JP2023086104A - Cell evaluation method and cell evaluation device - Google Patents

Abstract

To select better cells from cells having a linear structure.SOLUTION: A cell evaluation method comprises: an imaging step of imaging a cell having a linear structure to obtain an image; a data acquisition step of acquiring, from the image, data indicating a growth state of a linear portion extending from the cell body of the cell; and a determination step of determining the growth state of the linear portion from the data. The data includes data on the area occupancy of the linear portion and the reaching distance, which are calculated by processing the image.SELECTED DRAWING: Figure 2

Description

TECHNICAL FIELD The present invention relates to a cell evaluation method and cell evaluation device, and more particularly to a cell evaluation method and cell evaluation device for determining the growth state of cells having linear structures.

Conventionally, cells having a linear structure such as nerve cells were stained and visually observed to evaluate their growth state. However, there are problems such as that cells are damaged by staining, and evaluation by visual observation may lack objectivity.
In recent years, cell evaluation methods for determining the growth state of neurons by extracting objects from captured images and using the length and area of neurites have been studied (for example, Patent Documents 1 and 2).

JP-A-2002-207993 Japanese Patent Application Laid-Open No. 2021-27836

As a result of further studies by the present inventors, for the transmission of cell information, nerve fibers (neurites) composed of dendrites and/or axons are thick, It is important to have a large number and spread. From this, it was clarified that by evaluating the area occupancy rate and tip position of nerve fibers in nerve cells, it becomes possible to select nerve cells that are more excellent in information transmission.

The present invention has been made in view of the circumstances described above, and an object of the present invention is to provide a cell evaluation method and a cell evaluation device for selecting better cells from among cells having a linear structure.

The present invention has been made based on the above findings, and an object of the present invention is to advantageously solve the above problems. A first aspect of the present invention is to image a cell having a linear structure. a data acquisition step of acquiring data indicating the growth state of the linear portion extending from the cell body of the cell from the image; and a growth state of the linear portion from the data. and a determination step of determining the cell evaluation method.
By such a method, cells having linear structures such as nerve cells and hair cells can be selected for superior cells.

In the above aspect, the cell having the linear structure may be a nerve cell, and the linear portion extending from the cell body may be a dendrite and/or an axon.
This makes it possible to select better neurons.

Further, in the above aspect, the determining step is performed by machine learning based on a linear portion database in which linear portion growth stages, their area occupancy rates, and reaching distances are associated with each other. may include calculating a score representing the
By doing so, it is possible to select cells that are predicted to have a high degree of growth in advance.

Further, in the above aspect, the data further includes image data of cell body shape, and the determining step is performed by machine learning based on a cell body shape database that associates cell body growth stages with cell shapes. It may include determining the growth state of the nerve cells from the body shape image data.
By doing so, it is possible to evaluate the growth state of nerve cells and to select nerve cells with a high degree of growth.

In the above aspect, the determining step may include determining the growth state of the nerve cell from the data by machine learning based on a nerve fiber database in which nerve fibers and their characteristics are associated. .
By doing so, it becomes difficult to be affected by noise and foreign matter that cannot be removed by image processing alone, and the growth state of nerve cells can be determined with higher accuracy.

In addition, a second aspect of the present invention shows an imaging unit that acquires an image by imaging a cell having a linear structure, and the growth state of the linear part extending from the cell body of the cell from the image. A cell evaluation device, comprising: a data acquisition unit that acquires data; and a determination unit that determines the growth state of the linear portion from the data, wherein the data is calculated by processing the image. The cell evaluation device includes data on the area occupancy of the linear portion and the reaching distance.
With such an apparatus, better cells can be selected from cells having linear structures such as nerve cells and hair cells.

In the above aspect, the cell having the linear structure may be a nerve cell, and the linear portion extending from the cell body may be a dendrite and/or an axon.
This makes it possible to select better neurons.

According to the present invention, better cells can be selected from cells having a linear structure.

FIG. 1 is a schematic diagram for explaining steps of image processing. FIG. 2 is a schematic diagram illustrating an image after image processing.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail.
The method according to the first aspect of the present invention comprises an imaging step of imaging a cell having a linear structure to obtain an image; A cell evaluation method, comprising: a data acquisition step of acquiring data indicating the growth state of the linear portion; and a determination step of determining the growth state of the linear portion from the data, wherein the data is obtained by processing the image One of the characteristics is that data of the calculated area occupation ratio and reaching distance of the linear portion is included.
By using such a cell evaluation method, it becomes possible to select functionally superior cells.
Also, each of the above steps may be included in a computer program stored in computer-readable means and executed by a computer. Another aspect of the present invention is a computer program medium containing the above computer program.

A method according to a second aspect of the present invention includes an imaging unit that acquires an image by imaging a cell having a linear structure; a data acquisition unit that acquires data indicating the
One of the characteristics of the cell evaluation apparatus, wherein the data includes data of the area occupation ratio and the reaching distance of the linear portion, which are calculated by processing the image.
By using such a cell evaluation device, it becomes possible to select cells that are more functionally excellent.
In the above aspect, the imaging unit is, for example, a microscope device, and the data acquisition unit is, for example, a computer having a central processing unit (CPU), a semiconductor memory, a hard disk, and the like. and a determination unit in which one aspect of the determination program of the present invention is installed in a storage medium.

(Cells with linear structure)
Examples of cells having a linear structure include, but are not limited to, nerve cells, hair cells, and the like, as long as they have a linear structure and the extension part of the cell extends as it grows.
A linear part of a cell having a linear structure refers to a nerve fiber when the cell is a nerve cell, and the nerve fiber is composed of dendrites and axons.
Cells to be evaluated may be cells of a single type, or may be a group of cells composed of a plurality of types of cells.

(imaging)
For imaging, for example, a bright-field microscope equipped with an image sensor such as a CCD (Charge-Couple Device) image sensor or a CMOS (Complementary Metal-Oxide Semiconductor) image sensor is used to image the cells housed in the culture vessel. and outputs an image of the captured cell. Specifically, a bright-field image of the cells housed in the culture vessel is formed on an imaging device, and the bright-field image is output from the imaging device as a cell image.

As the imaging device, an imaging device provided with RGB color filters may be used, or a monochrome imaging device may be used. The microscope apparatus is not limited to a bright field microscope apparatus, and may be a phase contrast microscope apparatus, a differential buffer microscope apparatus, or the like. The cell image to be captured may be an image in which a plurality of cells are dispersed and distributed, an image in which cells are aggregated into a clump such as a spheroid, or an image such as an organoid. It may be a three-dimensional image of a cell.
The culture vessel used for imaging is not particularly limited and may be a commercially available cell culture dish or the like as long as the target cells can be cultured.

(data acquisition)
The captured cell image is subjected to the following image processing as necessary (for example, when the image contains noise, foreign substances, cell bodies, and other portions other than the linear portion), and then the data is acquired.
Specifically, for example, as shown in FIG. 1, an image (image 1) is generated by subjecting the above cell image (original image) to threshold cutoff processing. Threshold cut-off processing involves setting a predetermined threshold for the brightness (luminance) of each pixel, and for pixels with brightness above (brighter) than the threshold, the brightness is reduced to the threshold, and pixels below (dark) the threshold are reduced. This is a process for maintaining the luminance of a pixel having a luminance of . This process masks portions other than the target portion for data acquisition (linear portions of cells having linear structures, e.g., nerve fibers), such as foreign substances and cell bodies of cells having linear structures. can do.

Next, an image (image 2) is generated by subjecting image 1 to adaptive thresholding. Adaptive threshold processing is a process of setting a threshold for each set area range and binarizing, and can reduce the effects of shadows and noise in an image. This image 2 is inverted (image 3), and only the portion of interest for data acquisition (for example, nerve fibers) is extracted.

After the image processing as described above is performed as necessary, the growth state of the linear portion, including the area occupancy of the linear portion and the reaching distance obtained from the position in the culture vessel, is shown by image processing calculation. Ask for data.
The position in the culture vessel is, for example, a position detected by dividing the captured cell image into 8 columns×8 columns and defining 64 sections, as shown in FIG. 2 . By doing so, the nerve fiber elongation start position (branch point between the cell body and the nerve fiber) and the tip position (reaching position) are detected to calculate the reaching distance (elongation distance) of the nerve fiber, and The area occupancy of nerve fibers in each compartment can be calculated.

(data)
The data indicating the growth state of the cell body of the cell and the linear part extending from the cell body include the area occupancy rate of the part (for example, nerve fiber) extracted by image processing calculation, the reaching distance, the cell body shape data, etc. However, it is not limited to these, and may include the length, thickness, density, etc. of the extracted portion.

(judgement)
From the above data, the growth state of the linear portion is determined. When the cells are nerve cells, cells with a high area occupancy of nerve fibers in the captured cell image and with a long reaching distance are judged to be in a high state of growth.
Specifically, for example, when judging from a cell image taken at a magnification of 20 times, the area occupancy rate of nerve fibers is 25 to 33%, and if the reaching distance is equivalent to the distance between cell bodies, the growth state is judged to be high. The above determination criteria vary depending on the type and concentration of target cells.

The above determination may be made by machine learning in a computer.
In the determination step of one embodiment, machine learning based on a linear portion database in which growth stages of linear portions and their area occupancy, reaching distances, number of culture days, culture environments, etc. are associated with each other is performed using the above image processing operations. Based on the data obtained, a score representing the predicted growth rate of the cells is calculated.

The determination may be performed by a predictor machine-learned in advance so as to predict the degree of growth of the cell to be determined and output a score based on cell data of various growth stages. Machine learning of the predictor is performed, for example, as follows.
First, from a plurality of cell images acquired during a predetermined cell culture period, evaluation results such as the area occupancy rate and the reachable distance are acquired by the above-described image processing calculations.
Next, the above evaluation results and data such as culture environment data such as culture days, culture temperature, culture humidity, and CO ₂ concentration are used as learning data, and scored by random forest. The score may be expressed on a scale of 10, for example, "score 10" for cells with good growth and "score 1" for cells with poor growth.
The evaluation result obtained from the cell image in this manner, the number of culture days, the culture environment data, and the correspondence relationship between the scores are machine-learned by the predictor and used for the determination.

The machine learning algorithm is not limited to random forest, but linear regression, decision tree, k-nearest neighbor method (k-NN), neural network, naive Bayes, logistic regression, support vector machine (SVM), etc. can be used. can.

In the determination step of another embodiment, the data further includes image data of the cell body shape, and the cell body shape is determined by machine learning based on a cell body shape database in which cell body growth stages and their shapes are associated. Determine the growth state of neurons from the data of

The above judgment is performed by a predictor that has undergone machine learning (deep learning) in advance so as to predict the growth degree of the nerve cell to be judged based on the image data of the cell body shape at various growth stages and output the growth degree. It may be performed by Machine learning of the predictor is performed, for example, as follows.
First, at least two or more cell images of various growth stages are acquired, and cell bodies present in the cell images are annotated (teaching data is created) to create (1) learning image data.
Next, for the cell having the annotated cell body, (2) the degree of growth, such as the area occupancy rate and reaching distance of the portion (for example, nerve fiber) extracted by the image processing operation, is calculated.
The correspondence relationship between (1) learning image data and (2) the degree of growth acquired in this manner is machine-learned by the predictor and used for the above determination.

As the machine learning method, CNN (Convolutional Neural Network) may be used. Specifically, R-CNN (Regions with Convolutional Neural Network) for object detection, YOLO (You Only Look Once), SSD (Single Shot Multi Box Detector) or the like may be used.

In yet another embodiment, the determining step determines the growth state of nerve cells by machine learning based on a nerve fiber database in which nerve fibers and their properties are associated.

The above determination may be performed by a predictor that has undergone machine learning (deep learning) in advance so as to determine the growth state of the nerve cell to be determined based on nerve fibers in various stages of growth and their characteristic data. good. Machine learning of the predictor is performed, for example, as follows.
First, a plurality of cell images are obtained, and the evaluation result of the growth state of nerve cells is obtained by the image processing operation described above.
Next, each of the plurality of cell images is labeled based on the evaluation results. For example, “region without nerve fibers”, “with nerve fibers (small area occupation ratio)”, “with nerve fibers (large area occupation ratio)”, etc. are classified according to the area occupation ratio of nerve fibers and the thickness of nerve fibers. You can divide it.
It should be noted that the evaluation of the growth state of nerve cells may be evaluated by a method different from the image processing calculation described as an example in this embodiment. may be used to input the evaluation result.
A predictor machine-learns the correspondence between the cell images obtained in this manner and their labeling, and is used for the above determination.

As the machine learning method, CNN (Convolutional Neural Network) may be used. Specifically, R-CNN (Regions with Convolutional Neural Network) for object detection, YOLO (You Only Look Once), SSD ( Single Shot Multi Box Detector) or the like may be used.

Further, the calculation of the score representing the predicted degree of cell growth and the evaluation of the growth state of nerve cells in each of the above embodiments may be used in combination for determination.

EXAMPLES The present invention will be specifically described below with reference to Examples, but the present invention is not limited to these Examples.

(Example I)
Culture conditions Commercially available healthy iPS-derived motor neurons and ALS patient-derived motor neurons (manufactured by iXCells) were placed in a U-bottom 96-well plate (EZ-BindShut II, manufactured by ADC Technoglass) with a low adhesion surface at 6.4×. Each spheroid was prepared by seeding at a concentration of 10 ⁴ cells/well.
After culturing the spheroids for 7 days, they were seeded in one well of a microchannel plate coated with a commercially available mouse tumor-derived basement membrane extract (prepared by the method described in WO 2021/015213). The cells were cultured until the axons were sufficiently elongated, and the cells were evaluated on the 21st day.

Evaluation Method A microchannel plate was placed in a microscope (manufactured by Leica, model number: DMi8), and an 8-bit cell image was obtained by imaging with an objective lens magnification of 10 times. The resulting cell images were entered into the original application programmatically using MATLAB® functions.
In the application, first, only the range in which axons exist was selected from among the captured cell images, and the signal intensity of the image was uniformed by standardizing the selected range. After that, adaptive thresholding was performed on the standardized image, and the brightness of the resulting image was inverted to obtain an image in which only the nerve fiber portions were highlighted.
The obtained image was divided into 64 divisions of 8 vertical×8 horizontal, and the area occupation ratio of the axons in each division was calculated. For the longest axon, the axonal reach was calculated from the elongation start position and the tip position.

Results The longest axonal reach of healthy iPS-derived motor neurons was 0.200 mm. In addition, Table 1 shows the area occupancy rate of each section thus obtained.
The maximum axonal reach of ALS patient-derived motor neurons was 0.175 mm. In addition, Table 2 shows the area occupation ratio of each section obtained.
From these results, it can be determined that healthy iPS-derived motor cells are in a better state of axonal growth than ALS patient-derived motor neurons.

(Example II)
Culture conditions Commercially available dopaminergic neurons (manufactured by FUJIFILM Cellular Dynamics) were seeded in a poly-D-lysine-coated flat-bottom 96-well plate (3516, manufactured by Corning) at a concentration of 4×10 ⁴ cells/well and cultured for 13 days. After that, each was evaluated.
Evaluation Method A well plate was placed in a microscope apparatus, and an 8-bit cell image was obtained by imaging with an objective lens magnification of 20 times. The resulting cell images were input into an original application program using MATLAB (registered trademark) functions to obtain images in which only nerve fiber (dendritic and axonal) portions were highlighted.
The obtained image was divided into 35 divisions of 5×7, and the area occupation ratio of nerve fibers in each division was calculated. For the longest nerve fiber, the reaching distance of the nerve fiber was calculated from the elongation start position and the tip position.
Results The maximum fiber reach of the dopaminergic nerve was 0.167 mm. In addition, Table 3 shows the area occupation ratio of each section obtained. From these results, it can be seen that the nerve cells elongate the nerve fibers uniformly without aggregation or detachment, which is a good result.

(Example III)
For the cells cultured in Example I, the growth state was determined by a deep learning predictor.
First, a predictor was created as follows.
The day when the spheroids were seeded on the microchannel plate was defined as day 0 of the culture, and the cell images for 37 wells on the 27th day and the 43rd day of culture were each divided into 64 columns of 8×8, and the edge of the plate was reflected. 32 images for 2 days x 37 wells = 2368 images were secured as training data. This learning data was visually classified into three classes: A: area without nerve fibers, B: areas with nerve fibers (small area occupation), and C: areas with nerve fibers (large area occupation). A predictor was created by learning the correspondence relationship between the cell image and the label using CNN.
Cell images of healthy iPS-derived motor neurons and ALS patient-derived motor neurons on day 21 of culture imaged in Example I were input to this predictor. In the healthy iPS-derived motor neuron image, A: 17 images, B: 20 images, and C: 27 images among 64 divided images. On the other hand, in the ALS patient-derived motor neuron image, A: 30 images, B: 32 images, and C: 2 images among the 64-divided images.
From the above results, it was possible to determine that healthy iPS-derived motor cells had a better axon growth state than ALS patient-derived motor neurons.
Machine learning based on the above-mentioned nerve fiber database made it possible to determine the growth state of nerve cells.

以上のように、培養データと培養１３日目の成長度スコアの対応関係を学習させて、予測器を作成した。
続いて、実施例ＩＩで培養した細胞について、培養１日目の細胞画像を取得し、予測される培養１３日目の成長度スコアを算出した。
具体的には、まずウェルプレートを顕微鏡装置に設置し、対物レンズ２０倍で撮像することにより１０ウェル分の８ｂｉt画像を得た。これらの画像をＭＡＴＬＡＢ（商標登録）関数を使用したプログラムによるオリジナルアプリに入力し、神経線維（樹状突起および軸索）部分のみをハイライトさせた画像を得た。これらの得られた画像を縦５×横７の３５分割の区画に分け神経線維の面積占有率を算出し、最も長い神経線維について到達距離を算出した。
上記算出結果と培養日数、培養温度などの培養データを予測器に入力したところ、「面積占有率スコア」２×「到達距離スコア」３＝「成長度スコア」６の結果を得た。この結果から、成長が良好に進むと判断し、培養を継続したところ、予測された通り、実施例ＩＩに記載の良好な成長状態の神経細胞を得た。
上記のような神経線維の成長段階とその面積占有率および到達距離が関連付けられた神経線維データベースに基づく機械学習により、神経細胞の成長を予測し、判定することができた。 (Example IV)
Regarding the cells cultured in Example II, the day on which the cells were seeded was defined as day 0 of culture, and cell images were obtained on day 1 of culture to predict the degree of growth on day 13 of culture.
A predictor for predicting the degree of growth was created as follows.
As learning data, a plurality of cell images 1 on day 1 of culture and cell image 2 on day 13 of culture taken at the same position as cell image 1 were prepared. Cell image 1 and cell image 2 were input into an original application program using MATLAB (registered trademark) functions, and the area occupancy rate and reaching distance were calculated. Based on this calculation result (area occupancy, reaching distance), culture days, culture temperature, culture humidity, _CO2 concentration, cell type, cell concentration, surface modifier type, surface modifier liquid concentration, and cell culture plate material These culture data were scored by random forest as learning data. The scores were determined based on the criteria shown in Table 4, where "area occupancy score" x "reaching distance score" = "growth score", and it was predicted that the larger the value of "growth score", the better the growth. .
The area occupancy rate, which is the basis for scoring, refers to the average area occupancy rate of the area occupancy rates of each section calculated from the cell image, and the reaching distance refers to the reaching distance of the longest nerve fiber in the cell image. In addition, with respect to the reaching distance score criteria, values based on cell culture in a 96-well plate are shown in this example.

As described above, the predictor was created by learning the correspondence relationship between the culture data and the growth score on the 13th day of culture.
Subsequently, for the cells cultured in Example II, a cell image was obtained on day 1 of culture, and a predicted growth score on day 13 of culture was calculated.
Specifically, first, the well plate was placed in a microscope apparatus, and an 8-bit image for 10 wells was obtained by imaging with a 20-fold objective lens. These images were input into the original application programmatically using MATLAB (registered trademark) functions to obtain images in which only the nerve fiber (dendritic and axonal) portions were highlighted. The obtained images were divided into 35 divisions of 5 vertical×7 horizontal, and the area occupancy rate of the nerve fibers was calculated, and the reaching distance was calculated for the longest nerve fiber.
When the above calculation results and culture data such as culture days and culture temperature were input to the predictor, a result of "area occupancy score" 2 x "reaching distance score" 3 = "growth score" 6 was obtained. Based on this result, it was judged that the growth proceeded well, and the culture was continued. As expected, the neurons in the well-growing state described in Example II were obtained.
It was possible to predict and determine the growth of nerve cells by machine learning based on the nerve fiber database in which the growth stages of nerve fibers, their area occupancy rates, and reaching distances are associated as described above.

(Example V)
Regarding the cells cultured in Example II, the day of cell seeding was defined as day 0 of culture, and cell images were obtained on day 1 of culture, and the growth rate on day 13 of culture was predicted from the cell body shape.
A predictor for predicting the degree of growth was created as follows.
As learning data, cell image 1 on day 1 of culture and cell image 2 on day 3 of culture taken at the same position as cell image 1, and similarly imaged on days 5, 7, and 9 of culture. A plurality of cell images 3, 4, 5, 6, 7, and 8 were prepared for each day, 11th day, 13th day, and 18th day. These cell images 1 to 8 were input into an original application based on a program using MATLAB (registered trademark) functions to obtain evaluation results (area occupancy rate and reaching distance). In addition, cell bodies present in cell images 1 to 8 were labeled to obtain annotation images. The above evaluation results and annotation images were scored by CNN as learning data. Scoring was performed in the same manner as in Example III.
Subsequently, for the cells cultured in Example II, cell images were obtained on day 1, day 3, and day 5 of culture, and the predicted growth score on day 13 of culture was calculated.
Specifically, first, the well plate was placed in a microscope apparatus, and an 8-bit image for 10 wells was obtained by imaging with a 20-fold objective lens. These images were input into the original application programmatically using MATLAB (registered trademark) functions to obtain images in which only the nerve fiber (dendritic and axonal) portions were highlighted. The obtained images were divided into 35 divisions of 5 vertical×7 horizontal, and the area occupancy rate of nerve fibers was calculated, and the reaching distance was calculated for the longest nerve fiber.
As a result of inputting this calculation result and the cell image into the predictor, a result of area occupancy rate score 3×reaching distance score 3=growth score 9 was obtained. Based on this result, it was judged that the growth proceeded well, and the culture was continued, and as expected, the neurons in the well-growing state described in Example II were obtained.
It was possible to predict and determine the growth of nerve cells by machine learning based on a database in which cell growth stages and their shapes and the like are associated as described above.

(Comparative example)
From the cell images of healthy iPS-derived motor neurons and ALS patient-derived motor neurons captured in Example I, the average value of the axon length in both cells was calculated using the above app. The average axon length of healthy iPS-derived motor neurons was 1.12 mm, and the average length of the axons of ALS patient-derived motor neurons was 1.15 mm. From these results, it was not possible to distinguish between the growth states of the two cells simply by comparing the average length of the axons.

Claims (7)

an imaging step of imaging a cell having a linear structure to obtain an image;
a data acquisition step of acquiring, from the image, data indicating a growth state of a linear portion extending from the cell body of the cell;
a determination step of determining the growth state of the linear portion from the data;
A cell evaluation method comprising
The cell evaluation method, wherein the data includes data on the area occupancy of the linear portion and the reaching distance calculated by processing the image. the cell having a linear structure is a nerve cell,
The cell evaluation method according to claim 1, wherein the linear part extending from the cell body is a dendrite and/or an axon. The determination step includes
Calculating a score representing the predicted growth degree of the cell from the data by machine learning based on a linear part database in which the growth stage of the linear part and its area occupancy and reaching distance are associated. Item 1. The cell evaluation method according to item 1. The data further includes image data of cell body shape,
The determination step includes
3. The method according to claim 2, comprising determining the growth state of the nerve cell from the image data of the cell body shape by machine learning based on a cell body shape database in which cell body growth stages and their shapes are associated. cell evaluation method. The determination step includes
3. The cell evaluation method according to claim 2, comprising determining the growth state of the nerve cell from the data by machine learning based on a nerve fiber database in which nerve fibers and their properties are associated. an imaging unit that acquires an image by imaging a cell having a linear structure;
a data acquisition unit that acquires, from the image, data indicating a growth state of a linear portion extending from the cell body of the cell;
a determination unit that determines the growth state of the linear portion from the data;
A cell evaluation device comprising
The cell evaluation device, wherein the data includes data on the area occupancy of the linear portion and the reaching distance calculated by processing the image. the cell having a linear structure is a nerve cell,
The cell evaluation device according to claim 6, wherein the linear part extending from the cell body is dendrite and/or axon.

Images