JP7428173B2 - Cell observation device and cell observation method - Google Patents

Description

The present invention relates to a cell observation device and method for observing the state of cells, and more specifically, to non-invasively determining the state of cells during the process of culturing pluripotent stem cells (ES cells and iPS cells) and culturing them. The present invention relates to a cell observation device and method suitable for checking whether there are any defects or the like.

In the field of regenerative medicine, research using pluripotent stem cells such as iPS cells and ES cells has been actively conducted in recent years. In research and development of regenerative medicine using such pluripotent stem cells, it is necessary to culture large quantities of undifferentiated cells that maintain pluripotency. Therefore, it is necessary to select an appropriate culture environment and to stably control the environment, and it is also necessary to frequently check the state of cells during culture. For example, if the cells in a cell colony deviate from their undifferentiated state, all cells in the cell colony have the ability to differentiate, so eventually all the cells in the colony become undifferentiated. Transition to a deviation state. Therefore, the operator must check daily to see if any cells that have deviated from the undifferentiated state (already differentiated cells or cells that are about to differentiate, hereinafter referred to as "undifferentiated deviant cells") are present in the cultured cells. If undifferentiated deviant cells are found, they must be promptly removed. Furthermore, foreign matter such as dust may get mixed into the culture medium during the culture process, and it is necessary to remove such foreign matter as quickly as possible.

Whether or not pluripotent stem cells maintain an undifferentiated state can be reliably determined by staining with an undifferentiated marker. However, since the stained cells die, undifferentiated marker staining cannot be used to determine pluripotent stem cells for regenerative medicine. Therefore, in the current field of cell culture for regenerative medicine, operators determine whether cells are undifferentiated cells based on morphological observation of cells using a phase contrast microscope. A phase contrast microscope is used because cells are generally transparent and difficult to observe with a normal optical microscope.

Furthermore, as disclosed in Non-Patent Document 1, devices that use holography technology to obtain observation images of cells have also been put into practical use. As disclosed in Patent Documents 1 to 4, this device performs data processing such as phase recovery and image reconstruction on hologram data obtained with a digital holographic microscope, allowing cells to be clearly observed. This method creates an easy-to-read phase image (hereinafter referred to as an "IHM phase image" because an in-line holographic microscope (IHM) is used). Digital holographic microscopes can calculate phase information at any distance during the calculation process after acquiring hologram data, so there is no need to focus each time a photograph is taken, reducing measurement time. There is an advantage.

However, although cells can be observed with a certain degree of clarity using a phase contrast microscopic image or an IHM phase image, considerable skill is required for an operator to accurately determine undifferentiated cells and the like visually. Therefore, the number of workers who can make such judgments is quite limited. Furthermore, since it is based on human judgment, it is inevitable that the judgment will vary. Further, even for a skilled worker, it takes time to make an accurate determination, making it difficult to process a large number of images. Therefore, although these conventional methods are acceptable at the research level at universities and research institutes, they are not suitable for industrial mass production of pluripotent stem cells.

International Patent Publication No. 2017/203718 pamphlet International Patent Publication No. 2017/204013 pamphlet International Patent Publication No. 2016/084420 pamphlet Japanese Patent Application Publication No. 10-268740

"Cell culture analysis device CultureScanner CS-1", [online], Shimadzu Corporation, [searched on February 14, 2018], Internet <URL: https://www.an.shimadzu.co.jp/bio /cell/cs1/index.htm＞

The present invention has been made to solve the above problems, and its purpose is to judge the state and quality of cells based on observation images obtained of cells, and to detect inappropriate conditions such as foreign matter contamination. To provide a cell observation device that allows even an operator with little experience or skill to carry out work such as confirming whether there is any abnormal state, and allows such work to be carried out efficiently.

The cell observation device according to the present invention, which has been made to solve the above problems, includes:
a) an observation image acquisition unit that acquires a microscopic observation image of cells as an observation image;
b) For the entire observation image acquired by the observation image acquisition unit or a part thereof, a machine learning model is used that outputs an image in which two or more characteristic regions in the observation image are divided in response to input of the observation image. a region identification unit that identifies characteristic regions each representing two or more cell states;
c) Create an area identification image that depicts the two or more characteristic areas identified by the area identification unit in a visually distinguishable manner, and display the area identification image and the observed image on the display unit. an image display processing unit that displays on the same screen;
It is characterized by having the following.

When an observation image is obtained by the observation image acquisition unit, the region identification unit identifies two or more characteristic regions each indicating a different cell state from the entire observation image or, for example, a partial observation image within a range specified by the user. identify. The area identification algorithm used in the area identification unit is not particularly limited, but it is useful to use supervised machine learning as part of the algorithm.

Supervised machine learning includes Support Vector Machine (SVM), Random Forest, AdaBoost, Naive Bayes, k-nearest neighbor method, and even deep learning including neural networks. Any suitable method can be used. Furthermore, instead of learning the image itself, it may be possible to learn feature quantities extracted from the image by texture analysis or the like.

The above characteristic regions differ depending on the type of cells to be observed and the purpose of observation, but for example, if the cells to be observed are pluripotent stem cells including human iPS cells, the different cell states are undifferentiated cells and undifferentiated cells. These include deviant cells. Therefore, in the present invention, one of the characteristic regions divided in the observation image is a region where cells or cell colonies are present, and where the cells or cell colonies are undifferentiated cells ; Another one of the characteristic regions is a region where cells or cell colonies are present, and where the cells or cell colonies are undifferentiated deviant cells , and yet another one of the characteristic regions is a region where foreign substances are present. This is an area in which it exists . In addition, the different cell states may be cells differentiated into cells of a target organ (for example, myocardium for heart transplantation) and cells other than that (for example, nerve cells).

The image display processing unit creates a region identification image in which two or more characteristic regions identified as described above are divided and depicted in a visually distinguishable manner. Then, the area identification image is displayed on the screen of the display unit. Here, the "visually distinguishable aspect" typically refers to filling processing using mutually different display colors, that is, color-coding of characteristic regions. Of course, the characteristic regions may be divided by different shadings on the gray scale, or the characteristic regions may be divided by different graphic patterns in the filled portions.

In the region identification image displayed in this way, for example, an undifferentiated cell region, an undifferentiated deviant cell region, a foreign substance region, etc. are each shown in a different color. As a result, even inexperienced workers who cannot distinguish the state of cells from observation images can check whether undifferentiated deviant cells exist in a cell colony, and if so, distinguish between undifferentiated cells and undifferentiated cells. At a glance, you can see where the border with differentiated cells is, or whether there are any foreign substances mixed into the culture medium.

In the present invention, preferably, the area identification unit generates information that identifies the characteristic area on a pixel basis, and the image display processing unit generates information on a pixel basis or a small area unit of a plurality of adjacent pixels, It is preferable to have a configuration in which different characteristic regions are divided and depicted in a visually distinguishable manner.

For example, in an observation image with a small magnification, one cell may be about the size of one pixel of the image, but if one cell is determined to be an undifferentiated deviant cell, one pixel corresponding to the cell It is often difficult to understand even if the pixels are displayed in a different color from the surrounding pixels on the area identification image. In such a case, if any one of the plurality of pixels included in one small area is identified as a certain feature area, not in units of pixels but in units of small areas that combine multiple adjacent pixels. By showing the small area in a color associated with the characteristic area, it becomes easier to visually understand that the characteristic area exists. This can prevent an operator from overlooking a very small characteristic region when checking an image.

Further, in the present invention, it is preferable that the image display processing section displays information indicating the state of cells to be identified by the region identification section and the distinguishable aspect of each characteristic region together with the region identification image.

Specifically, when each characteristic region in a region identification image is shown in a different display color, information such as "red: undifferentiated deviation, green: undifferentiated" is displayed in a part of the image or in the vicinity of the image. It is recommended to display the This allows the operator to grasp the state of the cells in the image at a glance.

Further, in the present invention, the image display processing section may be configured to enlarge or reduce the area identification image displayed on the screen of the display section in accordance with a user's operation.
This allows the operator to accurately grasp the state of cells in a very small range, for example, about one cell.

Further, in the above configuration, it is preferable that the image display processing unit displays the enlarged image and the area identification image before enlargement on the same screen.
This allows the operator to grasp both the distribution state of the cell state in a wide range on the sample and the detailed state of some cells therein.

Further, in the above configuration, it is preferable that the image display processing section displays information indicating the state of cells to be identified by the region identification section and the distinguishable aspect of each characteristic region together with the enlarged image.
This allows the operator to accurately grasp the state of the cells in the enlarged image just by looking at the enlarged image.

In the present invention, the observed image is a phase image obtained from hologram data obtained with an in-line holographic microscope.

According to this configuration, information at an arbitrary focal position can be obtained by calculation at the stage of calculating phase information and intensity information based on hologram data, which is a tedious task at the stage of photographing a sample containing cells with a holographic microscope. There is no need to perform focusing. Therefore, the sample can be photographed smoothly in a short time, making it suitable for observing cells in a living state. Additionally, compared to general optical microscopic images, it is possible to compare on the same screen a phase image in which cells can be observed relatively clearly, an intensity image in which foreign objects can be observed relatively clearly, and a region identification image. Therefore, while comparing these images, it is possible to check whether or not the characteristic regions are appropriately identified by automatic identification processing.

Further, in the above configuration, when a predetermined range is selected on the area identification image, the image display processing unit is configured to correspond to an enlarged image of the range and the enlarged image within the phase image and/or the intensity image. It is advisable to display the images of the selected range on the same screen.

Furthermore, in the above configuration, when an operation of enlarging/reducing is performed on either the enlarged image and the enlarged image of the phase image and/or intensity image, the image display processing section It is best to enlarge or reduce all images in conjunction with each other.
Thereby, it is possible to always compare the region identification image and the phase image or the intensity image in the same range, so it is possible to accurately confirm whether the region identification is appropriate or not. Furthermore, since the operator does not have to individually enlarge or reduce each image, observation work can be performed efficiently.

Further, in the above configuration, the image display processing section may be configured to display a composite image in which the area identification image and the phase image or the intensity image are superimposed.

Further, in that case, the image display processing section may be configured to adjust the transparency of at least one of the superimposed images in accordance with an operation by the user.

According to such a configuration, the operator can easily and visually grasp the relationship between the pattern or boundary on the phase image or the intensity image and the characteristic region. Of course, it is convenient for the user to be able to switch between a display in which a plurality of images are arranged side by side and a display in which a plurality of images are superimposed.

According to the present invention, for example, at a site where pluripotent stem cells such as iPS cells and ES cells are cultured, an operator can determine whether the cells being cultured are maintaining an undifferentiated state or are in an undifferentiated state. It is possible to reduce the labor involved in determining whether or not undifferentiated deviant cells exist in a cell colony or whether or not a foreign substance is mixed into a culture medium. Further, even if the worker in charge of such work is insufficient in skill or skill, it is possible to reduce variations in judgment, reduce judgment errors and oversights, and improve judgment accuracy. Further, a large amount of observation images can be efficiently processed, and work efficiency can be improved. For this reason, quality control of cells during culture can be easily controlled, and productivity in cell culture can be improved.

1 is a schematic configuration diagram of a cell observation device that is an embodiment of the present invention. 3 is a flowchart showing the flow of region identification processing of observed images in the cell observation apparatus of the present embodiment. FIG. 3 is a schematic diagram showing an example of region identification processing in the cell observation apparatus of the present embodiment. FIG. 3 is a diagram showing one mode of displaying a region identification image in the cell observation device of the present example. FIG. 7 is a diagram showing another aspect of displaying a region identification image in the cell observation apparatus of the present embodiment.

An embodiment of the cell observation device according to the present invention will be described below with reference to the accompanying drawings.
FIG. 1 is a schematic diagram of the cell observation apparatus of this example.

The cell observation apparatus of this embodiment includes a microscopic observation section 10, a control/processing section 20, and an input section 30 and a display section 40, which are user interfaces.
The microscopic observation unit 10 is an in-line holographic microscope, and includes a light source unit 11 including a laser diode, etc., and an image sensor 12, and a cell colony (or single cell) 14 is placed between the light source unit 11 and the image sensor 12. A culture plate 13 containing the cells is arranged.

The control/processing unit 20 controls the operation of the microscopic observation unit 10 and processes data acquired by the microscopic observation unit 10, and includes an imaging control unit 21, a hologram data storage unit 22, and a phase information calculation unit. unit 23, image reconstruction unit 24, reconstructed image data storage unit 25, area identification processing unit 26, area identification result data storage unit 27, area identification image creation unit 28, display processing unit 29, is provided as a functional block. Details of the area identification processing section 26 will be described later.

Usually, the entity of the control/processing unit 20 is a computer system including a personal computer with predetermined software installed, a high-performance workstation, or a high-performance computer connected to such a computer via a communication line. be. That is, the functions of each block included in the control/processing unit 20 are implemented by executing software installed in a single computer or a computer system including multiple computers, or by executing software stored in the computer or computer system. It can be realized by processing using various types of data.

The cell observation device of this example can observe various cells, but here, observation images of cells are acquired for the purpose of cell observation necessary when culturing human iPS cells, and the observation images are A process for identifying an undifferentiated cell region and an undifferentiated deviant cell region in a cell colony and outputting the identification results will be described.

FIG. 2 is a flowchart showing the flow of region identification processing for observed images in the cell observation apparatus of this embodiment, and FIG. 3 is a schematic diagram showing an example of the region identification processing.
The operator sets the culture plate 13 containing the cell colony 14 at a predetermined position and performs a predetermined operation using the input unit 30 . In response to this operation, the imaging control section 21 controls the microscope observation section 10 to acquire hologram data as follows (step S1).

That is, the light source section 11 irradiates a predetermined region of the culture plate 13 with coherent light having a minute angle spread of about 10 degrees. The coherent light (object light 16) that has passed through the culture plate 13 and the cell colony 14 reaches the image sensor 12 while interfering with the light (reference light 15) that has passed through the area close to the cell colony 14 on the culture plate 13. . The object light 16 is light whose phase has changed when it passes through the cell colony 14, while the reference light 15 does not pass through the cell colony 14, so it is not subjected to a phase change caused by the colony 14. Therefore, an image is formed on the detection plane (image plane) of the image sensor 12 by interference fringes between the object light 16 whose phase has changed due to the cell colony 14 and the reference light 15 whose phase has not changed.

Note that the irradiation area (observation area) of the coherent light emitted from the light source section 11 is smaller than the size of the culture plate 13. Therefore, while moving the light source section 11 and the image sensor 12 together by a predetermined distance in the X-axis direction and the Y-axis direction using a moving mechanism (not shown), the culture plate 13 is irradiated with coherent light emitted from the light source section 11. , an interference image of a part of the culture plate 13 is repeatedly photographed. Thereby, hologram data (two-dimensional light intensity distribution data of the hologram formed on the detection surface of the image sensor 12) covering a wide two-dimensional area on the culture plate 13 can be acquired.

As described above, the hologram data obtained by the microscope observation section 10 is sequentially sent to the control/processing section 20 and stored in the hologram data storage section 22. In the control/processing unit 20, the phase information calculation unit 23 reads the hologram data from the hologram data storage unit 22, and calculates the phase information of the entire observation area (imaging area) by executing predetermined calculation processing for phase recovery. do. Then, the image reconstruction unit 24 creates an IHM phase image based on the calculated phase information, and stores the image data in the reconstructed image data storage unit 25 (step S2). When calculating such phase information and creating an IHM phase image, well-known algorithms disclosed in Patent Documents 3 and 4 may be used. In general, IHM phase images make it easier to see the outlines (boundaries) of cells and their internal patterns, which are transparent and difficult to see with an optical microscope.

Note that in addition to phase information, intensity information, pseudo-phase information, etc. are also calculated based on the hologram data, and the image reconstruction unit 24 creates a reconstructed image (IHM intensity image, IHM pseudo-phase image) based on these. You can also do that. An IHM intensity image may be used instead of an IHM phase image. Further, the IHM pseudo phase image is an image that is a combination of an IHM phase image and an IHM intensity image, and corresponds to a phase contrast microscopic image obtained with a phase contrast microscope. Further, when creating an IHM phase image or the like based on hologram data, it is possible to create each IHM phase image or the like at a plurality of focal positions.

The region identification processing unit 26 performs a process of identifying two characteristic regions, an undifferentiated cell region and an undifferentiated deviant cell region in the cell colony, on the created IHM phase image (step S3). Here, this area identification processing will be explained.

The region identification processing unit 26 includes a classifier constructed in advance through learning using a large amount of training data, and uses this classifier to identify a plurality of feature regions in the input image pixel by pixel, and performs the classification. It outputs the results. What kind of feature region to identify depends on the teacher data used during learning. Therefore, here, we will use a large number of IHM phase images prepared in advance and the results of the identification of undifferentiated cell areas and undifferentiated deviant cell areas by skilled workers (that is, correct images) as a teacher. Perform learning using data.

The correct image is a labeled image in which undifferentiated cell regions, undifferentiated deviant cell regions, and background regions (regions where no cells exist) on a given IHM phase image are labeled in pixel units. Roughly speaking, the label image is a "1" for pixels determined to be undifferentiated cell regions, a "2" for pixels determined to be undifferentiated deviant cell regions, and a "2" for pixels determined to be other background regions. This is an image in which a predetermined pixel value such as "0" is assigned to a pixel. In other words, this is information that associates the position information of each pixel with a pixel value of "0", "1", or "2". be. If this label image is rendered by assigning different display colors to different labels, an image will be obtained in which the undifferentiated cell region, the undifferentiated deviant cell region, and the background region are displayed in different colors.

Prepare a large number of pairs of IHM phase images and correct images as training data, and use these IHM phase images as input images to construct a machine learning learning model (in other words, a classifier) so that the output image is as close to the correct image as possible. do. Various methods are known as such supervised machine learning algorithms, such as support vector machine, random forest, AdaBoost, naive Bayes, k-nearest neighbor method, and even deep learning including neural networks. method may also be used. Furthermore, even when an image is input, an algorithm may be used that does not learn the image itself, but rather learns feature quantities extracted from the image by texture analysis or the like.

The function of constructing the above-mentioned classifier (trained machine learning model) may be built into the control/processing unit 20, but in general, calculations for such learning processing are quite heavy on the computer, and high performance is required. If you don't have a modern computer, it will take a long time. Therefore, a classifier capable of highly accurate region identification is constructed by performing learning processing using another high-performance computer, and it is stored in a memory that is a part of the region identification processing section 26. . The area identification processing unit 26 generally performs only identification processing using the classifier.

Note that in step S3, the area identification process may be performed on the entire IHM phase image created in step S2, but the area identification process may be performed on only a part of the IHM phase image specified by the operator. may be implemented.

The region identification result by the region identification processing unit 26 using the above-mentioned classifier is obtained by classifying the original IHM image into each characteristic region (here, an undifferentiated cell region, an undifferentiated deviant cell region, and other background regions) pixel by pixel. ) is a label image labeled correspondingly. This identification result is stored in the area identification result data storage section 27. The area identification image creation unit 28 receives such data and creates an area identification image by coloring the pixels included in each labeled feature area with the display color assigned in advance to each label (step S4). ).

The display color assigned to each feature region is predetermined by default, but can also be arbitrarily designated by the operator from the input unit 30. Specifically, for example, if the undifferentiated cell area is displayed in green, the undifferentiated deviant cell area is displayed in red, and the background area is displayed in black, the border between the background and the cell colony (in other words, the cell The outline of the colony) and the boundaries between undifferentiated cells and undifferentiated cells within the cell colony are clear.

FIG. 3 is a schematic diagram showing an example of a region identification image for an IHM phase image. In this figure, since colors cannot be expressed, each feature region is divided by differences in shading on a gray scale. In this way, it is possible to display not only the division of each feature region by color display, but also various aspects as long as the boundaries of each feature region are displayed in a visually easy-to-see manner, such as differences in shading on a gray scale or differences in fill patterns. You can take it.

The display processing unit 29 creates a display screen in which the region identification image created in step S5 and the IHM phase image are arranged side by side, and displays this on the screen of the display unit 40 (step S5). FIG. 4 is a diagram showing an aspect of such a display screen.

In FIG. 4, an area identification image display frame 51 and an IHM phase image display frame 52 are provided within the identification result display screen 50, and the respective images are displayed within the display frames 51 and 52. Both images are images of the same area on the culture plate 13, and for example, an operator can enlarge or enlarge the image using the input unit (specifically, a pointing device such as a mouse) 30 on either image. When an operation such as reduction, movement, rotation, etc. is performed, the display processing unit 29 that recognizes the operation reads data necessary for display from the reconstructed image data storage unit 25 and the area identification result data storage unit 27, and displays the area identification image. The images displayed in the display frame 51 and the IHM phase image display frame 52 are updated. This allows the operator to always check the IHM phase image and region identification image of the same observation range.

Further, instead of arranging the region identification image and the IHM phase image side by side, one of the images can be made semi-transparent and displayed overlappingly. FIG. 5 is a diagram showing an aspect of such a display screen. In FIG. 5, a superimposed image display frame 53 and a slider 54, which is an operator for specifying the transparency of the image to be made translucent, are provided in the identification result display screen 50. An image in which a region identification image and an IHM phase image of the same observation range are superimposed is displayed. The operator compares the region identification image and the IHM phase image using either the display screen shown in FIG. 4 or FIG. 5, and confirms, for example, whether the division into the undifferentiated cell region and the undifferentiated deviant cell region is correct. Can be done.

Further, instead of displaying the region identification image and the IHM phase image side by side or overlapping each other, only the region identification image may be displayed. The user may be able to select the format in which the area identification image is to be displayed using, for example, a GUI widget such as a radio button or a check box placed on the same screen.

In addition, information indicating the correspondence between the display color and the cell state indicated by the display color (that is, undifferentiated cell state, undifferentiated deviant cell state) is displayed in the region identification image display frame 51 or near the outside of the frame. May be indicated in text. This allows the operator to grasp at a glance the range of undifferentiated deviant cells in the displayed region identification image.

As mentioned above, feature region identification is performed pixel by pixel, and the identification results are also recorded pixel by pixel, so it is possible to color code pixel by pixel even in region identification images, but it is possible to color code by pixel, not by pixel by pixel. Each characteristic region may be color-coded in units of small regions made up of a plurality of pixels. In this case, for example, if any one of the multiple pixels included in one small region is identified as an undifferentiated deviant cell region, that small region is colored in a color associated with the undifferentiated deviant cell region. It would be good to show it. Thereby, even when displaying a low-magnification image of the entire culture plate, for example, the presence of one or a small number of undifferentiated deviant cells can be clearly shown visually. As a result, it is possible to prevent an operator from overlooking an undifferentiated cell region when checking an image.

In the above embodiment, the IHM phase image and the like are divided into three characteristic regions including the background region, but the number of regions to be identified, such as a foreign object region, may be increased as appropriate.

In addition, in the cell observation apparatus shown in FIG. 1, all processing is performed in the control/processing unit 20, but in general, the calculation of phase information based on hologram data and the imaging of the calculation results require a huge amount of processing. calculation is required. In addition, the processing load of region identification using machine learning is also large. Therefore, by using a computer system in which a personal computer connected to the microscope observation section 10 is used as a terminal device, and this terminal device and a server, which is a high-performance computer, are connected via a communication network such as the Internet or an intranet, the above-mentioned If complicated calculations and processing such as these are performed by a high-performance computer, roles such as controlling the microscope observation unit 10 and display processing using the processed data are performed by a relatively low-performance personal computer. good.

Further, the above-mentioned embodiments are merely examples of the present invention, and it is clear that further modifications, modifications, and additions may be made as appropriate within the scope of the spirit of the present invention and still fall within the scope of the claims of the present application.

10...Microscopic observation unit 11...Light source unit 12...Image sensor 13...Culture plate 14...Cell colony 15...Reference light 16...Object light 20...Control/processing unit 21...Photographing control unit 22...Hologram data storage unit 23...Phase information Calculation section 24...Image reconstruction section 25...Reconstructed image data storage section 26...Area identification processing section 27...Area identification image creation section 28...Area identification result data storage section 29...Display processing section 30...Input section 40...Display section

Claims (11)

a) an observation image acquisition unit that acquires a microscopic observation image of cells as an observation image;
b) Regarding the whole or part of the observation image acquired by the observation image acquisition unit, the observation image is identified by identifying three or more characteristic regions for each pixel in the observation image in response to the input of the observation image. an area identification unit that identifies the three or more feature regions for each pixel on the image and labels each pixel by using a machine learning model that outputs an image divided into three or more feature regions;
c) Create an area identification image in which the three or more feature areas labeled for each pixel on the image by the area identification unit are divided and depicted in a visually distinguishable manner, and the area identification image is created. and an image display processing unit that displays the observed image and the observed image on the same screen of a display unit;
one of the characteristic regions divided in the observation image is a cell region and is an area of undifferentiated cells, and another one of the characteristic regions is a region of fine cells. A cell observation device characterized in that the region is a cell region and is an undifferentiated deviant cell, and another one of the characteristic regions is a region in which a foreign substance is present. The cell observation device according to claim 1,
When a predetermined range is selected on the area identification image displayed on the display unit by a user's operation, the image display processing unit displays an enlarged image of the area identification image of the predetermined range and the observation image. or an image of a range corresponding to the enlarged image within the observation image, and is displayed on the same screen. The cell observation device according to claim 2,
When an enlargement/reduction operation is performed on either the enlarged image of the area identification image or the enlarged image of the observation image, the image display processing unit displays the area identification image and the observation image in accordance with the operation. A cell observation device characterized by expanding and contracting in conjunction with each other. The cell observation device according to any one of claims 1 to 3,
The cell observation apparatus is characterized in that the image display processing section displays a composite image obtained by superimposing the region identification image and the observation image. The cell observation device according to claim 4,
The cell observation apparatus is characterized in that the image display processing section adjusts the transparency of at least one of the superimposed images according to an operation by a user. The cell observation device according to any one of claims 1 to 5,
The machine learning model is created by performing machine learning using an observed image of cells and a correct image, which is a label image in which three or more characteristic regions in the observed image are labeled in pixel units, as training data. A cell observation device characterized by being a trained model. The cell observation device according to any one of claims 1 to 6 ,
The cell observation device is characterized in that the image display processing unit divides and depicts different characteristic regions in a visually distinguishable manner in units of pixels or in units of small regions made up of a plurality of adjacent pixels. . The cell observation device according to any one of claims 1 to 7,
The cell observation apparatus is characterized in that the image display processing unit displays information indicating the state of cells to be identified by the area identification unit and the distinguishable aspect of each characteristic area together with the area identification image. The cell observation device according to any one of claims 1 to 8,
When a predetermined range is selected on the area identification image by a user's operation, the image display processing unit displays an enlarged image of the range and an area identification image before the enlargement on the same screen. A cell observation device featuring: The cell observation device according to any one of claims 2, 3, or 9,
The cell observation apparatus is characterized in that the image display processing unit displays information indicating the state of cells to be identified by the region identification unit and the distinguishable aspect of each characteristic region together with the enlarged image. a) an observation image acquisition step of acquiring a microscopic observation image of cells as an observation image;
b) Regarding the entire or part of the acquired observation image, identifying three or more characteristic regions for each pixel in the observation image with respect to the input of the observation image, the observation image can be divided into three or more characteristic regions. a region identification step of identifying the three or more feature regions for each pixel on the image and labeling each pixel using a machine learning model that outputs an image divided into;
c) Create an area identification image in which the three or more feature areas labeled for each pixel on the image in the area identification step are divided and depicted in a visually distinguishable manner, and the area identification image is and the observed image on the same screen of a display unit;
One of the characteristic regions divided in the observed image is a cell region and an undifferentiated cell region, and another one of the characteristic regions is a cell region and an undifferentiated cell region. A cell observation method, characterized in that the cell region is a region of undifferentiated deviant cells, and another one of the characteristic regions is a region in which a foreign substance is present.

Images