue as a specimen by using a computer, so that a portion approximated to a specific sample tissue can be distinguished. In a biological tissue recognition method using image processing to obtain an identification image that can be recognized, while inputting a normal sample tissue image from a specific normal sample tissue,
Inputting an abnormal sample tissue image from a specific abnormal sample tissue, extracting a normal sample environmental factor indicating a texture characteristic of the normal sample tissue based on the input normal sample tissue image, and inputting the abnormal sample tissue image Extracting an abnormal sample environment factor indicating a texture characteristic of the abnormal sample tissue, inputting a sample tissue image from the sample tissue, and having a moving characteristic factor indicating a predetermined moving characteristic based on a predetermined initial characteristic. Generating a large number of virtual creatures with defined energy values so as to be distributed on the sample tissue image, and for each generated virtual creature, the texture characteristics of the sample tissue image around the location of the virtual creature A sample environmental factor that indicates that the similarity between the normal environmental sample and the abnormal sample environmental Applying the sample environmental factor to the virtual creature, and (a) moving the virtual creature on the specimen tissue image based on the movement characteristic factor, and determining the energy consumption value consumed by this movement, (b) Intake to be taken by the virtual creature based on the similarity between the sample environment factor indicating the texture characteristics of the sample tissue image around the position after the movement of the virtual creature and the sample environment factor of the virtual creature The process of determining the energy value. (C) The energy value of the virtual creature is reduced by the consumed energy value and the added energy value is added to update the energy value. If the virtual creature reaches the upper limit, the virtual creature is determined to have died. (A), (b), and (c) are repeatedly performed a predetermined number of times, ie, a process of generating a new virtual creature having a sample environmental factor in the vicinity. Based on the distribution of virtual organisms with living normal environmental factors, a discriminating image of normal tissue is created, and the abnormal living environmental factors that are finally living on the specimen tissue image are created. Creating an identification image of the abnormal specimen tissue based on the distribution of the virtual creatures.

(3) The third aspect of the present invention is the above-described second aspect.
In the biological tissue recognition method using image processing according to the aspect, when performing the process (b), the similarity between the sample environmental factor and the sample environmental factor of the virtual creature and the sample without the virtual creature The intake energy value to be taken by the virtual creature is determined based on both the environmental factor and the dissimilarity between the sample environmental factor.

(4) The fourth aspect of the present invention is the above-mentioned first aspect.
In the biological tissue recognition method using image processing according to the third to third aspects, the average, variance, energy, and entropy of the density value of each pixel in a predetermined reference region are set as environmental factors indicating the characteristics of the texture of the tissue image. , A plurality of parameter values selected from the group of contrasts are used, and the similarity between environmental factors is determined based on the closeness of the plurality of parameter values.

(5) The fifth aspect of the present invention is the above-mentioned first aspect.
In the biological tissue recognition method using the image processing according to the fourth to fourth aspects, a color image is used as each tissue image, and each pixel in a predetermined reference area is used as an environmental factor indicating a texture feature of the tissue image. Parameter values for the density value of each color component of the three primary colors are used,
The similarity between environmental factors is determined based on the angle difference between three-dimensional vectors defined based on parameter values for each color component.

(6) A sixth aspect of the present invention is the above-mentioned first aspect.
In the biological tissue recognition method using image processing according to the fifth to fifth aspects, as the movement characteristic factor, a parameter indicating a movement characteristic of the virtual creature in the front-back direction and a movement characteristic in the left-right direction and a rotation characteristic is used. When each virtual creature is generated, a randomly determined movement characteristic factor is added.

(7) A seventh aspect of the present invention is the above-mentioned first aspect.
In the biological tissue recognition method using image processing according to the sixth to sixth aspects, the age is defined for each virtual creature,
Each time the three processes (a), (b), and (c) are performed, the age of each virtual creature is increased, and virtual creatures whose age has reached a predetermined life are removed as dead. Things.

(8) The eighth aspect of the present invention is the first aspect of the present invention.
-In the biological tissue recognition method using image processing according to the seventh aspect, when growing the virtual creature in the process (c), the virtual creature of the parent to be multiplied is removed, and a plurality of new virtual creatures are added in the vicinity thereof. A virtual creature of a child is generated, a predetermined initial energy value is given to the virtual creature of the child, and a sample environmental factor and a movement characteristic factor of the virtual creature of the parent are inherited.

(9) The ninth aspect of the present invention is the above-mentioned eighth aspect.
In the biological tissue recognition method using image processing according to the aspect, when inheriting the movement characteristic factor that the parent virtual creature had, to cause a sudden displacement with a predetermined probability,
When a sudden displacement occurs, the movement characteristic factor is randomly changed and then inherited.

(10) According to a tenth aspect of the present invention, there is provided the biological tissue recognition method using the image processing according to the first to ninth aspects, wherein the near reference area centering on the existence position of each virtual organism is provided. And a distant reference area that includes the near reference area and further extends the boundary to a far distance, and defines a nearby environment factor that indicates the characteristics of the texture in the near reference area and a distant reference that indicates the characteristics of the texture in the distant reference area. When the similarity with the environmental factor is low, a process of reducing the energy value of the virtual creature or the energy value of the virtual creature is further added.

(11) An eleventh aspect of the present invention is the biological tissue recognition method using the image processing according to the first to tenth aspects, wherein the environmental factor indicating the characteristic of the texture of the tissue image is specified. Define the contrast value for the density value of each pixel in the reference area of the difference between the contrast value of the sample tissue image around the position where the virtual creature exists and the contrast value defined as the sample environmental factor of the virtual creature The absolute value of is calculated for all living creatures that are alive. It is further added.

(12) According to a twelfth aspect of the present invention, there is provided a program for causing a computer to execute each step constituting the biological tissue recognition method using image processing according to the first to eleventh aspects described above. It is recorded on a computer-readable recording medium.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below based on an embodiment shown in the drawings.

§1. Basic Concept of the Present Invention First, the basic concept of the present invention will be briefly described. Assume that tissue images as shown in FIGS. 1A and 1B are prepared. The tissue image shown here is a mammary gland tissue image of an animal, and FIG. 1 (a) shows a normal mammary gland tissue image, and FIG.
(b) shows an example of a cancerous mammary gland tissue image. FIG.
In the normal tissue shown in (a), the cell nucleus region has an annular tissue structure, whereas in the cancerous tissue shown in FIG. 1 (b),
The state where the cell nucleus region is enlarged can be confirmed. However, for convenience of explanation, each tissue is shown as a very clarified pattern, but in reality, each tissue is not so clearly recognizable. In the case of a tissue image acquired in actual clinical practice, it is difficult to clearly recognize individual tissues unless a pathologist with a certain level of skill is used. Therefore, if a tissue image as shown in FIGS. 1 (a) and 1 (b) is taken into a computer as digital data and subjected to predetermined image processing, if an identification image capable of distinguishing and recognizing only the cell nucleus region can be automatically generated. It is convenient.

The present invention relates to a technique for realizing such automatic generation of an identification image by a computer. According to the present invention, for example, by performing predetermined image processing on tissue image data as shown in FIGS. 1A and 1B, as shown in FIGS. 2A and 2B. An identification image can be generated. FIG. 2 (a) shows only the cell nucleus region recognized from the tissue image of FIG. 1 (a), and the recognized cell nucleus region is colored with a predetermined color (eg, green: indicated by hatching in the figure). It is. On the other hand, FIG.
(b) recognizes only the cell nucleus region from the tissue image of FIG. 1 (b) and colors the recognized cell nucleus region with a predetermined color (for example,
(Red: shown by hatching in the figure). As described above, the normal breast tissue and the cancerous breast tissue are displayed in different colors by the image processing, so that the respective tissues can be clearly recognized.
For example, if the method according to the present invention is applied to a tissue image in which normal breast tissue and cancerous breast tissue are mixed, the normal breast tissue region is displayed in green, and the cancerous breast tissue is displayed. Since the region is displayed in red, it is possible to easily recognize which portion is affected by cancer.

Of course, the identification image obtained by the method according to the present invention does not necessarily indicate a correct recognition result, and may cause erroneous recognition. However, in the field of clinical medicine, it is impossible to perform a completely correct diagnosis in the first place, and it is necessary to make a correct diagnosis as much as possible by making a comprehensive judgment based on information obtained from various diagnosis support devices. In such a point, the biological tissue recognition method according to the present invention has a sufficient industrial value as a technology that enables a computer to be used as a diagnosis support device.

Next, the basic concept of the present invention will be described with reference to FIG. A feature of the present invention resides in that a method of recognizing a portion approximated to a specific specimen tissue from an image of a biological tissue to be a specimen by utilizing the environmental adaptability of a virtual organism is adopted. Now, consider a state in which a virtual environment as shown in FIG. 3A is defined, and a large number of virtual creatures are distributed on this virtual environment. As shown in the figure, baits are arranged in the central area of this virtual environment, and an environment suitable for the survival of each virtual creature is secured. No bait is present and is unsuitable for the survival of individual virtual creatures. Therefore, it is assumed that virtual creatures placed in an environment without food will eventually die, virtual creatures placed in an environment with sufficient food will breed, and that each virtual creature is allowed to move with some degree of freedom. I will allow it. Then, even if the virtual creatures were initially distributed almost uniformly throughout the virtual environment as shown in Fig. 3 (a),
Surrounding virtual creatures without food die or move to the central part, while virtual creatures in the central bait continue to breed. As a result, as shown in FIG. 3 (b), the virtual creature density in the surrounding part gradually decreases, and the virtual creature density in the central part gradually increases. Eventually, as shown in FIG. 3 (c), the virtual creature survives only in the central part where food is present.

In this model, even if the existence area of the bait itself could not be recognized,
As shown in (c), if the distribution area of the living virtual creature can be finally recognized, the existence area of the bait itself can be indirectly recognized.

To apply this principle to a tissue image of an arbitrary specimen, the following method may be used. First,
A large number of virtual creatures are generated so as to be substantially uniformly distributed on a tissue image serving as a specimen. Here, it is assumed that each virtual creature has a movement characteristic factor and a sample environment factor. The movement characteristic factor is a factor indicating the direction in which the tissue moves in the tissue image, and a random movement characteristic factor is given to each virtual creature.
For example, one virtual creature tends to move upward on a two-dimensional tissue image, while another virtual creature tends to move diagonally downward and leftward on this image. Various movement characteristics are defined for each.

On the other hand, the sample environment factor is a factor indicating the texture characteristic of the tissue to be sampled. For example, if the purpose is to recognize a portion similar to “cell nucleus region in normal mammary gland tissue” from a tissue image to be sampled, this “cell nucleus region in normal mammary gland tissue”
Is the specimen tissue. Therefore, this sample tissue is prepared in advance, and a work of extracting a sample environment factor showing a characteristic of texture from this sample tissue is performed, and the extracted sample environment factor is given to each virtual creature. The substance of the sample environment factor is a parameter indicating the characteristic of the texture pattern in a predetermined reference region in the sample tissue, as will be described later, and can be expressed as a numerical value. For example,
When the “cell nucleus region in normal mammary gland tissue” is a sample tissue, the parameter values indicating the characteristics of the texture pattern in the cell nucleus region shown in FIG. Will be done. In other words, these virtual creatures can adapt to the environment having the same characteristics as the texture pattern in the cell nucleus region shown in Fig. 1 (a), and survive and grow further. But can't adapt to other environments and will die.

Each virtual creature defines an initial energy value. As described above, each virtual creature can move on the basis of the movement characteristic factor, but consumes predetermined energy each time it moves. On the other hand, if the environment after the movement matches the sample environmental factor of the user, predetermined energy can be taken. Then, when the remaining energy value reaches a predetermined lower limit (for example, zero), the virtual creature is removed as dead, and when it reaches the predetermined upper limit, the virtual creature proliferates. As a result, a plurality of new virtual creatures having the same sample environment factor are generated.

In such a setting, if the movement of each virtual creature, the consumption and ingestion of energy, the death and the multiplication are repeatedly performed, the virtual creatures will eventually survive only in an area suitable for the environment. In the case of the above example, initially, even if virtual creatures are generated so as to be uniformly distributed over the entire tissue image as a specimen, the virtual creature eventually becomes a cell nucleus region (a region having a texture pattern similar to the sample tissue). , And if the distribution area of the virtual creatures is colored and displayed, a target identification image can be obtained. This is the fundamental principle of the present invention.

In the present invention, by preparing a plurality of types of specimen tissues, it is possible to obtain a plurality of types of identification images. For example, a cell nucleus region of normal breast tissue shown in FIG. 1 (a) is prepared as a first sample tissue, and a cell nucleus region of cancerous breast tissue shown in FIG. 1 (b) is prepared as a second sample tissue. In this case, in the tissue image prepared as the specimen, a portion approximating the first sample tissue is displayed in green, and a portion approximating the second sample tissue is displayed in red. It is also possible to obtain such an identification image. In this case, two types of virtual creatures may be prepared in advance. The first kind of virtual creature (herein referred to as a green creature) is a living creature whose normal environment is the cell nucleus region of normal mammary gland tissue, and the cell nucleus region of normal mammary gland tissue shown in FIG. Specimen environmental factors having the characteristics of the texture pattern. On the other hand, the second kind of virtual creature (referred to as a red creature in this case) is a creature that uses the cell nucleus region of cancerous mammary gland tissue as a compatible environment.
The specimen environment factor has the characteristic of the texture pattern of the cell nucleus region of the cancerous mammary gland tissue shown in FIG. If these two types of virtual creatures are evenly distributed on the tissue image prepared as a specimen and repeated movement, energy consumption and ingestion, death and proliferation, the green creature will eventually become a normal mammary gland tissue. The red organisms are gathered in the cell nucleus region, and the red organisms are gathered in the cell nucleus region of the cancerous mammary gland tissue, and a target identification image is obtained.

§2. Implementation using a single type of sample tissue
State Next, the biological tissue recognition method using the image processing according to the present invention, a description of a specific procedure. Here, first,
Procedure using a single type of sample tissue (for example, Fig. 1 (a)
In the case where a cell nucleus region of a normal mammary gland tissue shown in (1) is prepared as a specimen tissue. FIG. 4 is a flowchart showing a basic flow of this procedure. An object of the present invention is to obtain an identification image by performing predetermined image processing using a computer on an image of a biological tissue to be a specimen so that a portion approximating a specific sample tissue can be distinguished and recognized. Each procedure shown in the flowchart of FIG. 4 is executed using a computer.

First, in step S1, a sample tissue image is input. For example, if a "cell nucleus region of a normal breast tissue" as shown in FIG. 1A is used as a sample tissue, a two-dimensional image of the entire "normal breast tissue" is input to a computer as digital data. Just do the work. Generally, as a method for obtaining a two-dimensional image of a biological tissue, a method using hematoxylin and eosin staining is known, and a microscope image of the biological tissue stained with hematoxylin and eosin is obtained by CCD.
What is necessary is just to perform the process which takes in as digital data via a camera etc.

In the following step S2, based on the sample tissue image thus input, a process of extracting a sample environmental factor indicating the texture characteristic of the sample tissue is performed.
For example, when a two-dimensional image of the entire “normal breast tissue” as shown in FIG. 1A is captured in step S1, only the “cell nucleus region” in the sample tissue becomes the sample tissue. Therefore, it is necessary to extract a sample environmental factor from the “cell nucleus region”. Therefore, here, as shown in FIG. 5, an input for designating several points inside the “cell nucleus region” is performed (the X-marked portions in the figure are designated points),
A method of quantitatively evaluating the texture characteristics of the area near each of these designated points will be adopted. Of course, it is necessary to determine which part of the “cell nucleus region” to be the specimen tissue is based on the judgment of a skilled pathologist. If possible, in a state where a two-dimensional image of the entire “normal breast tissue” as shown in FIG. 5 is displayed on the display screen, the feature as the “cell nucleus region” appears most prominently to this pathologist. Ask them to input the point that seems to be a designated point.

The following method is used to quantitatively evaluate the characteristics of the texture in the area near the designated point. Here, as an example, an area composed of 5 × 5 pixels as shown in FIG. 6 is considered as a reference area for a designated point. That is, since the image data captured by the computer is composed of a large number of pixels, a reference area including a set of 25 pixels as shown in FIG. 6 can be defined in the area near the designated point. The pixel at the center indicated by the thick line frame is a pixel corresponding to the position of the designated point,
Here, the region including all the 25 pixels including the periphery thereof is set as the reference region, but a wider region may be set as the reference region.

Each pixel has a predetermined density value.
In the example shown in FIG. 6, it is assumed that the input sample tissue image is a monochrome image, and the density value of the i-th pixel is Q
(I). Normally, the density value of a pixel has a predetermined gradation. For example, when the gradation is expressed by 8 bits, the number of gradations L = 256. In this case, each pixel takes one of the density values Q from 0 to 255. For example, five parameters as shown in FIG. 7 are known as parameters that quantitatively express the characteristics of the texture composed of such a gradation image. That is, average (Ave.), dispersion (Var.), Energy (En)
gy. ), Entropy (Ent.) And contrast (Cont.). Since these parameters are widely used in the field of image processing by a computer, detailed descriptions thereof are omitted, but the values of the parameters are mathematically defined by the illustrated equations. Here, Qa,
Qv, Qn, Qe, and Qc denote parameter values of average, variance, energy, entropy, and contrast, respectively. Q is the density value of each pixel (Q = 0 to 255 in the above example), M is the total number of pixels in the reference area (M = 25 in the above example), and N (Q) is the reference area. , The number of pixels having the density value Q, P (Q) is the probability of taking the density value Q in the reference area (here, P (Q) = N
(Calculated as (Q) / M).

In practice, as shown in FIG. 5, a plurality of points are designated as designated points indicating a specimen tissue, a reference region is defined for each designated point, and the above-described five patterns are defined for each reference region. Is preferably calculated, and the average of the same parameter values obtained for each designated point is used as the final parameter value. After all, in step S2, Qa, Qv, Qn, Qe, and Qc are used as sample environment factors indicating the characteristics of the texture of the sample tissue image.
The following five parameter values are obtained.

Subsequently, in step S3, an input process of a sample tissue image is performed. This processing is performed in step S1.
Is exactly the same as the input processing of the sample tissue image in the above, and a process of capturing a microscope image of a biological tissue stained with hematoxylin and eosin as digital data via a CCD camera or the like may be performed. The specimen tissue image is
In contrast to an image obtained from a tissue to be a typical sample like a pathological sample, a sample tissue image is an image obtained from the tissue of an individual sample to be examined.

Next, in step S4, a process of generating a large number of virtual creatures having the movement characteristic factor and the sample environment factor so as to be distributed on the specimen tissue image input in step S3 is performed. Of course, the generation of this virtual creature is performed as a simulation on a computer.
In the two-dimensional coordinate system, predetermined coordinate values (x, y) are defined, and the following processes are performed on the assumption that one virtual creature exists at a position having the coordinate values. Become. In the embodiment shown here, the coordinate values (x, y) are randomly defined using random numbers so that a large number of virtual creatures are generated as uniformly distributed as possible on the specimen tissue image. . In the case of this embodiment, each virtual creature has the same sample environment factor. In the example described above, five parameter values Qa, Qv, Qn, Qe, and Qc are used as sample environmental factors, and all virtual creatures generated in step S4 have the same parameter value. Will be done. In other words, all the virtual creatures generated in step S4 are
In either case, the texture of the specimen tissue image input in step S1 is a living organism whose self-adaptive environment.
However, in the embodiment shown here, only one set of the sample environment factors (five parameter values) indicating the characteristics of the average points of the textures near the eleven designated points indicated by X in FIG. 5 is used. Independent sample environmental factors are defined for each designated point, and any one of 11 sets of sample environmental factors
A set may be given to each virtual creature at random.

On the other hand, the movement characteristic factor is a parameter indicating the movement characteristic of the virtual creature, and is set to have a different value for each virtual creature. Although a specific example of the parameter used as the movement characteristic factor will be described later, a random parameter value is given to each virtual creature using a random number.

Each virtual creature is given a predetermined initial energy value. The energy value of a virtual creature is
This is used as a criterion for determining whether the virtual creature dies or multiplies.
The following description will be made assuming that an initial energy value of 1000 is uniformly applied to each virtual creature generated in step 4.

The processing in steps S5, S6, and S7 is as follows:
This process is repeated several times after step S8.
First, the process of step S5 is a process of moving all the living virtual creatures at that time on the specimen tissue image based on the respective movement characteristic factors, and determining the energy consumption value consumed by this movement. . As described above, each virtual creature is given a random movement characteristic factor, and each virtual creature moves in an arbitrary direction based on the movement characteristic factor. Although specific moving processing will be described later, in the case of this embodiment, no matter what movement is performed (even if no movement is performed), the energy consumption value is uniformly set to 300. And so on. Of course, the energy consumption value can be changed according to the moving distance.

In the next step S6, a sample environment factor indicating the texture characteristic of the sample tissue image around the position after the movement of the virtual creature is obtained, and the sample environment factor is compared with the sample environment factor of the virtual creature. Based on similarity,
This is a process for determining an energy value to be taken by the virtual creature. As a specific method for obtaining the sample environmental factors,
Although a reference region as shown in FIG. 6 is defined near a designated point on the specimen tissue, and an example of obtaining five parameter values as shown in FIG. 7 for 25 pixels in the reference region has been described.
The sample environmental factors required in step S6 are also:
Similarly, it is constituted by five parameter values. That is, 25 pixels are extracted as shown in FIG. 6 with the pixel corresponding to the position after movement of each virtual creature as the center pixel, and five parameter values as shown in FIG. 7 are obtained for these pixels. It is. Here, for convenience of explanation, five parameter values constituting the sample environmental factor are Q
a, Qv, Qn, Qe, and Qc, and five parameter values constituting the sample environmental factor are QQa, QQv, and QQ.
It will be referred to as n, QQe, QQc. of course,
Five parameter values Qa and Q constituting the sample environmental factor
In this embodiment, v, Qn, Qe, and Qc are common to all the virtual creatures and do not change at all by the movement of the virtual creature, but five parameter values QQa and QQv that constitute the sample environmental factor. , QQn, QQe,
QQc is a value determined depending on the current position of each virtual creature. Naturally, QQc differs for each virtual creature, and is a value that changes each time the same virtual creature moves.

In step S6, the similarity between the sample environmental factor and the sample environmental factor is determined. Specifically, the similarity between the above-described five parameter values is determined. FIG. 8 is a table showing a comparison between parameter values to be subjected to such similarity determination. Since each parameter value has some numerical value, the similarity can be determined based on the similarity of the parameter values. The approximation of the parameter value can be determined based on the difference between the two, or can be determined based on the ratio between the two. That is, it can be determined that the smaller the difference or ratio between the two, the higher the approximation.

It is preferable that the similarity between the sample environmental factor and the sample environmental factor is determined by comprehensively evaluating the individual similarities of the five parameter values. For example, even if the value of one parameter (for example, the average (Ave.)) is exactly the same (for example, Qa = QQa), the sample environmental factor and the sample environmental factor are similar for the same reason. It is not preferable to judge. Various methods can be considered as a method of comprehensively determining similarity. For example, for each parameter, based on the difference or ratio, the degree of approximation 3 (very similar), the degree of approximation 2
(Approximately approximate), approximation 1 (slightly approximate), and approximation 0 (not at all) are defined in four stages, and the approximation is defined for each of the five parameters. It is also possible to evaluate the total similarity as a total similarity.

The purpose of step S6 is to determine an intake energy value to be taken by the virtual creature based on the overall similarity thus obtained. When the overall similarity is evaluated as a numerical value by the method as in the above-described example, when the similarity is higher, a larger intake energy value can be determined. However, the process of determining the intake energy needs to be performed separately for each virtual creature, and needs to be performed every time each virtual creature moves. For this reason, when performing simulation by moving a large number of virtual creatures many times, if the calculation of the similarity requires a complicated calculation, the entire calculation load becomes enormous. In particular, as described in §3, when the tissue image is a color image, the computational load becomes considerably heavy.

Therefore, in the present embodiment, only two similarities, “similar” and “dissimilar”, are defined by the following simple method. Sets the energy intake to 1200 and dissimilar
When it was determined that the intake energy value was 0.
Which of the two similarities is “approximate” or “not approximate” for each of the five parameter values
The two similarities are defined (for example, “approximate” may be used only when the parameter value difference is within a predetermined value), and three or more types of parameters determined to be “approximate” exist. In this case, the sample environmental factor is comprehensively determined to be “similar” to the sample environmental factor. If there are only two or less parameters determined to be “approximate”, the sample environmental factor is determined. Makes a comprehensive judgment of "not similar" to the sample environmental factor. For example, in the table shown in FIG. 8, when Qa = QQa, the parameter indicating the average (Ave.) is naturally determined to be “approximate”, but the other four parameters are “not approximate”. If the decision is made,
Since only one parameter is determined to be “approximately”, the overall determination is made that the sample environment factor is “not similar” to the sample environment factor.

In step S7, a process of killing or multiplying by increasing or decreasing the energy is performed for each virtual creature. That is, since the energy consumption is determined in step S5 and the energy intake is determined in step S6, the energy consumption value is reduced by adding the energy consumption value to the energy value of the virtual creature, thereby obtaining the energy value. Will be updated. In the case of the present embodiment, regardless of the presence or absence and the form of movement, the energy consumption is uniformly 300, and based on the similarity, either 1200 or 0 intake energy is given, so that the In S6, if it is determined that the sample environment factor around the position after the movement is “similar” to the sample environment factor, an energy value of the subtraction 900 is added,
If determined to be "not similar", the energy value of 300 will be reduced.

The processes of steps S5, S6, and S7 are repeatedly executed until the termination condition in step S8 (for example, whether a predetermined number of repetitions has been exceeded) is satisfied. As a result, each virtual creature moves around on the sample tissue image and updates the energy value each time. Therefore, in step S7, when the updated energy value reaches the predetermined lower limit, the virtual creature is removed as being dead, and when the updated energy value reaches the predetermined upper limit, A process of generating a plurality of new virtual creatures having the same sample environmental factor as that of the virtual creatures will be performed. Specifically, virtual creatures whose updated energy value has reached the lower limit value 0 are removed as dead, and virtual creatures whose updated energy value has reached the upper limit value 2000 are assumed to be the same as those having proliferated. Are replaced with two new virtual creatures (the initial energy value is 1000) having the sample environmental factor of (1).

If such processing is repeatedly executed, a virtual creature present at a point having a sample environmental factor similar to the sample environmental factor (in other words, a surrounding texture at a point similar to the texture of the sample tissue) In contrast, a virtual creature that exists at a point having a sample environmental factor that is not similar to the specimen environmental factor (in other words, a point where the surrounding texture is not similar to the texture of the sample tissue) is higher. It is more likely to die. Thus, ultimately, a state in which the majority of virtual creatures have survived in a portion approximating the specimen tissue. As an end condition of step S8, an appropriate condition is set such that a change in the distribution of the virtual creatures is expected to be almost eliminated, and is repeated a sufficient number of times.

Finally, in step S9, based on the distribution of virtual living organisms finally living on the specimen tissue image,
A process of creating a target identification image is performed. For example, the identification image may be created by coloring and displaying the distribution of the living virtual creatures. Such a colored region in the identification image indicates a portion of the sample tissue image that is similar to the sample tissue.

§3. When the tissue image is a color image
Embodiment In the above-described §2, an example in which the tissue image is a monochrome image has been described. However, actually, a more accurate result can be obtained by using a color tissue image. Therefore, here, a method of handling environmental factors when a color tissue image is used will be briefly described.

In this case, both the sample tissue image input in step S1 of FIG. 4 and the sample tissue image input in step S3 are color images. Usually, a color image on a computer is handled as density value data for each color component of the three primary colors. Therefore, as environmental factors indicating the characteristics of the texture of each tissue image, each color component of the three primary colors of each pixel in the predetermined reference region (here, the three primary colors of red R, green G, and blue B are used) May be used as the parameter value regarding the density value. For example, as shown in FIG. 9, a density value is defined for each color component of the three primary colors in each pixel constituting a reference area composed of 5 × 5 pixels. Here, Qr (i), Qg (i), Qb
(I) shows the density value of the red component, the density value of the green component, and the density value of the blue component of the i-th pixel, respectively. Therefore, the five types of parameters shown in FIG. 7 may be obtained separately and independently for each of these color components.

As described above, five kinds of parameters which are sample environmental factors for each color component are obtained for the sample tissue image, and similarly, five kinds of parameters which are sample environmental factors for each color component are obtained for the sample tissue image. In such a case, in order to determine the similarity between the sample environmental factor and the sample tissue image, it is necessary to compare 15 types of parameter values as shown in the table of FIG. For example, the average (A
ve. ), A parameter Qra relating to a red component, a parameter Qga relating to a green component, and a parameter Qba relating to a blue component are obtained as parameters constituting a sample environment factor. , A parameter QQga for the green component and a parameter QQba for the blue component.

In order to determine the similarity of the environmental factors composed of the parameters for the three color components obtained in this way, in the present embodiment, the angle difference between the three-dimensional vectors defined based on the parameter values for each color component is determined. I use it. For example, when comparing the average (Ave.) parameter values, first, based on the three primary color parameters Qra, Qga, and Qba constituting the sample environment factors, a three-dimensional vector V (as shown in FIG. Av
e. ) Is defined. This three-dimensional vector V (Ave.)
Is a vector defined on an RGB three-dimensional coordinate system in which the density values of the three primary colors R, G, and B are indicated as coordinate axes, and is a vector connecting the origin O and the point Q (Ave.). Here, the point Q (Ave.) is represented by coordinate values (Qra, Q
ga, Qba). Similarly, the three primary color parameters QQra, QQ constituting the sample environmental factors
Based on ga and QQba, a three-dimensional vector VV (Ave.) as shown in FIG. 11B is defined. The three-dimensional vector VV (Ave.) is also a vector defined on an RGB three-dimensional coordinate system in which the density values of the three primary colors R, G, and B are shown as coordinate axes, respectively, and has an origin O and a point QQ (Av.
e. ). Here, the point QQ (Av
e. ) Is a point at a position represented by coordinate values (QQra, QQga, QQba).

When two vectors V (Ave.) And VV (Ave.) Can be defined for the average (Ave.) parameter in this manner, the angle difference θ ( Ave.)
The angle difference θ (Ave.) is a value indicating the similarity in consideration of the three color components (the smaller the angle difference, the higher the similarity). Similarly, the variance (Va
r. ), Energy (Engy.), Entropy (E
nt. ) And contrast (Cont.) Are also defined as three-dimensional vectors, and the angle difference θ (Va
r. ), Angle difference θ (Engy.), Angle difference θ (En
t. ) And the angle difference θ (Cont.).
FIG. 12 shows an example in which the angle difference is obtained for each of the five types of parameters.

The overall similarity between the sample environmental factor and the sample environmental factor may be determined comprehensively based on these five types of angle differences. As described above, it is possible to quantitatively evaluate the similarity and determine the value of the ingested energy according to the quantitative value, but in the case of the embodiment shown here, to reduce the computational load, For the evaluation of similarity, only two types, “similar” and “dissimilar”, are prepared. In the case of “similar”, the intake energy value is set to 1200; And
In this embodiment, the following handling is performed in order to perform two types of evaluation. First, a value of 5 ° is determined as a threshold value of the angle difference θ. If the angle difference θ is less than 5 °, the parameter is determined to be “approximate” and the angle difference θ is 5 ° or more. In this case, it is determined that the parameter is “not approximate”. Such a judgment 5
When there are three or more parameters that are determined for each type of parameter and finally determined to be “approximately”, “similar” is determined as a comprehensive similarity determination between the sample environmental factor and the sample environmental factor. Is determined, and the parameter finally determined to be “approximately” is 2
If there is no more than the type, a comprehensive judgment of "not similar" is made.

§4. For moving / killing / proliferating virtual creatures
As described in §2, the moving process of the virtual creature is performed in step S5 in the flowchart shown in FIG. 4, and the death / proliferation process of the virtual creature is performed in step S7. Therefore, here, the movement /
Specific embodiments regarding the killing / proliferation process will be described.

First, a specific example of the process of moving a virtual creature will be described. As described above, each virtual creature has a unique movement characteristic factor (given at the time of occurrence of step S4), and in the movement processing of step S5, the movement characteristic factor of each virtual creature is considered. Will be done. In the present embodiment, a movement characteristic factor composed of 6 bits as shown in FIG. 13 is defined. Although the information consists of only 6 bits, the content of the movement characteristic factor differs for each virtual creature, so it is necessary to prepare 6 bits of information for each virtual creature. Actually, step S
In 4, at the time of generating the virtual creature, random 6-bit data may be given to each virtual creature using a random number.

Of the 6-bit movement characteristic factor, the first 2 bits indicate the movement characteristic X in the X direction, the center 2 bits indicate the movement characteristic Y in the Y direction, and the last 2 bits.
The bit indicates the rotation characteristic φ. The specific meaning of each bit is as shown in FIG. That is, regarding the movement characteristics, the bit “00” does not move, and the bit “0” does not move.
"1" moves in the + direction, bit "10" moves in the-direction,
Bit “11” indicates a random movement, and regarding the rotation characteristics, bit “00” does not rotate, bit “01” rotates 90 ° after the movement, bit “10” rotates 180 ° after the movement, and bit “10” rotates. 11 ”rotates 270 ° after moving,
(Here, the rotation angle is defined clockwise.)

Therefore, the movement characteristic factor composed of 6-bit data “010111” shown as a specific example in FIG. 13 moves in the + X direction, moves in the + Y direction, and rotates 270 ° after the movement. Is shown. Note that the concepts of the X direction and the Y direction regarding the movement characteristics are relative to the direction of the virtual creature, and are different from the absolute two-dimensional coordinate system defined on the specimen tissue image. is there. That is, the front direction of the current virtual creature is the + Y direction, the back direction is the −Y direction, and the right direction is + X.
The direction and the left direction are defined as -X directions. When the virtual creature rotates, each direction also rotates at the same time.
In other words, of the 6-bit movement characteristic factor, the first two bits indicate the movement characteristics of the virtual creature in the front-rear direction, and the center two bits indicate the movement characteristics of the virtual creature in the left-right direction. Will be.

FIG. 14 shows a process of actually moving a virtual creature based on the movement characteristics indicated by such 6-bit data “010111”. A black triangle P1 indicates the position and orientation of the virtual creature before moving. First, it is moved to the position of the white triangle P2 by movement in the + X direction (instruction of the first 2 bits), and then is moved in the + Y direction (instruction of the center 2 bits).
Finally, by performing a rotation of 270 ° (indicating the last two bits), a state of a white triangle P3 is obtained. The moving distance may be set in advance so as to advance by a predetermined number of pixels, or may be set to advance by a random number of pixels based on a random number each time (an upper limit of the moving distance is set in advance. To).

However, in practice, it is only necessary to be able to determine the final moving position of the virtual creature.
It is not necessary to separately perform the movement in the direction. Therefore, actually, the final moving position is determined by the following method. For example, the 6-bit data "01011
In the case of a virtual creature having the movement characteristic indicated by “1”, as shown in FIG. 15, a region of 7 × 7 pixels around the current position is extracted as a movable pixel.
Pixels located farther than this are too far to be reached by one movement. Here, the black triangle in the figure indicates the current position and orientation of the user. According to the movement characteristic factor, this virtual creature has a property of performing movement in the + X direction and movement in the + Y direction. Therefore, a candidate pixel that can be a final movement position to be reached when the movement according to such movement characteristics is performed is obtained. In the case of this example, the pixels indicated by circles in the figure are the candidate pixels. That is, the movement in the + X direction and the +
When both the movement in the Y direction and the movement in the Y direction are performed, only the nine candidate pixels indicated by the circles can be the final movement positions among the 7 × 7 movable pixels. Therefore, an arbitrary one pixel is randomly selected from these nine types of candidate pixels, and the selected pixel position is set as a final movement position.

When the final movement position is determined by such a method, it is possible to avoid a situation where a plurality of virtual creatures move on the same pixel at the same time. That is, among the candidate pixels obtained based on the moving characteristic factors,
If a pixel that has already been the final movement position of another virtual creature is included, avoid overlapping movement on the same pixel by excluding such a pixel from candidate pixels. Can be. Note that if such a measure to avoid the overlapping movement would result in all candidate pixels being excluded (for example,
In the example of FIG. 15, if all of the nine candidate pixels indicated by the circles have already been determined as the final movement positions of the other nine virtual creatures), What is necessary is just not to perform the movement of the time.

Next, a specific embodiment relating to the process of killing / proliferating virtual creatures performed in step S7 will be described.
As described above, whether to perform the process of killing / proliferating the virtual creature is determined based on the energy value remaining in the virtual creature at that time. That is, when the energy value updated by reducing the energy consumed by the movement and adding the intake energy given based on the similarity of the environmental factors reaches a predetermined lower limit (for example, 0), the animal has died. And a predetermined upper limit (for example, 20
00), it is regarded as having proliferated. As described above, when the killing / proliferation process is performed based only on the updated energy value, as long as the energy value fluctuates between the upper limit value and the lower limit value, neither the death nor the proliferation occurs. It will be a halfway state. Of course, virtual creatures that maintain such an incomplete state are expected to eventually settle into either a state of death or proliferation, but in practice, after a certain number of movements, If a virtual creature that remains in a half-finished state is positively removed, a stable result can be obtained more quickly.

From such a viewpoint, it is actually preferable to adopt the concept of death based on the number of times of movement in addition to death based on the energy value. Specifically, the age is defined for each virtual creature, and step S8 in FIG.
(In other words, each time the procedure of steps S5 to S7 is repeated), the age of each virtual creature is increased, and the virtual creature whose age has reached a predetermined life is removed as dead. What is necessary is just to add processing which performs. For example, the age of the virtual creature at the time of occurrence is 0 years old, and the age of one year old is increased every time the process goes through step S8.
The virtual creature that has reached the age of 80 may be removed as dead. When the concept of lifespan is adopted in this way, each virtual creature dies when its energy value reaches the lower limit (so-called starvation), and when it reaches its lifespan, it dies (so-called senile death). As a result, as described above, it is not possible to continue living for a long period of time in an incomplete state.

Next, a specific multiplication process will be described.
Proliferation processing is processing for the purpose of increasing the number of virtual creatures whose updated energy value has reached a predetermined upper limit,
Any method may be used as long as a new virtual creature having the same sample environmental factor as the virtual creature can be generated nearby. In the present embodiment, the following method is used.
This multiplication process is performed. That is, when the virtual creature is propagated, the parent virtual creature to be multiplied is removed, and a plurality of new child virtual creatures are generated in the vicinity of the parent virtual creature. A value is given, and the sample environment factor and the migration characteristic factor that the parent virtual creature has are inherited.

For example, as shown in FIG. 16, consider the case where the i-th virtual creature G (i) becomes the parent virtual creature to be multiplied (in the above example, this virtual creature G (i)).
The updated energy value of (i) has reached 2000). In this case, the parent virtual creature G (i) is 2
Virtual creature G
(I) is removed, and (i + 1) -th virtual creatures G (i + 1) a and G (i + 1) b, which are children, are generated. At this time, each child virtual creature G (i + 1) a,
G (i + 1) b has an initial energy value of 100
0 (or the parent virtual creature G
If the energy value of (i) exceeds 2,000, half of the value may be used as the initial energy value of each child.) The virtual environment G (i + 1) a and G (i + 1) b, which are children, are provided with the sample environment factor and the movement characteristic factor of the virtual creature G (i) as the parent. .

As described above, two children that inherit the property of the parent as they are are generated, and the generation position and orientation of these two children are determined on any pixel near the parent. What is necessary is just to determine at random so that it may turn to that direction. However, pixels in which other virtual creatures already exist are avoided. In the example shown in FIG. 16, a 7 × 7 neighborhood area is defined around the pixel where the parent exists, and a child oriented in an arbitrary direction is generated on any pixel in this neighborhood area. ing. Here, an example in which one parent is split into two children is shown, but it is needless to say that three parents or more may be split. Also, in the above example,
Although the parent has been removed after propagation, of course, a new child is generated with the parent left as it is (in this case, for example, processing such as returning the energy value to the initial value will be necessary) It is also possible to multiply in such a way that parents and children coexist.

When the above-described multiplication process is performed,
Since the child inherits the sample environmental factor of the parent as it is, generally, the environment in the vicinity of the generation position is a favorable environment adapted to itself. Further, since the parent's movement characteristic factor is inherited as it is, there is a high possibility that the parent's movement behavior toward a favorable environment is directly followed. In this way, a certain portion of the specimen tissue image has a tendency for virtual organisms having genes adapted to the environment to more and more proliferate. However, if descendants of a virtual creature having a certain gene always keep the same gene as the parent, there may be a case where the gene loses diversity and all descendants become extinct. Therefore, in the present embodiment, when the movement characteristic factor is inherited, the movement characteristic factor held by the parent virtual creature is suddenly changed at a predetermined probability. For example, the probability of sudden displacement is 1
When set to 0%, when the child is inherited the mobility characteristic factor, the child is inherited as it is up to nine times out of ten times, but in one of the ten times, the mobility characteristic factor is randomly changed. Will be inherited from. Of course, a sudden displacement can also be caused for the sample environmental factor.

§5. Implementation using multiple types of sample tissues
Each of the embodiments described above is an example in which a single type of sample tissue is used, and the same sample environmental factor is given to all virtual organisms. This is the case, for example, when recognizing a cell nucleus region of normal breast tissue as shown in FIG. 1 (a), or when recognizing a cell nucleus region of cancerous breast tissue as shown in FIG. 1 (b). Is effective. However, when it is necessary to recognize both of these simultaneously, two samples having different sample environmental factors are used.
Simulations using different types of virtual creatures are required. Practically, for example, when a specimen tissue image of a breast tissue is given, a normal cell nucleus area in the specimen tissue image is colored green, and a cancerous cell nucleus area is colored red. It would be convenient if the form became possible.

Here, a specific embodiment using two types of specimen tissues will be described with reference to the flowchart of FIG. The procedure shown in the flowchart of FIG. 17 is essentially the same as the procedure shown in the flowchart of FIG. Therefore, here, the differences between the two will be mainly described.

First, in step S11, a normal specimen tissue image and an abnormal specimen tissue image are input. For example, the cell nucleus region of normal breast tissue shown in FIG. 1A may be input as a normal specimen tissue image, and the cell nucleus region of cancerous breast tissue shown in FIG. 1B may be input as an abnormal specimen tissue image. . Then, in the following step S12, each sample environment factor indicating the characteristics of the normal sample texture and the abnormal sample texture is extracted. That is, based on the image of the cell nucleus region of the normal mammary gland tissue, a normal specimen environmental factor indicating the characteristic of the texture is extracted, and based on the image of the cell nucleus region of the cancerous mammary gland tissue, an abnormality indicating the characteristic of the texture is extracted. Sample environmental factors are extracted. Next, in step S13, a specimen tissue image is input. The purpose of each of the following procedures is to identify the specimen tissue image in order to recognize a portion approximating the cell nucleus region of a normal breast tissue sample and a portion approximating the cell nucleus region of a cancerous breast tissue sample. Creating an image.

Subsequently, a process of generating a large number of virtual creatures having a predetermined movement characteristic factor, a predetermined sample environment factor, and a predetermined initial energy value so as to be distributed on the specimen tissue image is performed. Done. Here, the movement characteristic factor may be randomly assigned using a random number, and the initial energy value may be, for example, uniformly defined as 1000. On the other hand, regarding the sample environmental factors, in the embodiment described here, it is necessary to selectively assign a normal sample environmental factor or an abnormal sample environmental factor. The virtual creature to which the normal specimen environmental factor has been assigned will have properties that match the environment of the image texture of the cell nucleus region of the normal breast tissue given as a specimen,
Eventually, they will collect and proliferate in a portion of the normal mammary tissue similar to the nucleus region. In contrast, a virtual creature to which an abnormal specimen environmental factor has been added will have properties that match the environment of the image texture of the cell nucleus region of cancerous mammary gland tissue given as a specimen, Proliferate by gathering in the portion of the transformed mammary tissue close to the cell nucleus region.

Whether to give a normal specimen environmental factor or an abnormal specimen environmental factor to each generated virtual creature is as follows.
For example, it is also possible to determine them completely at random using random numbers. However, in the embodiment described here, in order to further increase the efficiency of the simulation, it is determined which sample environmental factor should be given in consideration of the environment in the vicinity of the occurrence location of each virtual creature. . In other words, at the stage of generating each virtual creature, the sample environmental factors are not given yet, and after the generation, the sample environmental factors near the occurrence location are obtained, and the sample environmental factors and the normal sample environmental factors are abnormal. It is determined which one of the sample environment factors is similar to the sample environment factor, and a sample environment factor having a higher similarity is assigned to the virtual creature. If the sample environment factors are provided in this manner, each virtual creature is provided with a sample environment factor that is more likely to survive at each occurrence location, and the efficiency of the simulation can be increased.

Specifically, first, in step S14,
A process is performed to generate a virtual creature in which the movement characteristic factor and the initial energy value are defined so as to be distributed on the sample tissue image. In the subsequent step S15, for each generated virtual creature, The process of assigning the sample environmental factor is performed. In this step S15, a sample environment factor showing the texture characteristic of the sample tissue image around the existence position of each virtual creature is examined, and the similarity to the examined sample environment factor among the normal sample environment factor and the abnormal sample environment factor is examined. Is given to the virtual creature.

Hereinafter, the processing of steps S16 to S18
It is repeatedly executed until the end condition of step S19 is satisfied. Here, the processing in steps S16 to S18 is as follows:
This corresponds to the processing of steps S5 to S7 in the flowchart of FIG. That is, in step S16, the virtual creature is moved on the specimen tissue image based on the movement characteristic factor, and a process of determining the energy consumption consumed by this movement is performed. In step S17,
Based on the similarity between the sample environment factor indicating the texture characteristic of the sample tissue image around the position after the movement of the virtual creature and the sample environment factor of the virtual creature, the intake energy value to be taken by the virtual creature is calculated. In step S18, the energy value is updated by subtracting the consumed energy value from the energy value of the virtual creature and adding the intake energy value to the energy value of the virtual creature. When the value reaches the value, the virtual creature is removed as being dead, and when the updated energy value reaches a predetermined upper limit, the virtual creature is assumed to have proliferated and has a new sample environmental factor having the same environmental factor. A process for generating a virtual creature is performed. Of course, when performing the processing of step S18, as described in §4,
The killing process may be performed in consideration of the life of each virtual creature.

The sample environmental factor to be subjected to the similarity determination in step S17 is one of the sample environmental factors given to the virtual creature in step S15. For example, the sample environmental factor of the virtual creature having the normal sample environmental factor is In such a case, if the similarity between the normal sample environmental factor of the subject and the sample environmental factor at the current position is high, the energy intake can be obtained, and even if the virtual organism has an abnormal sample environmental factor. For example, if the similarity between the abnormal sample environmental factor of the subject and the sample environmental factor at the current position is high, it is possible to acquire the intake energy. Note that, as in the above-described embodiment, when processing using two types of mutually opposite sample environmental factors is performed (for example, when one is an environmental factor relating to normal cells and the other is an environmental factor relating to cancerous cells). ), The virtual creature based on both the similarity between the sample environmental factor and the sample environmental factor of the virtual creature and the dissimilarity between the sample environmental factor that the virtual creature does not have and the sample environmental factor. May determine the intake energy value to be taken.

For example, when determining the intake energy value of a virtual creature having a normal sample environmental factor, the similarity between the normal sample environmental factor possessed by the subject and the sample environmental factor is determined, and the abnormal sample environmental factor not possessed by the subject is determined. And the dissimilarity of the sample environmental factors. The higher the similarity and dissimilarity obtained in this way, the higher the energy intake, the greater the similarity to the sample environmental factor that the user has and the sample environmental factor that the user does not have. Can be determined in consideration of both the dissimilarity of the energy and the intake energy value, and a more accurate simulation can be performed. For example, for both the sample environment factors that happen to have oneself and the sample environment factors that one does not have,
If the specimen environment is somewhat similar, the virtual creature is not particularly in a self-adapted environment. In such a case, similarity to the sample environmental factor owned by the user is recognized, but dissimilarity to the sample environmental factor not owned by the user is not recognized. Can be taken.

If the processes of steps S16 to S18 are repeatedly executed many times, the virtual creature having the normal specimen environmental factor and the virtual creature having the abnormal specimen environmental factor will eventually gather in the environmental parts suitable for each. They will breed. Therefore, finally, in step S20, an identification image of a normal sample tissue is created based on the distribution of virtual organisms having a normal sample environmental factor that finally survives on the sample tissue image (for example, a normal image). Based on the distribution of virtual organisms with abnormal sample environmental factors that finally survive on the specimen tissue image, the area where the virtual organisms with sample environmental factors are distributed is displayed in green.
An identification image of the abnormal specimen tissue may be created (for example, a region where virtual creatures having abnormal specimen environmental factors are distributed is displayed in red).

§6. Other Useful Embodiments Finally, a method for enabling a more practical simulation by adding predetermined processing to the various embodiments described above will be described.

First, a process for lowering the survival probability of the virtual creature near the boundary of the region will be described. An object of the present invention is to recognize an area similar to a specimen tissue on a given specimen tissue image, so that a large number of virtual creatures are finally gathered in a specific area. Is preferred. However, in the embodiments described above, since the movement characteristic factor of each virtual creature is determined at random, even though the person arrived at an environment suitable for himself, he again moved to an environment unsuitable for himself. There are not a few cases where they do. For this reason, there is a possibility that the result of the simulation does not converge forever. Therefore, it is preferable to add a process for lowering the survival probability of the virtual creature existing near the boundary of the region to suppress the phenomenon that the virtual creature flows out of the boundary.

More specifically, a near reference area centered on the location of each virtual creature and a distant reference area including the near reference area and further extending the boundary to a farther place are defined. When the similarity between the nearby environment factor indicating the texture feature and the distant environment factor indicating the texture feature in the distant reference area is low, the energy value of the virtual creature or the energy value of the virtual creature ingested May be further added.

For example, in the example shown in FIG. 18, 5 ×
A near reference area A1 composed of an area of 5 pixels and a distant reference area A2 composed of an area of 7 × 7 pixels are defined.
Therefore, a nearby environment factor indicating the feature of the texture in the near reference region A1 and a far environment factor indicating the feature of the texture in the far reference region A2 are obtained. This can be done by calculating the parameter values as shown in FIG. 7 using the density values of the pixels in each area. When the similarity between the nearby environmental factor and the distant environmental factor is low, the virtual creature is charged with extra energy that is different from the energy required for moving, or Measures are taken to reduce the intake energy value that the virtual creature originally intended to acquire.

In the example shown in FIG. 18, since a boundary as an image exists between the pixels in the second row and the pixels in the third row, the illustrated virtual creature exists near the boundary.
As described above, in the virtual creature near the boundary, the similarity between the nearby environmental factor and the distant environmental factor obtained by the above-described method is reduced. For example, in the example of FIG. 18, since the density values of the pixels located in the upper two rows affect the calculation of the distant environment factor, a parameter value considerably different from the calculation result of the nearby environment factor is obtained. Will be. As described above, when the similarity between the nearby environmental factor and the distant environmental factor is low, it is determined that the virtual creature exists near the boundary, and the process of reducing the survival probability is performed as described above. What should I do?

Next, an additional process focusing on the contrast among the environmental factors will be described. As shown in FIG. 7, a parameter called contrast (Cont.) Is used as one of the parameters constituting the environmental factor. The additional processing described here can be said to be a processing of comparing the contrast value between the sample tissue image and the sample tissue image and, when the absolute values are significantly different, reducing the survival probability of the virtual creature. The main purpose is to kill virtual creatures existing in an area different from the original recognition target area at an early stage. For example, suppose that the image shown in FIG. 1A is given as a specimen tissue image, and a cell nucleus region is a region to be recognized. In this case, as shown in the figure, it can be seen that there is a large difference in contrast between the cell nucleus region and the background region. In general biological tissues, there is often such a large contrast difference between the cell nucleus region and the background region. Therefore, in a simulation in which a cell nucleus region is to be recognized (in other words, in a simulation in which a cell nucleus region is provided as a sample tissue), a process of thinning out virtual creatures existing in a background portion region at a stretch based on a contrast value is performed. It is effective.

As described in §3, when the tissue image is a color image, the similarity is determined based on the angle difference between the three-dimensional vectors defined based on the parameter values for each color component. Was. However, the similarity determination based on such an angle difference is not always an appropriate determination. For example, when two three-dimensional vectors V1 and V2 are considered as shown in FIG.
2 functions as one index indicating the similarity between the two vectors V1 and V2. That is, the smaller the angle difference θ12 is, the more similar the environmental factors represented by the two vectors V1 and V2 are generally. However, when there is a large difference between the absolute values of the two vectors V1 and V2, it may be more appropriate to determine that both environmental factors are dissimilar even if the angle difference θ12 is small. In particular, the contrast (Con
t. This tendency is remarkable with regard to the parameter).

For example, in the example shown in FIG. 20, three-dimensional vectors V (Cont.) And VV (Cont.) Defined based on parameters indicating the contrast of the three primary colors.
FIG. Here, the vector V (Cont.) Is a vector indicating the sample environmental factor, and the vector VV (Ct.
ont. ) Is a vector indicating a sample environmental factor. Although the angle difference θ (Cont.) Between the two vectors is quite small, the absolute value of the two vectors has a large difference, and in fact, the overall contrast between the two greatly differs. Therefore,
For example, as shown, the vector VV (Cont.)
), The absolute value of the difference vector D (Cont.) Can be used as an index indicating the overall contrast difference. The point of the method described here is to perform a process of thinning out virtual creatures existing in the background area at a stretch based on the absolute value of the difference vector D (Cont.).

Specifically, the following method may be adopted. First, the absolute value of the difference between the contrast value of the sample tissue image around the position where the virtual creature exists and the contrast value defined as the sample environment factor of the virtual creature is
Ask for all living virtual creatures. In practice, the absolute value of D (Cont.) Of the above-described difference vector may be obtained. Then, for a certain percentage of virtual creatures in the descending order of the absolute value of D (Cont.) Of the difference vector, a process of reducing the energy value of the virtual creature or the energy value taken by the virtual creature is performed. I do. In practice, as shown in the table of FIG.
nt. ), Sort all virtual creatures by magnitude of absolute value,
The energy value of the virtual creature corresponding to the upper fixed ratio may be reduced or the energy value to be consumed may be reduced. For example, when the number of all virtual creatures is k and it is determined that the processing to reduce the energy value is performed on the virtual creatures corresponding to the ratio of 50%, the difference vector D is calculated based on the table of FIG. (Co
nt. ) From the virtual creature with the largest absolute value to the (k / 2) th virtual creature in total (k / 2)
Energy reduction processing may be performed on one virtual creature.

By such processing, the energy value of a virtual creature existing in a region different from the region to be originally recognized can be reduced, and death can be promoted. Of course, in the table shown in FIG. 21, a setting may be made so that the amount of reduction in energy is increased for a virtual creature ranked higher, and for a virtual creature with a specific rank or higher. It is permissible to carry out a process that kills immediately.

Although the biological tissue recognition method using image processing according to the present invention has been described based on several embodiments, the present invention is not limited to these embodiments. Can be implemented in various forms. Further, some embodiments described so far may be executed in any combination. The biological tissue recognition method using image processing according to the present invention is based on the premise that processing is performed using a computer.
The procedure shown in the flowchart shown in FIG. 7 is executed by a computer. Therefore, in order for a computer to execute each of these procedures, a predetermined program is prepared, but the program may be recorded on a computer-readable recording medium such as a magnetic disk or an optical disk and distributed. It is possible, and it is also possible to transfer via a network.

[0088]

As described above, according to the biological tissue recognition method using image processing according to the present invention, a specific tissue can be identified by performing image processing using a computer on a two-dimensional image of biological tissue. Automatic recognition becomes possible.

[Brief description of the drawings]

FIG. 1 is a diagram showing an example of images of normal breast tissue and cancerous breast tissue of an animal.

FIG. 2 is a diagram showing a state in which only each cell nucleus region is recognized from the tissue image shown in FIG. 1, and the recognized cell nucleus region is colored and displayed in a predetermined color.

FIG. 3 is a model diagram illustrating a basic concept of the present invention for recognizing a biological tissue by simulation using a virtual creature.

FIG. 4 is a flowchart showing a basic flow of an embodiment of the present invention using a single type of specimen tissue.

FIG. 5 is a diagram showing an operation of performing an input for designating several points inside a cell nucleus region on an image of a normal breast tissue.

FIG. 6 is a diagram showing a pixel configuration in a reference area defined near one point on an image of a biological tissue.

FIG. 7 is a diagram illustrating an example of parameter values to be used as environmental factors for a texture in a reference area including a plurality of pixels.

FIG. 8 is a table showing a comparison between a sample environmental factor composed of five parameters and a sample environmental factor.

FIG. 9 is a diagram showing a color pixel configuration in a reference region defined near one point on an image of a biological tissue.

FIG. 10 is a table showing a comparison between a sample environment factor and a sample environment factor each including five parameters for each color component of the three primary colors.

FIG. 11 is a diagram illustrating an example of a method of performing a similarity determination between a sample environment factor and a sample environment factor including parameters for each color component of three primary colors.

FIG. 12 is a diagram illustrating an example of a technique for performing a similarity determination between a sample environment factor and a sample environment factor including five parameters for each color component of three primary colors.

FIG. 13 is a diagram showing an example in which a movement characteristic factor given to each virtual creature is composed of 6-bit data.

FIG. 14 is a view showing a movement mode of a virtual creature based on the movement characteristic factors shown in FIG.

15 is a diagram showing a specific method of moving a virtual creature based on the movement characteristic factors shown in FIG.

FIG. 16 is a diagram showing one mode of propagation of a virtual creature.

FIG. 17 is a flowchart showing a basic flow of an embodiment of the present invention using two types of specimen tissues.

FIG. 18 is a diagram illustrating a procedure for recognizing a virtual creature located near a boundary of a region.

FIG. 19 is a diagram illustrating a method of determining the similarity of environmental factors as a three-dimensional vector angle difference.

FIG. 20 is a diagram illustrating an example of a technique for obtaining a difference between absolute values of three-dimensional vectors.

FIG. 21 is a diagram illustrating an example of a technique for reducing energy to a virtual creature existing in an environment in which the absolute value of contrast is significantly different from that of its own environmental factor.

[Explanation of Codes] A1: near reference area A2: far reference area Ave. ... Parameter indicating the average of the pixel values constituting the texture Cont. ... A parameter indicating the contrast of the pixel values constituting the texture D (Cont.)... Vector VV (Cont.) And V
(Cont.) With the difference vector Engy. ... a parameter indicating the energy of the pixel value constituting the texture Ent. ... parameter G (i) ... i-th generation virtual creature G (i + 1) a, G (i + 1) b ... (i + 1) th generation virtual creature Var. ... Parameters indicating the variance of pixel values constituting the texture P1, P2, P3... Positions of virtual creatures Q (Ave.), QQ (Ave.)... Vector end points Q (Cont.), QQ (Cont.). Vector end point Q (Engy.), QQ (Engy.) ... Vector end point Q (Ent.), QQ (Ent.) ... Vector end point Q (Var.), QQ (Var.) ... Vector Tip point Q (1) to Q (25): density value of each pixel Qb (1) to Qb (25): density value of blue component of each pixel Qg (1) to Qg (25): green component of each pixel Qr (1) to Qr (25): density value of red component of each pixel Qa, Qra, Qga, Qba... Average value of pixel values constituting texture of sample tissue image Qc, Qrc, Qgc, Qbc. Texture of specimen tissue image Contrast values of constituent pixel values Qn, Qrn, Qgn, Qbn: Energy values of pixel values forming texture of sample tissue image Qe, Qre, Qge, Qbe: Entropy values of pixel values forming texture of sample tissue image Qv , Qrv, Qgv, Qbv: variance values of pixel values constituting the texture of the sample tissue image QQa, QQra, QQga, QQba: average value of the pixel values constituting the texture of the sample tissue image QQc, QQrc, QQgc, QQbc: sample Contrast values of pixel values constituting the texture of the tissue image QQn, QQrn, QQgn, QQbn... Energy values of the pixel values constituting the texture of the sample tissue image QQe, QQre, QQge, QQbe... Pixels constituting the texture of the sample tissue image Value entropy value QQv, QQrv , QQgv, QQbv: variance values of pixel values constituting the texture of the sample tissue image V1, V2 ... three-dimensional vector V (Ave.): Vector V (Cont. )... Vector V (Energy.) Indicating the contrast of the pixel values constituting the texture of the sample tissue image V. Vector (Ent.) Indicating the energy of the pixel values forming the texture of the sample tissue image Vector V (Var.) Indicating the entropy of the constituent pixel values. Vector VV (Ave.) indicating the variance of the pixel values forming the texture of the sample tissue image. Indicates the average of the pixel values forming the texture of the sample tissue image. Vector VV (Cont.): Indicates the contrast of pixel values constituting the texture of the sample tissue image Vector VV (Engy. )... Vector VV (Ent.) Indicating the energy of the pixel value forming the texture of the sample tissue image V.sub.V (Var.) Indicating the entropy of the pixel value forming the texture of the sample tissue image A vector indicating the variance of the pixel values constituting the pixel X: a parameter indicating the movement characteristic of the virtual creature in the front-rear direction Y: a parameter indicating the movement characteristic of the virtual creature in the left-right direction φ: a parameter indicating the rotation characteristic of the virtual creature θ12: the vector V1, Angle difference between V2 θ (Ave.): Angle difference between both vectors θ (Cont.): Angle difference between both vectors θ (Engy.): Angle difference between both vectors θ (Ent.): Angle difference between both vectors θ (Var.): Angle difference between both vectors

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Yasunori Fujisawa 14-8 Tokiwacho, Muroran-shi, Hokkaido F-term (reference) 2G045 AA24 AA26 CB01 CB02 FA19 GB10 JA01 JA06 4B063 QA01 QA07 QA08 QQ08 QQ42 QQ52 QQ58 QR33 QR82 QS02 QX015 AA07 BA28 CA01 CB01 CC03 CD02 CD03 CE20 CH11 DC01 DC32

Claims (12)

[Claims] 1. An image of a biological tissue as a specimen is subjected to predetermined image processing using a computer,
What is claimed is: 1. A method for obtaining an identification image capable of distinguishing and recognizing a portion approximating a specific sample tissue, comprising: a step of inputting a sample tissue image from a specific sample tissue; and Extracting a sample environment factor indicating a characteristic of the texture of the sample, a step of inputting a sample tissue image from a sample tissue, and having a movement characteristic factor indicating a predetermined movement characteristic and the sample environment factor, a predetermined initial energy value. Generating a plurality of virtual creatures defined as being distributed on the specimen tissue image; and (a) moving the virtual creature on the specimen tissue image based on the movement characteristic factor thereof; (B) a sample environment factor indicating the texture characteristic of the sample tissue image around the position after the movement of the virtual creature; A process of determining an energy value to be consumed by the virtual creature based on the similarity to the sample environmental factor of the creature; (c) reducing the consumed energy value to the energy value of the virtual creature, By adding the energy value, the energy value is updated, and when the updated energy value reaches the predetermined lower limit, the virtual creature is removed as dead and the updated energy value is set to the predetermined upper limit. When the value reaches the value, a process of generating a new virtual creature having the same sample environmental factor in the vicinity as that the virtual creature has multiplied, the following three processes (a), (b), and (c), Repeatedly executing a predetermined number of times, and generating a target identification image based on a distribution of virtual living organisms finally living on the specimen tissue image, an image characterized by comprising: Biological tissue recognition method using the physical. 2. An image of a biological tissue as a specimen is subjected to predetermined image processing using a computer,
This is a method of obtaining an identification image that allows a portion similar to a specific sample tissue to be distinguished and recognized.A normal sample tissue image is input from a specific normal sample tissue, and an abnormal sample tissue image is input from a specific abnormal sample tissue. Inputting, based on the input normal sample tissue image, and extracting a normal sample environmental factor indicating the characteristics of the texture of the normal sample tissue, based on the input abnormal sample tissue image,
Extracting an abnormal sample environment factor indicating a texture characteristic of the abnormal sample tissue; inputting a sample tissue image from the sample tissue; having a transfer characteristic factor indicating a predetermined transfer characteristic, and having a predetermined initial energy value. Generating a plurality of defined virtual creatures so as to be distributed on the sample tissue image; and for each generated virtual creature, characterizing the texture of the sample tissue image around the existence position of the virtual creature. Examining the sample environment factor to be shown, of the normal sample environment factor and the abnormal sample environment factor, providing a sample environment factor having a higher similarity to the sample environment factor examined to the virtual organism; A) moving a virtual creature on the specimen tissue image based on the movement characteristic factor, and determining a value of energy consumption consumed by the movement; Based on the similarity between the sample environment factor indicating the texture characteristic of the sample tissue image around the position after the movement of the virtual creature and the sample environment factor of the virtual creature, the intake energy value to be taken by the virtual creature (C) reducing the consumed energy value with respect to the energy value of the virtual creature and adding the ingested energy value to update the energy value, and the updated energy value becomes a predetermined lower limit value If the virtual creature reaches the upper limit, the virtual creature is removed as dead, and if the updated energy value reaches a predetermined upper limit, it is assumed that the virtual creature has grown and a new one having the same sample environmental factor is obtained. Repeating the three processes (a), (b), and (c) a predetermined number of times to generate virtual creatures in the vicinity; Based on the distribution of virtual organisms having normal sample environmental factors, a discrimination image for a normal sample tissue is created, and a virtual sample having an abnormal sample environmental factor that finally survives on the sample tissue image is prepared. Creating an identification image of the abnormal specimen tissue based on the distribution of organisms, and a biological tissue recognition method using image processing. 3. The biological tissue recognition method according to claim 2, wherein in the processing (b), the similarity between the sample environment factor and the sample environment factor of the virtual organism and the sample environment without the virtual organism. A biological tissue recognition method using image processing, wherein an intake energy value to be taken by the virtual creature is determined based on both the factor and the dissimilarity between the sample environment factor. 4. The biological tissue recognition method according to any one of claims 1 to 3, wherein:
The average, variance, energy, entropy, and density of each pixel in a predetermined reference area are determined by using a plurality of parameter values selected from the group consisting of contrasts. A biological tissue recognition method using image processing, wherein the biological tissue recognition method is determined based on the similarity of parameter values. 5. The biological tissue recognition method according to claim 1, wherein a color image is used as each tissue image, and a predetermined reference area is used as an environmental factor indicating a texture characteristic of the tissue image. Parameter values relating to the density values of the three primary colors of each pixel in each of the three primary colors are used. And a biological tissue recognition method using image processing. 6. The biological tissue recognizing method according to claim 1, wherein a parameter indicating a movement characteristic of the virtual creature in the front-rear direction, a movement characteristic in the left-right direction, and a rotation characteristic is used as the movement characteristic factor. A biological tissue recognition method using image processing, wherein a randomly determined moving characteristic factor is added when each virtual creature is generated. 7. The biological tissue recognition method according to claim 1, wherein an age is defined for each virtual creature, and three processes are performed.
Each time (a), (b), or (c) is performed, the age of each virtual creature is increased, and virtual creatures whose age has reached a predetermined life are removed as dead. Biological tissue recognition method using image processing. 8. The biological tissue recognizing method according to claim 1, wherein, in the step (c), when the virtual creature is propagated, the parent virtual creature to be proliferated is removed, and the vicinity thereof is removed. To generate a plurality of new child virtual creatures, giving the child virtual creatures a predetermined initial energy value and inheriting the sample environmental factors and migration characteristic factors that the parent virtual creatures had. A biological tissue recognition method using image processing as a feature. 9. The biological tissue recognizing method according to claim 8, wherein when the parental virtual creature inherits the movement characteristic factor, a sudden displacement occurs at a predetermined probability, and the sudden displacement occurs. In this case, a biological tissue recognition method using image processing is characterized in that a moving characteristic factor is randomly changed and then inherited. 10. The biological tissue recognizing method according to claim 1, wherein a near reference area centered on the existence position of each virtual creature, and a boundary including the near reference area and further extending the boundary to a further distant place. Define a distant reference region, the nearby environment factor indicating the characteristics of the texture in the near reference region, and the distant environment factor indicating the characteristics of the texture in the far reference region, if the similarity is low, A biological tissue recognition method using image processing, further comprising a process of reducing the energy value of the virtual creature or the energy value of the virtual creature. 11. The biological tissue recognition method according to any one of claims 1 to 10, wherein:
Define a contrast value for the density value of each pixel in the predetermined reference region, the contrast value of the sample tissue image around the existence position of the virtual living thing, and the contrast value defined as a sample environmental factor of the virtual living thing A process of calculating the absolute value of the difference for all living virtual creatures, and reducing the energy value of the virtual creature or the energy value of the virtual creature for a certain percentage of the virtual creatures in the descending order of the absolute value. A biological tissue recognition method using image processing, characterized by further adding. 12. A computer-readable recording medium on which a program for causing a computer to execute each step constituting the biological tissue recognition method using the image processing according to claim 1 is recorded.