Cell profiling analysis methods

Abstract

This cell image analysis method comprises: a step for acquiring a cell image (10) in which a cell (90) is shown; a step for generating, by performing filter processing on the cell image, a background component image (14) from which a distribution of luminance components of the background (91) is extracted; a step for generating, on the basis of the cell image and the background component image, a corrected cell image (11) in which correction has been performed in order to reduce luminance unevenness; and a first estimation step for estimating whether the cell in the image is a normal cell or an abnormal cell by using the corrected cell image and a trained model (6) trained with cell analysis.

Description

Technical Field

[0001] This invention relates to a method for analyzing cell images, and particularly to a cell analysis method for analyzing cells using a pre-trained model.

Background Art

[0002] Conventionally, a cell analysis method for analyzing cells using a pre-trained model is known. Such a cell analysis method is disclosed, for example, in International Publication No. 2019 / 171546.

[0003] International Publication No. 2019 / 171546 discloses a cell image analysis method for analyzing an image of cells captured by an imaging device. Specifically, International Publication No. 2019 / 171546 discloses a configuration for acquiring a cell image by photographing cells cultured on a culture plate with a photographing device such as a microscope. Further, the cell image analysis method disclosed in International Publication No. 2019 / 171546 classifies a normal cell region and an abnormal cell region using the analysis result of a pre-trained model. Also, International Publication No. 2019 / 171546 discloses a configuration for classifying a normal cell region and an abnormal cell region by performing a segmentation process for determining to which category each pixel of the cell image belongs.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Although not disclosed in International Publication No. 2019 / 171546, when culturing cells in a culture plate (culture vessel), brightness unevenness may occur in the background of the cell image due to the culture solution. Specifically, near the edges of the culture vessel, surface tension causes the liquid level of the culture solution to change, increasing as you approach the edges of the vessel. When the liquid level of the culture solution changes, brightness unevenness occurs in the background. When analyzing cell images with brightness unevenness in the background using a trained model, there is a disadvantage in that the analysis accuracy decreases. Therefore, there is a need for a cell image analysis method that can suppress the decrease in analysis accuracy by a trained model even when brightness unevenness occurs in the background of the cell image.

[0006] This invention was made to solve the above-mentioned problems, and one of its objectives is to provide a cell image analysis method that can suppress the decrease in analysis accuracy by a trained model even when brightness unevenness occurs in the background of the cell image. [Means for solving the problem]

[0007] To achieve the above objective, the cell image analysis method according to the first aspect of this invention comprises the steps of: acquiring a cell image in which cells are visible; generating a background component image by filtering the acquired cell image to extract the distribution of background brightness components from the cell image; generating a corrected cell image by correcting the cell image to reduce brightness unevenness based on the cell image and the background component image; and a first estimation step of estimating whether the cells in the image are normal cells or abnormal cells using the corrected cell image and a trained model trained to analyze cells, wherein the cell image is a cultured cell cultured in a culture solution filled in a culture vessel, and includes cultured cells in a region where the height of the liquid level of the culture solution changes relative to the center of the culture vessel due to surface tension, and in the step of generating the corrected cell image, Based on the cell image and the background component image generated by filtering,Corrected cell images are generated by correcting the cell images to reduce brightness unevenness caused by changes in the culture solution level in the region where the liquid level of the culture solution in the culture vessel changes relative to the center of the culture vessel, due to surface tension. [Effects of the Invention]

[0008] The cell image analysis method in the first phase described above comprises the steps of: generating a corrected cell image by correcting the cell image to reduce brightness unevenness; and a first estimation step of estimating whether the cells in the image are normal or abnormal cells using the corrected cell image and a trained model that has been trained to analyze cells. The cell image is a cultured cell cultured in a culture solution filled in a culture vessel, and includes cultured cells in a region where the height of the liquid level of the culture solution changes relative to the center of the culture vessel due to surface tension. In the step of generating the corrected cell image, a corrected cell image is generated by correcting the cell image to reduce brightness unevenness caused by the change in the height of the culture solution due to surface tension in the region where the height of the liquid level of the culture solution in the culture vessel changes relative to the center of the culture vessel. As a result, it is possible to estimate whether the cells in the image are normal or abnormal cells using the corrected cell image with reduced background brightness unevenness and the trained model. Consequently, it is possible to suppress the decrease in the analysis accuracy by the trained model due to brightness unevenness in the background of the cell image. Furthermore, in order to achieve the above objective, the cell image analysis method according to the second aspect of this invention comprises the steps of: acquiring a cell image in which cells are visible; generating a background component image by filtering the acquired cell image to extract the distribution of background brightness components from the cell image; generating a corrected cell image by correcting the cell image to reduce brightness unevenness based on the cell image and the background component image; a first estimation step of estimating whether the cells in the image are normal or abnormal cells using the corrected cell image and a trained model trained to analyze cells; and receiving an input for whether or not to generate a corrected cell image. The cell image is filled into a culture vessel. The image is of cultured cells cultured in a culture solution, and includes cultured cells in a region where the liquid level of the culture solution changes relative to the center of the culture vessel due to surface tension. When generating a corrected cell image, in the step of generating the corrected cell image, the cell image is corrected to reduce brightness unevenness caused by the change in the liquid level of the culture solution due to surface tension in the region where the liquid level of the culture solution in the culture vessel changes relative to the center of the culture vessel. The first estimation step is performed using the corrected cell image and the trained model. If a corrected cell image is not generated, the step of generating a background component image is not performed, and the cell image and the trained model are used. Whether the cells in the image are normal or abnormal It further includes a second estimation step. [Brief explanation of the drawing]

[0009] [Figure 1] This is a schematic diagram showing the overall configuration of a cell image analysis device according to one embodiment. [Figure 2] This is a schematic diagram to explain cell images. [Figure 3] This is a schematic diagram illustrating how brightness variations occur in the background of cell images due to the culture solution filled in the culture vessel. [Figure 4] This is a schematic diagram illustrating a method for training a learning model according to one embodiment, and a method for analyzing cell images using the trained learning model. [Figure 5]This is a schematic diagram illustrating a configuration in which an image processing unit according to one embodiment generates a background component image. [Figure 6] This is a schematic diagram illustrating a configuration in which an image processing unit according to one embodiment generates a corrected cell image. [Figure 7] This is a schematic diagram illustrating a configuration in which an image analysis unit according to one embodiment generates a first region-estimated image using a first trained model. [Figure 8] This is a schematic diagram illustrating a configuration in which an image processing unit according to one embodiment generates a first superimposed cell image using a corrected cell image and a first region estimation image. [Figure 9] This is a schematic diagram illustrating superimposed cell images using comparative examples. [Figure 10] This is a schematic diagram illustrating a configuration in which an image processing unit according to one embodiment generates a second region-estimated image using a second trained model. [Figure 11] This is a schematic diagram illustrating a configuration in which an image processing unit according to one embodiment generates a second superimposed cell image using a cell image and a second region estimation image. [Figure 12] This is a flowchart illustrating the process by which a cell image analysis device according to one embodiment displays superimposed cell images. [Figure 13] This is a flowchart illustrating the process by which a cell image analysis device according to one embodiment generates a background component image. [Modes for carrying out the invention]

[0010] The following describes embodiments of the present invention based on the drawings.

[0011] Referring to Figure 1, the configuration of a cell image analysis device 100 according to one embodiment will be described.

[0012] (Configuration of the cell image analysis system) As shown in Figure 1, the cell image analysis device 100 comprises an image acquisition unit 1, a processor 2, a storage unit 3, a display unit 4, and an input receiving unit 5.

[0013] The image acquisition unit 1 is configured to acquire a cell image 10. The cell image 10 is an image in which cells 90 (see FIG. 2) are captured. Specifically, the cell image 10 is an image obtained by photographing cultured cells 90 cultured in a culture solution 81 (see FIG. 3) filled in a culture vessel 80 (see FIG. 3). In the present embodiment, the image acquisition unit 1 is configured to acquire the cell image 10 from a device that photographs the cell image 10, such as a microscope 7 to which an imaging device is attached. The image acquisition unit 1 includes, for example, an input / output interface.

[0014] The processor 2 is configured to analyze the acquired cell image 10. The processor 2 includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), a GPU (Graphics Processing Unit), or an FPGA (Field-Programmable Gate Array) configured for image processing. Further, the processor 2 composed of a CPU or the like as hardware includes a control unit 2a, an image analysis unit 2b, an image processing unit 2c, and a superimposed cell image generation unit 2d as functional blocks of software (program). The processor 2 functions as the control unit 2a, the image analysis unit 2b, the image processing unit 2c, and the superimposed cell image generation unit 2d by executing a program stored in the storage unit 3. The control unit 2a, the image analysis unit 2b, the image processing unit 2c, and the superimposed cell image generation unit 2d may be individually configured by hardware by providing dedicated processors (processing circuits).

[0015] The control unit 2a is configured to control the cell image analysis device 100. Further, the control unit 2a is configured to perform control to display the superimposed cell image 50 on the display unit 4. Details of the superimposed cell image 50 will be described later.

[0016] In this embodiment, the image analysis unit 2b analyzes whether the cells 90 (see Figure 2) in the cell image 10 are normal cells or abnormal cells. Specifically, the image analysis unit 2b is configured to analyze whether the cells 90 in the cell image 10 are normal cells or abnormal cells using a trained model 6 that has been trained to analyze cells 90. The trained model 6 includes a first trained model 6a used to analyze a corrected cell image 11 (see Figure 6), which will be described later, and a second trained model 6b used to analyze the cell image 10. Details of normal cells, abnormal cells, the first trained model 6a, and the second trained model 6b will be described later.

[0017] The image processing unit 2c is configured to generate a background component image 14 (see Figure 5) by extracting the distribution of luminance components of the background 91 (see Figure 3) from the cell image 10. The image processing unit 2c is also configured to generate a corrected cell image 11 (see Figure 6) by correcting the cell image 10 to reduce luminance unevenness. Details of the process by which the image processing unit 2c generates the background component image 14 and the corrected cell image 11 will be described later.

[0018] The superimposed cell image generation unit 2d is configured to generate superimposed cell images 50 that can distinguish between normal cells and abnormal cells. Details of the configuration by which the superimposed cell image generation unit 2d generates the superimposed cell images 50 will be described later.

[0019] The memory unit 3 is configured to store the cell image 10, the first trained model 6a, and the second trained model 6b. The memory unit 3 is also configured to store various programs executed by the processor 2. The memory unit 3 includes, for example, a storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0020] The display unit 4 is configured to display superimposed cell images 50 and the like generated by the superimposed cell image generation unit 2d. The display unit 4 includes, for example, a display device such as a liquid crystal monitor.

[0021] The input receiving unit 5 is configured to accept operation input from the operator. The input receiving unit 5 includes, for example, an input device such as a mouse or keyboard.

[0022] (Cell image) Referring to Figure 2, the cell image 10 will be described. The cell image 10 is an image of cultured cells 90. In this embodiment, the cell image 10 is a microscopic image taken by a microscope 7 to which an imaging device is attached. The cultured cells 90 are images of cells 90 that have differentiation potential. For example, cells 90 include iPS cells (induced Pluripotent Stem Cells) and ES cells (Embryonic Stem Cells). The image analysis unit 2b is configured to analyze whether the cells 90 (see Figure 3) shown in the cell image 10 are undifferentiated cells or undifferentiated stray cells. An undifferentiated cell is a cell that has differentiation potential. An undifferentiated stray cell is a cell that has started to differentiate into a specific cell and does not have differentiation potential. In this embodiment, an undifferentiated cell is considered a normal cell. An undifferentiated stray cell is considered an abnormal cell.

[0023] (Uneven background brightness caused by the culture solution) Next, referring to Figure 3, we will explain that brightness unevenness may occur in the background 91 of the cell image 10 due to the culture solution 81. In this embodiment, among the cell images 10, the image in which brightness unevenness occurs in the background 91 is designated as cell image 10a, and the image in which brightness unevenness does not occur in the background 91 is designated as cell image 10b.

[0024] As shown in Figure 3, the cells 90 are cultured cells cultured in a culture solution 81 filled in a culture vessel 80. The liquid level 81a of the culture solution 81 gradually changes in height due to surface tension near the edge 70 of the culture vessel 80. In the example shown in Figure 3, the height of the liquid level 81a of the culture solution 81 increases as you approach the edge 80a of the culture vessel 80. In this case, due to the difference in the height of the culture solution 81, brightness unevenness occurs in the background 91, as shown in cell image 10a in Figure 3. Note that cell image 10a is an image that includes cultured cells 90 near the edge 70 of the culture vessel 80. In addition, in cell image 10a shown in Figure 3, the brightness unevenness of the background 91 is represented by hatching with different spacing. Cell image 10a shown in Figure 3 is an example in which brightness unevenness occurs where the brightness value (pixel value) of the background 91 decreases from the left to the right of the image. Furthermore, the term "near the end of the culture vessel 80 70" includes both the position of the end 80a of the culture vessel 80 itself and the area near the position of the end 80a of the culture vessel 80.

[0025] Furthermore, as shown in Figure 3, the height of the liquid level 81a of the culture solution 81 is constant in the central part 71 of the culture vessel 80. When the height of the liquid level 81a of the culture solution 81 is constant, no brightness unevenness occurs in the background 91. That is, as shown in cell image 10b, in an image including the cultured cells 90 in the central part 71 of the culture vessel 80, no brightness unevenness occurs in the background 91. Note that cell image 10b is an image that shows only the cultured cells 90 in the central part 71 of the culture vessel 80.

[0026] If a cell image 10a with brightness unevenness in the background 91 is used as is for analysis by the trained model 6, the analysis accuracy may decrease due to the brightness unevenness in the background 91. Therefore, in this embodiment, the image processing unit 2c generates a corrected cell image 11 with reduced brightness unevenness in the background 91. The image analysis unit 2b is configured to perform analysis using the corrected cell image 11 generated by the image processing unit 2c and the trained model 6.

[0027] (Image analysis method) Next, with reference to Figure 4, a method for analyzing the cell image 10 using the cell image analysis method according to this embodiment will be described. In this embodiment, the cell image analysis device 100 (see Figure 1) analyzes the cell image 10 to determine whether the cells 90 in the cell image 10 are normal cells or abnormal cells. In this embodiment, the cell image analysis device 100 determines whether the cells 90 in the cell image 10 are normal cells or abnormal cells by analyzing the cell image 10 using a trained model 6 (see Figure 1). Upon receiving the cell image 10 as input, the trained model 6 outputs a region estimation image 12. The region estimation image 12 is an image that displays normal cell regions 20 (see Figure 7) and abnormal cell regions 21 (see Figure 7) in a distinguishable manner.

[0028] Figure 4 is a block diagram showing the image processing flow according to this embodiment. As shown in Figure 4, in this embodiment, the cell image analysis method is broadly divided into an image analysis method 101, a method for generating a first trained model 6a 102, and a method for generating a second trained model 6b 103.

[0029] (Generation of a learning model) The method 102 for generating the first pre-trained model 6a according to this embodiment generates the pre-trained model 6 (first pre-trained model 6a) by training the first pre-trained model 7a using the corrected cell image 11. Specifically, the method 102 for generating the first pre-trained model 6a generates the first pre-trained model 6a by training the first pre-trained model 7a using the training corrected cell image 30 and the training region estimation image 31. That is, the method 102 for generating the first pre-trained model 6a uses the training corrected cell image 30 as the input data for the corrected cell image 11, and outputs an image in which the normal cell region 20, the abnormal cell region 21, and the background region are assigned different label values ​​as the training region estimation image 31. As a result, the method 102 for generating the first pre-trained model 6a trains the first pre-trained model 7a to classify each pixel of the input image as either a normal cell, an abnormal cell, or the background. Specifically, the method 102 for generating the first trained model 6a includes the steps of inputting a training correction cell image 30 to the first trained model 7a (step 102a) and training the first trained model 7a to output a training region estimation image 31, which is a label image corresponding to the training correction cell image 30 (step 102b). The first trained model 6a is, for example, a convolutional neural network (CNN) as shown in Figure 4, or includes a convolutional neural network in part. The first trained model 6a generated by training the first trained model 7a is stored in the memory unit 3 (Figure 1) of the cell image analysis device 100. The training correction cell image 30 is a cell image 10 with reduced brightness unevenness in the background 91.

[0030] Furthermore, the method 103 for generating the second trained model 6b generates the second trained model 6b by training the second trained model 7b using the training cell image 32 and the training region estimation image 33. In this embodiment, the second trained model 6b analyzes whether the cell 90 is suitable for analysis of whether it is a normal cell or an abnormal cell in the uncorrected cell image 15, which is the cell image 10 that has not undergone correction processing. The uncorrected cell image 15 is an example of the "cell image that has not undergone correction processing" in the claims.

[0031] In other words, the method 103 for generating the second pre-trained model 6b takes a training cell image 32 as an uncorrected cell image 15 as input data, and outputs an image as a training region estimation image 33 in which normal cell regions 20 (see Figure 7), abnormal cell regions 21 (see Figure 7), and background regions are assigned different label values. Thus, the method 103 for generating the second pre-trained model 6b trains the second pre-trained model 7b to classify each pixel of the input image as either a normal cell, an abnormal cell, or the background. Specifically, the method 102 for generating the second pre-trained model 6b includes the step 103a of inputting the training cell image 32 to the second pre-trained model 7b, and the step 103b of training the second pre-trained model 7b to output the training region estimation image 33, which is a label image corresponding to the training cell image 32. The second pre-trained model 6b is, for example, a convolutional neural network as shown in Figure 4, or includes a convolutional neural network in part. The second trained model 6b, generated by training the second learning model 7b, is stored in the memory unit 3 (Figure 1) of the cell image analysis device 100. The training cell image 32 is a cell image 10 that has not undergone any processing to reduce brightness unevenness (brightness unevenness correction processing), regardless of whether or not there is brightness unevenness in the background 91.

[0032] (Image analysis method) The image analysis method 101 according to this embodiment is an image analysis method that classifies cells 90 captured in a cell image 10 acquired by the image acquisition unit 1 from a microscope 7 (see Figure 1) or the like into normal cells and abnormal cells. The image analysis method 101 according to this embodiment includes the steps of: acquiring a cell image 10 in which cells 90 (see Figure 2) are captured; generating a background component image 14 (see Figure 5) by extracting the distribution of brightness components of the background 91 (see Figure 2) from the cell image 10; generating a corrected cell image 11 (see Figure 6) by correcting the cell image 10 to reduce brightness unevenness; and a first estimation step of estimating whether the cells 90 in the image are normal cells or abnormal cells using the corrected cell image 11 and a trained model 6 (first trained model 6a) that has been trained to analyze cells 90. Detailed processing of each step of the image analysis method 101 will be described later.

[0033] In this embodiment, the step of acquiring a cell image 10 is performed by the image acquisition unit 1. The image acquisition unit 1 acquires the cell image 10 from an image acquisition device such as a microscope 7 (see Figure 1). The image acquisition unit 1 also outputs the acquired cell image 10 to the image analysis unit 2b. The image acquisition unit 1 also outputs the acquired cell image 10 to the image processing unit 2c.

[0034] Furthermore, in this embodiment, as shown in Figure 4, the step of analyzing the uncorrected cell image 15 is performed by the image analysis unit 2b. The image analysis unit 2b acquires a region estimation image 12 by inputting either the corrected cell image 11 or the uncorrected cell image 15 to the trained model 6. Specifically, the image analysis unit 2b acquires a first region estimation image 12a (see Figure 7) by inputting the corrected cell image 11 to the first trained model 6a. The image analysis unit 2b also acquires a second region estimation image 12b (see Figure 10) by inputting the uncorrected cell image 15 to the second trained model 6b. Whether the image analysis unit 2b performs the analysis using the first trained model 6a or the second trained model 6b is determined by the control unit 2a. The image analysis unit 2b also outputs the acquired region estimation image 12 (first region estimation image 12a or second region estimation image 12b) to the superimposed cell image generation unit 2d.

[0035] The control unit 2a determines, based on the operator's input, whether to perform the analysis using the first pre-trained model 6a or the second pre-trained model 6b. Specifically, the control unit 2a determines, based on the operator's input of whether or not to generate a corrected cell image 11, whether to perform the analysis using the first pre-trained model 6a or the second pre-trained model 6b. That is, the control unit 2a decides to use the first pre-trained model 6a for the corrected cell image 11 and not to use the first pre-trained model 6a for the uncorrected cell image 15. Furthermore, the control unit 2a decides to use the second pre-trained model 6b for the uncorrected cell image 15 and not to use the second pre-trained model 6b for the corrected cell image 11.

[0036] The superimposed cell image generation unit 2d generates a superimposed cell image 50 based on the cell image 10 or the corrected cell image 11 and the region estimation image 12. The superimposed cell image generation unit 2d also displays the generated superimposed cell image 50 on the display unit 4.

[0037] (Background component image) Next, with reference to Figure 5, a configuration in which the image processing unit 2c generates the background component image 14 will be described. In this embodiment, the image processing unit 2c generates a background component image 14 by filtering the cell image 10 acquired by the image acquisition unit 1, thereby extracting the distribution of brightness components of the background 91 from the cell image 10. Specifically, the image processing unit 2c generates the background component image 14 by applying a median filter to the cell image 10. In this embodiment, the image processing unit 2c smooths the cell image 10 by applying a median filter. In this embodiment, the image processing unit 2c smooths the image so that the cells 90 in the cell image 10 are almost invisible, and as a result, extracts the background components.

[0038] The median filter process obtains the median value of each pixel within a predetermined region (kernel) centered on the pixel of interest. Specifically, the median filter process obtains the pixel value of each pixel within the kernel centered on the pixel of interest, sorts them in order of pixel value, and obtains the median. This obtained median is then used as the pixel value of the pixel of interest. This process is repeated for each pixel of the cell image 10, changing the pixel of interest. Therefore, the median filter process increases the processing load.

[0039] Therefore, in this embodiment, the image processing unit 2c (see Figure 1) generates a reduced cell image 10c by reducing the size of the cell image 10a. Specifically, the image processing unit 2c generates the cell image 10c by performing a lossless compression process on the cell image 10a. In this embodiment, for example, the image processing unit 2c generates the reduced cell image 10c by reducing the size of the cell image 10a to 1 / 8 of its original size.

[0040] The image processing unit 2c then generates a reduced background component image 13 by filtering the reduced cell image 10c. The reduced background component image 13 is an image in which the brightness unevenness (background component) of the background 91 has been extracted by filtering. In the reduced cell image 10c, the fact that the cells 90 are almost invisible is indicated by showing the cells 90 with dashed lines.

[0041] In this embodiment, the kernel size of the median filter is set to a size that allows for the extraction of the luminance component of the background 91 and smooths the image to the point where the cells 90 in the cell image 10a are almost completely obscured. That is, the kernel size of the median filter is set according to the size of the cells 90 in the cell image 10a. In this embodiment, the image processing unit 2c applies the median filter to the cell image 10c, which is a scaled-down version of the cell image 10a. Therefore, the kernel size of the median filter is set based on the magnification of the cell image 10c. This makes it possible to perform filtering with a filter with a kernel size appropriate for the magnification of the cell image 10c, thereby suppressing an increase in processing load due to an excessively large kernel size, which would increase the number of times pixel values ​​are extracted and sorted. It also suppresses the inability to accurately extract background components due to an excessively small kernel size.

[0042] The image processing unit 2c then generates a background component image 14 by enlarging the reduced background component image 13. Specifically, the image processing unit 2c enlarges the reduced background component image 13 so that the size of the background component image 14 is the same as the size of the cell image 10a. In this embodiment, for example, the image processing unit 2c generates the background component image 14 by enlarging the reduced background component image 13 to eight times its original size. In the background component image 14, the background 91 component is extracted by filtering with a median filter, resulting in a state where the cells 90 are almost completely obscured. In the background component image 14 shown in Figure 5, the cells 90 are shown with dashed lines to indicate that the cells 90 are almost completely obscured.

[0043] (Corrected cell image) Next, referring to Figure 6, the configuration in which the image processing unit 2c (see Figure 1) generates the corrected cell image 11 will be described. In this embodiment, the image processing unit 2c generates a corrected cell image 11 by correcting the cell image 10 to reduce brightness unevenness based on the cell image 10a and the background component image 14. Specifically, as shown in Figure 6, the image processing unit 2c subtracts the background component image 14 from the cell image 10a. However, when the background component image 14 is subtracted from the cell image 10a, the pixel value of the entire image becomes smaller, and the contrast of the image may decrease. Therefore, in this embodiment, the image processing unit 2c generates the corrected cell image 11 by subtracting the background component image 14 from the cell image 10a and then adding a predetermined brightness value. The predetermined pixel value is, for example, half the value of the grayscale of the pixel value of the cell image 10a. In this embodiment, for example, if the pixel value of the cell image 10a is 256 grayscale, the image processing unit 2c adds 128 as the predetermined pixel value.

[0044] As shown in Figure 6, the corrected cell image 11 shows reduced brightness unevenness in the background 91. The corrected cell image 11 shown in Figure 6 shows that the brightness unevenness in the background 91 has been reduced by not adding hatching to the background 91.

[0045] (Estimated image of the first region) Next, with reference to Figure 7, a configuration in which the image analysis unit 2b acquires the first region estimation image 12a will be described. In this embodiment, the image analysis unit 2b acquires the first region estimation image 12a by inputting the corrected cell image 11 into the first trained model 6a. Specifically, the image analysis unit 2b estimates the normal cell region 20, which is the region of normal cells, and the abnormal cell region 21, which is the region of abnormal cells, based on the estimation results of the first trained model 6a. In this embodiment, the image analysis unit 2b generates an image as the first region estimation image 12a in which the normal cell region 20, the abnormal cell region 21, and the background region can be distinguished by coloring the normal cell region 20, the abnormal cell region 21, and the background region with different display colors. In the example shown in Figure 7, the image analysis unit 2b generates a first region estimation image 12a that displays the normal cell region 20 in black, the abnormal cell region 21 in gray, and the background region in white, thereby making the normal cell region 20, the abnormal cell region 21, and the background region distinguishable. In the first region estimation image 12a, the abnormal cell region 21 is shown to be displayed in gray by hatching. In the example shown in Figure 7, the first region estimation image 12a shows that the cells 90 in region 92 of the corrected cell image 11 are abnormal cells.

[0046] (Image of the first superimposed cell) Next, with reference to Figure 8, the configuration in which the superimposed cell image generation unit 2d (see Figure 1) generates the superimposed cell image 50 will be described. In this embodiment, the image processing unit 2c generates the superimposed cell image 50 by superimposing a first marker 51 indicating a normal cell region 20 (see Figure 7) and a second marker 52 indicating an abnormal cell region 21 (see Figure 7) onto the corrected cell image 11. Specifically, the superimposed cell image generation unit 2d generates a first superimposed cell image 50a based on the corrected cell image 11 and the first region estimation image 12a. The superimposed cell image generation unit 2d acquires the first marker 51 based on the normal cell region 20 in the first region estimation image 12a. The superimposed cell image generation unit 2d also acquires the second marker 52 based on the abnormal cell region 21 in the first region estimation image 12a. The superimposed cell image generation unit 2d then generates a first superimposed cell image 50a by superimposing the first marker 51 and the second marker 52, which were acquired based on the first region estimation image 12a, onto the corrected cell image 11.

[0047] Here, one possible method for distinguishing whether cell 90 is a normal cell or an abnormal cell is to stain normal cells and abnormal cells with different colors. However, this method of staining normal cells and abnormal cells with different colors destroys cell 90. If cell 90 is destroyed, it cannot be used for subculturing or other procedures. Therefore, as shown in Figure 8, the superimposed cell image generation unit 2d generates a first superimposed cell image 50a in which the normal cell region 20 and the abnormal cell region 21 can be distinguished by displaying the first marker 51 and the second marker 52 in a distinguishable manner. The superimposed cell image generation unit 2d displays the normal cell region 20 and the abnormal cell region 21 in a distinguishable manner by coloring them with different display colors. For example, the superimposed cell image generation unit 2d colors the first marker 51 in blue and the second marker 52 in red. In the example shown in Figure 8, the hatching of the first marker 51 and the hatching of the second marker 52 are made different to show the difference in color between the normal cell region 20 and the abnormal cell region 21 in the first superimposed cell image 50a.

[0048] (Region estimation image based on comparative example) Next, with reference to Figure 9, the region estimation image 60 from the comparative example will be explained. The region estimation image 60 from the comparative example is an image in which normal cell regions 20 (see Figure 7) and abnormal cell regions 21 (see Figure 7) are distinguished by analyzing cell image 10, in which brightness unevenness occurred in the background 91 (see Figure 3), using a trained model. Note that the region estimation image 60 from the comparative example is an image in which normal cell regions 20 and abnormal cell regions 21 were estimated by the trained model without performing any correction processing on cell image 10a (see Figure 3), which was used to generate the first region estimation image 12a (see Figure 8). The region estimation image 60 from the comparative example shows a decrease in estimation accuracy due to the brightness unevenness in the background 91. Specifically, the estimation accuracy on the left side of the image is reduced due to the brightness unevenness in the background 91.

[0049] Specifically, when comparing the first region estimation image 12a shown in Figure 8 with the region estimation image 60 from the comparative example shown in Figure 9, the region estimation image 60 from the comparative example contains regions that are not estimated as abnormal cell regions 21 (see Figure 7), such as region 61 shown by a dashed line. Also, there are regions within the normal cell region 20 (see Figure 7) that are not estimated as normal cell regions 20, such as region 62 shown by a solid line. In other words, the region estimation image 60 from the comparative example includes regions 61 that do not accurately classify abnormal cell regions 21, and regions 62 that do not accurately classify normal cell regions 20. Therefore, it can be seen that the accuracy of analysis by the trained model is reduced in the comparative example. On the other hand, the first region estimation image 12a according to this embodiment is able to suppress the reduction in the accuracy of classifying normal cell regions 20 and abnormal cell regions 21.

[0050] (Estimated image of the second region) Next, with reference to Figure 10, the configuration in which the image analysis unit 2b acquires the second region estimation image 12b will be described. In this embodiment, the image analysis unit 2b acquires the second region estimation image 12b by inputting the uncorrected cell image 15, which is an image that has not undergone correction processing, into the second trained model 6b. Specifically, the image analysis unit 2b estimates the normal cell region 20, which is the region of normal cells, and the abnormal cell region 21, which is the region of abnormal cells, based on the estimation results of the second trained model 6b. The image analysis unit 2b generates an image as the second region estimation image 12b in which the normal cell region 20, the abnormal cell region 21, and the background region can be distinguished. Specifically, in the second region estimation image 12b, the image analysis unit 2b displays the normal cell region 20, the abnormal cell region 21, and the background region in a state where they are colored with different display colors, thereby making the normal cell region 20, the abnormal cell region 21, and the background region distinguishable. In the example shown in Figure 10, the image analysis unit 2b generates a second region estimation image 12b that displays the normal cell region 20 in black, the abnormal cell region 21 in gray, and the background region in white, thereby making the normal cell region 20, the abnormal cell region 21, and the background region distinguishable. In the second region estimation image 12b, the abnormal cell region 21 is shown to be displayed in gray by hatching. In the example shown in Figure 10, the second region estimation image 12b shows that the cell 90 in region 93 of the uncorrected cell image 15 is an abnormal cell.

[0051] (Image of the second superimposed cell) Next, with reference to Figure 11, the configuration in which the superimposed cell image generation unit 2d (see Figure 1) generates the second superimposed cell image 50b will be described. In this embodiment, the superimposed cell image generation unit 2d generates the second superimposed cell image 50b by superimposing a first marker 51 indicating a normal cell region 20 (see Figure 10) and a second marker 52 indicating an abnormal cell region 21 (see Figure 10) onto the uncorrected cell image 15. Specifically, the superimposed cell image generation unit 2d acquires the first marker 51 based on the normal cell region 20 of the second region estimation image 12b. The superimposed cell image generation unit 2d also acquires the second marker 52 based on the abnormal cell region 21 of the second region estimation image 12b. Then, the superimposed cell image generation unit 2d generates the second superimposed cell image 50b by superimposing the first marker 51 and the second marker 52 acquired based on the second region estimation image 12b onto the uncorrected cell image 15.

[0052] As shown in Figure 11, the superimposed cell image generation unit 2d generates a second superimposed cell image 50b in which the normal cell region 20 and the abnormal cell region 21 can be distinguished by displaying the first marker 51 and the second marker 52 in a distinguishable manner. The superimposed cell image generation unit 2d displays the normal cell region 20 and the abnormal cell region 21 in a distinguishable manner by coloring them with different display colors. For example, the superimposed cell image generation unit 2d colors the first marker 51 in blue and the second marker 52 in red. In the example shown in Figure 11, the difference in color between the normal cell region 20 and the abnormal cell region 21 is displayed in the second superimposed cell image 50b by making the hatching of the first marker 51 and the hatching of the second marker 52 different.

[0053] (Superimposed cell image display processing) Next, referring to Figure 12, the process by which the cell image analysis device 100 displays the superimposed cell image 50 will be described.

[0054] In step 200, the control unit 2a receives an operation input indicating whether or not to generate the corrected cell image 11. In this embodiment, the control unit 2a receives the operation input from the operator via the input receiving unit 5 (see Figure 1) to indicate whether or not to generate the corrected cell image 11. The control unit 2a also sets whether or not to generate the corrected cell image 11 based on the operator's operation input. The control unit 2a also stores the setting for whether or not to generate the corrected cell image 11 in the storage unit 3.

[0055] In step 201, the image acquisition unit 1 acquires a cell image 10 in which the cell 90 is visible.

[0056] In step 202, the control unit 2a determines whether or not to generate a corrected cell image 11 based on the operation input received in step 200. If a corrected cell image 11 is to be generated, the process proceeds to step 203. If a corrected cell image 11 is not to be generated, the process proceeds to step 206. In this embodiment, the control unit 2a determines whether or not to generate a corrected cell image 11 by referring to the setting stored in the storage unit 3 regarding whether or not to generate a corrected cell image 11.

[0057] In step 203, the image processing unit 2c generates a background component image 14 by filtering the acquired cell image 10 to extract the distribution of the brightness component of the background 91 from the cell image 10.

[0058] In step 204, the image processing unit 2c generates a corrected cell image 11 by correcting the cell image 10 to reduce brightness unevenness based on the cell image 10 and the background component image 14. In this embodiment, as shown in Figure 6, the image processing unit 2c generates the corrected cell image 11 in the processing of step 204 by subtracting the background component image 14 from the cell image 10 and adding a predetermined brightness value.

[0059] In step 205, the image analysis unit 2b performs a first estimation step to estimate whether the cells 90 in the image are normal or abnormal, using the corrected cell image 11 and the trained model 6 which has been trained to analyze cells 90. In this embodiment, when the image analysis unit 2b generates the corrected cell image 11, it uses the first trained model 6a, which was generated using the corrected cell image 11 (training corrected cell image 30), as the trained model 6 to perform the first estimation step on the corrected cell image 11.

[0060] Furthermore, if the process proceeds from step 202 to step 206, in step 206, the image analysis unit 2b performs the second estimation step on the cell image 10 using the second trained model 6b generated using the uncorrected cell image 15 (training cell image 32).

[0061] In step 207, the image processing unit 2c estimates the normal cell region 20, which is the region of normal cells, and the abnormal cell region 21, which is the region of abnormal cells, based on the estimation results of the trained model 6. Specifically, in step 207, the image processing unit 2c obtains a region estimation image 12 in which the normal cell region 20 and the abnormal cell region 21 can be distinguished.

[0062] In step 208, the superimposed cell image generation unit 2d generates a superimposed cell image 50 by superimposing a first marker 51 indicating a normal cell region 20 and a second marker 52 indicating an abnormal cell region 21 onto the corrected cell image 11.

[0063] In step 209, the control unit 2a displays the superimposed cell image 50 on the display unit 4. That is, by displaying the superimposed cell image 50 on the display unit 4, the control unit 2a displays the normal cell region 20 and the abnormal cell region 21 in a distinguishable manner.

[0064] In this embodiment, when a corrected cell image 11 is generated, the cell image analysis device 100 performs the process of generating the corrected cell image 11 in step 204, and then performs the first estimation step (processing in step 205) using the corrected cell image 11 and the trained model 6. If a corrected cell image 11 is not generated, the cell image analysis device 100 does not perform the process of generating the background component image 14 in step 203, and instead performs the second estimation step (processing in step 206) which estimates using the cell image 10 and the trained model 6.

[0065] (Background component image generation process) Next, referring to Figure 13, the process by which the image processing unit 2c generates the background component image 14 will be described.

[0066] In step 203a, the image processing unit 2c reduces the size of the cell image 10a. Specifically, the image processing unit 2c reduces the size of the cell image 10a by performing a lossless compression process on it, thereby generating a reduced cell image 10c.

[0067] In step 203b, the image processing unit 2c generates a reduced background component image 13 by filtering the reduced cell image 10c. In this embodiment, the image processing unit 2c generates a reduced background component image 13 by filtering the reduced cell image 10c with a median filter.

[0068] In step 203c, the image processing unit 2c enlarges the reduced background component image 13. That is, the image processing unit 2c generates a background component image 14 by enlarging the reduced background component image 13. The process then proceeds to step 204.

[0069] (Effects of this embodiment) In this embodiment, the following effects can be obtained.

[0070] In this embodiment, as described above, the cell image analysis method includes the steps of: acquiring a cell image 10 in which cells 90 are visible; generating a background component image 14 by filtering the acquired cell image 10 to extract the distribution of brightness components of the background 91 from the cell image 10; generating a corrected cell image 11 by correcting the cell image 10 to reduce brightness unevenness based on the cell image 10 and the background component image 14; and a first estimation step of estimating whether the cells 90 in the image are normal cells or abnormal cells using the corrected cell image 11 and a trained model 6 that has been trained to analyze cells 90.

[0071] As a result, the corrected cell image 11, in which brightness unevenness in the background 91 has been reduced, and the trained model 6 can be used to estimate whether the cells 90 in the image are normal or abnormal. Consequently, the reduction in the analysis accuracy by the trained model 6 due to brightness unevenness in the background 91 of the cell image 10 can be suppressed.

[0072] Furthermore, the above embodiment provides the following additional benefits by configuring it as described below.

[0073] In other words, in this embodiment, as described above, the cell image 10 is an image that includes cultured cells 90 near the end 70 of the culture vessel 80, cultured in a culture solution 81 filled in the culture vessel 80. Here, near the end 70 of the culture vessel 80, due to surface tension, the height of the liquid surface 81a of the culture solution 81 changes so that the height of the liquid surface 81a of the culture solution 81 increases as you approach the end 70 of the culture vessel 80. When the height of the liquid surface 81a of the culture solution 81 changes, brightness unevenness occurs in the background 91 of the cell image 10 due to the difference in the height of the culture solution 81. Therefore, by applying the present invention to the cell image 10 that includes cultured cells 90 near the end 70 of the culture vessel 80, even when analyzing the cell image 10 in which brightness unevenness occurs in the background 91 due to a change in the height of the liquid surface 81a of the culture solution 81 near the end 70 of the culture vessel 80, it is possible to easily suppress a decrease in the analysis accuracy by the trained model 6.

[0074] Furthermore, in this embodiment, as described above, the filtering process is a process that generates a background component image 14 by applying a median filter to the cell image 10. Here, when applying, for example, a Gaussian filter or an averaging filter as the filtering process, it may not be possible to accurately obtain the luminance components of the background 91 due to stray light with extremely high pixel values ​​or foreign objects with extremely low pixel values. Therefore, by applying a median filter that smooths the image using the median value of the pixel values ​​of each pixel within a predetermined area, the luminance components of the background 91 can be accurately obtained even when stray light or foreign objects are present. As a result, a corrected cell image 11 with accurately corrected luminance unevenness in the background 91 can be obtained.

[0075] Furthermore, in this embodiment, as described above, the step of reducing the cell image 10 is further included, and in the step of generating the background component image 14, a reduced background component image 13 is generated by filtering the reduced cell image 10c, and the reduced background component image 13 is further enlarged. This makes it possible to suppress an increase in the processing load of filtering by applying filtering to the reduced cell image 10c. Moreover, when applying a median filter that smooths by the median of the pixel values ​​of each pixel within a predetermined region, the increase in the processing load due to filtering can be suppressed even further.

[0076] Furthermore, in this embodiment, as described above, the system further includes a step of receiving an input for whether or not to generate a corrected cell image 11. If a corrected cell image 11 is to be generated, the system executes the step of generating the corrected cell image 11 and the first estimation step using the corrected cell image 11 and the trained model 6 (first trained model 6a). If a corrected cell image 11 is not to be generated, the system further includes a second estimation step in which the system does not execute the step of generating the background component image 14 and instead uses the cell image 10 (uncorrected cell image 15) and the trained model 6 (second trained model 6b) to perform estimation. As a result, when a corrected cell image 11 is to be generated, the system executes the first estimation step using the corrected cell image 11 and the first trained model 6a, thereby suppressing a decrease in the analysis accuracy of the trained model 6 (first trained model 6a) due to brightness unevenness in the background 91 of the cell image 10. Also, if a corrected cell image 11 is not to be generated, the system executes the second estimation step using the uncorrected cell image 15 and the second trained model 6b, so the system can analyze the uncorrected cell image 15 without generating the background component image 14. As a result, by not generating the background component image 14, the increase in processing load due to filtering can be suppressed.

[0077] Furthermore, in this embodiment, as described above, when generating a corrected cell image 11, the first estimation step for the corrected cell image 11 is performed using the first trained model 6a generated using the corrected cell image 11 as the trained model 6. When a corrected cell image 11 is not generated, the second estimation step for the cell image 10 is performed using the second trained model 6b generated using the cell image 10 that has not undergone correction processing (uncorrected cell image 15). As a result, when analyzing the corrected cell image 11, the first estimation step is performed, so that the analysis can be performed using the first trained model 6a which is suitable for analyzing the corrected cell image 11. As a result, when analyzing a cell image 10 in which brightness unevenness occurs in the background 91, a decrease in analysis accuracy can be suppressed. Here, if filtering (correction processing) is performed on the cell image 10 to reduce brightness unevenness, the contour lines of the cells 90 become blurred, and the contrast of the cells 90 decreases. Therefore, when analyzing cell images 10 that have not undergone correction processing (uncorrected cell images 15), a second estimation step is performed using the uncorrected cell images 15 and the second trained model 6b. This allows the second trained model 6b to analyze the uncorrected cell images 15 that have not undergone filtering. As a result, the analysis of the uncorrected cell images 15 can be performed while suppressing the decrease in contrast of cells 90 by not performing filtering.

[0078] Furthermore, in this embodiment, as described above, in the step of generating the corrected cell image 11, the corrected cell image 11 is generated by subtracting the background component image 14 from the cell image 10 and adding a predetermined brightness value. This differs from the configuration in which the background component image 14 is simply subtracted from the cell image 10, as a predetermined brightness value is added, thereby suppressing a decrease in the contrast of the corrected cell image 11.

[0079] Furthermore, in this embodiment, as described above, in the step of estimating whether the cells 90 shown in the corrected cell image 11 are normal cells or abnormal cells, the embodiment further includes the step of estimating a normal cell region 20, which is the region of normal cells, and an abnormal cell region 21, which is the region of abnormal cells, based on the estimation results of the trained model 6, and displaying the normal cell region 20 and the abnormal cell region 21 in a distinguishable manner. As a result, in the cell image 10 in which brightness unevenness of the background 91 is suppressed, the normal cell region 20 and the abnormal cell region 21 are displayed in a distinguishable manner, so that the operator can easily distinguish between the normal cell region 20 and the abnormal cell region 21.

[0080] Furthermore, in this embodiment, as described above, the method further includes the step of generating a superimposed cell image 50 by superimposing a first marker 51 indicating a normal cell region 20 and a second marker 52 indicating an abnormal cell region 21 onto the corrected cell image 11. This makes it easy to distinguish between the normal cell region 20 and the abnormal cell region 21 by the first marker 51 and the second marker 52 displayed in the superimposed cell image 50.

[0081] Furthermore, this embodiment further includes a step of generating a trained model 6 (first trained model 6a) by training the first learning model 7a using the corrected cell image 11, as described above. This makes it possible to generate a trained model 6 (first trained model 6a) suitable for analyzing the corrected cell image 11 with reduced brightness unevenness in the background 91. As a result, it is possible to suppress a decrease in the accuracy of the analysis of the corrected cell image 11.

[0082] [Differentiation] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of the present invention is indicated by the claims rather than the description of the embodiments above, and further includes all modifications (exceptions) within the meaning and scope of the claims.

[0083] For example, the above embodiment shows an example of a configuration in which the image processing unit 2c generates a background component image 14 by applying a median filter to the cell image 10, but the present invention is not limited thereto. For example, the image processing unit 2c may be configured to generate a background component image 14 by applying a Gaussian filter or an averaging filter to the cell image 10. However, when the image processing unit 2c generates a background component image 14 by applying a Gaussian filter or an averaging filter, it may not be possible to accurately acquire the components of the background 91 due to stray light with extremely high pixel values ​​or foreign objects with extremely low pixel values. Therefore, it is preferable that the image processing unit 2c be configured to generate a background component image 14 by applying a median filter.

[0084] Furthermore, while the above embodiment shows an example of a configuration in which the image processing unit 2c generates a background component image 14 by reducing the size of the cell image 10a and applying a filter to the reduced cell image 10c, the present invention is not limited to this. The image processing unit 2c may be configured to generate a background component image 14 by applying a filter to the cell image 10a without reducing the size of the cell image 10a. However, if the image processing unit 2c is configured to generate a background component image 14 without reducing the size of the cell image 10a, the processing load due to the filtering process increases. Therefore, it is preferable that the image processing unit 2c is configured to generate a background component image 14 by applying a filter to the reduced cell image 10c.

[0085] Furthermore, while the above embodiment shows an example of a configuration in which the control unit 2a determines whether or not to generate a corrected cell image 11 based on an operation input indicating whether or not to generate a corrected cell image 11, the present invention is not limited thereto. For example, the control unit 2a may be configured to acquire whether or not there is brightness unevenness in the background 91 of the cell image 10, and if there is brightness unevenness in the background 91 of the cell image 10, it may determine to generate a corrected cell image 11, and if there is no brightness unevenness in the background 91 of the cell image 10, it may determine not to generate a corrected cell image 11.

[0086] Furthermore, while the above embodiment shows an example of a configuration in which the image processing unit 2c switches whether or not to generate a corrected cell image 11 based on the determination of the control unit 2a, the present invention is not limited thereto. The image processing unit 2c may be configured to always generate a corrected cell image 11, regardless of the determination of the control unit 2a. However, if the image processing unit 2c is configured to always generate a corrected cell image 11, filtering will be performed even on cell images 10b where there is no brightness unevenness in the background 91. This increases the processing load. Therefore, it is preferable that the image processing unit 2c be configured to switch whether or not to generate a corrected cell image 11 based on the determination of the control unit 2a.

[0087] Furthermore, in the above embodiment, an example was shown in which the image processing unit 2c generates a corrected cell image 11 by applying a correction process to cell image 10a in which brightness unevenness occurs in the background 91, and does not apply a correction process to cell image 10b in which brightness unevenness does not occur in the background 91. However, the present invention is not limited to this. The image processing unit 2c may be configured to apply a correction process to cell image 10 regardless of whether or not brightness unevenness occurs in the background 91. However, when a correction process is applied to cell image 10 in which brightness unevenness occurs in the background 91, smoothing is also performed on the cells 90, which reduces the contrast of the cells 90. Therefore, it is preferable that the image processing unit 2c is configured to apply a correction process only to cell image 10 in which brightness unevenness occurs in the background 91.

[0088] Furthermore, although the above embodiment shows an example of a configuration in which the image processing unit 2c generates a corrected cell image 11 by subtracting the background component image 14 from the cell image 10a, the present invention is not limited thereto. For example, the image processing unit 2c may be configured to generate a corrected cell image 11 by dividing the background component image 14 from the cell image 10a. Any method may be used for the image processing unit 2c to generate the corrected cell image 11.

[0089] Furthermore, while the above embodiment shows an example of a configuration in which the image processing unit 2c generates a corrected cell image 11 by subtracting the background component image 14 from the cell image 10a and adding a predetermined brightness value, the present invention is not limited thereto. For example, the image processing unit 2c may be configured to generate a corrected cell image 11 without subtracting the background component image 14 from the cell image 10a and adding a predetermined brightness value. However, if the image processing unit 2c generates a corrected cell image 11 without subtracting the background component image 14 from the cell image 10a and adding a predetermined brightness value, the contrast of the corrected cell image 11 will decrease. Therefore, it is preferable that the image processing unit 2c be configured to generate a corrected cell image 11 by subtracting the background component image 14 from the cell image 10a and adding a predetermined brightness value.

[0090] Furthermore, in the above embodiment, an example was shown in which the superimposed cell image generation unit 2d generates a superimposed cell image 50 that displays the normal cell region 20 and the abnormal cell region 21 in a distinguishable manner by coloring the normal cell region 20 blue and the abnormal cell region 21 red. However, the present invention is not limited to this. The superimposed cell image generation unit 2d may color the normal cell region 20 and the abnormal cell region 21 with any color as long as the normal cell region 20 and the abnormal cell region 21 are distinguishable.

[0091] Furthermore, while the above embodiment shows an example of a configuration in which the cell image analysis device 100 generates a first trained model 6a and a second trained model 6b, the present invention is not limited thereto. For example, the cell image analysis device 100 may be configured to use the first trained model 6a and the second trained model 6b generated by an analysis device separate from the cell image analysis device 100.

[0092] Furthermore, while the above embodiment shows an example in which the image analysis unit 2b performs analysis of the corrected cell image 11 using the first trained model 6a and also performs analysis of the uncorrected cell image 15 using the second trained model 6b, the present invention is not limited to this. For example, the image analysis unit 2b does not have to perform analysis using the second trained model 6b. However, if the image analysis unit 2b does not perform analysis using the second trained model 6b, the uncorrected cell image 15 will be analyzed using the first trained model 6a. In that case, the contrast of the cells 90 shown in the uncorrected cell image 15 may decrease. Therefore, it is preferable to configure the image analysis unit 2b to perform analysis of the corrected cell image 11 using the first trained model 6a and also perform analysis of the uncorrected cell image 15 using the second trained model 6b.

[0093] Furthermore, although the above embodiment shows an example in which the image acquisition unit 1 acquires the cell image 10 as part of the process in step 201, the present invention is not limited thereto. For example, the image processing unit 2c may be configured to acquire the cell image 10 that has been previously acquired by the image acquisition unit 1 and stored in the storage unit 3.

[0094] Furthermore, while the above embodiment shows an example in which the image analysis unit 2b analyzes whether the cells 90 in the cell image 10 are undifferentiated cells or undifferentiated stray cells, the present invention is not limited thereto. For example, the image analysis unit 2b may be configured to analyze cancer cells and non-cancer cells. The cells analyzed by the image analysis unit 2b are not limited to undifferentiated cells and undifferentiated stray cells.

[0095] [Pattern] Those skilled in the art will understand that the exemplary embodiments described above are specific examples of the following embodiments.

[0096] (Item 1) Steps to obtain a cell image showing the cells, The process involves a step of generating a background component image by filtering the acquired cell image to extract the distribution of background brightness components from the cell image, The steps include generating a corrected cell image by correcting the cell image to reduce brightness unevenness based on the cell image and the background component image, A cell image analysis method comprising: a first estimation step of estimating whether the cells in the image are normal or abnormal cells using the corrected cell image and a trained model trained to analyze the cells.

[0097] (Item 2) The cell image analysis method according to item 1, wherein the cell image is an image including cultured cells near the end of the culture vessel that have been cultured in a culture solution filled in the culture vessel.

[0098] (Item 3) The cell image analysis method according to item 1 or 2, wherein the filtering process is a process of generating the background component image by applying a median filter to the cell image.

[0099] (Item 4) The step further comprises reducing the size of the cell image, In the step of generating the background component image, the reduced background component image is generated by the filtering process applied to the reduced cell image. A cell image analysis method according to any one of items 1 to 3, further comprising the step of enlarging the reduced background component image.

[0100] (Item 5) The system further includes a step of receiving an input for whether or not to generate the corrected cell image, When generating the corrected cell image, the step of generating the corrected cell image is performed, and the first estimation step is performed using the corrected cell image and the trained model. A cell image analysis method according to any one of items 1 to 4, further comprising a second estimation step of estimating using the cell image and the trained model, without performing the step of generating the background component image if the corrected cell image is not generated.

[0101] (Item 6) When generating the corrected cell image, the first estimation step for the corrected cell image is performed using the first trained model generated using the corrected cell image as the trained model. The cell image analysis method according to item 5, wherein, if the corrected cell image is not generated, the second estimation step for the cell image is performed using a second trained model generated using the cell image that has not undergone correction processing.

[0102] (Item 7) A cell image analysis method according to any one of items 1 to 6, wherein in the step of generating the corrected cell image, the corrected cell image is generated by subtracting the background component image from the cell image and adding a predetermined brightness value.

[0103] (Item 8) In the step of estimating whether the cells shown in the corrected cell image are normal cells or abnormal cells, the normal cell region, which is the region of normal cells, and the abnormal cell region, which is the region of abnormal cells, are estimated based on the estimation results of the trained model. A cell image analysis method according to any one of items 1 to 7, further comprising the step of displaying the normal cell region and the abnormal cell region in a distinguishable manner.

[0104] (Item 9) The cell image analysis method according to item 8, further comprising the step of generating a superimposed cell image by superimposing a first marker indicating a normal cell region and a second marker indicating an abnormal cell region onto the corrected cell image.

[0105] (Item 10) The cell image analysis method according to item 1, further comprising the step of generating the trained model by training a first learning model using the corrected cell images. [Explanation of Symbols]

[0106] 6. Pre-trained models 6a First pre-trained model 6b Second pre-trained model 10 Cell images 10c Reduced cell image 13. Reduced background component image 14 Background component image 15. Uncorrected cell images (cell images that have not undergone any correction processing) 20 Normal cell area 21 Abnormal cell area 50 superimposed images 51 1st sign 52 Second sign 70 Near the edge of the culture vessel 80 Culture vessels 80a End of culture vessel 81 Culture solution 90 cells (cultured cells) 91 Background

Claims

1. Steps to obtain a cell image showing the cells, The process involves a step of generating a background component image by filtering the acquired cell image to extract the distribution of background brightness components from the cell image, The steps include generating a corrected cell image by correcting the cell image to reduce brightness unevenness based on the cell image and the background component image, The system includes a first estimation step of estimating whether the cells in the image are normal or abnormal, using the corrected cell image and a trained model trained to analyze the cells. The cell image is of cultured cells cultured in a culture solution filled in a culture vessel, and includes the cultured cells in a region where the height of the liquid level of the culture solution changes relative to the center of the culture vessel due to surface tension. A cell image analysis method comprising the step of generating the corrected cell image, wherein the cell image is corrected based on the cell image and the background component image generated by the filtering process to reduce the brightness unevenness caused by changes in the height of the culture solution in the culture vessel due to surface tension in a region where the height of the liquid level of the culture solution in the culture vessel changes relative to the central part of the culture vessel.

2. The cell image analysis method according to claim 1, wherein the filtering process is a process of generating the background component image by applying a median filter to the cell image.

3. The step further comprises reducing the size of the cell image, In the step of generating the background component image, the reduced background component image is generated by the filtering process applied to the reduced cell image. The cell image analysis method according to claim 1, further comprising the step of enlarging the reduced background component image.

4. Steps to obtain a cell image showing the cells, The process involves a step of generating a background component image by filtering the acquired cell image to extract the distribution of background brightness components from the cell image, The steps include generating a corrected cell image by correcting the cell image to reduce brightness unevenness based on the cell image and the background component image, A first estimation step involves using the corrected cell image and a trained model that has been trained to analyze the cells to estimate whether the cells in the image are normal or abnormal. The system includes a step of receiving an input for whether or not to generate the corrected cell image, The cell image is of cultured cells cultured in a culture solution filled in a culture vessel, and includes the cultured cells in a region where the height of the liquid level of the culture solution changes relative to the center of the culture vessel due to surface tension. When generating the corrected cell image, in the step of generating the corrected cell image, the cell image is corrected to reduce the brightness unevenness caused by the change in the height of the culture solution due to surface tension in a region where the height of the liquid level of the culture solution in the culture vessel changes relative to the central part of the culture vessel, and the first estimation step is performed using the corrected cell image and the trained model. A cell image analysis method further comprising, if the corrected cell image is not generated, a second estimation step of estimating whether the cells in the image are normal cells or abnormal cells using the cell image and the trained model, without performing the step of generating the background component image.

5. When generating the corrected cell image, the first estimation step for the corrected cell image is performed using the first trained model generated using the corrected cell image as the trained model. The cell image analysis method according to claim 4, wherein, if the corrected cell image is not generated, the second estimation step for the cell image is performed using a second trained model generated using the cell image that has not undergone correction processing.

6. The cell image analysis method according to claim 1, wherein in the step of generating the corrected cell image, the corrected cell image is generated by subtracting the background component image from the cell image and adding a predetermined brightness value.

7. In the step of estimating whether the cells shown in the corrected cell image are normal cells or abnormal cells, the normal cell region, which is the region of normal cells, and the abnormal cell region, which is the region of abnormal cells, are estimated based on the estimation results of the trained model. The cell image analysis method according to claim 1, further comprising the step of displaying the normal cell region and the abnormal cell region in a distinguishable manner.

8. The cell image analysis method according to claim 7, further comprising the step of generating a superimposed cell image by superimposing a first marker indicating a normal cell region and a second marker indicating an abnormal cell region onto the corrected cell image.

9. The cell image analysis method according to claim 1, further comprising the step of generating the trained model by training the trained model using the corrected cell images.

Images