JP5740101B2 - Cell evaluation apparatus, incubator, cell evaluation method, cell evaluation program, and cell culture method - Google Patents

Description

  The present invention relates to a cell evaluation apparatus, an incubator, a cell evaluation method, a cell evaluation program, and a cell culture method.

  In general, techniques for evaluating the culture state of cells have become fundamental technologies in a wide range of fields including advanced medical fields such as regenerative medicine and drug screening. For example, in the field of regenerative medicine, there are processes for growing and differentiating cells in vitro. And in said process, in order to manage the success or failure of cell differentiation, the canceration of a cell, and the presence or absence of infection, it is indispensable to evaluate the culture state of a cell exactly. As an example, a cancer cell evaluation method using a transcription factor as a marker has been disclosed (see Patent Document 1).

  On the other hand, universal cells such as ES (Embryonic Stem) cells or iPS (Induced Pluripotent Stem) cells are theoretically infinite while maintaining the pluripotency that differentiates into all tissues. Because of its ability to grow rapidly, attention has been focused on pharmaceutical development and application to regenerative medicine.

JP 2007-195533 A

  When applying iPS cells to regenerative medicine, it is necessary to extract only iPS cells in good condition, and therefore it is necessary to evaluate the state of the cultured iPS cells. However, the conventional technique has a problem that the state of iPS cells cannot be evaluated without staining iPS cells with a staining marker.

  Therefore, the present invention has been made in view of the above problems, and provides a cell evaluation apparatus, an incubator, a cell evaluation method, a cell evaluation program, and a cell culture method that estimate the state of a cell without staining the cell. This is the issue.

  In order to solve the above problem, the present invention provides a target image in which a first type of cells and a second type of cells that are present around the first type of cells and are cultured together are imaged. A region extracting unit for extracting a region occupied by the first type of individual cells and cell aggregates and a region occupied by the second type of cells; and a second region extracted by the region extracting unit. A first feature amount calculation unit for calculating a feature amount indicating a morphological feature of the second type of cell existing in an area occupied by the type of cell; and a morphological feature of the second type of cell. A cell evaluation apparatus comprising: an evaluation processing unit that evaluates a state of a first type of cell based on a correspondence relationship between a feature amount and the state of the first type of cell.

  The present invention also includes a temperature-controlled room that accommodates a culture container for culturing cells and that can maintain the interior in a predetermined environmental condition, and an imaging device that captures an image of the cell contained in the culture container in the temperature-controlled room And a cell evaluation device according to any one of claims 1 to 7.

  Further, the present invention provides the first image in the target image obtained by imaging the first type of cells and the second type of cells that are present around the first type of cells and are cultured together. A region extraction procedure for extracting a region occupied by individual types of cells and cell aggregates and a region occupied by the second type of cells, and a region occupied by the second type of cells extracted in the region extraction procedure A first feature amount calculation procedure for calculating a feature amount indicating a morphological feature of the second type of cells existing in the second type cell, a feature amount indicating a morphological feature of the second type of cell, An evaluation processing procedure for evaluating a state of a first type of cell based on a correspondence relationship with a state of one type of cell.

  Further, the present invention provides the first image in the target image obtained by imaging the first type of cells and the second type of cells that are present around the first type of cells and are cultured together. A first step of extracting an area occupied by individual types of cells and cell aggregates and an area occupied by the second type of cells; and the second type of cells extracted in the first step A second step of calculating a feature amount indicating a morphological feature of the second type of cells present in the occupied region, a feature amount indicating a morphological feature of the second type of cells, and the first A cell evaluation program for causing a computer to execute the third step of evaluating the state of the first type of cell based on the correspondence relationship with the state of the type of cell.

  The present invention also provides the second type of cell from an image obtained by imaging the first type of cells and the second type of cells that are present around the first type of cells and are cultured together. A feature amount indicating a morphological feature of a cell is extracted, a state of the first type cell is estimated based on a feature amount indicating a morphological feature of the second type cell, and the state of the first cell The cell culturing method is characterized in that the culture conditions of the first type of cells or the second type of cells are changed.

  According to the present invention, the state of a cell can be estimated without staining the cell.

It is a front view of the incubator which is an embodiment of the present invention. It is a top view of the incubator which is an embodiment of the present invention. It is a block block diagram of the incubator which is 1st Example of this invention. It is a figure for demonstrating the process which extracts the area | region of the colony of an iPS cell, and the area | region of a feeder cell from the image by which the feeder cell co-cultured with the iPS cell was imaged. It is a figure for demonstrating the area | region of the colony of the extracted iPS cell, and the area | region of a feeder cell. It is a figure for demonstrating each feature-value of the form of a cell. It is a figure for demonstrating the table which stores the parameter of the calculation model about a certain feature-value. It is a figure for demonstrating the outline | summary of Artificial Neural Network. It is a figure for demonstrating the outline | summary of a Fuzzy Neural Network. It is a figure for demonstrating the sigmoid function in Fuzzy Neural Network. It is a flowchart which shows an example of the observation process in an incubator. It is a flowchart for demonstrating calculation of the teacher value for a calculation model. It is a flowchart for demonstrating construction of a calculation model. It is a flowchart for estimating the differentiation degree of the iPS cell using a calculation model. It is a block block diagram of the incubator which is 2nd Example of this invention. It is a figure for demonstrating the decision tree of a colony + feeder model according to an image. It is a figure for demonstrating the decision tree of the model classified by image and colony feeder. It is a figure for demonstrating the decision tree of the model classified by cell and colony feeder. It is a flowchart for demonstrating calculation of the teacher value for decision tree calculation. It is a flowchart for demonstrating calculation of the feature-value for decision tree calculation. It is a flowchart for demonstrating calculation of the decision tree according to an image. It is a flowchart for demonstrating the decision tree calculation for every cell. It is a flowchart for demonstrating the estimation of the state of the iPS cell using the image and colony + feeder model shown in FIG. It is a flowchart for demonstrating the estimation of the state of the iPS cell using the image classification and colony feeder classification model shown in FIG. It is a flowchart for demonstrating the estimation of the state of the iPS cell using the cell classification and colony feeder classification model shown in FIG. It is a flowchart for demonstrating the culture | cultivation method of the cell using the cell evaluation program of Example 2 of this invention.

  Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the drawings. In the embodiment of the present invention, iPS cells and one type of feeder cells that exist around iPS cells and supply nutrients and the like to iPS cells are cultured together.

<Example 1>
In general, ES cells or iPS cells are considered to have better quality as their differentiation ability increases. Therefore, in Example 1 of the present invention, in order to estimate the state of iPS cells, the ratio of differentiated cells in the total cells (hereinafter referred to as the degree of differentiation) is used as an index.
FIG. 1 is a front view of an incubator according to an embodiment of the present invention. FIG. 2 is a plan view of the incubator according to the embodiment of the present invention.

A large door 16, a middle door 17 and a small door 18 are arranged on the front surface of the temperature-controlled room 15. Large door 16
Covers the front surfaces of the upper casing 12 and the lower casing 13. The middle door 17 covers the front surface of the upper casing 12 and isolates the environment between the temperature-controlled room 15 and the outside when the large door 16 is opened. The small door 18 is a door for carrying in and out a culture vessel 19 for culturing cells, and is attached to the middle door 17. It is possible to suppress environmental changes in the temperature-controlled room 15 by carrying the culture container 19 in and out of the small door 18. Large door 16, middle door 17, small door 18
The airtightness is maintained by the packings P1, P2, and P3, respectively.

In the temperature-controlled room 15, a stocker 21, an observation unit 22, a container transfer device 23, and a transfer table 24 are arranged. Here, the conveyance stand 24 is disposed in front of the small door 18, and carries the culture container 19 in and out of the small door 18.

The stocker 21 is disposed on the left side of the temperature-controlled room 15 when viewed from the front surface of the upper casing 12 (the lower side in FIG. 2). The stocker 21 has a plurality of shelves, and each shelf of the stocker 21 can store a plurality of culture vessels 19. Each culture container 19 contains cells to be cultured together with a medium.

Here, the observation unit 22 is disposed by being fitted into the opening of the base plate 14 of the upper casing 12. The observation unit 22 includes a sample stage 31, a stand arm 32 protruding above the sample stage 31, and a main body portion 33 incorporating a microscopic optical system for phase difference observation and an imaging device (34). . The sample stage 31 and the stand arm 32 are disposed in the temperature-controlled room 15, while the main body portion 33 is accommodated in the lower casing 13.

The observation unit 22 is disposed on the right side of the temperature-controlled room 15 when viewed from the front surface of the upper casing 12. The observation unit 22 can execute time-lapse observation of cells in the culture vessel 19.

Here, the observation unit 22 is disposed by being fitted into the opening of the base plate 14 of the upper casing 12. The observation unit 22 includes a sample stage 31, a stand arm 32 protruding above the sample stage 31, and a main body portion 33 incorporating a microscopic optical system for phase difference observation and an imaging device (34). . The sample stage 31 and the stand arm 32 are disposed in the temperature-controlled room 15, while the main body portion 33 is accommodated in the lower casing 13.

The sample stage 31 is made of a translucent material, and the culture vessel 19 can be placed thereon. The sample stage 31 is configured to be movable in the horizontal direction, and the position of the culture vessel 19 placed on the upper surface can be adjusted. The stand arm 32 includes an LED light source 35. And the imaging device 34 acquires the microscope image of a cell by imaging the cell of the culture container 19 permeate | transmitted and illuminated from the upper side of the sample stand 31 with the stand arm 32 through the optical system of a microscope.

The container transport device 23 is disposed in the center of the temperature-controlled room 15 when viewed from the front surface of the upper casing 12. The container transport device 23 delivers the culture container 19 between the stocker 21, the sample table 31 of the observation unit 22, and the transport table 24.

  As shown in FIG. 2, the container transport device 23 includes a vertical robot 34 having an articulated arm, a rotary stage 35, a mini stage 36, and an arm unit 37. The rotary stage 35 is attached to the tip of the vertical robot 34 via a rotary shaft 35a so as to be capable of rotating 180 ° in the horizontal direction. Therefore, the rotary stage 35 can make the arm portions 37 face the stocker 21, the sample table 31, and the transport table 24.

  The mini stage 36 is attached to the rotation stage 35 so as to be slidable in the horizontal direction. An arm part 37 that holds the culture vessel 19 is attached to the mini stage 36.

FIG. 3 is a block diagram of the incubator according to the first embodiment of the present invention.
The incubator 11 of the first embodiment has an upper casing 12 and a lower casing 13. In the assembled state of the incubator 11, the upper casing 12 is placed on the lower casing 13. Note that the internal space between the upper casing 12 and the lower casing 13 is vertically divided by a base plate 14.

The upper casing 12 includes a temperature adjustment device 15a, a humidity adjustment device 15b, a container transport device 23, and an LED light source 35.
First, an outline of the configuration of the upper casing 12 will be described. Inside the upper casing 12,
A temperature-controlled room 15 for culturing cells is formed. The temperature-controlled room 15 includes a temperature adjusting device 15a and a humidity adjusting device 15b.

The temperature adjusting device 15a and the humidity adjusting device 15b maintain the inside of the temperature-controlled room 15 in an environment suitable for cell culture (for example, a temperature of 37 ° C. and a humidity of 90%) (Note that the temperature adjusting device 15a in FIGS. 1 and 2) The illustration of the humidity adjusting device 15b is omitted).
The LED light source 35 illuminates the culture vessel 19.

Next, the outline of the configuration of the lower casing 13 will be described. Inside the lower casing 13 are housed an imaging device 34 of the observation unit 22 and a control device (cell evaluation device) 41 of the incubator 11.
The imaging device 34 acquires a microscopic image of the cell by imaging the cell of the culture vessel 19 that is transmitted and illuminated from above through the optical system of the microscope. The imaging device 34 supplies the microscope image of the cell to the CPU 42.

The control device (cell evaluation device) 41 includes a temperature adjusting device 15a, a humidity adjusting device 15b, an observation unit 22,
Each is connected to the container transport device 23. The control device 41 comprehensively controls each part of the incubator 11 according to a predetermined program.

As an example, the control device 41 controls the temperature adjustment device 15a and the humidity adjustment device 15b, respectively, to maintain the inside of the temperature-controlled room 15 at a predetermined environmental condition. The control device 41 controls the observation unit 22 and the container transport device 23 based on a predetermined observation schedule, and automatically executes the observation sequence of the culture vessel 19. Furthermore, the control device 41 executes a culture state evaluation process for evaluating a cell state based on an image acquired in the observation sequence.

The control device 41 is configured using a CPU 42 and a storage unit 43. The CPU 42 is a processor that executes various arithmetic processes of the control device 41.
The storage unit 43 is configured by a hard disk, a nonvolatile storage medium such as a flash memory, or the like. The storage unit 43 holds management data related to each culture vessel 19 stored in the stocker 21 and data of a microscope image captured by the imaging device. Furthermore, the storage unit 43 holds a program executed by the CPU 42.

  The management data includes (a) index data indicating individual culture containers 19, (b) the storage position of the culture container 19 in the stocker 21, and (c) the type and shape of the culture container 19 (well plate, (Dish, flask, etc.), (d) type of cells cultured in culture vessel 19, (e) observation schedule of culture vessel 19, (f) imaging conditions during time-lapse observation (magnification of objective lens, observation in vessel) Point, etc.). In addition, for the culture container 19 that can simultaneously culture cells in a plurality of small containers such as a well plate, management data is generated for each small container.

  In addition, the storage unit 43 holds microscopic image data of a sample prepared in advance. Further, the data of the microscope image for model construction prepared in advance and the teacher value (for example, the degree of differentiation) of the image are held.

  The CPU 42 reads out and executes a program stored in the storage unit 43, thereby executing an area extraction unit 44, a feature amount calculation unit 45, a frequency distribution calculation unit 46, an index extraction unit 47, and an evaluation processing unit 48. And realize the function of each part.

FIG. 4 is a diagram for explaining processing for extracting an iPS cell colony region and a feeder cell region from an image obtained by imaging feeder cells co-cultured with iPS cells.
The area extraction unit 44 reads from the storage unit 43 data of a microscope image of a sample prepared in advance. Then, the region extraction unit 44 designates an image 81 to be processed from among the microscopic images of the above samples. Then, the region extraction unit 44 applies an Open-Close filter to the image 81 to generate an image 82 that has been subjected to the Open-Close filter.

  Then, the region extraction unit 44 generates a background correction image 83 obtained by subtracting the image 82 that has been subjected to the Open-Close filter generated from the image 81. Thereby, the background luminance of the image 81 can be made uniform.

  The region extraction unit 44 extracts cells included in the image from the background correction image 83. For example, when a cell is imaged with a phase-contrast microscope, a halo appears in the vicinity of a region with a large phase difference change such as a cell wall. Therefore, the region extraction unit 44 extracts the halo corresponding to the cell wall by a known edge extraction method, and estimates the closed space surrounded by the edges by the contour tracking process as a cell. Thereby, individual cells can be extracted from the above-mentioned microscopic image.

  Then, the region extraction unit 44 generates an image 84 obtained by extracting a colony region of iPS cells from the background correction image 83. Further, the region extraction unit 44 generates an image 85 obtained by extracting the feeder cell region from the background correction image 83.

FIG. 5 is a diagram for explaining an extracted iPS cell colony region and a feeder cell region.
The region extraction unit 44 extracts the iPS cell colony region 91 from the image 84 obtained by extracting the iPS cell colony region. The region extraction unit 44 extracts the feeder cell region 92 from the image 85 obtained by extracting the feeder cell region.
The area extraction unit 44 supplies the feature amount calculation unit 45 with information on the iPS cell colony region 91 and information on the feeder cell region 92 at each observation time.

FIG. 6 is a diagram for explaining each feature amount of the cell form. Specifically, the feature amount of the cell morphology is, for example, as follows.
“Total area” (see FIG. 6A) is a value indicating the area of the cell of interest. For example, the cell shape feature quantity calculation unit 45 can obtain the value of “Total area” based on the number of pixels in the cell area of interest.

“Hole area” (see FIG. 6B) is a value indicating the area of Hole in the cell of interest. Here, Hole refers to a portion where the brightness of the image in the cell is equal to or greater than a threshold value due to contrast (a portion that is close to white in phase difference observation). For example, stained lysosomes of intracellular organelles are detected as Hole.
Further, depending on the image, a cell nucleus and other organelles can be detected as Hole. It should be noted that the cell form feature quantity calculation unit 45 may detect a group of pixels whose luminance value in the cell is equal to or greater than a threshold value as a Hole, and obtain a value of “Hole area” based on the number of pixels of the Hole.

  “Relative hole area” (see FIG. 6C) is a value obtained by dividing the value of “Hole area” by the value of “Total area” (relative hole area = Hole area / Total area). The “relative hole area” is a parameter indicating the ratio of the organelle in the cell size, and the value varies depending on, for example, enlargement of the organelle or deterioration of the shape of the nucleus.

  “Perimeter” (see FIG. 6D) is a value indicating the length of the outer periphery of the cell of interest. For example, the cell form feature amount calculation unit 45 can acquire the value of “Perimeter” by the contour tracking process when extracting cells.

“Width” (see (e) in FIG. 6) is a value indicating the length of the cell of interest in the horizontal direction (X direction) of the image.
“Height” (see (f) in FIG. 6) is a value indicating the length of the cell of interest in the image vertical direction (Y direction).

“Length” (see (g) of FIG. 6) is a value indicating the maximum value (the total length of the cell) of the lines crossing the cell of interest.
“Breadth” (see (h) in FIG. 6) is a value indicating the maximum value (lateral width of the cell) among the lines orthogonal to “Length”.

  “Fiber Length” (see (i) of FIG. 6) is a value indicating the length when the target cell is assumed to be pseudo linear. The cell form feature amount calculation unit 45 obtains the value of “Fiber Length” by the following equation (1).

  However, in the formula of this specification, “P” indicates the value of Perimeter. Similarly, “A” indicates the value of Total Area.

  “Fiber Breath” (see (j) of FIG. 6) is a value indicating a width (a length in a direction orthogonal to Fiber Length) when the target cell is assumed to be pseudo linear. The cell form feature quantity calculation unit 45 obtains the value of “Fiber Breath” by the following equation (2).

  “Shape Factor” (see (k) of FIG. 6) is a value indicating the circularity (cell roundness) of the cell of interest. The cell form feature quantity calculation unit 45 obtains the value of “Shape Factor” by the following equation (3).

  “Elliptical form factor” (see (l) in FIG. 6) is a value obtained by dividing the value of “Length” by the value of “Breadth” (Elliptical form Factor = Length / Breadth), and the length of the cell of interest. It becomes a parameter indicating the degree of.

“Inner radius” (see (m) in FIG. 6) is a value indicating the radius of the inscribed circle of the cell of interest.
“Outer radius” (see (n) in FIG. 6) is a value indicating the radius of the circumscribed circle of the cell of interest.

“Mean radius” (see (o) of FIG. 6) is a value indicating an average distance between all the points constituting the outline of the cell of interest and its centroid point.
“Equivalent radius” (see (p) of FIG. 6) is a value indicating the radius of a circle having the same area as the cell of interest. The parameter “Equivalent radius” indicates the size when the cell of interest is virtually approximated to a circle.

  Here, the feature quantity calculation unit 45 of the cell morphology may obtain each feature quantity by adding an error to the number of pixels corresponding to the cell. At this time, the feature quantity calculation unit 45 of the cell morphology may obtain the feature quantity in consideration of the imaging conditions of the microscope image (imaging magnification, aberration of the microscopic optical system, etc.). When calculating “Inner radius”, “Outer radius”, “Mean radius”, and “Equivalent radius”, the feature calculation unit 45 of the cell morphology calculates the center of gravity of each cell based on a known center of gravity calculation method. It is only necessary to obtain each parameter based on the center of gravity.

  The cell morphology feature quantity calculation unit 45 receives the information of the iPS cell colony region 91 and the information of the feeder cell region 92 supplied from the region extraction unit 44 at each observation time.

The cell morphology feature amount calculation unit 45 calculates the morphological feature amount of the feeder cell exemplified in FIG. 6 in each feeder cell region 92.
Here, the morphological feature amount of the feeder cell may include the following feature amount. “Luminance value corresponding to unevenness of feeder cell itself”, “Luminance distribution in feeder cell”, and the like.

In addition, the feature quantity calculation unit 45 of the cell morphology may further calculate the morphological feature quantity of the iPS cell colony illustrated in FIG. 6 in the colony region 91 of each iPS cell.
Here, the morphological feature amount of the colony of iPS cells is “the luminance value corresponding to the unevenness of the colony itself”, “the luminance distribution in the colony”, “the shape of the colony group that considered the colony as one object” The feature amount of “distribution” may be included.

  The cell morphology feature quantity calculator 45 may calculate not only the morphological feature quantity of the iPS cell colony but also the morphological feature quantity of each iPS cell exemplified in FIG. 6. Here, the morphological feature amount of the iPS cell may include a feature amount such as “a luminance value corresponding to the unevenness of the iPS cell itself” and “a luminance distribution in the iPS cell”.

  Then, the morphological feature amount of each feeder cell, the morphological feature amount of the colony of each iPS cell, and the morphological feature amount of each iPS cell at each calculated observation time are recorded in the storage unit 43.

  The frequency distribution calculation unit 46 reads from the storage unit 43 the morphological feature amount of each feeder cell, the morphological feature amount of each iPS cell colony, and the morphological feature amount of each iPS cell at each observation time. . Further, the frequency distribution calculation unit 46 reads the teacher value of the observation image from the storage unit 43.

  The frequency distribution calculation unit 46 calculates the frequency distribution of feature amounts for each type of feature amount calculated for each microscope image. Therefore, the frequency distribution calculation unit 46 obtains, for example, 57 (= 19 [type feature amount] × 3 [type cell]) type feature value frequency distribution for one microscope image. Further, the frequency distribution calculation unit 46 calculates the number of cells corresponding to each segment of the feature value as a frequency (%) in each frequency distribution.

  The frequency distribution calculation unit 46 normalizes the frequency classification in the frequency distribution using a standard deviation for each type of feature amount. Here, as an example, a case will be described in which a division in the frequency distribution of “Shape Factor” is determined.

  First, the frequency distribution calculation unit 46 calculates the standard deviation of all “Shape Factor” values calculated from the respective microscope images. Next, the frequency distribution calculation unit 46 substitutes the value of the standard deviation into the Fisher equation to obtain a reference value for the frequency classification in the frequency distribution of “Shape Factor”. At this time, the frequency distribution calculation unit 46 divides the standard deviation (S) of all “Shape Factor” values by 4 and rounds off the third decimal place to obtain the above-described reference value. In addition, the frequency distribution calculation part 46 in one Embodiment shall draw the area for 20 series on a monitor etc., when frequency distribution is illustrated as a histogram.

As an example, when the standard deviation S of “Shape Factor” is 259, 259/4
= 64.750 to “64.75” is the reference value. Then, "S"
When calculating the frequency distribution of the “shape factor”, the frequency distribution calculating unit 46 selects “Shap”.
According to the value of “e Factor”, the cells may be classified into each class set in increments of 0 to 64.75, and the number of cells in each class may be counted.

As described above, the frequency distribution calculation unit 46 normalizes the frequency distribution division with the standard deviation for each type of feature amount, and thus the tendency of the frequency distribution can be roughly approximated between different feature values. Therefore, in one embodiment, it is relatively easy to obtain a correlation between a cell culture state and a change in frequency distribution between different feature values.

  The frequency distribution calculation unit 46 histograms all the morphological feature amounts. The frequency distribution calculation unit 46 stores all histograms in the storage unit 43.

The index extraction unit 47 reads all histograms from the storage unit 43.
The index extraction unit 47 compares the histograms over time for all combinations over time for each morphological feature amount, and calculates the absolute value of the area difference of the histogram over time.

FIG. 7 is a diagram for explaining a table that stores parameters of a calculation model for a certain feature amount. The table 100 shows an area difference between the histogram of the feature quantity at time 1 and the histogram of the feature quantity at time 2 for time 1, time 2, and each feature quantity f (j) (j is an integer from 1 to n). It is configured using a difference between ΔS _{f (j)} and a teacher value (for example, the degree of differentiation).
Here, as an example, five frequency distributions having the same type of morphological feature amount and different shooting times (0, 24, 48, 72, 96th hour) are assumed.

  The index extraction unit 47 calculates the absolute value of the area difference of the frequency distribution (integrated absolute value of the difference of the frequency distribution between images) by the following formula (4) for all the 10 combinations. calculate. That is, when focusing on one type of feature amount for one set, the index extraction unit 47 calculates a total of 10 area differences for the frequency distribution of the feature value.

  However, in Expression (4), “Control” indicates the frequency (number of cells) for one section in the initial frequency distribution (8th hour). “Sample” indicates the frequency of one section in the frequency distribution to be compared. Further, i is a variable indicating a frequency distribution interval.

The index extraction unit 47 can obtain 10 types of frequency distribution change amounts for all feature amounts by performing the above calculation for each type of feature amount. That is, with respect to one set of microscope images, 19 types × 10 combinations of 190 frequency distributions can be taken. Hereinafter, one of the combinations of the frequency distributions will be referred to as an index. The index extraction unit 47 stores the table 100 in the storage unit 43 for each morphological feature amount.

  The index extraction unit 47 designates one or more indices among the 190 indices that well reflect the cell culture state by multivariate analysis. In addition to the nonlinear model equivalent to a fuzzy neural network described later, the use of a linear model is also effective in selecting the combination of indices according to the complexity of the cell and its form. Along with the selection of such index combinations, the index extraction unit 47 calculates a calculation model for estimating the state of iPS cells from the microscope image using one or more of the specified indices.

  Specifically, the average value (input value 1) of 19 types of indicators in each time form, the standard deviation (input value 2) of 19 types of indicators in each time format, and the 19 types of indicators in each time format There are four input values, that is, the distribution value itself (input value 3) and the area difference (input value 4) over time of the distribution of the 19 kinds of distributions in the form of each time. The degree of differentiation (teacher value) of each image cell is related to the four input values. Therefore, the index extraction unit 47 finds a combination of input values that can predict the teacher value with the highest accuracy.

  Here, the index extraction unit 47 obtains the above calculation model by analyzing a fuzzy neural network (FNN).

  FNN is a method in which an artificial neural network (ANN) and fuzzy inference are combined. In this FNN, an ANN is incorporated in fuzzy inference to automatically determine the membership function, which is a drawback of fuzzy inference. ANN (see FIG. 10), which is one of the learning machines, is a mathematical model of a neural network in a living brain and has the following characteristics.

FIG. 8 is a diagram for explaining the outline of the Artificial Neural Network.
The learning in the ANN uses learning data (input value: X) having a target output value (teacher value), and uses a back propagation (BP) method to calculate the teacher value and the output value (Y). This is a process of constructing a model so that the coupling load in a circuit connecting between nodes (indicated by circles in FIG. 8) is changed so that the error becomes small, and the output value approaches the teacher value.

  Using this BP method, the ANN can automatically acquire knowledge by learning. Finally, by inputting data that is not used for learning, the versatility of the model can be evaluated. Conventionally, membership function determination has relied on human senses, but by incorporating ANN as described above into fuzzy inference, membership function can be automatically identified. This is FNN.

FIG. 9 is a diagram for explaining the outline of the Fuzzy Neural Network.
In the FNN, like the ANN, the input / output relationship given to the network by using the BP method can be automatically identified and modeled by changing the coupling weight. FNN has a characteristic that knowledge can be acquired as a linguistic rule that is easy for humans to understand like fuzzy reasoning (see the balloon at the lower right in FIG. 9 as an example) by analyzing a model after learning.

  In other words, FNN automatically determines the optimal combination of fuzzy reasoning in the combination of variables such as numerical values representing the morphological characteristics of the cell from its structure and characteristics, and provides an estimate of the prediction target and an effective index for prediction. Rules that indicate combinations can be generated simultaneously.

  The structure of the FNN is “input layer”, “membership function part (preceding part)” that determines parameters Wc and Wg included in the sigmoid function, Wf is determined, and the relationship between input and output is taken out as a rule It consists of four possible layers of “fuzzy rule part (consequent part)” and “output layer” (see FIG. 9).

FIG. 10 is a diagram for explaining a sigmoid function in the Fuzzy Natural Network.
There are Wc, Wg, and Wf as bond loads that determine the model structure of FNN. The combined load Wc determines the center position of the sigmoid function used for the membership function, and Wg determines the inclination at the center position (see FIG. 10).

  Returning to FIG. 9, in the model, the input value is expressed by a fuzzy function with flexibility close to human sensation (see the balloon at the lower left of FIG. 9 as an example). The combined load Wf represents a contribution to the estimation result of each fuzzy region, and a fuzzy rule can be derived from Wf. That is, the structure in the model can be deciphered later and written as a rule (see the balloon at the lower right in FIG. 9 as an example).

  A Wf value, which is one of the coupling loads, is used to create a fuzzy rule in the FNN analysis. If the Wf value is positive and large, the unit contributes greatly to being determined to be “effective for prediction”, and the index applied to the rule is determined to be “effective”. If the Wf value is negative and small, the unit has a large contribution to being determined as “invalid for prediction”, and the index applied to the rule is determined as “invalid”.

  As a more specific example, the index extraction unit 47 obtains the calculation model by the following processes (A) to (H).

(A) The index extraction unit 47 selects one index from 190 indexes.
(B) The index extraction unit 47 obtains the state (predicted value) of iPS cells in each set of microscopic images by a calculation formula using the amount of change in the frequency distribution according to the index selected in (A) as a variable. .

Assuming that a calculation formula for obtaining the state of iPS cells from one index is “Y = αX ₁ ” (where “Y” is a calculated value of iPS cells (for example, a value indicating the number of iPS cells increased), “X ₁ ”represents the amount of change in the frequency distribution corresponding to the selected index, and“ α ”represents the coefficient value corresponding to X ₁ ). In this case, the index extraction unit 47 is configured to assign any value to alpha, if the values are the variation of the frequency distribution in each set to X _1. Thereby, the calculated value (Y) of the cancer cell in each set is obtained.

(C) The index extraction unit 47 obtains an error between the calculated value Y obtained in (B) and the actual number of iPS cells (teacher value) for each set of microscope images. Note that the above-described teacher value is obtained by the index extraction unit 47 based on information on the number of read iPS cells.
Then, the index extraction unit 47 corrects the coefficient α of the above calculation formula by supervised learning so that the error of the calculated value in each set of microscope images becomes smaller.

(D) The index extraction unit 47 repeats the processes (B) and (C), and obtains a calculation formula model that minimizes the average error of the calculated values for the index (A).
(E) The index extraction unit 47 repeats the processes (A) to (D) for 190 indices. Then, the index extraction unit 47 compares the average errors of the calculated values of the 190 indices, and sets the index having the lowest average error as the first index used for generating the evaluation information.

  (F) The index extraction unit 47 obtains a second index to be combined with the first index obtained in (E) above. At this time, the index extraction unit 47 pairs the first index and the remaining 189 indices one by one. Next, the index extraction unit 47 obtains a cancer cell prediction error by a calculation formula in each pair.

Suppose that the calculation formula for obtaining the number of cancer cells from two indices is “Y = αX ₁ + βX ₂ ” (where “Y” corresponds to the calculated value of cancer cells, and “X ₁ ” corresponds to the first index. The amount of change in the frequency distribution, “α” is a coefficient value corresponding to X ₁ , “X ₂ ” is the amount of change in the frequency distribution corresponding to the selected index, and “β” is the coefficient value corresponding to X _2. Respectively).
Then, the index extraction unit 47 obtains the values of the coefficients α and β so that the average error of the calculated values is minimized by the same processing as the above (B) and (C).

  Thereafter, the index extraction unit 47 compares the average errors of the calculated values obtained for each pair, and obtains the pair having the lowest average error. Then, the index extraction unit 47 sets the pair index having the lowest average error as the first and second indices used for generating the evaluation information.

  (G) The index extraction unit 47 ends the calculation process when a predetermined end condition is satisfied. For example, the index extraction unit 47 compares the average errors of the respective calculation formulas before and after increasing the index. Then, the index extraction unit 47 calculates the average error based on the calculation formula after increasing the index is higher than the average error based on the calculation formula before increasing the index (or the difference between the two is within the allowable range). The process ends.

  (H) On the other hand, when the termination condition is not satisfied in (G), the index extraction unit 47 further increases the number of indices by 1, and repeats the same processing as in (F) and (G). Thereby, when obtaining the above calculation model, the index is narrowed down by stepwise variable selection.

  The index extraction unit 47 records the information of the calculation model obtained above (information indicating each index used in the calculation formula, information on coefficient values corresponding to each index in the calculation formula, etc.) in the storage unit 43.

Note that the index is not limited to the above index, and the index extraction unit 47 may perform the 0th, 24th, 48th, 72th, 96th, 120th, 148th, and Using each of the 168th hour as a reference, 36 (= 9 × 8 ÷ 2) indexes are obtained for one type of feature value, and a total of 684 (19 types × 36 type) for 19 types of feature value. May be obtained.
By using such indexes with different reference times, the index extraction unit 47 can construct a calculation model with higher prediction accuracy.

The evaluation processing unit 48 reads the input value 1 to the input value 4 of the target image and the information of the calculation model from the storage unit 43.
The evaluation processing unit 48 selects from the input value 1 to the input value 4 only the index value used in the calculation formula of the already constructed calculation model.

For example, the evaluation processing unit 48, in the target image, “the standard deviation of the cell length at the 8th hour” and “the area difference between the cell roundness histogram at the 96th hour and the histogram of the cell roundness at the 12th hour”. And select.
The evaluation processing unit 48 inputs the selected index into the calculation model and calculates the estimated degree of differentiation.

  The CPU 42 controls to change the culture conditions (for example, temperature, humidity, etc.) of the iPS cells or feeder cells based on the state of the iPS cells estimated by the evaluation processing unit 48. For example, the CPU 42 supplies a control signal to the temperature adjustment device 15a and controls the temperature inside the temperature-controlled room 15 to increase. In addition, the CPU 42 supplies a control signal to the humidity adjusting device 15b to control the humidity in the temperature-controlled room 15 to decrease.

<Example of iPS cell observation>
FIG. 11 is a flowchart showing an example of observation processing in the incubator. FIG. 11 shows an operation example in which the culture container 19 carried into the temperature-controlled room 15 is time-lapse observed according to a registered observation schedule.

  In step S <b> 101, the CPU 42 compares the observation schedule of the management data in the storage unit 43 with the current date and time to determine whether or not the observation start time of the culture vessel 19 has come. When it is the observation start time (step S101 YES), the CPU 42 shifts the process to S102. On the other hand, when it is not the observation time of the culture container 19 (NO in step S101), the CPU 42 waits until the time of the next observation schedule.

  Next, in step S102, the CPU 42 instructs the container transport device 23 to transport the culture container 19 corresponding to the observation schedule. Then, the container transport device 23 carries the instructed culture container 19 out of the stocker 21 and places it on the sample stage 31 of the observation unit 22. Note that, when the culture vessel 19 is placed on the sample stage 31, an entire observation image of the culture vessel 19 is captured by a bird view camera (not shown) built in the stand arm 32.

  Next, in step S103, the CPU 42 instructs the observation unit 22 to capture a microscopic image of the cell. The observation unit 22 turns on the LED light source 35 to illuminate the culture vessel 19 and drives the imaging device 34 to take a microscopic image of the cells in the culture vessel 19.

  At this time, the imaging device 34 captures a microscope image based on the imaging conditions (magnification of the objective lens, observation point in the container) designated by the user based on the management data stored in the storage unit 43. For example, when observing a plurality of points in the culture container 19, the observation unit 22 sequentially adjusts the position of the culture container 19 by driving the sample stage 31 and picks up a microscope image at each point. The microscopic image data acquired in S103 is read by the control device 41 and recorded in the storage unit 43 under the control of the CPU.

  Next, in step S104, the CPU 42 instructs the container transport device 23 to transport the culture container 19 after the observation schedule is completed. Then, the container transport device 23 transports the designated culture container 19 from the sample stage 31 of the observation unit 22 to a predetermined storage position of the stocker 21. Thereafter, the CPU 42 ends the observation sequence and returns the process to S101. Above, description of this flowchart is complete | finished.

<Example of teacher value calculation for calculation model>
FIG. 12 is a flowchart for explaining calculation of a teacher value for a calculation model. In this calculation model generation process, the control device 41 determines the combination of frequency distributions used for estimating the state of the iPS cell from a combination of a plurality of frequency distributions having different image capturing times and different types of feature values.

  First, in step S <b> 201, the imaging device 34 continuously acquires a reference image in which iPS cells and feeder cells existing around the iPS cells and cultured together are acquired and stored in the storage unit 43.

Next, in step S202, after culturing for a certain period, the user stains the imaged cells with a staining marker (for example, SSEA-4, SSEA-1, and nanog which are staining markers indicating the degree of differentiation).
The CPU 42 may control to add a staining marker previously placed in a container to the culture solution after culturing for a certain period. Thereby, the imaged cell can be stained with the staining marker.

  The imaging device 34 images stained iPS cells imaged by a microscopic optical system for phase difference observation. When the staining marker is labeled with a fluorescent substance, the imaging device 34 images the stained iPS cells imaged by a microscopic optical system for fluorescence observation (not shown). The imaging device 34 stores the stained image in the storage unit 43.

Next, in step S <b> 203, the CPU 42 reads the dyed image from the storage unit 43. The CPU 42 determines the presence or absence of staining for each iPS cell.
Next, in step S204, the CPU 42 calculates a teacher value (differentiation degree) of the image. For example, if the number of stained cells is 72 and the total number of iPS cells is 100, the teacher value (differentiation degree) of the image is the number of stained cells / total number of cells = 72 (%).

  Next, in step S205, when the processing from step S201 to step S204 has not been performed for all the samples a predetermined number of times (NO in step S201), the above steps were co-cultured with other iPS cells and feeder cells. Dozens or hundreds of each sample is performed.

  For example, if the user has a degree of differentiation near 20 (30 samples), a degree of differentiation near 50 (30 samples), a degree of differentiation near 80 (30 samples), a sample near the degree of differentiation 100 (30 samples), Images are taken every 24 hours for 4 days.

  In the continuous culture image, 4 [conditions] × 30 [samples] = 120 [sheets] exist for 4 days, and therefore, for each 120 images, the 0th hour, the 24th hour, the 48th hour, There are 5 points at 72 hours and 96 hours. For each image, the CPU 42 calculates a teacher value (degree of differentiation) and stores it in the storage unit 43.

  Next, in step S205, when the process from step S201 to step S204 is performed a predetermined number of times for all samples (YES in step S201), this flowchart ends.

FIG. 13 is a flowchart for explaining the construction of the calculation model.
First, in step S <b> 301, the CPU 42 reads from the storage unit 43 microscopic image data of a sample prepared in advance. Note that the CPU 42 in S301 also acquires information indicating the total number of cells and the number of iPS cells corresponding to each image at this time.

  Next, in step S302, the CPU 42 designates an image to be processed from the microscopic images of the above samples. Here, it is assumed that the CPU 42 in S302 sequentially designates all of the sample microscope images prepared in advance as processing targets.

  Next, in step S303, the region extraction unit 44 extracts the iPS cell colony region and the feeder cell region included in the processing target microscope image.

  Next, in step S304, the feature amount calculation unit 45 calculates a feature amount indicating the morphological features of each iPS cell colony, each iPS cell, and each feeder cell included in the microscope image to be processed. . For example, for each 120 images (= 4 [conditions] × 30 [sheets]) for each time, the feature value for each time is calculated.

Next, in step S <b> 305, the feature amount calculation unit 45 stores all the calculated feature amounts in the storage unit 43.
Next, in step S306, when the feature amount of all images is not calculated for all images (NO in step S306), the process returns to step S302. When the feature amount of all the images has been calculated (YES in step S306), the process proceeds to step S307.

  Next, in step S307, the frequency distribution calculation unit 46 calculates the average value of the feature values over time, the standard deviation of the feature values over time, and the frequency distribution of the feature values over time. In other words, the frequency distribution calculation unit 46 collects the morphological information of all the cells in each image and forms a histogram as a group.

Next, in step S308, the index extraction unit 47 calculates, for each feature amount, the difference in the area of the frequency distribution of the feature amount over time for all combinations over time.
For example, it is assumed that there are five points at the 0th, 24th, 48th, 72th, and 96th hours for each of 120 images. Then, there are the following 10 combinations over time. The combination over time is expressed in the format of [Time]-[Time]. 24-0, 48-0, 72-0, 96-0, 48-24, 72-24, 96-24, 72 -48, 96-48, 96-72.

  Next, in step S309, when the difference in the area of the frequency distribution of the feature quantity over time is not calculated for all the feature quantities for all combinations over time (NO in step S309), the process returns to step S308. In step S309, when the difference in the area of the frequency distribution of the feature amount over time is calculated for all the feature amounts with respect to all combinations over time (YES in step S309), the process proceeds to step S310.

  Next, in step S310, the index extraction unit 47 designates one or more indices among the 190 indices that well reflect the state of iPS cells by multivariate analysis. In addition to the nonlinear model equivalent to the above-described fuzzy neural network, the use of a linear model according to the complexity of the cell and its form is also effective in selecting the combination of indices. Along with the selection of such index combinations, the index extraction unit 47 obtains a calculation model for deriving the state of iPS cells from the microscope image using one or more of the specified indices.

  Next, in step S311, the index extraction unit 47 records calculation model information (information indicating each index used in the calculation formula, coefficient value information corresponding to each index in the calculation formula, and the like) in the storage unit 43. . Above, this flowchart is complete | finished.

<Example of iPS cell state estimation processing>
FIG. 14 is a flowchart for estimating the degree of differentiation of iPS cells using a calculation model.
First, in step S <b> 401, the evaluation processing unit 48 reads data of a plurality of microscope images (target images) to be evaluated from the storage unit 43. Here, it is assumed that the target image is acquired by observing the culture container 19 in which the cell group is cultured with the incubator 11 in the same field of view under the same imaging conditions. In this case, the time-lapse observation is performed every 24 hours until the 96th hour from the beginning of the culture in order to align the conditions with the example of FIG.

Next, in step S402, the evaluation processing unit 48 reads information on the calculation model in the storage unit 43 (recorded in S311 in FIG. 13).
Next, in step S403, the feature amount calculation unit 45, the frequency distribution calculation unit 46, and the evaluation processing unit 48 obtain the change amount of the frequency distribution for each index corresponding to the variable of the calculation model.

  Next, in step S304, the evaluation processing unit 48 performs calculation by substituting the change amount of the frequency distribution of each index obtained in step S403 into the calculation model read out in S302. And the evaluation process part 48 produces | generates the estimation information which shows the differentiation degree of the iPS cell in the microscope image of evaluation object based on said calculation result. Thereafter, the evaluation processing unit 48 displays this estimation information on a monitor (not shown) or the like. Above, this flowchart is complete | finished.

  According to the first embodiment, the state of iPS cells can be estimated from the morphological feature amount of the iPS cell colony, the morphological feature amount of the iPS cell, and the morphological feature amount of the feeder cell.

In the first example, the iPS cell differentiation degree was estimated from the morphological feature amount of the iPS cell colony, the morphological feature amount of the iPS cell, and the morphological feature amount of the feeder cell. Not exclusively.
The differentiation degree of iPS cells may be estimated only from the temporal change of the frequency distribution of the morphological feature amount of the feeder cells.
Also, from the time change of the frequency distribution of the morphological feature amount of feeder cells and the time change of the frequency distribution of the morphological feature amount of individual iPS cells or the time change of the frequency distribution of the morphological feature amount of colonies of iPS cells. The degree of iPS cell differentiation may be estimated.

  In the first embodiment, the degree of iPS cell differentiation is estimated. However, the present invention is not limited to this, but the degree of iPS cell degradation, the degree of iPS cell proliferation, the direction of iPS cell differentiation, and the ease of tumor formation of iPS cells. Etc.

<Example 2>
In Example 1 described above, the state of iPS cells was evaluated using continuous values. In Example 2, the state of iPS cells is roughly classified as a level. The method of the second embodiment is particularly useful when continuous value evaluation is not suitable depending on the reliability of staining evaluation and the estimation accuracy of the first embodiment. Hereinafter, a second embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 15 is a block diagram of an incubator according to the second embodiment of the present invention. The incubator 11a is configured using an upper casing 12 and a lower casing 13a. The lower casing 13a is configured using an imaging device 34 and a control device 41a. The control device (cell evaluation device) 41a is configured using a CPU 42a and a storage unit 43a.
Except for the CPU 42a and the storage unit 43a, the configuration is the same as the block configuration diagram of FIG.

The storage unit 43a is configured by a non-volatile storage medium such as a hard disk or a flash memory. The storage unit 43a holds management data regarding each culture container (not shown) stored in a stocker (not shown) and data of a microscope image captured by the imaging device.
Furthermore, the storage unit 43a holds a program executed by the CPU 42a and information on the discrimination model calculated by the index extraction unit 47a.

  The CPU 42a reads out and executes a program stored in the storage unit 43a, thereby executing a region extraction unit 44, a feature amount calculation unit 45a, a frequency distribution calculation unit 46, an index extraction unit 47a, and an evaluation processing unit 48a. And realize the function of each part.

  Since the processing of the region extraction unit 44 and the frequency distribution calculation unit 46 is the same as that of the first embodiment, description thereof is omitted.

  As in the first embodiment, the feature amount calculation unit 45a obtains feature amounts indicating the morphological features of the cells for each image object extracted from the image by the region extraction unit 44. As an example, the feature quantity calculation unit 45a obtains 19 types of feature quantities for each cell.

  The feature amount calculation unit 45a obtains an average value and an image standard deviation of 19 types of feature amounts for the teacher image to be calculated. Then, the feature amount calculation unit 45a records the in-image average value and the in-image standard deviation of each feature amount in the storage unit 43a.

  As a result, regarding the five teacher images having the same cell passage number and different observation times (0, 24, 48, 72, and 96 hours), 38 feature values (19 2 types (average value, standard deviation)) are obtained.

  The feature amount calculation unit 45a groups the teacher images having the same passage number by the following processes (A1) to (A3), and calculates the feature amount between the two teacher images in the grouped teacher image group. Find the amount of change.

  (A1) The feature amount calculation unit 45a groups the six teacher images having the same cell passage number and different observation times (0, 24, 48, 72, and 96 hours). Although not particularly limited, it is preferable that the teacher images to be grouped are five microscopic images having the same culture container and captured position (field of view).

  (A2) The feature amount calculation unit 45a combines two teacher images at different imaging points in the teacher image group grouped in (A1). Here, since there are five teacher images in one group, there are ten combinations of two teacher images. Further, since there are 10 combinations of two teacher images, there are 380 types of change amounts determined from one group of teacher images (19 feature values × 2 patterns (average value, standard deviation) × 10 combinations). Become.

  (A3) For each combination of (A2), the feature amount calculation unit 45a includes 19 types of change amounts indicating changes in the average value of the respective feature amounts between the two teacher images, and two teacher images. 19 kinds of change amounts indicating changes in the standard deviation in the image of each feature amount in between are obtained.

Next, the process of the index extraction unit 47a will be described.
The index extraction unit 47 a calculates the area difference of each time of the frequency distribution of the morphological feature amount calculated by the frequency distribution calculation unit 46.

  The index extraction unit 47a classifies cells showing different differentiation into different classifications based on the morphological feature amount calculated by the feature amount calculation unit 45a and the area difference of the frequency distribution of the morphological feature amount. Build a model. The index extraction unit 47a in the second embodiment employs a decision tree analysis as one form of the discrimination model, and classifies the microscopic images according to the number of cell passages. Build up.

<How to construct a decision tree>
Hereinafter, a method by which the index extraction unit 47a constructs a decision tree will be described.
The index extraction unit 47a obtains a parameter type (index type) and a parameter value (index threshold) to be applied when classifying cells having different passage numbers at the root node of the decision tree. Note that the type of the parameter is selected from each feature amount (150 types) and feature amount change amount (380 types) at each observation period (0, 24, 48, 72, and 96 hours).

  Hereinafter, an example of how to obtain the above-described index type and index threshold will be described. First, the index extraction unit 47a selects a parameter to be calculated from the above 530 (= 150 + 380) types of parameters. At this time, the index extraction unit 47a previously extracts a teacher image group corresponding to the calculation target parameter. The extracted teacher image group includes teacher images having different cell passage numbers.

Next, the index extraction unit 47a sets a search range (upper limit and lower limit) for the parameter to be calculated, and divides the search range for the parameter to be calculated into a plurality of arbitrary increments. The index extraction unit 47a classifies the cells of the teacher image group corresponding to the parameter to be calculated into two sets using each step (parameter value) of the parameter to be calculated as a threshold value. As a result, the classification result of the cells of the teacher image group is obtained at each step of the parameter to be calculated.
At this time, the index extraction unit 47a obtains the distribution of the number of cells according to the number of passages of cells in each set of classification results.

  The index extraction unit 47a performs the same processing for all the 450 types of parameters described above, and acquires the classification result of the teacher image group at each step of each parameter. The index extraction unit 47a then selects the parameter type (index type) most suitable for classification of cells having different passage numbers from among all the classification results, and a threshold value for the parameter (index threshold value or index threshold value). To decide.

  As an example, the index extraction unit 47a determines the index type and the threshold of the index that can completely separate cells of any passage number into one set from among the combinations of the parameter type and the threshold. When there are a plurality of combinations that meet the above conditions, the index extraction unit 47a sets the index type and the threshold value of the index in preference to those that can separate only one kind of cells of any passage number into one set. Just decide.

  When there is no combination that meets the above conditions, the index extraction unit 47a may determine the index type and the index threshold in the following manner. For example, the index extraction unit 47a includes two sets of teacher images having the same passage number, but one set can extract only one type of passage number of teacher images, and one set can extract teacher images. The type of the index and the threshold value of the index may be determined by giving priority to the image having the largest number of images.

  Next, the index extraction unit 47a designates the node to be calculated for obtaining the index type and the index threshold. For example, the index extraction unit 47a designates the left node among the child nodes that branch left and right from the parent node. When the index type and the index threshold are already obtained in all nodes below the left child node, the index extracting unit 47a specifies a node that branches from the parent node to the right side.

  The index extraction unit 47a determines whether or not the current computation target node is a terminal node that cannot branch further. For example, the index extraction unit 47a determines that it is a terminal node when cells of one passage number can be separated at the current computation target node.

  The index extraction unit 47a determines whether the calculation is completed. The index extraction unit 47a determines that the calculation is completed when the index type and the index threshold value have been obtained in all the nodes other than the terminal node.

The index extraction unit 47a obtains the index type and the index threshold at the current calculation target node. Thereafter, the index extraction unit 47a changes the calculation target node. Thereafter, the index type and index threshold value are recursively obtained by the loop.
The index extraction unit 47a records the data of the discriminant model constructed by the above processing in the storage device 43a.

<Indicator calculation method>
The index extraction unit 47a calculates an index for classifying the state of iPS cells using the following two methods.
As a first method, the index extraction unit 47a evaluates the state of cells by staining or the like, and roughly classifies them into “levels”. For example, the index extraction unit 47a divides into a low differentiation degree, a middle differentiation degree, and a high differentiation degree. That is, the index extraction unit 47a classifies the items having the order and superiority into several “levels”.

  As a second method, the index extraction unit 47a evaluates the quality of cells by staining or the like, and roughly classifies them into “groups”. For example, the index extraction unit 47a classifies into a bone differentiation group, a cartilage differentiation group, and a fat differentiation group. In other words, the index extraction unit 47a classifies each group, which is not a sequence or superiority and is a parallel group, and cannot be expressed well with numbers or the like into several “groups”.

The index extraction unit 47a creates the following two types of models using the decision tree method.
The index extraction unit 47a groups data using an input value for each image and a teacher value for each image as a model for each image. The index extraction unit 47a classifies each “level” or each “group”, classifies the level of cell differentiation, and classifies each differentiation type.

  The index extraction unit 47a groups data using an input value for each cell and a teacher value for each cell as a model for each cell. The index extraction unit 47a classifies the level and group of each cell for each cell in the image. As a result, it is possible to know what type of combination of indicators corresponds to what level or group in each cell.

In each of the above classifications, the index extraction unit 47a applies to each “level” or “group” so that the prepared “level” or “group” group can be separated with the highest accuracy. Select a combination of indicators and a threshold for that indicator.
The index extraction unit 47a stores the decision tree index and the threshold value of the index in the storage unit 43a.

The evaluation processing unit 48a reads the index of the decision tree and the threshold value of the index from the storage unit 43a. The evaluation processing unit 48a extracts the index value from the feature amount of the target image. The evaluation processing unit 48a compares the index value extracted from the target image with a threshold value for each index of the decision tree node. Accordingly, the evaluation processing unit 48a classifies the target image or the target iPS cell into “level” or “group”. The evaluation processing unit 48a calculates an estimation result of the state of iPS cells based on the classification result.
Hereinafter, specific processing examples of the index extraction unit 47a and the evaluation processing unit 48a will be described.

<By image, colony + feeder model>
FIG. 16 is a diagram for explaining a decision tree for each image, colony + feeder model.
The decision tree in FIG. 16 is an example in which differentiation groups are divided. In the target image, if the area difference between the histograms of the feeder cells at 96 hours and 24 hours is 20 or more, it belongs to group 1 (bone differentiation group).

In the target image, if the area difference between the histograms of the feeder cells at 96 hours and 24 hours is less than 20, and the average roundness of colonies at 24 hours is 30 or more, group 2 (cartilage differentiation group) Belonging to.
In the target image, if the area difference between the histograms of the feeder cells at 96 hours and 24 hours is less than 20 and the average roundness of colonies at 24 hours is less than 30, group 3 (fat differentiation group) Belonging to.

The index extraction unit 47a calculates the type of index of the decision tree using the iPS cell colony feature amount and the feeder cell feature amount as the image-specific model.
For example, the index extracting unit 47a selects index types “area difference between histograms of feeder cells at 96 hours and 24 hours” and “average value of roundness of colonies at 24 hours”.

In addition, the index extraction unit 47a determines a threshold value (index threshold value) for determining which value is greater than or less than each index type.
For example, the index extraction unit 47a determines the index threshold 20 for the type of the index “area difference between the histograms of the feeder cells at 96 hours and 24 hours”, and sets the index type “colony at 24 hours”. The threshold value 30 of the index is determined with respect to the “average value of roundness”.

<By image, colony feeder model>
FIG. 17 is a diagram for explaining a decision tree for each image and each colony / feeder model.
FIG. 17A is a diagram for explaining a decision tree (hereinafter referred to as a colony model) based on the feature amount of a colony of iPS cells for each image. The decision tree in FIG. 17A is an example in which differentiation groups are divided. In the target image, if the average value of roundness of colonies over 24 hours is 20 or more, it belongs to group 1 (bone differentiation group).

In the target image, if the average value of the roundness of the colonies at 24 hours is less than 20, and the amount of change in the length of the colonies at 96 hours and 24 hours is 100 or more, group 2 (cartilage differentiation group) Belongs.
On the other hand, if the average value of the roundness of the colonies for 24 hours is less than 20 and the change in the length of the colonies at 96 hours and 24 hours is less than 100 in the target image, group 3 (fat differentiation group) ).

  FIG. 17B is a diagram for explaining a decision tree (hereinafter referred to as a feeder model) based on the feature amount of a feeder cell for each image. The decision tree in FIG. 17B is an example in which the level of iPS cell degradation is divided. In the target image, if the average value of the roundness of the feeder cells for 48 hours is 40 or more, it belongs to group 1 (deterioration degree: large).

In the target image, if the average value of the roundness of the 48-hour feeder cells is less than 40 and the area difference in the histogram of the roundness of the feeder cells at 96 hours and 24 hours is 50 or more, group 2 (degradation degree) : Medium).
On the other hand, in the target image, if the average value of the roundness of the 48-hour feeder cells is less than 40 and the area difference between the histograms of the roundness of the feeder cells at 96 hours and 24 hours is less than 50, group 3 ( Degradation degree: small).

The index extraction unit 47a calculates indexes of different decision trees using the feature amount of the iPS cell colony and the feature amount of the feeder cell as the image-specific model.
For example, the index extraction unit 47a sets the types of indices “average value of roundness of colonies at 24 hours” and “change amount of colony lengths at 96 hours and 24 hours” to the colony model. select.
In addition, the index extraction unit 47a refers to the “average value of the roundness of the feeder cells at 48 hours” and the “area difference between the histograms of the roundness of the feeder cells at 96 hours and 24 hours” for the feeder model. Select the type of indicator.

In addition, the index extraction unit 47a determines an index threshold value for each index type.
For example, the index extraction unit 47a determines the index threshold 20 for the index type “average value of roundness of colonies at 24 hours”, and sets the index types “96th and 24th colonies at 24 hours”. An index threshold 100 is determined for the “length variation”.

  Further, the index extraction unit 47a determines an index threshold 40 for the index type “average value of roundness of feeder cells at 48 hours”, and sets the index types “feeders at 96 hours and 24 hours”. An index threshold value 50 is determined for “the area difference of the cell roundness histogram”.

The evaluation processing unit 48a performs complex estimation such as group 1 (differentiation group: bone) in the colony model and group 3 (degradation degree: small) in the feeder model.
Note that the evaluation processing unit 48a may make a reliable prediction only when the group names match in both the group 1 (differentiated group bone) in the colony model and the group 1 (differentiated group bone) in the feeder model. .

<Models by cell and colony / feeder>
FIG. 18 is a diagram for explaining a decision tree of a model for each cell and each colony / feeder. FIG. 18A is a diagram for explaining a colony model for each cell. The decision tree in FIG. 18A is an example in which differentiation groups are divided. In the colony of iPS cells with the target image, the average value of the roundness of the colonies at 24 hours is 80 or more, and the area difference in the histogram of the roundness of the colonies at 96 hours and 24 hours at 96 hours and 24 hours Is less than 20, the iPS cell belongs to group 1 (bone differentiation group).

If the colony of the iPS cell with the target image has a colony roundness of 24 hours or more of 80 or more and the area difference in the histogram of the roundness of colonies of 96 hours and 24 hours is 20 or more, the iPS cells are It belongs to group 2 (cartilage differentiation group).
On the other hand, in a colony of iPS cells with a target image, if the roundness of colonies for 24 hours is less than 80, the iPS cells belong to group 3 (adipose differentiation group).

  FIG. 18B is a diagram for explaining a feeder model for each cell. The decision tree in FIG. 18B is an example in which the differentiation groups are divided. In the target image, in the feeder cells existing around the iPS cell whose state is to be estimated, if the length of the feeder for 48 hours is 20 or more and the brightness of the feeder for 24 hours is 30 or more, the iPS cells are at the level. 1 (proliferation ability: high).

In the target image, in the feeder cells existing around the iPS cell whose state is to be estimated, if the length of the feeder for 48 hours is 20 or more and the brightness of the feeder for 24 hours is 30 or more, the iPS cells are at the level. 2 (proliferative ability: medium).
On the other hand, in the target image, if the feeder cell existing around the iPS cell whose state is to be estimated has a length of the feeder for 48 hours of 20 or more and the luminance of the feeder for 24 hours is 30 or more, the iPS cell Belongs to level 3 (proliferative ability: small).

The index extraction unit 47a creates a decision tree using the iPS cell colony feature amount and a decision cell using the feeder cell feature amount in order to estimate the state of the iPS cell at the individual cell level.
The index extraction unit 47a verifies the combination of colony model indices for each colony of iPS cells, and calculates an index for the colony.

In addition, the index extraction unit 47a verifies the combination of the indices of the feeder model for each feeder cell, and calculates an index for the feeder cells.
In addition, the index extraction unit 47a determines an index threshold value for each index type.

For example, the index extracting unit 47a determines the index threshold 80 for the index type “roundness of colonies at 24 hours” and sets the index types “rounds of colonies at 96 hours and 24 hours”. An index threshold 20 is determined for the “area difference of the histogram”.
In addition, the index extraction unit 47a determines the index threshold 20 for the index type “length of feeder cells at 48 hours”, and for the index type “brightness of feeder cells at 24 hours”. The threshold value 30 of the index is determined.

When estimating the state of a certain iPS cell in the target image, the evaluation processing unit 48a estimates the state of the iPS cell by combining the results of the two models.
For example, in the colony model, if the area difference in the histogram of the roundness of colonies at the 24th hour of the colonies of iPS cells is not less than 20 and the roundness of the colonies at the 96th and 24th hours is less than 20, the evaluation processing unit 48a Classifies iPS cells into group 1. Thus, the evaluation processing unit 48a estimates that the iPS cell has the ability to differentiate into bone.

Further, for example, among the morphological features of feeder cells existing around the iPS cells, the length of the feeder cells at 48 hours is 20 or more and the brightness of the feeder cells at 24 hours is 30 or more. The evaluation processing unit 48a classifies iPS cells into level 1. Thus, the evaluation processing unit 48a estimates that the iPS cell proliferation ability is high.
Then, the evaluation processing unit 48a has a capability of differentiating iPS cells into bones and estimates that the proliferation capability is high.

  The CPU 42a controls to change the culture conditions (for example, temperature, humidity, etc.) of the iPS cells or feeder cells based on the state of the iPS cells estimated by the evaluation processing unit 48a. For example, the CPU 42a supplies a control signal to the temperature adjustment device 15a and controls the temperature inside the temperature-controlled room 15 to increase. In addition, the CPU 42a supplies a control signal to the humidity adjusting device 15b to control the humidity in the temperature-controlled room 15 to decrease.

FIG. 19 is a flowchart for explaining calculation of a teacher value for calculating a decision tree.
First, in step S501, the imaging device 34 continuously acquires a reference image in which iPS cells and feeder cells existing around the iPS cells and cultured together are acquired and stored in the storage unit 43a.

Next, in step S502, after culturing for a certain period, the user stains the imaged cells with a staining marker (for example, SSEA-4, SSEA-1, nanog, etc., which are staining markers indicating the degree of differentiation).
The CPU 42a may control to add the staining marker previously placed in the container to the culture solution after culturing for a certain period. Thereby, the imaged cell can be stained with the staining marker.

Next, in step S503, the CPU 42a determines the presence or absence of staining for each cell from the stained image.
In the next step S504 and step S505, the CPU 42a evaluates the evaluation result as “level” or “group”, and creates the following teacher value for each image and teacher value for each cell. Note that the teacher value may be based not only on the staining result but also on the evaluation by an expert.

<Teacher value by image>
First, in step S504, for each image, the CPU 42a determines the presence or absence of staining for each object in the image recognized as a cell included in the image. When determining the staining, the CPU 42a determines “stained” if the luminance value of the object is greater than a predetermined threshold. On the other hand, the index extraction unit 47a determines “no staining” when the luminance value of the object is equal to or lower than a predetermined threshold.

  Then, the CPU 42a calculates the ratio of the number of cells determined as “stained” to the total number of cells. The CPU 42a determines the classification name of “level” or “group” for each image based on the ratio. For example, the CPU 42a determines the classification name “differentiation group: bone” for a certain image.

<Teaching value by cell>
Next, in step S505, the CPU 42a stains each object in the image recognized as a cell in each sample (= data under a certain experimental condition composed of a plurality of images). Give each cell a “level” or “group” name based on the presence or absence and the experience of the skilled person. Here, a group such as “undecidable” and “trash” may be created.

  Next, in step S506, when the process from step S501 to step S505 has not been performed a predetermined number of times under all conditions (NO in step S506), the process returns to step S501. On the other hand, in step S506, when the processing from step S501 to step S505 is performed a predetermined number of times under all conditions (YES in step S506), this flowchart ends.

FIG. 20 is a flowchart for explaining calculation of a feature amount for determining a decision tree. The process of the flowchart in FIG. 20 is executed by the CPU 42a in response to an instruction from the user.
First, in step S601, the CPU 42a reads data of a microscope image of a sample prepared in advance from the storage unit 43a. Note that the CPU 42a in S601 also acquires information indicating the total number of cells and the number of iPS cells corresponding to each image at this time.

  Next, in step S602, the CPU 42a designates an image to be processed from among the microscopic images of the sample. Here, it is assumed that the CPU 42a in S602 sequentially designates all the microscope images of the sample prepared in advance as processing targets.

  Next, in step S603, the region extraction unit 44 extracts the iPS cell colony region and the feeder cell region included in the image from the microscopic image to be processed.

  Next, in step S604, the feature amount calculation unit 45a obtains feature amounts indicating the morphological features of the cells for each image object extracted from the image by the region extraction unit 44, as in the first embodiment. . As an example, the feature quantity calculation unit 45a obtains the 19 types of feature quantities shown in FIG. 6 for each cell.

  Next, in step S605, the feature amount calculation unit 45a obtains the average value and the standard deviation within the image of the 19 types of feature amounts for the teacher image to be calculated. Then, the feature amount calculation unit 45a records the in-image average value and the in-image standard deviation of each feature amount in the storage unit 43a.

  Next, in step S606, when the feature amount calculation unit 45a has not calculated the feature amounts of all the images (NO in S606), the process returns to step S602. On the other hand, when the feature amount calculation unit 45a calculates the feature amounts of all the images (S606 YES), this flowchart ends.

FIG. 21 is a flowchart for explaining calculation of a decision tree for each image. In advance, in the flowchart of FIG. 20, the feature amount calculation unit 45a calculates the average value (input value 1) of each morphological feature amount for each time period and the standard deviation (input value 2) of each morphological feature amount for each time period. It shall be.
First, in step S701, the frequency distribution calculation unit 46 calculates the frequency distribution (input value 3) of each morphological feature amount over time.

Next, in step S702, the index extraction unit 47a calculates, for each morphological feature amount, the difference in the area of the frequency distribution of the feature amount (input value 4) in all combinations over time.
Next, in step S703, when the index extraction unit 47a has not calculated the difference in the area of the frequency distribution of the feature amount over time for all the feature amounts (NO in S703), the index extraction unit 47a The process returns to S702. On the other hand, when the index extraction unit 47a calculates the difference in the area of the frequency distribution of the feature amount over time for all feature amounts (NO in S703), the process proceeds to step S704.

Next, in step S704, the index extraction unit 47a groups the data using the four input values for each image and the teacher value for each image to construct a decision tree. Here, the teacher value is, for example, high, medium or small if it is a level indicating the degree of differentiation of a cell, and is bone, cartilage or fat if it is a group indicating the type of differentiation.
Thus, each “level” or each “group” is divided. For example, a level indicating the degree of cell differentiation is classified, or a group indicating each type of differentiation is classified.

  Next, in step S705, the index extraction unit 47a stores the structure of the decision tree, the type of index when grouped, and the index threshold for each index type in the storage unit 43a. Above, this flowchart is complete | finished.

FIG. 22 is a flowchart for explaining calculation of a decision tree for each cell.
First, in step S801, the index extraction unit 47a stores a shape index for each object recognized as a cell in each image calculated by the feature amount calculation unit 45a for each experimental condition. Read from.

  Next, in step S802, the index extraction unit 47a groups the data using the four input values for each cell and the teacher value for each cell to construct a decision tree. Thus, the “level” or “group” of each cell is classified for each cell in the target image. Therefore, in each cell, the “level” or “group” of the cell can be specified from the combination of the shape indicators.

  Next, in step S803, the index extraction unit 47a stores the structure of the decision tree, the index type when grouped, and the index threshold for each index type in the storage unit 43a. Above, this flowchart is complete | finished.

  Next, three methods for estimating the state of an iPS cell from a target image using a decision tree that is a discrimination model will be described. First, a method for estimating the state of an iPS cell population included in a target image from the morphological feature amount of a feeder cell and the feature amount of a colony of iPS cells will be described.

FIG. 23 is a flowchart for explaining the estimation of the state of iPS cells using the image-specific colony + feeder model shown in FIG.
First, in step S901, the imaging device 34 continuously acquires a reference image in which iPS cells and feeder cells existing around the iPS cells and cultured together are acquired and stored in the storage unit 43a.

Next, in step S902, the feature amount calculation unit 45a calculates the four input values.
Next, in step S903, the evaluation processing unit 48a reads from the storage unit 43a the type of index used for each image, the colony + feeder model, and the threshold value of the index.

Next, in step S904, the evaluation processing unit 48a reads a value corresponding to the read index type from the four input values calculated by the feature amount calculation unit 45a.
Then, the evaluation processing unit 48a classifies the iPS cells into a predetermined group by comparing a value corresponding to the read index type with a threshold value of the index. Thereby, the evaluation processing unit 48a can estimate the state of the iPS cell.

For example, the evaluation processing unit 48a corresponds to the index types “area difference between histograms of feeder cells at 96 hours and 24 hours” and “average roundness value of colonies at 24 hours” from the above input values. Each value is read (see FIG. 16).
Then, if the value corresponding to “the area difference between the histograms of the feeder cells at 96 hours and 24 hours” is equal to or greater than the threshold value (20) of the index, the evaluation processing unit 48a converts the iPS cells into “bones”. Classify into a group that has the ability to differentiate.

  Thus, the type of iPS cell differentiation can be estimated from the temporal change in the frequency distribution of the morphological feature amount of the feeder cell population existing around the iPS cell.

  On the other hand, the evaluation processing unit 48a determines that the value corresponding to the “area difference between the histograms of the feeder cells at 96 hours and 24 hours” of the input values is less than the threshold (20) of the index, If the value corresponding to the “average value of the roundness of colonies at 24 hours” is equal to or greater than the threshold (30) of the index, the iPS cells are classified into a group “having the ability to differentiate into cartilage”.

  This makes it possible to estimate the type of iPS cell differentiation from the temporal change of the frequency distribution of the morphological feature amount of the feeder cell population existing around the iPS cell and the morphological feature amount of the colony population of iPS cells. it can. Above, this flowchart is complete | finished.

Second, a method for estimating the state of the iPS cell population included in the target image from the morphological feature amount of the feeder cell and the feature amount of the colony of the iPS cell will be described.
FIG. 24 is a flowchart for explaining the estimation of the state of iPS cells using the image-specific and colony-feeder-specific models shown in FIG.

First, step S1001 to step S1002 are the same as step S901 to step S902, and thus description thereof is omitted.
Next, in step S1003, the evaluation processing unit 48a reads from the storage unit 43a the types of indices used in the image-specific and colony / feeder-specific models and the threshold values of the indices.

Next, in step S1004, the evaluation processing unit 48a reads a value corresponding to the index type of the colony model from the four input values calculated by the feature amount calculation unit 45a.
Then, the evaluation processing unit 48a classifies the iPS cells into a predetermined group by comparing a value corresponding to the read index type with a threshold value of the index.

For example, the evaluation processing unit 48a corresponds to the index types “the average value of the roundness of the colonies at the 24th hour” and “the amount of change in the length of the colonies at the 96th and 24th hours” from the input values. Each value is read (see FIG. 17A).
If the value corresponding to the “average value of roundness of colonies at the 24th hour” among the input values is equal to or greater than the threshold value (20) of the index, the evaluation processing unit 48a differentiates iPS cells into “differentiated into bones”. Into the group “Having the ability to do”.

  Thereby, the type of iPS cell differentiation can be estimated from the morphological feature of the colony population of iPS cells.

Next, in step S1005, the evaluation processing unit 48a reads a value corresponding to the type of index of the feeder model from the four input values calculated by the feature amount calculation unit 45a.
Then, the evaluation processing unit 48a classifies the iPS cells into a predetermined level by comparing a value corresponding to the read index type with a threshold value of the index.

For example, the evaluation processing unit 48a uses the above-described input values as index types “the average value of the roundness of the feeder cells at 48 hours” and “the area difference between the histograms of the rounds of the feeder cells at 96 hours and 24 hours”. Each of the values corresponding to is read (see FIG. 17B).
If the value corresponding to “the average roundness value of the feeder cells at 48 hours” among the input values is equal to or greater than the threshold value (40) of the index, the evaluation processing unit 48 a ”.

  Thereby, the degree of iPS cell degradation can be estimated from the morphological feature of the feeder cell population.

Next, in step S1006, the evaluation processing unit 48a collectively estimates the state of the iPS cell based on the classified group and level.
For example, since the evaluation processing unit 48a is classified into group 1 (differentiation group: bone) in the colony model and level 1 (degradation degree: large) in the feeder model, the iPS cells are differentiated into bone, but the deterioration is severe. A complex estimation is made. Above, this flowchart is complete | finished.

Finally, a method for estimating the state of the iPS cell population included in the target image from the morphological feature amount of the feeder cell and the feature amount of the colony of the iPS cell will be described.
FIG. 25 is a flowchart for explaining the estimation of the state of iPS cells using the cell-specific and colony-feeder-specific models shown in FIG.

First, Steps S1101 to S1102 are the same as Steps S901 to S902, and thus description thereof is omitted.
Next, in step S1103, the evaluation processing unit 48a reads out from the storage unit 43a the types of indices used in the cell-by-cell and colony-feeder models and the threshold values of the indices.

Next, in step S1104, the evaluation processing unit 48a reads a value corresponding to the index type of the colony model from the four input values calculated by the feature amount calculation unit 45a.
Then, the evaluation processing unit 48a classifies the iPS cells into a predetermined group by comparing a value corresponding to the read index type with a threshold value of the index.

For example, the evaluation processing unit 48a determines a value corresponding to the index types of “the roundness of the colony at the 24th hour” and “the area difference between the histograms of the roundness of the colonies at the 96th and 24th hours” from the input value. Are read out (see FIG. 18A).
If the value corresponding to the “roundness of the colony at the 24th hour” among the input values is less than the threshold value (80) of the index, the evaluation processing unit 48a “is capable of differentiating iPS cells into fat. Group.

  Thus, the type of iPS cell differentiation contained in the colony of the iPS cell can be estimated from the morphological feature amount of one colony of the iPS cell.

Next, in step S1105, the evaluation processing unit 48a reads a value corresponding to the type of index of the feeder model from the four input values calculated by the feature amount calculation unit 45a.
Then, the evaluation processing unit 48a classifies the iPS cells into a predetermined level by comparing a value corresponding to the read index type with a threshold value of the index.

For example, the evaluation processing unit 48a reads values corresponding to the index types “the length of the feeder cell at the 48th hour” and “the brightness of the feeder cell at the 24th hour” from the input values (FIG. 18B). )reference).
And the evaluation process part 48a will call an iPS cell "low proliferation ability" if the value applicable to "the length of the feeder cell of the 48th hour" among input values is less than the threshold value (20) of the parameter | index. Classify into levels.

  This makes it possible to estimate the type of iPS cell differentiation present in the surrounding area from the morphological feature of one feeder cell.

Next, in step S1106, the evaluation processing unit 48a collectively estimates the state of the iPS cell based on the classified group and level.
For example, since the evaluation processing unit 48a is classified into group 3 (differentiation group: fat) in the colony model and level 3 (proliferation ability: low) in the feeder model, the iPS cells have differentiation ability to differentiate into fat. However, it makes a complex estimate that the proliferation ability is low. Above, this flowchart is complete | finished.

As described above, according to the second embodiment of the present invention, the iPS cell population included in the target image is determined from the morphological feature amount of the feeder cell population included in the target image. It can be divided into “groups”.
Furthermore, from the morphological feature amount of the feeder cell population included in the target image and the morphological feature amount of the iPS cell population included in the target image, the iPS cell population included in the target image is set to a predetermined “level”. Or a predetermined “group”.

Further, one iPS cell in the target image can be classified into a predetermined “level” or a predetermined “group” based on the morphological feature amount of one feeder cell included in the target image.
Furthermore, from the morphological feature amount of one feeder cell included in the target image and the morphological feature amount of one iPS cell included in the target image, one iPS cell in the target image is given a predetermined “ It can be divided into “levels” or predetermined “groups”.

In the second embodiment, separate decision trees are constructed for the iPS cell colony feature amount and the feeder cell feature amount. However, the present invention is not limited to this, and a decision tree using both feature amounts is used. May be constructed.
In the second example, the direction of iPS cell differentiation, the degree of iPS cell degradation, or the degree of iPS cell proliferation was estimated. However, the present invention is not limited to this. It may be good.

<Example of cell culture method>
FIG. 26 is a flowchart for explaining a cell culturing method using the cell evaluation program of Example 2 of the present invention. In this flowchart, it is assumed that a petri dish containing iPS cells to be grown is placed in the incubator 11.

  First, in step S1201, the CPU 42a controls the temperature adjustment device 15a and the humidity adjustment device 15b so as to culture cells under predetermined culture conditions (for example, a predetermined temperature, humidity, etc.). After a predetermined period, in step S1202, the CPU 42a executes the cell evaluation program of the embodiment of the present invention read from the storage unit 43a, and evaluates the state of the cells. Next, in step S1203, the CPU 42a determines whether to stop the culture, change the culture conditions, or continue the culture under the current culture conditions, based on the evaluation result of the cell state.

Next, when it is determined in step S1204 that the CPU 42a stops culturing (YES in step S1204), the CPU 42a displays information indicating that the culturing is stopped on a monitor or the like (not shown) and ends the process.
On the other hand, when the CPU 42a does not determine that the culture is to be stopped (NO in step S1204), the CPU 42a proceeds to the process of step S1205.

Next, when it is determined in step S1205 that the CPU 42a changes the culture condition (YES in step S1205), the CPU 42a changes to the culture condition (eg, predetermined temperature, humidity, etc.) associated with the evaluation result of the cell state. The temperature adjusting device 15a and the humidity adjusting device 15b are controlled so as to be performed (step S1206). The CPU 42a returns to the process of step S1202.
On the other hand, when the CPU 42a determines that the culture is continued under the current culture conditions (NO in step S1205), the CPU 42a proceeds to the process of step S1207.

  Next, in step S1207, the CPU 42a extracts an object corresponding to an iPS cell from an image captured by the imaging device 34 every time a predetermined time elapses, and counts the number of objects, that is, the number of cells.

Next, in step S1208, when the counted number of cells has not reached the predetermined number of cells (NO in step S1208), after the time corresponding to the counted number of cells has elapsed, the CPU 42a proceeds to the process of step S1202. Return.
On the other hand, when the counted number of cells has reached the predetermined number of cells (YES in step S1208), the CPU 42a displays information indicating that the culture of the cells is completed on a monitor or the like (not shown). Above, this flowchart is complete | finished.

As described above, if the cell culture method of the present invention is used, the culture conditions can be dynamically changed based on the evaluation of the cells, so that iPS cells can be easily proliferated.
In addition, although the cell culture conditions were changed using the cell evaluation program of Example 2 of this invention, it is not restricted to this, The cell culture conditions are changed using the cell evaluation program of Example 1 of this invention. May be.

In the embodiment of the present invention, iPS cells have been described. However, the present invention is not limited to this, and ES cells or other cells may be used.
Further, in the embodiment of the present invention, in order to estimate the state of iPS cells, the morphological features of other iPS cells are used as an index. The amount may be used as an index.

In the embodiment of the present invention, the feeder cell is described as one type of cell. However, the present invention is not limited to this, and the feeder cell may be configured using a plurality of types of cells. In that case, what is necessary is just to use the morphological feature-value of at least 1 type of feeder cell among the multiple types of feeder cells as a parameter | index.
In the embodiment of the present invention, two types of cells, iPS cells and feeder cells, are cultured together. However, the present invention is not limited to this, and other types of cells are cultured together with iPS cells and feeder cells. It may be.

<Supplementary items of the embodiment>
(1) In the above embodiment, the example in which the information processing device is incorporated in the control device 41 or the control device 41a included in the incubator 11 or the incubator 11a has been described. You may comprise the external independent computer which acquires a microscope image from the incubator 11a and performs analysis (illustration in this case is abbreviate | omitted).

  (2) In the above embodiment, the region extraction unit 44, the feature amount calculation unit 45 or the feature amount calculation unit 45a, the frequency distribution calculation unit 46, the index extraction unit 47 or the index extraction unit 47a, the evaluation processing unit 48 or the evaluation processing unit Although an example has been described in which each function of 48a is realized by software as a program, these processes may of course be realized by hardware using an ASIC.

  (3) In the above-described one embodiment, the example in which the control device 41 or the control device 41a obtains the calculation model of the evaluation information and the index thereof by the FNN has been described. However, the culture state evaluation apparatus of the present invention may obtain a calculation model of evaluation information and its index by other multivariate analysis such as multiple regression analysis.

  Furthermore, the culture state evaluation apparatus of the present invention may generate final evaluation information by combining a plurality of calculation models and performing majority (or weighted average) of calculation results based on these calculation models. In this case, for example, data with a low mixing ratio has a strong MRA, and data with a high mixing ratio has a strong FNN, so that one of the calculation models can cover a situation with low accuracy by another model. The accuracy of evaluation information can be further increased.

  (4) Moreover, when the culture state evaluation apparatus calculates | requires the calculation model of evaluation information, you may make it increase / decrease a parameter | index by the variable increase / decrease method by calculating the calculation result by combining a several parameter | index first. Further, all indicators may be used if the accuracy of the data is high.

  (5) In the above embodiment, the example in which the amount of change in the frequency distribution is obtained using the sum of absolute values of the differences between the two frequency distributions has been described. However, the frequency distribution is calculated from the sum of squares of the differences between the two frequency distributions. The amount of change may be obtained. Further, the calculation formula shown in the above-described one embodiment is merely an example, and may be, for example, a quadratic or higher order n-order equation.

  (6) The feature values shown in the above-described embodiments and examples are merely examples, and it is of course possible to employ other feature value parameters according to the type of cells to be evaluated.

  As mentioned above, although embodiment of this invention was explained in full detail with reference to drawings, the concrete structure is not restricted to this embodiment, The design etc. of the range which does not deviate from the summary of this invention are included.

DESCRIPTION OF SYMBOLS 11 Incubator 11a Incubator 15 Temperature-controlled room 19 Culture container 22 Observation unit 34 Imaging device 41 Control device 41a Control device 42 CPU
42a CPU
43 storage unit 43a storage unit 44 region extraction unit 45 feature quantity calculation unit 45a feature quantity calculation unit 46 frequency distribution calculation unit 47 index extraction unit 47a index extraction unit 48 evaluation processing unit 48a evaluation processing unit

Claims (11)

In the target image obtained by imaging the first type of cells and the second type of cells that are present around the first type of cells and are cultured together, the individual cells of the first type and A region extraction unit that extracts a region occupied by an aggregate of cells and a region occupied by the second type of cells;
A first feature amount calculation unit that calculates a feature amount indicating a morphological feature of the second type of cells present in a region occupied by the second type of cells extracted by the region extraction unit;
Evaluation that evaluates the state of the first type of cell based on the correspondence between the temporal change of the feature quantity indicating the morphological characteristics of the second type of cell and the state of the first type of cell A processing unit;
A cell evaluation apparatus comprising: A first frequency distribution calculating unit that calculates a frequency distribution of the feature amount of the second type of cells for each target image;
The evaluation processing unit extracts a change in the frequency distribution calculated by the first frequency distribution calculation unit between target images acquired at different times, and the first type according to the extracted change The cell evaluation apparatus according to claim 1, wherein the state of the cell is evaluated. A second frequency distribution calculating unit that calculates the frequency distribution of the feature quantity of the first type of cells for each target image;
The evaluation processing unit calculates a change in the frequency distribution calculated by the first frequency distribution calculating unit between the target images acquired at different times and the second frequency distribution calculating unit between the target images. The cell evaluation apparatus according to claim 2, wherein a change in the frequency distribution is extracted, and a state of the first type of cell is evaluated based on the extracted change. A storage unit for storing a reference image obtained by imaging a state in which cells of the same type as the first type of cells and the second type of cells are cultured together;
An index for classifying the quality of the first type of cell based on the information of the time when the reference image was captured and the feature quantity indicating the morphological feature of the second type of cell in the reference image A first index calculation unit for calculating;
Further comprising
The evaluation processing unit collates the index calculated by the first index calculation unit with a feature amount indicating a morphological feature of the second type of cells extracted from the target image, and The cell evaluation apparatus according to claim 1, wherein the state of one type of cell is evaluated. Information on the time at which the reference image was captured, feature quantities indicating morphological characteristics of the first type of cells in the reference image, and morphological characteristics of the second type of cells in the reference image A second index calculating unit that calculates an index for classifying the quality of the first type of cell based on the feature amount;
Further comprising
The evaluation processing unit includes the index calculated by the first index calculation unit, the index calculated by the second index calculation unit, and the form of the first type of cells extracted from the target image The cell evaluation apparatus according to claim 4, wherein the state of the first type of cell is evaluated by collating with a feature amount indicating a characteristic feature.   The cell evaluation apparatus according to any one of claims 1 to 5, wherein the first type of cell is an ES cell or an iPS cell.   The cell evaluation apparatus according to any one of claims 1 to 6, wherein the second type of cell is a feeder cell. A temperature-controlled room that accommodates a culture vessel for culturing cells and can maintain the interior in a predetermined environmental condition;
An imaging device for capturing an image of the cells contained in the culture vessel in the temperature-controlled room;
The cell evaluation apparatus according to any one of claims 1 to 7,
An incubator comprising: In the target image obtained by imaging the first type of cells and the second type of cells that are present around the first type of cells and are cultured together, the individual cells of the first type and A region extraction procedure for extracting a region occupied by an aggregate of cells and a region occupied by the second type of cells;
A first feature amount calculation procedure for calculating a feature amount indicating a morphological feature of the second type of cells present in the region occupied by the second type of cells extracted in the region extraction procedure;
Evaluation that evaluates the state of the first type of cell based on the correspondence between the temporal change of the feature quantity indicating the morphological characteristics of the second type of cell and the state of the first type of cell Processing procedure and
The cell evaluation method characterized by having. In the target image obtained by imaging the first type of cells and the second type of cells that are present around the first type of cells and are cultured together, the individual cells of the first type and A first step of extracting a region occupied by an aggregate of cells and a region occupied by the second type of cells;
A second step of calculating a feature amount indicating a morphological feature of the second type of cells present in an area occupied by the second type of cells extracted in the first step;
The first type of cell state is evaluated based on the correspondence between the temporal change in the feature amount indicating the morphological characteristics of the second type of cells and the state of the first type of cells. 3 steps,
Cell evaluation program to make computer execute. The morphological characteristics of the second type cells are obtained from an image obtained by imaging the second type cells existing around the first type cells and the first type cells and cultured together. Extract the feature quantity to show,
Estimating the state of the first type of cell based on the temporal change of the characteristic amount indicating the morphological characteristics of the second type of cell;
A method for culturing a cell, characterized by changing a culture condition of the first type of cell or the second type of cell based on the state of the first cell.