Cell quality evaluation device, cell quality evaluation method, and program

The device evaluates cell quality by analyzing cell shape and force information from images, addressing the limitations of existing methods by incorporating mechanical properties for improved accuracy and efficiency.

WO2024247727A1PCT designated stage expired Publication Date: 2024-12-05NIKON CORP +1

Patent Information

Application Number
PCT/JP2024/017956
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-30
Filing Date
2024-05-15
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Current methods for evaluating cell quality, such as analyzing cell proliferation and gene expression, lack the inclusion of mechanical properties like elastic modulus and stress field, which are crucial for accurate cell diagnosis and are invasive, costly, and have low throughput.

Method used

A device and method that evaluates cell quality using cell shape and force information by receiving cell images, extracting contours, deriving shape and force information, and using this data to assess cell quality non-invasively with high throughput.

Benefits of technology

Enables comprehensive cell quality evaluation by incorporating mechanical properties, improving accuracy and reducing costs and invasiveness compared to existing methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2024017956_05122024_PF_FP_ABST
    Figure JP2024017956_05122024_PF_FP_ABST
Patent Text Reader

Abstract

A cell quality evaluation device comprising: a reception unit that receives a cell image; a contour extraction unit that extracts the contour of a cell included in the cell image received by the reception unit; a force information derivation unit that derives force information for the cell on the basis of the contour of the cell extracted by the contour extraction unit; a shape information derivation unit that derives shape information for the cell on the basis of the contour of the cell extracted by the contour extraction unit; and a quality evaluation unit that evaluates the quality of the cell on the basis of the shape information for the cell derived by the shape information derivation unit and the force information for the cell derived by the force information derivation unit.
Need to check novelty before this filing date? Find Prior Art

Description

Cell quality evaluation device, cell quality evaluation method, and program

[0001] This application claims priority to Japanese Patent Application No. 2023-088762, filed May 30, 2023, the contents of which are incorporated herein by reference.

[0002] Currently, cell diagnosis and quality assessment are primarily performed by analyzing cell proliferation, viability, gene expression, and other parameters. Cell morphology is also important for understanding cell status. Cell biology and molecular biology are exploring new indicators for evaluating cell properties in addition to these, and the importance of mechanical properties and conditions of cells, such as elastic modulus and stress field, is particularly noteworthy. Therefore, it is anticipated that adding evaluation of such mechanical properties of cells to existing cell analysis methods will lead to improved accuracy in cell diagnosis and assessment.

[0003] There are a variety of laboratory techniques for capturing the mechanical characteristics of cells. For example, using a laser cutter to cut the edge of a cell and then analyzing the tension on that edge, or using specialized platforms such as micropillars or Traction Force Microscopy (TFM), are some of the established techniques. While these have been studied for a long time and are already well established, their application to actual cell evaluation poses challenges in terms of invasiveness, the cost of preparing specialized platforms, and throughput. One approach to address these challenges is to numerically estimate the stress field acting on cells by assuming a balance of forces between cells (see, for example, Non-Patent Document 1).

[0004] Goshi Ogita, Takefumi Kondo, Keisuke Ikawa, Tadashi Uemura, Shuji Ishihara, Kaoru Sugimura, “Image-based parameter inference for epithelial mechanics”, PLOS Computational Biology Published: June 23, 2022

[0005] An object of the present invention is to provide a cell quality evaluation device, a cell quality evaluation method, and a program that can evaluate the quality of a cell using shape information and force information of the cell.

[0006] In order to solve the above problems, one aspect of the present invention is a cell quality evaluation device comprising: a receiving unit that receives a cell image; a contour extraction unit that extracts the contour of a cell included in the cell image received by the receiving unit; a force information derivation unit that derives force information of the cell based on the contour of the cell extracted by the contour extraction unit; a shape information derivation unit that derives shape information of the cell based on the contour of the cell extracted by the contour extraction unit; and a quality evaluation unit that evaluates the quality of the cell based on the shape information of the cell derived by the shape information derivation unit and the force information of the cell derived by the force information derivation unit.

[0007] One aspect of the present invention is a cell quality evaluation method comprising: a reception step for receiving a cell image; a contour extraction step for extracting a contour of a cell included in the cell image received in the reception step; a force information derivation step for deriving force information of the cell based on the contour of the cell extracted in the contour extraction step; a shape information derivation step for deriving shape information of the cell based on the contour of the cell extracted in the contour extraction step; and a quality evaluation step for evaluating the quality of the cell based on the shape information of the cell derived in the shape information derivation step and the force information of the cell derived in the force information derivation step.

[0008] One aspect of the present invention is a program that causes a computer for evaluating the quality of a cell to execute a reception process for receiving a cell image, a contour extraction process for extracting a contour of a cell included in the cell image received in the reception process, a force information derivation process for deriving force information of the cell based on the contour of the cell extracted in the contour extraction process, a shape information derivation process for deriving shape information of the cell based on the contour of the cell extracted in the contour extraction process, and a quality evaluation process for evaluating the quality of the cell based on the shape information of the cell derived in the shape information derivation process and the force information of the cell derived in the force information derivation process.

[0009] According to the embodiments of the present invention, it is possible to provide a cell quality evaluation device, a cell quality evaluation method, and a program that can evaluate the quality of a cell using shape information and force information of the cell.

[0010] FIG. 1 is a block diagram showing an overview of an incubator including a cell evaluation device according to an embodiment. FIG. 2 is a front view of an incubator according to an embodiment. FIG. 3 is a plan view of an incubator according to an embodiment. FIG. 4 is a diagram showing an example of cell force. FIG. 5 is a diagram showing an example of the relative value of the tension on the cell adhesion surface and the pressure of the cell estimated from a cell image. FIG. 6 is a flowchart showing an example of the operation of the cell quality evaluation device according to the present embodiment. FIG. 7 is a block diagram showing an overview of an incubator including a cell evaluation device according to a modified embodiment. FIG. 8 is a diagram showing an example of the relationship between cell shape and estimation accuracy. FIG. 9 is a diagram for explaining a method for estimating a stress field of a cell. FIG. 10 is a diagram showing the flow of a 2D vertex model. FIG. 11 is a diagram showing T implemented in a 2D vertex model. 1 -transition. Shows a group of parameters used in a simulation model. Shows control parameters used in a simulation model. Shows an example of a simulation result. Shows a comparison result between simulation and estimation. Shows the parameter dependency of estimation accuracy. Shows the shape of a cell in parameter space. Shows an example of a cell shape. Shows the parameter dependency of the average value of each cell shape feature. Shows the parameter dependency of the standard deviation of each cell shape feature. Shows the correlation between estimation accuracy in all parameter sets and the average value of each cell shape feature. Shows the correlation between estimation accuracy in all parameter sets and the standard deviation of each cell shape feature. Shows an example of a learning device according to a modified embodiment. Shape-related value x m and the force value x s and the predicted value y i FIG. 10 is a diagram illustrating an example of a relationship between the above and the above. FIG. 11 is a flowchart illustrating an example of an operation of a cell quality evaluation device according to a modified example of an embodiment. FIG. 12 is a flowchart illustrating an example of an operation of a learning device according to a modified example of an embodiment. FIG. 13 is a flowchart illustrating an example of an operation of an information processing device according to a modified example of an embodiment. FIG. 14 is a diagram for explaining an example of an operation of an information processing device according to a modified example of an embodiment.

[0011] Embodiments Hereinafter, embodiments of the present invention will be described with reference to the drawings. Fig. 1 is a block diagram showing an outline of an incubator including a cell evaluation device according to an embodiment. Figs. 2 and 3 are a front view and a plan view of the incubator according to the embodiment.

[0012] The incubator 11 according to the embodiment has an upper casing 12 and a lower casing 13. When the incubator 11 is assembled, the upper casing 12 is placed on the lower casing 13. The internal spaces of the upper casing 12 and the lower casing 13 are divided into upper and lower spaces by a base plate 14.

[0013] First, an outline of the configuration of the upper casing 12 will be described. A temperature-controlled chamber 15 in which cells are cultured is formed inside the upper casing 12. This temperature-controlled chamber 15 has a temperature controller 15a and a humidity controller 15b, and the inside of the temperature-controlled chamber 15 is maintained in an environment suitable for cell culture (for example, an atmosphere with a temperature of 37°C and a humidity of 90%) (note that the temperature controller 15a and humidity controller 15b are not shown in Figures 2 and 3).

[0014] A large door 16, a middle door 17, and a small door 18 are arranged on the front of the temperature-controlled chamber 15. The large door 16 covers the front of the upper casing 12 and the lower casing 13. The middle door 17 covers the front of the upper casing 12 and isolates the temperature-controlled chamber 15 from the outside environment when the large door 16 is open. The small door 18 is a door for loading and unloading a culture vessel 19 for culturing cells, and is attached to the middle door 17. Loading and unloading the culture vessel 19 through this small door 18 makes it possible to suppress environmental changes in the temperature-controlled chamber 15. The large door 16, middle door 17, and small door 18 are kept airtight by packings P1, P2, and P3, respectively.

[0015] The temperature-controlled room 15 also contains a stocker 21, an observation unit 22, a container transport device 23, and a transport table 24. The transport table 24 is located in front of the small door 18, and the culture containers 19 are transported in and out through the small door 18.

[0016] The stocker 21 is disposed on the left side of the temperature-controlled room 15 when viewed from the front of the upper casing 12 (the lower side in FIG. 3 ). The stocker 21 has multiple shelves, and each shelf of the stocker 21 can accommodate multiple culture vessels 19. Each culture vessel 19 accommodates cells to be cultured together with a culture medium.

[0017] The observation unit 22 is disposed on the right side of the temperature-controlled chamber 15 when viewed from the front of the upper casing 12. This observation unit 22 is capable of performing time-lapse observation of cells in the culture vessel 19. Here, the observation unit 22 is disposed by being fitted into an opening in the base plate 14 of the upper casing 12. The observation unit 22 has a sample stage 31, a stand arm 32 that extends above the sample stage 31, and a main body portion 33 that incorporates a microscopic optical system and an imaging device (38) for phase-contrast observation. The sample stage 31 and the stand arm 32 are disposed in the temperature-controlled chamber 15, while the main body portion 33 is housed in the lower casing 13.

[0018] The sample stage 31 is made of a light-transmitting material, and the culture vessel 19 can be placed on it. The sample stage 31 is configured to be movable horizontally, allowing the position of the culture vessel 19 placed on its top surface to be adjusted. An LED light source 35 is built into the stand arm 32. The imaging device 38 can capture a microscopic image of the cells in the culture vessel 19, which are transmitted and illuminated from above the sample stage 31 by the stand arm 32, via a micro-optical system.

[0019] The container transport device 23 is disposed in the center of the temperature-controlled chamber 15 when viewed from the front of the upper casing 12. The container transport device 23 transfers the culture container 19 between the stocker 21, the sample stage 31 of the observation unit 22, and the transport stage 24.

[0020] 2, the container transport device 23 includes a vertical robot 34 with an articulated arm, a rotation stage 39, a mini-stage 36, and an arm unit 37. The rotation stage 39 is attached to the tip of the vertical robot 34 via a rotation shaft 39a so as to be rotatable 180° in the horizontal direction. Therefore, the rotation stage 39 allows the arm unit 37 to face the stocker 21, the sample stage 31, and the transport table 24, respectively.

[0021] The mini-stage 36 is attached so as to be slidable in the horizontal direction relative to the rotary stage 39. An arm part 37 for gripping the culture vessel 19 is attached to the mini-stage 36.

[0022] Next, an outline of the configuration of the lower casing 13 will be described. Inside the lower casing 13, the main body 33 of the observation unit 22 and the cell quality evaluation device 100 of the incubator 11 are housed.

[0023] The cell quality evaluation device 100 is connected to the temperature adjustment device 15a, the humidity adjustment device 15b, the observation unit 22, and the container transport device 23. The cell quality evaluation device 100 comprehensively controls each part of the incubator 11 according to a predetermined program.

[0024] As an example, the cell quality evaluation device 100 controls the temperature adjustment device 15a and the humidity adjustment device 15b to maintain predetermined environmental conditions inside the temperature-controlled room 15. Furthermore, the cell quality evaluation device 100 controls the observation unit 22 and the container transport device 23 based on a predetermined observation schedule to automatically execute an observation sequence for the culture container 19. Furthermore, the cell quality evaluation device 100 executes a culture state quality evaluation process that evaluates the quality of the culture state of the cells based on images acquired in the observation sequence.

[0025] The configuration of the cell quality evaluation device 100 will be described with reference to FIG. 1 . The cell quality evaluation device 100 is realized by a device such as a personal computer, a server, a smartphone, a tablet computer, or an industrial computer. The cell quality evaluation device 100 includes a receiving unit 102, a contour extraction unit 104, a shape information derivation unit 106, a force information derivation unit 108, a quality evaluation unit 110, and a storage unit 112. The storage unit 112 is configured with a non-volatile storage medium such as a hard disk or flash memory. The storage unit 112 stores management data related to each culture vessel 19 stored in the stocker 21 and data of microscopic images captured by the imaging device 38. The storage unit 112 also stores programs executed by the cell quality evaluation device 100.

[0026] The management data includes (a) index data indicating each culture vessel 19, (b) the storage position of the culture vessel 19 in the stocker 21, (c) the type and shape of the culture vessel 19 (well plate, dish, flask, etc.), (d) the type of cells being cultured in the culture vessel 19, (e) an observation schedule for the culture vessel 19, and (f) imaging conditions during time-lapse observation (magnification of the objective lens, observation point within the vessel, etc.). In addition, for culture vessels 19 that can simultaneously culture cells in multiple small vessels such as well plates, management data is generated for each of the small vessels.

[0027] The receiving unit 102 reads one or more images (hereinafter referred to as "cell images") of one or more cells cultured in a culture vessel, captured in time series, from the storage unit 112. The receiving unit 102 receives the read one or more cell images. For example, the receiving unit 102 receives one or more cell images obtained by bright-field observation or phase-contrast observation. The contour extraction unit 104 acquires the one or more cell images received by the receiving unit 102. The contour extraction unit 104 extracts the contour of one or more cells included in each of the acquired one or more cell images. For example, the contour extraction unit 104 extracts the contour of one or more cells by image processing such as binarization.

[0028] The shape information derivation unit 106 acquires information indicating the contours of one or more cells contained in each of the one or more cell images from the contour extraction unit 104. The shape information derivation unit 106 derives shape information of one or more cells based on the information indicating the contours of one or more cells contained in each of the acquired one or more cell images. For example, the shape information derivation unit 106 acquires circularity, perimeter, number of polygons, etc. as shape information of the cells.

[0029] The force information derivation unit 108 acquires information indicating the contours of one or more cells contained in each of the one or more cell images extracted by the contour extraction unit 104, and derives force information of one or more cells based on the information indicating the contours of one or more cells contained in each of the acquired one or more cell images. For example, the force information derivation unit 108 formulates simultaneous equations of force balance based on the cell images and estimates unknown tension and pressure by solving the simultaneous equations using Bayesian estimation. An example of cellular force is the relative value of the tension at the cell adhesion surface and the pressure at the cell. For example, the tension at the cell adhesion surface takes a positive value.

[0030] Figure 4A shows an example of cell forces (Non-Patent Document 1). Figure 4A shows the tension T2 at the adhesive surface between cells C1 and C3, between cells C1 and C2, between cells C2 and C3, between cells C3 and C1, the pressure P1 at the adhesive surface between cells C1 and C2, the pressure P2 at the adhesive surface between cells C2 and C3, and the pressure P3 at the adhesive surface between cells C1 and C3.

[0031] According to Bayes' theorem, the posterior distribution function is proportional to the product of the likelihood function and the prior distribution. In other words, equation (1) holds: P(force|form) ∝ P(form|force) P(force) (1) In equation (1), the form is expressed by a balance equation.

[0032] Figure 4B shows an example of the relative values ​​of the tension at the cell adhesion surface and the pressure of the cells estimated from a cell image (Reference 1). Figure 4B shows the tension at the cell adhesion surface of each of the multiple cells and the pressure of each of the multiple cells. Returning to Figure 1, we will continue the explanation.

[0033] The quality evaluation unit 110 acquires information indicating the shape of one or more cells included in each of the one or more cell images from the shape information derivation unit 106. The quality evaluation unit 110 acquires force information of one or more cells included in each of the one or more cell images from the force information derivation unit 108. The quality evaluation unit 110 evaluates the quality of one or more cells based on the information indicating the shape of one or more cells included in each of the acquired one or more cell images and the force information of one or more cells included in each of the one or more cell images. For example, the quality evaluation unit 110 derives a predicted value y based on an evaluation based on the shape information of the cells and an evaluation based on the force information of the cells, and comprehensively evaluates the cells.

[0034] The predicted value y will be explained. i For the predicted value y i is expressed by equation (2). i =w m f m (x m ) + w s f s (x s ) (2) In equation (2), the predicted value y i For example, x is 1 for normal cells and 0 for abnormal cells. m is a value related to the shape, for example, x m represents the roundness, perimeter, polygon, etc. m Ha x m is converted into a predicted value. m Ha x m is the weight for x s is a value related to the force, for example, x s represents pressure and tension. s Ha x s is the weight for f s Ha x s is converted into a predicted value. For example, w m , w s is 1.

[0035] The reception unit 102, the contour extraction unit 104, the shape information derivation unit 106, the force information derivation unit 108, and the quality evaluation unit 110 are realized, for example, by a hardware processor such as a CPU (Central Processing Unit) executing a computer program (software) stored in the storage unit 112.

[0036] Furthermore, some or all of these functional units may be realized by hardware (including circuitry) such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a GPU (Graphics Processing Unit), or may be realized by a combination of software and hardware. The computer program may be stored in advance in a storage device such as an HDD or flash memory, or may be stored on a removable storage medium such as a DVD (Digital Versatile Disc) or CD-ROM, and installed by inserting the storage medium into a drive.

[0037] (Operation of the cell quality evaluation device) Figure 5 is a flowchart showing an example of the operation of the cell quality evaluation device according to this embodiment. (Step S1-1) The receiving unit 102 receives one or more cell images. (Step S2-1) The contour extraction unit 104 acquires one or more cell images received by the receiving unit 102. The contour extraction unit 104 extracts the contours of one or more cells included in each of the acquired one or more cell images.

[0038] (Step S3-1) The shape information derivation unit 106 acquires information indicating the contours of one or more cells included in each of the one or more cell images from the contour extraction unit 104. The shape information derivation unit 106 derives shape information of one or more cells based on the information indicating the contours of one or more cells included in each of the acquired one or more cell images. (Step S4-1) The force information derivation unit 108 acquires information indicating the contours of one or more cells included in each of the one or more cell images extracted by the contour extraction unit 104, and derives force information of one or more cells based on the information indicating the contours of one or more cells included in each of the acquired one or more cell images.

[0039] (Step S5-1) The quality evaluation unit 110 acquires information indicating the shape of one or more cells included in each of one or more cell images from the shape information derivation unit 106. The quality evaluation unit 110 acquires force information of one or more cells included in each of one or more cell images from the force information derivation unit 108. The quality evaluation unit 110 evaluates the quality of one or more cells based on the information indicating the shape of one or more cells included in each of the acquired one or more cell images and the force information of one or more cells included in each of the acquired one or more cell images. In the flowchart shown in FIG. 5, the order of step S3-1 and step S4-1 may be reversed.

[0040] In the above-described embodiment, the case where the receiving unit 102 in the cell quality evaluation device 100 receives one or more cell images obtained by bright-field observation or phase-contrast observation has been described as an example, but the present invention is not limited to this example. For example, the receiving unit 102 may receive, as the cell image, a differential interference contrast image, a fluorescent image, or an image obtained by a confocal microscope. The receiving unit 102 may receive, as the cell image, a still image or a time-lapse image. Furthermore, the receiving unit 102 is not limited to receiving the above-described two-dimensional (2D) cell images, but may also receive three-dimensional (3D) cell images.

[0041] In the above-described embodiment, the outline extraction unit 104 in the cell quality evaluation device 100 extracts the outlines of the cells or individuals through image processing such as binarization, but this is not limiting. For example, the outline extraction unit 104 may extract the outlines of the cells or individuals through machine learning, rule-based algorithms, the Watershed algorithm, or the Superpixel algorithm. An example of machine learning is NIS.ai's convert.ai. The outline extraction unit 104 may extract the outlines of the cells or individuals through Voronoi division, or may acquire information indicating the outlines of the cells or individuals extracted manually.

[0042] In the above-described embodiment, the shape information derivation unit 106 of the cell quality evaluation device 100 acquires the circularity, perimeter, number of sides, etc., as the shape information of the cell. However, this is not a limitation. For example, the shape information derivation unit 106 may acquire the area, the circularity (= 4π × area / (perimeter × perimeter)), the length (the maximum value of the lines crossing the cell), the slope (the slope of the "length"), or the width (the maximum value of the lines perpendicular to the "length") as the shape information of the cell. The shape information derivation unit 106 may acquire the radius of the inscribed circle of the cell, the radius of the circumscribed circle of the cell, the average distance between all points constituting the cell's outline and its center of gravity, a value indicating the radius of a circle with the same area as the cell, the principal moments of inertia of the inertia tensor and their ratios, or the lengths and ratios of the major and minor axes of the bounding box.

[0043] In the embodiment described above, in cell quality evaluation device 100, the case has been described in which force information derivation unit 108 estimates cellular forces from cell images using Bayesian force estimation, but the present invention is not limited to this example. For example, force information derivation unit 108 may estimate cellular forces using pseudoinverse matrix calculations.

[0044] In the above-described embodiment, the quality evaluation unit 110 may evaluate the entire cell group, or may evaluate each individual cell. Examples of evaluations of the entire cell group include differentiation potential, proliferation potential, toxicity, metabolic level, chemotaxis, adhesiveness, and sealing function. Examples of evaluations of each individual cell include mutation detection, contamination detection, differentiated / undifferentiated state, differentiation potential, proliferation potential, metabolic level, chemotaxis, adhesiveness, and sealing function.

[0045] According to the incubator 11 of the embodiment, the cell quality evaluation device 100 includes a receiving unit 102 that receives a cell image, a contour extraction unit 104 that extracts the contours of the cells included in the cell image received by the receiving unit 102, a shape information derivation unit 106 that derives shape information of the cells based on the contours of the cells extracted by the contour extraction unit 104, a force information derivation unit 108 that derives force information of the cells based on the contours of the cells extracted by the contour extraction unit 104, and a quality evaluation unit 110 that evaluates the quality of the cells based on the shape information of the cells derived by the shape information derivation unit 106 and the force information of the cells derived by the force information derivation unit 108.

[0046] With this configuration, the cell quality evaluation device 100 can derive cell shape information based on the cell outline, and cell force information based on the cell outline, and can therefore evaluate the quality of the cell based on the derived cell shape information and cell force information. The quality evaluation of cells during culture generally uses the fluorescence intensity of specific proteins and cell shape information. However, there are other indices that represent cell properties, and the "force" acting on the cell is one such indices. In this embodiment, the cell quality can be comprehensively evaluated using the cell shape information and cell force information.

[0047] Appropriate control of substrate hardness and application of an appropriate force field to cells are essential for controlling directed cell differentiation and organoid formation. Cells convert substrate hardness into force magnitude, thereby changing differentiation control signals and inducing different differentiations. Because force is generated while maintaining the cell's shape and size, it has been difficult to identify differentiated cells using previous methods (fluorescent staining, shape information, etc.). This embodiment measures the force field, recognizes differentiated cells, and ultimately makes it possible to appropriately control the direction of differentiation. While there are several methods for measuring cell force fields, this embodiment, compared to these methods, enables non-invasive, high-throughput monitoring of force fields without the use of special reagents, plates, or equipment.

[0048] In the cell quality evaluation device 100, the force information derivation unit 108 derives the tension and pressure of the cell based on the outline of the cell. With this configuration, the cell quality evaluation device 100 can derive the tension and pressure of the cell based on the outline of the cell, and can therefore evaluate the quality of the cell based on the derived tension and pressure of the cell.

[0049] (Modification of the embodiment) Hereinafter, an incubator 11a according to a modification of the embodiment will be described with reference to the drawings. FIG. 6 is a block diagram showing an outline of an incubator including a cell evaluation device according to a modification of the embodiment. The incubator 11a according to the modification of the embodiment differs from the incubator 11 in that it includes a cell quality evaluation device 100a instead of the cell quality evaluation device 100. FIG. 6 shows a learning device 200 and an information processing device 300 in addition to the incubator 11a. The incubator 11a, the learning device 200, and the information processing device 300 communicate with each other via a network NW. The network NW includes, for example, the Internet, a wide area network (WAN), a local area network (LAN), a provider device, a wireless base station, etc.

[0050] The cell quality evaluation device 100a derives the accuracy of the cellular force information derived by the force information derivation unit 108 in the cell quality evaluation device 100, and evaluates the quality of the cell based on the accuracy of the derived cellular force information. The configuration of the cell quality evaluation device 100a will be described with reference to FIG. 6. The cell quality evaluation device 100a is realized by a device such as a personal computer, a server, a smartphone, a tablet computer, or an industrial computer. The cell quality evaluation device 100a includes a receiving unit 102, a contour extraction unit 104, a shape information derivation unit 106, a force information derivation unit 108, an accuracy information derivation unit 109a, a quality evaluation unit 110a, a storage unit 112, and a communication unit 114a.

[0051] The communication unit 114a is realized by a communication module. The communication unit 114a communicates with an external communication device via the network NW. The communication unit 114a communicates using a wireless communication method such as wireless LAN, Bluetooth (registered trademark), or LTE (registered trademark). The communication unit 114a holds communication information necessary for communicating with the learning device 200 and the information processing device 300 via the network NW. The communication unit 114a receives a trained model notification transmitted by the learning device 200. The trained model notification will be described later. The communication unit 114a receives a cell shape-force information acquisition accuracy map transmitted by the information processing device 300. The cell shape-force information acquisition accuracy map associates shape information of one or more cells with the accuracy of the cell force information for each of the shape information of one or more cells.

[0052] The accuracy information derivation unit 109a acquires a cell shape-force information acquisition accuracy map from the communication unit 114. The accuracy information derivation unit 109a acquires shape information of one or more cells included in each of one or more cell images from the shape information derivation unit 106. The accuracy information derivation unit 109a derives the accuracy of the force information of one or more cells based on the shape information of one or more cells included in each of the acquired one or more cell images. For example, the accuracy information derivation unit 109a acquires the accuracy of the force information of one or more cells associated with each of the acquired shape information of one or more cells from the cell shape-force information acquisition accuracy map. The creation of the cell shape-force information acquisition accuracy map will be described later.

[0053] The quality evaluation unit 110a acquires information indicating the shape of one or more cells included in each of one or more cell images from the shape information derivation unit 106. The quality evaluation unit 110a acquires force information of one or more cells included in each of one or more cell images from the force information derivation unit 108. The quality evaluation unit 110a acquires the accuracy of the force information of the cells associated with each of the shape information of one or more cells from the accuracy information derivation unit 109a. The quality evaluation unit 110a acquires cell lineage information, and calculates a shape-related value x from the trained model based on the acquired cell lineage information. m The basic weight of w m and the force value x s The basic weight of w s For example, the quality evaluation unit 110a acquires information indicating the number of cell types (homologous / heterologous) as the cell lineage information. m The basic weight of w m and the value of the force x s The basic weight of w s The details will be described later.

[0054] The quality evaluation unit 110a derives the estimation accuracy of shape information for each of one or more cells based on the shape information of one or more cells included in each of one or more acquired cell images. FIG. 7 is a diagram showing an example of the relationship between cell shape and estimation accuracy. In FIG. 7, a case will be described in which the circularity of a cell is used as an example of the cell shape. The circularity of a cell is derived by 4πS / p^2, where S is the area of ​​the cell and p is the perimeter of the cell. In FIG. 7, the horizontal axis represents the mean of circularity x c The vertical axis represents the estimation accuracy σ. Here, the root mean square error is used as an example of the estimation accuracy σ.

[0055] According to FIG. 7, the formulas (3) and (4) hold true: σ=σ th (if x c ≧x cth ) (3) σ = a c x c +b c (if x c <x cth(4) The quality evaluation unit 110a calculates the force-related value x based on the estimation accuracy σ. s The weight of s Recalculate the weight of the force w snew is expressed by equation (5). snew =w s ×σ^2 / σ th ^2 (5)

[0056] The quality evaluation unit 110a evaluates the shape-related value x m The basic weight of w m、 and the value of the force x s The weight of s The result of recalculating snew The shape-related value x m The basic weight of w m The weight w obtained by normalizing m The force x' is expressed by the formula (6). s The weight of snew The weight w obtained by normalizing s ' is expressed by equation (7). m ´=αw m (6) w s ´=αw snew (7) In equations (6) and (7), α is expressed by equation (8): α = 1 / (w m +w snew ) (8)

[0057] The quality evaluation unit 110a evaluates information indicating the contours of one or more cells included in each of one or more cell images, shape information of one or more cells included in each of one or more cell images, and a value x m The weight of m and the value of the force x s The weight of s For example, the quality evaluation unit 110a evaluates the quality of one or more cells based on the shape information of the one or more cells and the force information of the one or more cells. m The weight of m and the value of the force x s The weight of sA predicted value y is derived based on the result of multiplication by the above, and the cells are comprehensively evaluated.

[0058] The predicted value y will be explained. i For the predicted value y i is expressed by equation (9). i =w m f m (x m ) + w s f s (x s ) (9) In equation (9), the predicted value y i For example, x is 1 for normal cells and 0 for abnormal cells. m The weight of m is the weight w for the normalized shape m The value x relating to the shape may be m represents a value relating to the shape, such as circularity, perimeter, polygon, etc. m is the value related to the shape x m The force value x is converted into a predicted value. s The weight of s is the weight of the normalized force w s The value x relating to the force may be s represents a value related to force such as pressure or tension. s is the value of force x s The method of deriving the predicted value y and the weight w will be described later with reference to FIG.

[0059] The accuracy information derivation unit 109a and the quality evaluation unit 110a are realized, for example, by a hardware processor such as a CPU executing a computer program (software) stored in the storage unit 112. Furthermore, some or all of these functional units may be realized by hardware (including circuitry) such as an LSI, ASIC, FPGA, or GPU, or may be realized by a combination of software and hardware. The computer program may be stored in advance in a storage device such as an HDD or flash memory, or may be stored on a removable storage medium such as a DVD or CD-ROM, and installed by inserting the storage medium into a drive.

[0060] (Creation of cell shape-force information acquisition accuracy map) The cell shape-force information acquisition accuracy map is created by the information processing device 300. A method for estimating the stress field of a cell will be described. As an example, the stress field acting on the cell is estimated by solving the force balance equation around the vertex. An example of a cell will be described using an epithelial cell. In this estimation method, it is assumed that epithelial cells are densely packed together, and each cell is a polygon formed by edges connecting the vertices with straight lines. The position vector of the vertex is r i = (x i , y i )

[0061] Figure 8A is a diagram for explaining a method for estimating the stress field of a cell. Figure 8A shows the forces to be estimated around a certain vertex. The pressure acting from the center of the surrounding cell toward the vertex and the tension acting along the edge are shown. When there are k vertices around a certain vertex O that are connected to the vertex O by edges, the equations for the forces in the x and y directions acting on the vertex O can be obtained as follows: In equation (10), Ti is the tension on edge i connecting vertex O and vertex i, and Pi is the pressure on edge (i mod k) and the cell that has edge i as its side. By considering equation (10) for all vertices of the cell system, a vector F = (f 0 ^x,f 0 ^y,f 1 ^x,f 1 Using ^y, ..., equation (11) is obtained.

[0062] F = AP (11) In equation (11), P is a matrix that combines tension and pressure. 1 , T 2 , ..., P 1 , P 2 , ...), and A is a matrix that summarizes the coefficients. In this case, we can assume that the speed of cell deformation is sufficiently slow and quasi-static. Therefore, when considering the forces around a vertex, we can assume that the tension from the edge connecting to that vertex and the pressure from the cell are balanced. In other words, equation (12) holds.

[0063] AP = 0 (12) The unknown tension and pressure can be determined by solving the simultaneous equations in equation (12), but because there are more unknowns than in the balance equation, it cannot be solved uniquely. Specifically, there is an uncertainty of R + f + 1. Here, R is the number of cells on the outermost periphery, and f is the number of junctions in the four directions. A Bayesian approach was adopted to resolve the uncertainty in this equation. Here, the prior function assumes that tension exists in a Gaussian distribution around a certain positive value. Hyperparameters are calculated by maximizing the marginal likelihood, and then P is estimated by maximizing the posterior distribution.

[0064] Because epithelial cells are densely packed together, the edges connecting the vertices can be approximated by straight lines. A vertex model is an effective tool for simulating the shape of cells with such characteristics. We use a two-dimensional vertex model to simulate the shape of cells and verify stress field estimation techniques. In the vertex model, the position r of vertex i is i is obtained by equation (13). In equation (13), η is the friction coefficient, and U is the potential energy expressed by equation (14).

[0065] In equation (14), the first term represents the area elasticity. K is the area modulus, and A i is the area of ​​cell i, and A eq is the total area. The second term represents the circumferential elasticity. Γ is the circumferential elastic modulus, and p i is the perimeter of cell i. This term represents the sum of the contractile force of the actomyosin ring of each cell and the adhesion between cells at the junctions between cells. The perimeter elastic modulus Γ may have a range depending on the analysis. In that case, the probability density of the Gaussian distribution is defined. Γ ~ N(μ Γ , σ Γ ), where μ Γ is the mean value and σ Γ is the standard deviation. The third term represents the line tension. Λ j is the line tension coefficient for edge j, and L j is the length of edge j. The value of this line tension changes depending on the contractile force of actin-myosin.

[0066] The line tension is expressed by equation (15). In equation (15), Λ j ^c is the base of the original actin-myosin contractile force, and the average μ Λ , standard deviation σ Λ is expressed as a Gaussian distribution (Λ j ^c~N(μ Λ , σ Λ ^2)). w j (t) is the fluctuation component that changes over time and is expressed by the following formula: j (t k+1 ) = -w j(t k ) dt / τf+w norm_dist w norm_dist ~N(0,(σ f dt / τf)^2), where σ r is the fluctuation strength, τ r is the correlation time of the fluctuations.

[0067] The procedure for the simulation will now be explained. In the simulation, the vertex position of the cell is calculated sequentially from the potential energy according to equations (13) and (14). As explained in equation (15), this simulation is roughly divided into two processes depending on the definition of line tension. Specifically, the simulation includes a pre-calculation process (0≦t<t) that promotes relaxation of the cell shape. f ) and the process that determines the shape of the cell (t f ≦t≦t total ) and

[0068] Figure 8B shows the flow of the 2D vertex model. f In t, the system is relaxed by adding a fluctuation term to the line tension, and f ≦t≦t total By eliminating the fluctuation term in the system, the system transforms to a steady state that minimizes the energy.

[0069] When cells undergo structural changes, 1 Figure 8C shows the T transformation implemented in the 2D vertex model. 1 FIG. 8C shows the cell C 1 ~Cell C 4 According to FIG. 8C, cell C 2 and cell C 4 When the length of the edge ED approaches a certain threshold, the cell C 2 and cell C 4 and connect to cell C 1 and cell C 3 Specifically, the vertices approach each other and the edge length becomes L T1 When the edge length becomes shorter than 1.5L, the conversion shown in FIG. 8C occurs. T1 This is set to T1 Immediately after conversion, the next T 1 In all processes, periodic boundary conditions were applied in the x and y directions.

[0070] FIG. 9 shows a group of parameters used in the simulation model. For each parameter, a brief explanation, a typical value, and a unit are shown. eq ) is set as the unit length, and KA eq ^2 is set as unit energy, and η / K is set as unit time. The parameters Γ and μ are used to represent heterogeneous cells. Γ , σ Γ , μ Λ , σ Λ are control parameters. Figure 10 shows the control parameters used in the simulation model. Figure 10 also shows the range of each control parameter. The values ​​of these control parameters are not changed simultaneously, but rather some of them are changed depending on the analysis conditions.

[0071] The stress field estimation method was verified using simulations. In the simulations, pressure and tension are calculated as shown in Equation (16). In formula (16), p l , p m is the perimeter of the cell including edge j. Since the estimated value is a relative value, scaling is performed when comparing it with the true value. Scaling is performed with a coefficient c so that the average tension becomes 1. That is, T est = cT true In this case, the scaling coefficient is obtained by equation (17).

[0072] The same scaling coefficient c is used for the pressure, and the offset is set so that the average is 0. That is, equation (18) is used. Once comparable scaled true and estimated values ​​are obtained, the RMSE is calculated as the estimated accuracy σ est The definition of RMSE is given by equation (19).

[0073] To investigate the parameter dependence of estimation accuracy in the two-dimensional parameter space (Γ, Λ), we performed a total of 20 x 17 = 340 simulations by varying Γ between 0.04 and 0.20 in increments of 0.01 (17 steps) and Λ between -1.4 and 0.5 in increments of 0.1 (20 steps). For each parameter set, we performed simulations using a 2D vertex model to obtain cell morphology. Based on these, we estimated the stress field, calculated the estimation accuracy, and created a map.

[0074] Figure 11 shows an example of simulation results. Figure 11 shows a heat map of the estimation accuracy of the stress field estimation results for each parameter set, with the vertical axis representing the circumferential elastic modulus Γ and the horizontal axis representing the line tension Λ. In some parameter sets, large fluctuations in the vertices caused distortions in the cell shape, leading to a breakdown in the simulation midway through the calculation. Such parameter sets are shown in gray. Figure 12 shows the results of a comparison between simulation and estimation. Figure 12 shows a scatter plot comparing tension and pressure at four representative points in the map shown in Figure 11. In Figure 12, the vertical axis represents the estimated value, and the horizontal axis represents the true value obtained by simulation. The figure shows the correlation between the tensions of all edges and the pressure of all cells in each simulation and its estimation results.

[0075] In Figure 12, A shows the results of a comparison between simulation and estimation for the parameter set (Λ, Γ) = (-0.8, 0.11) (Parameter 1 in Figure 11). With this parameter set, the estimated values ​​and the true values ​​obtained by simulation have a low correlation, resulting in poor estimation accuracy. In Figure 12, B shows the results of a comparison between simulation and estimation for the parameter set (Λ, Γ) = (-0.3, 0.11) (Parameter 2 in Figure 11). With this parameter set, the estimated values ​​and the true values ​​obtained by simulation have a high correlation, resulting in high estimation accuracy.

[0076] In Figure 12, C shows the comparison results between simulation and estimation for the parameter set (Λ, Γ) = (-0.3, 0.16) (Parameter 3 in Figure 11). With this parameter set, the estimation accuracy is high, but the range of true values ​​of tension and pressure is small. In Figure 12, D shows the comparison results between simulation and estimation for the parameter set (Λ, Γ) = (0.1, 0.04) (Parameter 4 in Figure 11). With this parameter set, the estimation accuracy is high, similar to Parameter 2.

[0077] FIG. 13 shows the parameter dependency of estimation accuracy. In FIG. 13, A shows the relationship between estimation accuracy (RMSE) and Λ for multiple Γ (0.05, 0.10, 0.15, 0.20). It can be seen that as Λ increases, the RMSE value decreases, i.e., the estimation accuracy increases. In FIG. 13, B shows the relationship between estimation accuracy (RMSE) and Γ for multiple Λ (-1.1, -0.7, -0.3, 0.1). It can be seen that as Γ increases, the RMSE value increases. In both FIG. 13, A and B show a range in which RMSE changes nonlinearly and a range in which it does not change nonlinearly, with a threshold value near RMSE 0.1-0.2. Here, an RMSE of 0.2 or less is defined as the range in which estimation is possible.

[0078] To understand the parameter dependency of estimation accuracy, we investigated the shape of cells in parameter space. Figure 14 shows the shape of cells in parameter space. According to Reference 2, parameter space can be divided into three regions, Region I, Region II, and Region III, based on the base shape of the cell. The boundaries of each region are obtained by calculating the minimum value of the potential energy in Equation (14).

[0079] The gradient of the boundary between Region I and Region II depends on the minimum number of polygons of the cell group near the boundary. In the case of the simulation, the minimum number of polygons is 3, so the boundary between Region I and Region II is expressed by equation (20). The boundary between Region II and Region III depends on the maximum number of polygons of the cell group near the boundary. In the simulation, the shape of the cells near this boundary is approximately a regular hexagon, so the boundary between Region II and Region III is expressed by equation (21).

[0080] In the simulation, σ Λ is basically set to 0.05, so ±σ Λ The regions are shaded. The characteristics of cell shape in each region are explained below. In Region I, cells can take many local energy minima, resulting in a state in which the shape is not uniquely stable. In this state, shear stress is low and the cell shape tends to become elongated. In simulations, when the parameter set in this region was used, the vertices moved significantly, causing the cells to twist, and most simulations failed midway through the calculation. Region I and the gray region in Figure 11 almost overlap.

[0081] In Region II, cells can reach a fixed energy minimum and their shape is stable. The estimation accuracy depends on the parameters, and as defined above, a boundary of RMSE = 0.2 is drawn to separate Region II. The equation for this boundary is Γ = -(1 / 6)Λ. The estimation accuracy is low to the left of the boundary (Γ < -(1 / 6)Λ).

[0082] Figure 15 shows an example of cell shape. The region to the left of the boundary (Γ<-(1 / 6)Λ) is close to Region I, so the cell shape is elongated as shown in Figure 15A. On the other hand, the right side of Region II (Γ>-(1 / 6)Λ) can be estimated with high accuracy. The cell shape on the right side of Region II (Γ>-(1 / 6)Λ) is rounded as shown in Figure 15B.

[0083] In Region III, all cells are almost identical in shape, being regular hexagons. Even in this region, estimation can be performed with high accuracy. However, because the cells have almost the same shape, all tensions and pressures are almost constant, as shown in Figure 15C. Furthermore, in this region, the simulation failed in areas where Γ was small. From the above, it was found that within the investigated range of the two-dimensional parameter space (Γ, Λ), highly accurate stress field estimation is possible in the region where Γ > -(1 / 6)Λ.

[0084] When considering applying stress field estimation to actual cells, observers cannot directly know the cell parameters (Γ, Λ), and therefore cannot determine whether stress field estimation is possible for the currently observed cell based on current information. Therefore, we investigated the parameter dependency of "cell shape features" observable under a microscope to correlate cell shape features with estimation accuracy. In the same two-dimensional parameter space (Γ, Λ) as in Figures 11 and 14, we examined circularity, perimeter, and number of polygons as cell shape features. Figure 16A shows the parameter dependency of the average value of each cell shape feature. Figure 16B shows the parameter dependency of the standard deviation of each cell shape feature.

[0085] The case where circularity is used as the cell shape will be described. As shown in A of FIG. 16A, the circularity is low in the region to the left of Region II (Γ<-(1 / 6)Λ) shown in FIG. 14, and it can be seen that the circularity increases as (Γ, Λ) increases. This is because the shear stress of the cells is low on the left side of Region I and Region II, making it easier for the cells to move. As shown in D of FIG. 16B, the circularity increases as (Γ, Λ) increases, and therefore, in principle, the standard deviation of the circularity also decreases. In particular, in the region corresponding to Region III shown in FIG. 14, all the cells have similar shapes, regular hexagons, so the standard deviation approaches almost 0.

[0086] A case where perimeter is used as the cell shape will be described. As shown in B of FIG. 16A, the parameter dependency of the average perimeter has a similar tendency to the parameter dependency of the average circularity. In principle, the perimeter becomes longer in regions with low circularity, so the perimeter becomes shorter as (Γ, Λ) increases. As shown in E of FIG. 16B, no parameter dependency is observed for the standard deviation of perimeter within Region II. In Region III, all cells have similar shapes, so the standard deviation of perimeter is almost zero.

[0087] The case where the number of polygons is used as the cell shape will be described. As shown in C of FIG. 16A, the average number of polygons does not exhibit parameter dependency within Region II. As discussed in relation to circularity and perimeter in Region III, all cells are regular hexagons, so the average number of polygons is approximately 6 and the standard deviation of the number of polygons is approximately 0. As shown in F of FIG. 16B, the standard deviation of the number of polygons in Region II tends to decrease as (Γ, Λ) increases.

[0088] As described above, parameter dependence was observed for some cell shape characteristics. Next, the correlation between estimation accuracy and cell shape characteristics was investigated by comparing Figures 16A and 16B with Figure 11. Figure 17A shows the correlation between estimation accuracy and the average value of each cell shape characteristic for all parameter sets. Figure 17B shows the correlation between estimation accuracy and the standard deviation of each cell shape characteristic for all parameter sets.

[0089] 17A and 17B, the mean circularity (A in FIG. 17A), mean perimeter (B in FIG. 17A), and standard deviation of the number of polygons (F in FIG. 17B) showed particularly strong correlations with estimation accuracy. Since circularity and perimeter are related, if we assume that they represent essentially the same thing, then the mean circularity and standard deviation of the number of polygons can be considered good indicators for determining whether or not a cell can be estimated. At the previously determined estimation feasibility line RMSE = 0.2, the circularity is approximately 0.85 and the standard deviation of the number of polygons is 0.12. In other words, this stress field estimation method can be determined to be applicable when the circularity is 0.85 or higher and the standard deviation of the number of polygons is 0.12 or lower.

[0090] (The basic weight of the shape value w m and the basic weight w of the force value s 18 is a diagram illustrating an example of a learning device 200 according to a modified example of the embodiment. The learning device 200 acquires a plurality of pieces of learning data. The learning device 200 derives a value x relating to a shape included in each of the acquired pieces of learning data. m and the force value x s and the predicted value y i and the shape value x m and the value of the force x s and the shape value x m Weight for w m and the value of the force x s Weight for w s Basic weight data (hereinafter also referred to as the "trained model") is created by machine learning the relationship between

[0091] FIG. 19 shows the shape-related value x m and the force value x s and the predicted value y i 1 is a diagram illustrating an example of the relationship between a value x m By converting y into a predicted value, m (=f m (x m )) is obtained, and the value for the force x s By converting y into a predicted value, s (=f s (x s The predicted value y i is a value related to the shape x as shown in equation (9). m Weight for w m and the value x related to the shape m The value y converted into a predicted value m (=f m (x m )) and the value of force x s Weight for w s and the force value x s The value y converted into a predicted value s (=f s (x s )) and the sum of them. Returning to FIG. 18, the explanation will be continued.

[0092] The learning device 200 creates a trained model notification including the created trained model. The learning device 200 transmits the created trained model notification to the cell quality evaluation device 100a. In the cell quality evaluation device 100a, the communication unit 114a receives the trained model notification transmitted by the learning device 200. The quality evaluation unit 110a acquires the trained model included in the trained model notification received by the communication unit 114a, and stores and sets the acquired trained model.

[0093] (Learning device 200) The learning device 200 is realized by a device such as a personal computer, a server, a smartphone, a tablet computer, or an industrial computer. The learning device 200 creates a trained model. The learning device 200 includes a communication unit 202, a processing unit 206, a creation unit 208, an input unit 210, and a storage unit 250.

[0094] The communication unit 202 is realized by a communication module. The communication unit 202 communicates with an external communication device via the network NW. The communication unit 202 communicates using a wireless communication method such as wireless LAN, Bluetooth (registered trademark), or LTE (registered trademark). The communication unit 202 holds communication information necessary for communicating with the cell quality evaluation device 100 via the network NW. The communication unit 202 transmits the trained model notification created by the creation unit 208 to the cell quality evaluation device 100a.

[0095] The input unit 210 inputs information. As an example, the input unit 210 may have an operation unit such as a keyboard and a mouse. In this case, the input unit 210 inputs information according to an operation performed by a user on the operation unit. As another example, the input unit 210 may input information from an external device. The external device may be, for example, a portable storage medium. A plurality of pieces of training data (parameter sets) are input to the input unit 210. An example of the training data is a value x related to a shape. m and the force value x s and the predicted value y i The input unit 210 stores the input learning data in the storage unit 250.

[0096] The processing unit 206 acquires a plurality of pieces of learning data stored in the storage unit 250. The processing unit 206 calculates a value x relating to a shape included in each of the acquired plurality of pieces of learning data. m and the force value x s and the predicted value y i The processing unit 206 obtains the obtained shape-related value x m and the force value x s and the predicted value y i and the shape value x m and the value of the force x s and the shape value x m Weight for w m and the value of the force x s Weight for w s A trained model is created by machine learning the relationship between

[0097] For example, the processing unit 206 may calculate a value x relating to the shape included in the training data. m and the force value x s and the predicted value y i For every pair of x and , the shape value x m and the force value x s The combination of these is input to the input layer of the learning model, and the output value output from the output layer is calculated. m and the force value x s The predicted value y corresponding to the combination i (training data) and calculate the error, and adjust the learning model parameters (shape-related values ​​x m Weight for w m and the value of the force x s Weight for w s ) (train the learning model) to create a trained model.

[0098] The creation unit 208 acquires the trained model from the processing unit 206. The creation unit 208 creates a trained model notification including the acquired trained model. The creation unit 208 outputs the created trained model notification to the communication unit 202. The communication unit 202 transmits the trained model notification output by the creation unit 208 to the cell quality evaluation device 100a. The storage unit 250 is realized by a HDD, flash memory, RAM, ROM, etc., and stores information.

[0099] The processing unit 206 and the creation unit 208 are realized, for example, by a hardware processor such as a CPU executing a computer program (software) stored in the storage unit 250. Furthermore, some or all of these functional units may be realized by hardware (including circuitry) such as an LSI, ASIC, FPGA, or GPU, or may be realized by a combination of software and hardware. The computer program may be stored in advance in a storage device such as a HDD or flash memory, or may be stored on a removable storage medium such as a DVD or CD-ROM and installed by inserting the storage medium into a drive.

[0100] 20 is a flowchart showing an example of the operation of the cell quality evaluation device according to the modified embodiment. (Step S1-2) The quality evaluation unit 110 acquires information indicating the cell lineage. Based on the acquired information indicating the cell lineage, the quality evaluation unit 110 calculates a value x relating to the shape from the trained model. m The basic weight of w m and the force value x s The basic weight of w s (Step S2-2) The receiving unit 102 receives one or more cell images.

[0101] (Step S3-2) The contour extraction unit 104 acquires one or more cell images accepted by the accepting unit 102. The contour extraction unit 104 extracts the contours of one or more cells included in each of the acquired one or more cell images. (Step S4-2) The shape information derivation unit 106 acquires information indicating the contours of one or more cells included in each of the one or more cell images from the contour extraction unit 104. The shape information derivation unit 106 derives shape information of one or more cells based on the information indicating the contours of one or more cells included in each of the acquired one or more cell images.

[0102] (Step S5-2) Force information derivation unit 108 acquires information indicating the contours of one or more cells included in each of the one or more cell images extracted by contour extraction unit 104, and derives force information of one or more cells based on the information indicating the contours of one or more cells included in each of the acquired one or more cell images. (Step S6-2) Accuracy information derivation unit 109a acquires shape information of one or more cells included in each of the one or more cell images from shape information derivation unit 106. Accuracy information derivation unit 109a derives the accuracy of the force information of one or more cells from the cell shape-force information acquisition accuracy map based on the shape information of one or more cells included in each of the acquired one or more cell images.

[0103] (Step S7-2) The quality evaluation unit 110a acquires information indicating the shape of one or more cells included in each of one or more cell images from the shape information derivation unit 106. The quality evaluation unit 110a acquires force information of one or more cells included in each of one or more cell images from the force information derivation unit 108. The quality evaluation unit 110a acquires the accuracy of the force information of the cells associated with each of the shape information of one or more cells from the accuracy information derivation unit 109a. The quality evaluation unit 110a derives the estimation accuracy of the shape information for each of the one or more cells based on the shape information of one or more cells included in each of the acquired one or more cell images. The quality evaluation unit 110a calculates a force-related value x based on the estimation accuracy σ. s The weight of s Recalculate.

[0104] The quality evaluation unit 110a evaluates the shape-related value x m The basic weight of w m and the value of the force x s The weight of s The result of recalculating snew The quality evaluation unit 110a normalizes information indicating the contours of one or more cells included in each of one or more cell images, shape information of one or more cells included in each of one or more cell images, and a value x m The weight of m and the value of the force x s The weight of s The quality of one or more cells is evaluated based on the above. Specifically, as described above in the description of the relationship between cell shape and estimation accuracy in Fig. 7 and in equations (3) to (9), the order of step S4-2 and step S5-2 may be interchanged in the flowchart shown in Fig. 20, and the order of step S4-2 and steps S5-2 to S6-2 may be interchanged.

[0105] (Operation of the Learning Device) Fig. 21 is a flowchart showing an example of the operation of the learning device according to the modified embodiment. (Step S1-3) Learning data of a plurality of cell groups of the same type is input to the input unit 210. The input unit 210 inputs shape information of the cells included in the learning data (shape-related value x m (Step S2-3) The input unit 210 acquires the force information (force value x s ) to obtain the

[0106] (Step S3-3) The input unit 210 receives the correct value (the predicted value y i (Step S4-3) The input unit 210 stores the acquired cell shape information, cell force information, and correct answer value in the storage unit 250. The processes of steps S1-3 to S4-3 are performed for a plurality of cell groups.

[0107] (Step S5-3) The processing unit 206 acquires a plurality of pieces of learning data stored in the storage unit 250. The processing unit 206 calculates the shape-related value x mand the force value x s and the predicted value y i The processing unit 206 obtains the obtained shape-related value x m and the force value x s and the predicted value y i and the shape value x m and the value of the force x s and the shape value x m Weight for w m and the value of the force x s Weight for w s (Step S6-3) The processing unit 206 performs machine learning on the relationship between the shape-related value x m and the value of the force x s and the shape value x m Weight for w m and the value of the force x s Weight for w s The relationship (weight data) is stored in the storage unit 250. In the flowchart shown in FIG. 21, the order of steps S1-3 and S2-3 may be reversed.

[0108] (Operation of Information Processing Device) FIG. 22 is a flowchart showing an example of the operation of an information processing device according to a modified embodiment. FIG. 23 is a diagram for explaining an example of the operation of an information processing device according to a modified embodiment. An example of the operation of an information processing device 300 according to a modified embodiment will be described with reference to FIGS. 22 and 23 . The information processing device 300 is realized by a device such as a personal computer, a server, a smartphone, a tablet computer, or an industrial computer. (Step S1-4) The information processing device 300 performs a simulation using a two-dimensional vertex model. For example, the information processing device 300 performs a simulation by sequentially calculating the vertex position of a cell from the potential energy according to Equations (13) and (14). (Step S2-4) The information processing device 300 calculates the shape of the cell according to Equations (13) and (14) (A).

[0109] (Step S3-4) The information processing device 300 calculates the characteristics of the cell shape based on the calculation results of the cell shape. (Step S4-4) The information processing device 300 calculates the force distribution of the cell according to equations (13) and (14) (B). (Step S5-4) The information processing device 300 executes force estimation.

[0110] (Step S6-4) The information processing device 300 determines the force distribution (estimated value) (C). (Step S7-4) The information processing device 300 calculates the estimation accuracy. As described with reference to Figures 12 to 15, the information processing device 300 analyzes steps S1-4 to S7-4 for all parameter sets. (Step S8-4) The information processing device 300 derives the parameter dependency of the estimation accuracy of Figure 11 (D).

[0111] (Step S9-4) The information processing device 300 derives the parameter dependency of the cell shape shown in Figures 16A to 16B based on the cell shape feature quantities obtained in step S3-4 (E). (Step S10-4) The information processing device 300 creates the cell shape-force information acquisition accuracy map shown in Figures 17A to 17B (F). In the flowchart shown in Figure 22, the order of steps S2-4 to S3-4 and step 4-4 may be interchanged, and the order of step S8-4 and step 9-4 may be interchanged.

[0112] In the modified example of the embodiment described above, as an example, the quality evaluation unit 110a in the cell quality evaluation device 100a derives the estimation accuracy of the circularity of a cell as an example of the shape of the cell, but the present invention is not limited to this example. For example, the quality evaluation unit 110a may derive the estimation accuracy of the perimeter of a cell, the estimation accuracy of the number of polygons of a cell, or the estimation accuracy of the shape of a cell as described in the embodiment.

[0113] In the above-described modified example of the embodiment, the quality evaluation unit 110a in the cell quality evaluation device 100a acquires, as the cell lineage information, information indicating the number of cell types (homologous / heterologous), but the invention is not limited to this example. For example, the quality evaluation unit 110a may acquire, as the cell lineage information, information indicating the number of cells, information indicating the culture platform (flat culture / curved culture), information indicating the culture medium, information indicating the type of dish or the coating on the base, information indicating the addition of a drug, or information indicating the number of days of culture.

[0114] In a modification of the above-described embodiment, for example, a value x relating to the shape is calculated by machine learning. m The basic weight of w m and the force value x s The basic weight of w s However, the present invention is not limited to this example. For example, the shape-related value x m The basic weight of w m and the force value x s The basic weight of w s may be derived.

[0115] According to a cell quality evaluation device 100a according to a modified embodiment, the cell quality evaluation device 100a further includes an accuracy information derivation unit 109a that derives the accuracy of the cellular force information derived by the force information derivation unit 108. The quality evaluation unit 110a evaluates the quality of the cell further based on the accuracy of the cellular force information derived by the accuracy information derivation unit 109a. With this configuration, the cell quality evaluation device 100a can derive the accuracy of the cellular force information and evaluate the quality of the cell further based on the accuracy of the derived cellular force information, thereby improving the accuracy of the evaluation result of the cell quality compared to when the quality of the cell is evaluated not based on the accuracy of the cellular force information.

[0116] In the cell quality evaluation device 100a, the quality evaluation unit 110a weights the cell shape information and cell force information based on the accuracy of the cell force information. By configuring in this manner, the cell quality evaluation device 100a can weight the cell shape information and cell force information based on the accuracy of the cell force information, thereby improving the accuracy of the cell quality evaluation results compared to when the cell shape information and cell force information are not weighted.

[0117] In the cell quality evaluation device 100a, the weights of the cell shape information and cell force information are obtained by machine learning the relationship between the cell shape information and cell force information and the weights of the cell shape information and cell force information based on multiple combinations of the cell shape information and cell force information and the predicted values ​​of cell quality. With this configuration, the relationship between the cell shape information and cell force information and the weights of the cell shape information and cell force information can be obtained by machine learning, so the cell quality evaluation device 100a can weight the cell shape information and cell force information.

[0118] In cell quality evaluation device 100a, accuracy information derivation unit 109a derives the accuracy of the cellular force information derived by force information derivation unit 108 based on the cellular shape information derived by shape information derivation unit 106. With this configuration, cell quality evaluation device 100a can derive the accuracy of the cellular force information based on the cellular shape information, thereby improving the accuracy of the cell quality evaluation results compared to when cell quality is evaluated not based on the accuracy of the cellular force information.

[0119] In cell quality evaluation device 100a, accuracy information derivation unit 109a derives the accuracy of the cellular force information derived by the force information derivation unit based on the cell shape information derived by shape information derivation unit 106 and information correlating the accuracy of the cell shape information and the cellular force information. By being configured in this manner, cell quality evaluation device 100a can derive the accuracy of the cellular force information from the cell shape information and information correlating the accuracy of the cell shape information and the cellular force information based on the cell shape information, thereby improving the accuracy of the cell quality evaluation results compared to when cell quality is evaluated not based on the accuracy of the cellular force information.

[0120] Although the embodiments of the present invention have been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and the present invention also includes designs within the scope of the gist of the present invention.

[0121] Reference 1: Shuji Ishihara, Kaoru Sugimura, “Bayesian inference of force dynamics during morphogenesis”, Journal of Theoretical Biology, Volume 313, 21 November 2012, Pages 201-211 Reference 2: Reza Farhadifar, Jens-Christian Roper, Benoit Aigouy, Suzanne Eaton, and Frank Julicher, “The Influence of Cell Mechanics, “Cell-Cell Interactions, and Proliferation on Epithelial Packing”, Current Biology 17, 2095-2104, December 18, 2007

[0122] 11, 11a incubator, 12...upper casing, 13...lower casing, 15...temperature-controlled room, 15a...temperature control device, 15b...humidity control device, 16...large door, 17...middle door, 18...small door, 19...culture vessel, 21...storage container, 22...observation unit, 23...vessel transport device, 24...transport table, 31...sample table, 32...stand arm, 33...main body, 34...vertical robot, 35...LED light source, 36...mini stage , 37...Arm unit, 38...Imaging device, 39...Rotation stage, 100, 100a...Cell quality evaluation device, 102...Reception unit, 104...Contour extraction unit, 106...Shape information derivation unit, 108...Force information derivation unit, 109a...Accuracy information derivation unit, 110, 110a...Quality evaluation unit, 112...Memory unit, 200...Learning device, 202...Communication unit, 206...Processing unit, 208...Creation unit, 210...Input unit, 250...Memory unit, 300...Information processing device

Claims

1. A cell quality evaluation device comprising: a receiving unit that receives a cell image; a contour extraction unit that extracts a contour of a cell contained in the cell image received by the receiving unit; a force information derivation unit that derives force information of the cell based on the contour of the cell extracted by the contour extraction unit; a shape information derivation unit that derives shape information of the cell based on the contour of the cell extracted by the contour extraction unit; and a quality evaluation unit that evaluates quality of the cell based on the shape information of the cell derived by the shape information derivation unit and the force information of the cell derived by the force information derivation unit.

2. The cell quality evaluation device of claim 1, further comprising an accuracy information derivation unit that derives the accuracy of the force information of the cell derived by the force information derivation unit, wherein the quality evaluation unit evaluates the quality of the cell further based on the accuracy of the force information of the cell derived by the accuracy information derivation unit.

3. The cell quality evaluation device according to claim 2, wherein the quality evaluation unit weights the shape information of the cell and the force information of the cell based on the accuracy of the force information of the cell.

4. The cell quality evaluation device described in claim 3, wherein the weights of the shape information of the cell and the force information of the cell are obtained by machine learning the relationship between the cell shape information and cell force information and the weights of the cell shape information and cell force information based on multiple combinations of the cell shape information and cell force information and predicted values ​​of cell quality.

5. A cell quality evaluation device as described in claim 2, wherein the accuracy information derivation unit derives the accuracy of the force information of the cell derived by the force information derivation unit based on the shape information of the cell derived by the shape information derivation unit.

6. The cell quality evaluation device described in claim 2, wherein the accuracy information derivation unit derives the accuracy of the force information of the cell derived by the force information derivation unit based on the shape information of the cell derived by the shape information derivation unit and information correlating the shape information of the cell and the accuracy of the force information of the cell.

7. The cell quality evaluation device according to claim 1, wherein the force information deriving unit derives the tension and pressure of the cell based on the contour of the cell.

8. A cell quality evaluation method comprising: a reception process for receiving a cell image; a contour extraction process for extracting a contour of a cell contained in the cell image received in the reception process; a force information derivation process for deriving force information of the cell based on the contour of the cell extracted in the contour extraction process; a shape information derivation process for deriving shape information of the cell based on the contour of the cell extracted in the contour extraction process; and a quality evaluation process for evaluating the quality of the cell based on the shape information of the cell derived in the shape information derivation process and the force information of the cell derived in the force information derivation process.

9. A cell quality evaluation method as described in claim 8, wherein the weights of the shape information of the cell and the force information of the cell are obtained by machine learning the relationship between the cell shape information and cell force information and the weights of the cell shape information and cell force information based on multiple combinations of the cell shape information and cell force information and predicted values ​​of cell quality.

10. A program for causing a computer for evaluating the quality of a cell to execute the following steps: a reception process for receiving a cell image; a contour extraction process for extracting a contour of a cell contained in the cell image received in the reception process; a force information derivation process for deriving force information of the cell based on the contour of the cell extracted in the contour extraction process; a shape information derivation process for deriving shape information of the cell based on the contour of the cell extracted in the contour extraction process; and a quality evaluation process for evaluating the quality of the cell based on the shape information of the cell derived in the shape information derivation process and the force information of the cell derived in the force information derivation process.

11. The program of claim 10, wherein the weights of the shape information of the cell and the force information of the cell are obtained by machine learning the relationship between the cell shape information and the force information of the cell and the weights of the shape information and the force information of the cell based on multiple combinations of the cell shape information and the force information of the cell and the predicted values ​​of cell quality.

Citation Information

Patent Citations

  • Evaluation technique for forecasting cell migration direction, and image processing program and image processing apparatus of cell observation image

    JP2010022319A

  • Cultivation information processing apparatus, cultivation status evaluating device, cell cultivation method, and program

    JP2013027368A

  • Force measuring method, and force measuring apparatus

    JP2017072552A

  • Region detecting method and region detecting device related to cell aggregation

    WO2017090741A1

Cited By

  • AI-based auxiliary device for pathological detection

    CN120259293A