Content rate estimation device, content rate estimation method, and content rate estimation system

Abstract

A content rate estimation device 2 includes: a storage unit 17 for storing a trained model 220 which has been trained to output a feature amount that indicates whether or not a cell shown in a cell image is a specific cell when the cell image is input; an acquisition unit 22 for acquiring a plurality of cell images; an estimation unit 25 for estimating the content rate of the specific cell in cells in the plurality of cell images by using a distribution of feature amounts obtained by inputting the plurality of cell images into the trained model 220; and an output unit 26 for outputting a signal that indicates the content rate estimated by the estimation unit 25.

Description

Content rate estimation device, content rate estimation method, and content rate estimation system

[0001] The present disclosure relates to a content rate estimation device, a content rate estimation method, and a content rate estimation system.

[0002] Various techniques for improving the measurement method of gene transfer rate, which is an important indicator for genetically engineered cell medicines such as chimeric antigen receptor (CAR)-T cells, are known. For example, Patent Document 1 describes a method including a step of measuring cellular complexity in gene-transfected animal cells and a step of predicting the gene transfer rate based on the value measured in the complexity measurement step. The method described in Patent Document 1 measures cellular complexity and uses the measured complexity as an index, thereby enabling simple and efficient prediction of the gene transfer rate of animal cells. Note that the "gene transfer rate" refers to the rate of transfer of a foreign gene into cells, and is, for example, the proportion of cells into which the foreign gene has been introduced among all cells, the amount of foreign gene uptake in a cell population, or the expression rate of the foreign gene in the entire cell population. When the foreign gene is a CAR, it is sometimes referred to as the CAR-positive rate.

[0003] International Publication No. 2023 / 190974

[0004] FIG. 1 is a conceptual diagram showing the configuration of an inclusion rate estimation system including an inclusion rate estimation device according to an embodiment. FIG. 2 is a block diagram of a server shown in FIG. 1. FIG. 3 is a diagram showing the trained model shown in FIG. 2. FIG. 4 is a block diagram of the inclusion rate estimation device shown in FIG. 1. FIG. 5 is a flowchart of a learning image capture process executed by the inclusion rate estimation device shown in FIG. 4. FIG. 6 is a flowchart of a learning process executed by the inclusion rate estimation device shown in FIG. 4. FIG. 7A is a diagram (part 1) for explaining the process indicated in S204. FIG. 7B is a diagram (part 2) for explaining the process indicated in S204. FIG. 7C is a diagram (part 3) for explaining the process indicated in S204. FIG. 8A is a diagram showing the accuracy rate and loss in training data. FIG. 8B is a diagram showing the accuracy rate and loss in accuracy verification data. FIG. 9 is a flowchart of an inclusion rate estimation image capture process executed by the inclusion rate estimation device shown in FIG. 4. FIG. 10 is a flowchart of an inclusion rate estimation process executed by the inclusion rate estimation device shown in FIG. 4. FIG. 11 is a flowchart showing more detailed processing of the process indicated by S405 in FIG. 10 . FIGS. 12A to 12D are histograms showing the CAR gene expression rate of donor A, with FIG. 12A showing the distribution of feature amounts after CAR gene introduction. FIG. 12B showing the distribution of actual measured values ​​of feature amounts of CAR-T cells and T cells after CAR gene introduction. FIG. 12C showing the distribution of feature amounts of T cells before CAR gene introduction. FIG. 12D showing the distribution of actual measured values ​​and estimated values ​​of feature amounts of CAR-T cells and T cells after CAR gene introduction. FIGS. 13A to 13D are histograms showing the CAR gene expression rate of donor B, with FIG. 13A showing the distribution of feature amounts after CAR gene introduction. FIG. 13B showing the distribution of actual measured values ​​of feature amounts of CAR-T cells and T cells after CAR gene introduction. FIG. 13C showing the distribution of feature amounts of T cells before CAR gene introduction. Fig. 13D shows the distribution of measured and estimated values ​​of feature quantities of CAR-T cells and T cells after CAR gene transfer. Fig. 14A to Fig. 14D are histograms showing the CAR gene expression rate of donor C, and Fig. 14A shows the distribution of feature quantities after CAR gene transfer.FIG. 14B shows the distribution of actual measured values ​​of feature quantities of CAR-T cells and T cells after CAR gene introduction. FIG. 14C shows the distribution of feature quantities of T cells before CAR gene introduction. FIG. 14D shows the distribution of actual measured values ​​and estimated value of feature quantities of CAR-T cells and T cells after CAR gene introduction. FIGS. 15A to 15D are histograms showing the CAR gene expression rate of donor D, and FIG. 15A shows the distribution of feature quantities after CAR gene introduction. FIG. 15B shows the distribution of actual measured values ​​of feature quantities of CAR-T cells and T cells after CAR gene introduction. FIG. 15C shows the distribution of feature quantities of T cells before CAR gene introduction. FIG. 15D shows the distribution of actual measured values ​​and estimated value of feature quantities of CAR-T cells and T cells after CAR gene introduction. 16A to 16D are diagrams for explaining the content estimation process according to the modified example, in which FIG. 16A shows the total distribution of cellular feature amounts of donor A. FIG. 16B shows the total distribution of cellular feature amounts of donor B. FIG. 16C shows the total distribution of cellular feature amounts of donor C. FIG. 16D shows the total distribution of cellular feature amounts of donor D. FIG. 17 is a flowchart showing more detailed processing of S405 in the content estimation process according to the modified example, which is executed by the content estimation device shown in FIG. 4. FIG. 18 is a conceptual diagram showing the configuration of a content estimation system according to the modified example. FIG. 19 is a block diagram of the content estimation device shown in FIG. 18.

[0005] The content rate estimation device according to the present disclosure will be described below with reference to the drawings. However, please note that the technical scope of the present disclosure is not limited to the embodiments, but extends to the inventions set forth in the claims and their equivalents.

[0006] (Configuration and Function of Content Estimation System Including Content Estimation Device According to Embodiment) Fig. 1 is a conceptual diagram showing the configuration of a content estimation system including a content estimation device according to an embodiment. The content estimation system 100 has a microscope main body 101, a server 200, and a content estimation device 1. In Fig. 1, the optical axis of the optical system of the microscope main body 101 is indicated by a dashed dotted line L1, and illumination light from a light source 111 is indicated by a dashed two-dotted line L2.

[0007] The microscope main body 101 includes a transmitted illumination optical system 110 that irradiates illumination light onto the object S, a stage 120, an imaging optical system 130, and a detection unit 140. The microscope main body 101 captures a bright-field image of the object S placed on the stage 120 and outputs bright-field image data representing the captured bright-field image to the content estimation device 1. The content estimation device 1 may be a separate device, such as a computer, from the microscope main body 101. The bright-field image is an image captured by bright-field observation. Bright-field observation allows cells to be observed and photographed without losing cellular information, such as their internal structure. Phase-contrast observation is commonly used to observe adherent cells, but phase-contrast observation creates a bright, edging-like halo around the cell's outline, resulting in loss of information around the cell's outline. Furthermore, cells suspended in a culture medium are spherical and thick, and phase-contrast observation results in significant phase change around the cell's outline. As a result, a stronger halo is generated than in phase contrast observation of adherent cells, and cell information around the cell outline is lost. By photographing using bright-field observation, cells can be photographed without losing information around the cell outline. Note that floating cells include suspended cells, cells that are not directly attached to the bottom of the imaging container but are indirectly attached to the bottom of the imaging container via an anchor or the like and are suspended in the medium, and adherent cells that have detached from the adhesive surface and exist in a floating state in the medium.

[0008] The transmitted illumination optical system 110 includes a light source 111, a first lens 112, a bandpass filter 113, a field stop 114, a second lens 115, an aperture stop 116, and a condenser lens 117. The light source 111 includes a non-coherent light source device such as a halogen lamp, and emits illumination light L2 to illuminate the object S. The illumination light L2 emitted from the light source 111 is incident on the first lens 112. The illumination light L2 incident on the first lens 112 is refracted by the first lens 112 to become approximately parallel light, which exits the first lens 112 and is incident on the bandpass filter 113. Of the illumination light L2 incident on the bandpass filter 113, only light of wavelength components within a desired wavelength range is transmitted by the bandpass filter 113 and is incident on the field stop 114. The bandpass filter 113 can be retracted to a position outside the optical path.

[0009] The illumination light L2 incident on the field stop 114 has its beam diameter adjusted, exits the field stop 114, and is then incident on the second lens 115. The illumination light L2 incident on the second lens 115 is converged by the second lens 115, exits the second lens 115, and is then incident on the aperture stop 116. The illumination light L2 incident on the aperture stop 116 is converted so that its wavefront becomes spherical, exits the aperture stop 116, and is then incident on the condenser lens 117. The illumination light L2 incident on the condenser lens 117 is refracted by the condenser lens 117, and becomes light having a wavefront that is approximately perpendicular to the optical axis when irradiated onto the object S, and is then irradiated onto the object S placed on the stage 120.

[0010] The stage 120 is movable in a direction along an axis parallel to the optical axis of the objective lens 151, which is parallel to the optical axis of the optical system of the microscope main body 101, and in directions along two axes perpendicular to the optical axis of the objective lens 151. When photographing the object S, the stage 120 is electrically driven by a moving device such as a motor and moves in a direction along an axis parallel to the optical axis of the objective lens 151 and in directions along two axes perpendicular to the optical axis of the objective lens 151.

[0011] The subject S placed on the stage 120 includes a plurality of CAR-T cells and a plurality of T cells. The CAR-T cells are also referred to as first cells, and the T cells are also referred to as second cells. The CAR-T cells are an example of exogenous gene-expressing cells into which an exogenous gene has been introduced and a CAR gene has been expressed, while the T cells are an example of exogenous gene-non-expressing cells into which no exogenous gene has been introduced or into which an exogenous gene has been attempted but the CAR gene has not been expressed. The subject S is formed, for example, through a T cell collection step of collecting T cells from a donor, also referred to as a subject; a CAR gene introduction step of introducing a CAR gene into the T cells collected in the T cell collection step; and a cell culture step of culturing the T cells into which the CAR gene has been introduced in the CAR gene introduction step. The subject S is placed on the stage 120, for example, as a 20 μL cell suspension, contained in an imaging container.

[0012] The imaging optical system 130 has an objective optical system 150 and a relay optical system 160. The objective optical system 150 includes a plurality of objective lenses 151 with different numerical apertures NA, etc. The relay optical system 160 has an imaging lens 161, a beam splitter 162, mirrors 163a, 163b, and 163c, lenses 164a, 164b, and 164c, and an eyepiece lens 165.

[0013] The imaging lens 161 refracts light incident from the objective optical system 150 so as to form an image on the detection unit 140, and outputs the light to the beam splitter 162. The beam splitter 162 reflects a portion of the light incident from the objective optical system 150 to the detection unit 140, and transmits the remainder, outputting it to the mirror 163a. ​​The light reflected by the mirror 163a is reflected or refracted by the lenses in the order of lens 164a, mirror 163b, lens 164b, lens 164c, and mirror 163c, before entering the eyepiece 165. The light incident on the eyepiece 165 is refracted by the eyepiece 165 and enters the user's eye E to be perceived.

[0014] The detection unit 140 includes a detector such as an imaging element as a CCD or a CMOS, and captures an image of the object S. The detection unit 140 detects light reflected by the beam splitter 162 of the relay optical system 160. A detection signal corresponding to the detected light is A / D converted by an A / D converter (not shown) or the like, and output to the content rate estimation device 1.

[0015] FIG. 2 is a block diagram of the server 200 .

[0016] The server 200 has a server communication unit 201, a server storage unit 202, a server input unit 203, a server display unit 204, and a server processing unit 205, and stores a trained model 220 in the server storage unit 202. The server 200 executes a learning process for training the trained model 220 based on instructions from the content rate estimation device 1.

[0017] The server communication unit 201 has a communication interface circuit for connecting the server 200 to the content rate estimation device 1 and an external device (not shown) via a network (not shown). The server communication unit 201 supplies data received from the content rate estimation device 1 and the external device (not shown) via the network to the server processing unit 205. The server communication unit 201 also transmits the data supplied from the server processing unit 205 to the content rate estimation device 1 and the external device via the network.

[0018] The server storage unit 202 includes, for example, one of a semiconductor memory, a magnetic disk device, and an optical disk device. The server storage unit 202 stores an operating system program, a driver program, an application program, data, etc., used for processing by the server processing unit 205. For example, the server storage unit 202 stores, as driver programs, an input device driver program that controls the server input unit 203, an output device driver program that controls the server display unit 204, etc.

[0019] The server storage unit 202 also stores a trained model 220. When a cell image including a plurality of CAR-T cells and a plurality of T cells is input, the trained model 220 is trained to output a feature amount indicating whether the cell image is the CAR-T cell. A cell image is an image that represents a cell. A cell image that includes a certain cell can also be said to be a cell image that represents that cell.

[0020] FIG. 3 is a diagram showing the trained model 220.

[0021] The trained model 220 is a convolutional neural network and includes a feature representation unit 221 and a classifier 222. The feature representation unit 221 is formed by multiple convolutional layers and reduces the spatial extent of the input image while increasing the number of channels to activate specific features of the image. For example, the feature representation unit 221 reduces an image with a height of 224 pixels and a width of 224 pixels to a height and width of 7 pixels and converts it into two-dimensional data with 1,280 channels.

[0022] The classifier 222 includes a two-dimensional convolution layer 223, a pooling layer 224, a first activation function layer 225, and a second activation function layer 226. The two-dimensional convolution layer 223 slides the two-dimensional data input from the feature representation unit 221 and multiplies each element to convert the two-dimensional matrix of feature quantities into different map data. For example, the two-dimensional convolution layer 223 converts two-dimensional data having a height and width of 7 pixels and a number of channels of 1,280 into map data having a height and width of 7 pixels and a number of channels of 2.

[0023] The pooling layer 224 performs a pooling operation, such as an average pooling operation, to compress the amount of information in the map data converted by the two-dimensional convolution layer 223. For example, the pooling layer 224 converts map data having a height and width of 7 and two channels into map data having a height and width of 7 and two channels.

[0024] The first activation function layer 225 is, for example, a scaled exponential linear unit (SELU), and activates the feature values ​​of the channels of the map data converted by the pooling layer 224 to suppress gradient vanishing. The feature value of one channel activated by the first activation function layer 225 is output as the feature value of CAR-T cells, and the feature value of the other channel is output as the feature value of T cells. The histograms of the feature values ​​of CAR-T cells and T cells output from the first activation function layer 225 are used to determine the CAR-T content as the feature value of cells corresponding to the cell image in the content estimation process by the content estimation device. In the content estimation process by the content estimation device, the feature value of cells corresponding to the cell image is a numerical value that can be converted into a probability distribution indicating whether the cell is a CAR-T cell or a T cell. The first activation function layer 225 is a feature generation layer that generates the feature values ​​of CAR-T cells and T cells from the output of the convolution layer.

[0025] The second activation function layer 226 is a softmax function, and normalizes the feature quantities of the channels activated by the first activation function layer 225 so that the sum of the feature quantities of the channels becomes 1.0. The feature quantity of one channel normalized by the second activation function layer 226 is output as a probability distribution of CAR-T cells, and the feature quantity of the other channel is output as a probability distribution of T cells.

[0026] The server input unit 203 may be any device capable of operating the server 200, such as a keyboard or a touchpad. An operator can input characters, numbers, and the like via the server input unit 203. When operated by the operator, the server input unit 203 generates a signal corresponding to the operation. The generated signal is then supplied to the server processing unit 205 as an instruction from the operator.

[0027] The server display unit 204 may be any device capable of displaying video, images, text, etc., such as a liquid crystal display, an organic EL (Electro-Luminescence) display, etc. The server display unit 204 displays video corresponding to video data, images corresponding to image data, text corresponding to text data, etc., supplied from the server processing unit 205. The server display unit 204 may also display a graphical user interface (GUI) for operating the server 200.

[0028] The server processing unit 205 includes one or more processors and their peripheral circuits. The server processing unit 205 controls the overall operation of the server 200 and is, for example, a central processing unit (CPU). The server processing unit 205 controls the operation of the server communication unit 201, the server display unit 204, etc. so that various processes of the server 200 are executed in an appropriate order in accordance with the programs stored in the server storage unit 202 and the operations of the server input unit 203. The server processing unit 205 executes processes based on the programs (such as operating system programs, driver programs, and application programs) stored in the server storage unit 202. The server processing unit 205 can also execute multiple programs (such as application programs) in parallel.

[0029] FIG. 4 is a block diagram of the content rate estimation device 1.

[0030] The content estimation device 1, also referred to as a terminal device, has a communication unit 11, a memory unit 12, an input unit 13, a display unit 14, and a processing unit 20. Based on a program pre-stored in the memory unit 12, the content estimation device 1 executes various processes by referencing data stored in the memory unit 12. The content estimation device 1 also executes various processes in response to instructions input by an operator via the input unit 13 and outputs the execution results to the display unit 14. The content estimation device 1 is a computer. The content estimation device 1 estimates the content of cell images relating to specific cells (e.g., CAR-T cells) in a plurality of cell images using a feature distribution based on a plurality of feature amounts obtained by inputting cell images of a plurality of cells contained in an object S placed on a stage 120 into a trained model. The content estimation device 1 calculates the gene expression rate of the CAR gene using the estimation result of the content of cell images relating to CAR-T cells in the plurality of cell images.

[0031] The communication unit 11 has a communication interface circuit for connecting the content estimation device 1 to a network (not shown). The communication unit 11 supplies data received from the microscope main body 101 and an external device (not shown) via the network to the processing unit 20. The communication unit 11 also transmits data supplied from the processing unit 20 to the external device via the network. The communication unit 11 receives a detection signal A1 indicating light detected by the detection unit 140, and transmits a control signal A2 to the microscope main body 101 for controlling each of the components of the microscope main body 101, such as the stage 120 and the detection unit 140.

[0032] The storage unit 12 includes, for example, one of a semiconductor memory, a magnetic disk device, and an optical disk device. The storage unit 12 stores an operating system program, a driver program, an application program, data, and the like used for processing by the processing unit 20. For example, the storage unit 12 stores, as driver programs, an input device driver program that controls the input unit 13, an output device driver program that controls the display unit 14, and the like. The storage unit 12 also stores, as an application program, a learning image capture program that causes the processing unit 20 to execute a learning image capture process for capturing images of cells contained in the object S placed on the stage 120, for use when training the trained model 220. The storage unit 12 also stores, as an application program, a learning program that causes the processing unit 20 to execute a learning process for training the trained model 220. The storage unit 12 also stores, as an application program, an image capturing program for content rate estimation that causes the processing unit 20 to execute an image capturing process for content rate estimation that captures images of cells contained in the object S placed on the stage 120, for use in estimating the content rate of CAR-T cells. The storage unit 12 also stores, as an application program, a content rate estimation program that causes the processing unit 20 to execute a content rate estimation process that estimates the content rate of CAR-T cells. The learning image capturing program, the learning program, the image capturing program for content rate estimation, and the content rate estimation program may be installed into the storage unit 12 from a computer-readable portable storage medium, such as a CD-ROM or a DVD-ROM, using a known setup program or the like.

[0033] The storage unit 12 also stores various data used in the learning image capturing process, the learning process, the inclusion rate estimation image capturing process, and the inclusion rate estimation process. For example, the storage unit 12 stores UTD histogram information, virtual CAR-T cell first distribution information, and virtual CAR-T cell second distribution information used in the inclusion rate estimation process. The UTD is T cells before CAR gene introduction, and the UTD histogram information is a histogram showing the distribution of T cell features in cells collected from a donor before CAR gene introduction. The UTD histogram information is generated from cells collected from a donor before CAR gene introduction into the donor, and is generated for each donor from which cells are collected. The virtual CAR-T cell first distribution information and the virtual CAR-T cell second distribution information are information that approximates the distribution of CAR-T cell features using a Gaussian distribution. Each of the first virtual CAR-T cell distribution information and the second virtual CAR-T cell distribution information includes a mean value and a variance that define a Gaussian distribution. The first peak and the second peak of the distribution of the feature amounts of CAR-T cells have substantially the same median and half-width regardless of the donor, and therefore can be defined by a Gaussian distribution.

[0034] The input unit 13 may be any device, such as a keyboard or a touchpad, that can operate the content estimation device 1. An operator can input letters, numbers, and the like via the input unit 13. When operated by the operator, the input unit 13 generates a signal corresponding to the operation. The generated signal is then supplied to the processing unit 20 as an instruction from the operator.

[0035] The display unit 14 may be any device capable of displaying video, images, text, etc., such as a liquid crystal display, an organic EL display, etc. The display unit 14 displays video corresponding to video data, images corresponding to image data, text corresponding to text data, etc., supplied from the processing unit 20. The display unit 14 may also display a graphical user interface for operating the content rate estimation device 1.

[0036] The processing unit 20 has one or more processors and their peripheral circuits. The processing unit 20 is, for example, a CPU, and comprehensively controls the overall operation of the content rate estimation device 1. The processing unit 20 controls the operation of the communication unit 11, the display unit 14, etc. so that various processes of the content rate estimation device 1 are executed in an appropriate order in accordance with the programs stored in the storage unit 12, the operation of the input unit 13, etc. The processing unit 20 executes processes based on the programs stored in the storage unit 12. The processing unit 20 can also execute multiple programs in parallel.

[0037] The processing unit 20 includes an instruction unit 21, an acquisition unit 22, an image processing unit 23, a learning unit 24, an estimation unit 25, and an output unit 26. Each of these units is a functional module implemented by a program executed by a processor included in the processing unit 20. Alternatively, each of these units may be implemented in the content rate estimation device 1 as firmware.

[0038] (Learning Image Capturing Process by Content Estimation Device According to Embodiment) Fig. 5 is a flowchart of the learning image capturing process executed by the content estimation device 1. The learning image capturing process shown in Fig. 5 is executed mainly by the processing unit 20 in cooperation with each element of the content estimation system 100, based on an imaging program stored in advance in the storage unit 12. Note that the object S used in the learning image capturing process contains a fluorescent label that is used when visualizing CAR-T cells in which CAR molecules are expressed on the cell surface.

[0039] First, the instruction unit 21 outputs to the stage 120 an imaging area movement instruction indicating that the object S should be moved to an imaging position where a predetermined imaging area of ​​the object S can be imaged (S101). In response to the input of the imaging area movement instruction, the stage 120 moves along two axes perpendicular to the optical axis of the objective lens 151 of the objective optical system 150, and moves the object S to a position where a predetermined imaging area of ​​the object S can be imaged. Note that the instruction unit 21 may move the imaging optical system 130 without moving the stage 120, or may move both the stage 120 and the imaging optical system 130.

[0040] Next, the instruction unit 21 outputs a first imaging position movement instruction to the stage 120, instructing the stage 120 to move the object S to a first imaging position where a bright-field image will be captured (S102). The first imaging position is a position parallel to the optical axis of the objective lens 151 and shifted by a predetermined offset amount from the focal position of the objective lens 151 in a direction away from the objective lens 151. In response to the input of the first imaging position movement instruction, the stage 120 moves along an axis parallel to the optical axis of the objective lens 151 to move the object S to the first imaging position. For example, in the case of a spherical cell, the focal position is an intermediate position between the position closest to the objective lens 151 and the position farthest from the objective lens 151 on the contour of the cell, and is the center position of the cell thickness relative to the optical axis direction of the objective lens 151. The offset amount is determined before the imaging process is performed, based on the optical characteristics of the objective lens 151, such as the NA, the sizes of the CAR-T cells and T cells contained in the object S, and the like. In bright-field images captured at the focal position, the brightness values ​​of light and dark around the cell outlines are lost, resulting in uniform information around the cell outlines for all cells. Capturing images at a position shifted from the focal position results in changes in brightness values ​​around the cell outlines, allowing images to be captured without losing information around the outlines of each cell. The offset amount is determined, for example, by using cells identified as CAR-T cells or T cells by fluorescent labeling, acquiring bright-field images captured at a position shifted by mΔz from the focal plane, and calculating predetermined feature amounts from each bright-field image. Here, m is an integer greater than or equal to 0, and Δz is the absolute value of the focal depth of the objective lens 151. The offset amount is preferably greater than or equal to 2 times and less than or equal to 15 times the focal depth Δz of the objective lens 151, and more preferably greater than or equal to 6 times and less than or equal to 10 times. To move the imaging position, the instruction unit 21 may move the imaging optical system 130 without moving the stage 120, or may move both the stage 120 and the imaging optical system 130.

[0041] Next, the instructing unit 21 outputs a bright-field image capturing instruction to the detecting unit 140, which instructs the detecting unit 140 to capture an image of a predetermined photographing area of ​​the object S that has been moved to the photographing position (S103). In response to the input of the bright-field image capturing instruction, the detecting unit 140 outputs bright-field image data indicating a bright-field image of the predetermined photographing area of ​​the object S to the content rate estimation device 1.

[0042] Next, the acquisition unit 22 acquires a bright-field image corresponding to the bright-field image data in response to the input of the bright-field image data from the detection unit 140 (S104). The acquisition unit 22 stores bright-field image information indicating the acquired bright-field image in the storage unit 12.

[0043] Next, the instructing unit 21 outputs a second imaging position movement instruction to the stage 120, instructing the stage 120 to move the object S to a second imaging position where a fluorescent image is to be captured (S105). The second imaging position is the focal position of the objective lens 151. In response to the input of the second imaging position movement instruction, the stage 120 moves along an axis parallel to the optical axis of the objective lens 151, thereby moving the object S to the second imaging position.

[0044] Next, the instructing unit 21 outputs a fluorescence image capturing instruction to the detecting unit 140, which instructs the detecting unit 140 to capture an image of a predetermined capture area of ​​the object S that has been moved to the capturing position (S106). In response to the input of the fluorescence image capturing instruction, the detecting unit 140 irradiates the object S with light that has passed through a fluorescence filter that transmits light having wavelengths included in a predetermined band, and outputs fluorescence image data indicating a fluorescence image of the predetermined capture area of ​​the object S to the content rate estimation device 1.

[0045] Next, the acquisition unit 22 acquires a fluorescence image corresponding to the fluorescence image data in response to the input of the fluorescence image data from the detection unit 140 (S107). The acquisition unit 22 stores fluorescence image information indicating the acquired fluorescence image in the storage unit 12.

[0046] Next, the instruction unit 21 determines whether bright-field images and fluorescent images have been captured for all of the imaging regions (S108). The number of imaging regions for which bright-field images and fluorescent images are captured is determined based on the number of cells used as training data in the learning process for training the trained model 220.

[0047] The processing shown in S101 to S108 is repeated until the instruction unit 21 determines that bright-field images and fluorescent images have been taken for all of the photographing regions (S108-YES). When the instruction unit 21 determines that bright-field images and fluorescent images have been taken for all of the photographing regions (S108-YES), the learning image photographing processing ends.

[0048] (Learning Process by Content Rate Estimation Device According to Embodiment) Fig. 6 is a flowchart of the learning process executed by the content rate estimation device 1. The learning process shown in Fig. 6 is executed mainly by the processing unit 20 in cooperation with each element of the content rate estimation system 100, based on a learning program stored in advance in the storage unit 12.

[0049] First, the image processing unit 23 generates a quantitative phase image from the bright-field image captured in the imaging process (S201). The image processing unit 23 generates a quantitative phase image from the bright-field image and stores quantitative phase image information indicating the generated quantitative phase image in the storage unit 12. The quantitative phase image is generated using a known method such as the method described in International Publication WO 2019 / 097587.

[0050] Next, the image processing unit 23 generates multiple cell masks used to extract the periphery of each cell's outline contained in the quantitative phase image generated in the process shown in S201 (S202). Generating a cell mask requires an image in which the entire cell has uniform contrast. It is difficult to generate a mask from a bright-field image, in which intracellular contrast varies depending on the intracellular structure. Therefore, the mask is generated using a quantitative phase image in which intracellular contrast does not vary depending on the intracellular structure and the entire cell has uniform contrast. The image processing unit 23 binarizes the pixel values ​​of each pixel in the quantitative phase image and performs a filtering process to remove overlapping cells, cells located at the boundary of the captured region, and foreign matter, thereby generating a cell mask showing the outline of each of the multiple cells contained in the quantitative phase image.

[0051] Next, the learning unit 24 determines a cell label to be used when performing supervised learning for each of the multiple cells included in the imaging area (S203). First, the learning unit 24 reads the fluorescence image acquired in the process shown in S107. Next, the learning unit 24 reads the cell mask generated in the process shown in S202 and arranges it so as to be superimposed on the fluorescence image. Next, the learning unit 24 calculates the average fluorescence brightness value of each of the cells surrounded by the cell mask and arranges the calculated average fluorescence brightness values ​​in descending order from the cell with the largest average fluorescence brightness value to the cell with the smallest average fluorescence brightness value. Next, the learning unit 24 acquires expression rate information indicating the CAR gene expression rate. The expression rate corresponding to the expression rate information is determined using a flow cytometry device also known as FACS and stored in the storage unit 12. Then, the learning unit 24 assigns a cell label to each of the multiple cells. The learning unit 24 assigns a label indicating CAR-T cells to a number of cells corresponding to the CAR gene expression rate, starting from the cell with the largest average fluorescence brightness value. The learning unit 24 labels cells that have not been labeled as CAR-T cells with labels indicating T cells.

[0052] Next, the learning unit 24 isolates each of the cells included in the image capture area and generates an isolated cell image corresponding to each of the cells (S204). The isolated cell image is a cell image representing a single cell. The isolated cell image generated in the process shown in S204 is used as training data in the learning process for training the trained model 220.

[0053] FIG. 7A is a diagram (part 1) for explaining the processing shown in S204, FIG. 7B is a diagram (part 2) for explaining the processing shown in S204, and FIG. 7C is a diagram (part 3) for explaining the processing shown in S204.

[0054] First, as shown in FIG. 7A, the learning unit 24 reads the bright-field image acquired in the process shown in S104 so that one of the cells contained in the bright-field image is positioned at the center. Next, the learning unit 24 reads the cell mask generated in the process shown in S202 and positions it so that it is superimposed on the bright-field image. In FIG. 7B, the cell indicated by the symbol "C1" is the cell to be isolated, and the cells indicated by the symbols "C2" and "C3" are cells not to be isolated. Next, the learning unit 24 stores the brightness of each pixel corresponding to the cell to be isolated in association with the pixel position. Next, the learning unit 24 changes the brightness of pixels corresponding to cells not to be isolated to a predetermined background brightness. Then, as shown in FIG. 7C, the learning unit 24 generates an isolated cell image by superimposing pixels corresponding to the cell to be isolated on the image with the changed background brightness. The isolated cell image generated in this manner clearly depicts the cell to be analyzed. The isolated cell image may also depict other cells other than the cell to be analyzed. The learning unit 24 sequentially generates isolated cell images for all cells included in the imaging area.

[0055] Next, the learning unit 24 instructs the server 200 to execute a learning process in which the trained model 220 is trained so that, when a cell image containing a CAR-T cell and the T cell is input, the trained model 220 outputs a feature indicating whether or not the cell image is a CAR-T cell (S205). The learning unit 24 outputs a plurality of isolated cell images and cell labels corresponding to each of the plurality of isolated cell images, along with a learning instruction indicating the execution of the learning process, to the server 200 via the communication unit 11. In response to the input of the plurality of isolated cell images and cell labels along with the learning instruction, the server processing unit 205 executes a learning process in which the trained model 220 is trained so that, when a cell image containing a CAR-T cell and the T cell is input, the trained model 220 outputs a feature indicating whether or not the cell image is a CAR-T cell. The server processing unit 205 executes supervised learning by inputting the isolated cell images and the cell labels corresponding to the isolated cell images into the pre-learning learning model.

[0056] Then, the learning unit 24 determines the trained model 220 to be used (S206). The learning unit 24 determines the trained model 220 to be used based on the accuracy rate in the training data, the loss in the training data, the accuracy rate in the accuracy verification data, and the loss in the accuracy verification data. The learning unit 24 determines, as the trained model 220 to be used, for example, a trained model for which the learning process has been executed over the number of epochs that minimizes the loss in the accuracy verification data.

[0057] FIG. 8A is a diagram showing the accuracy rate and loss in the training data, and FIG. 8B is a diagram showing the accuracy rate and loss in the accuracy verification data. In FIGS. 8A and 8B, the horizontal axis represents the number of epochs. In FIG. 8A, the left vertical axis represents the accuracy rate in the training data, and the right vertical axis represents the loss in the training data. Waveform W201 represents the change in the accuracy rate in the training data as the number of epochs increases, and waveform W202 represents the change in the loss in the training data as the number of epochs increases. In FIG. 8B, the left vertical axis represents the accuracy rate in the accuracy verification data, and the right vertical axis represents the loss in the accuracy verification data. Waveform W101 represents the change in the accuracy rate in the accuracy verification data as the number of epochs increases, and waveform W102 represents the change in the loss in the accuracy verification data as the number of epochs increases.

[0058] The accuracy rate in the training data gradually increases as the number of epochs increases, and the loss in the training data gradually decreases as the number of epochs increases. The accuracy rate in the accuracy validation data monotonically increases to around 80% as the number of epochs increases up to about 50, but saturates between 75% and 80% when the number of epochs exceeds 50. The loss in the accuracy validation data gradually decreases to about 0.009 as the number of epochs increases up to 55, but gradually increases when the number of epochs exceeds 55.

[0059] The learning unit 24 decides to use the learning model for which the learning process was performed with 55 epochs, at which the loss in the accuracy verification data is minimized, as the trained model 220. The learning unit 24 outputs a storage instruction to the server 200 to store the learning model decided to be used as the trained model 220 in the server storage unit 202. In response to the input of the storage instruction, the server processing unit 205 stores the trained model 220 in the server storage unit 202.

[0060] (Image Capturing Process for Inclusion Rate Estimation by Inclusion Rate Estimation Device According to Embodiment) Fig. 9 is a flowchart of the image capturing process for inclusion rate estimation executed by the inclusion rate estimation device 1. The image capturing process for inclusion rate estimation shown in Fig. 9 is executed mainly by the processing unit 20 in cooperation with each element of the inclusion rate estimation system 100, based on an imaging program stored in advance in the storage unit 12. The processes shown in S301 to S304 are the same as the processes shown in S101 to S104, and therefore detailed description thereof will be omitted here.

[0061] When the process shown in S304 is completed, the instruction unit 21 determines whether a bright-field image including a predetermined number of cells has been captured (S305). A bright-field image is also called a cell image. The number of cells included in the bright-field image is determined according to the number of cells required to estimate the gene expression rate of the CAR gene introduced into T cells.

[0062] The processing shown in S301 to S305 is repeated until the instruction unit 21 determines that a bright-field image including a predetermined number of cells has been captured (S305-YES). By repeating the processing shown in S301 to S305, multiple cell images derived from the same donor are acquired. When the instruction unit 21 determines that a bright-field image including a predetermined number of cells has been captured (S305-YES), the content rate estimation image capturing processing ends.

[0063] (Content Estimation Process by Content Estimation Device According to Embodiment) Fig. 10 is a flowchart of content estimation process executed by the content estimation device 1. The content estimation process shown in Fig. 10 is an example of a content estimation method according to the embodiment, and is executed mainly by the processing unit 20 in cooperation with each element of the content estimation system 100, based on a content estimation program stored in advance in the storage unit 12.

[0064] The processes shown in S401 to S403 are similar to the processes shown in S201 to S202 and S204, and therefore detailed description thereof will be omitted here. Following the process shown in S403, the acquisition unit 22 acquires images of the multiple cells isolated in the process shown in S403 (S404).

[0065] Next, the estimation unit 25 estimates the content rate of cell images related to CAR-T cells in the plurality of cell images using a feature distribution based on a plurality of feature amounts obtained by inputting the plurality of cell images acquired in the process shown in S404 into the trained model 220, and estimates the gene expression rate of the CAR gene by estimating the content rate of CAR-T cells in the plurality of cell images (S405). The estimation unit 25 extracts a distribution of CAR-T cell features and a distribution of T cell features from the distribution of the plurality of feature amounts acquired by the acquisition unit 22, and estimates the content rate of CAR-T cells from the distribution of CAR-T cell features and the distribution of T cell features. Specifically, the estimation unit 25 estimates that the first peak, which is the largest peak in the distribution of the plurality of feature amounts acquired by the acquisition unit 22, is the peak of T cells, and estimates that the second peak, which is the peak next to the first peak in the distribution of the plurality of feature amounts acquired by the acquisition unit 22, is the peak of CAR-T cells. Based on the estimated first peak and second peak, the estimation unit 25 extracts, by statistical processing, a distribution of the feature amounts of CAR-T cells and a distribution of the feature amounts of T cells from the distributions of the plurality of feature amounts acquired by the acquisition unit 22. The estimation unit 25 estimates the content of CAR-T cells from the extracted distributions of the feature amounts of CAR-T cells and T cells.

[0066] Then, the output unit 26 outputs a content rate signal, which is a signal indicating the content rate of CAR-T cells estimated by the estimation unit 25 in the process shown in S405 (S406). The content rate corresponding to the content rate signal output by the output unit 26 is the gene expression rate of the CAR gene.

[0067] FIG. 11 is a flowchart showing the process shown in S405 in FIG. 10 in more detail.

[0068] First, the estimation unit 25 instructs the server 200 to input one of the cell images acquired in the process shown in S404 into the trained model 220 (S501). The estimation unit 25 outputs a cell image input instruction indicating that one of the cell images acquired in the process shown in S404 is to be input into the trained model 220 to the server 200, together with the cell image. In response to the input of the cell image input instruction, the server processing unit 205 inputs the cell image input together with the cell image input instruction into the trained model 220. In response to the input of the cell image, the trained model 220 outputs the feature of a cell corresponding to the input cell image. The server processing unit 205 stores feature information indicating the feature output from the trained model 220 in the server storage unit 202.

[0069] Next, the estimation unit 25 acquires features of cells included in the image input to the trained model 220 in the process shown in S501 (S502). The estimation unit 25 outputs a feature output instruction to the server 200 to output features of cells included in the cell image input to the trained model 220 in the process shown in S501. In response to the input of the feature output instruction, the server processing unit 205 outputs a feature signal corresponding to the feature information stored in the server storage unit 202 to the content rate estimation device 1. The estimation unit 25 acquires features corresponding to the feature signal input from the server 200 and stores the acquired feature information in the storage unit 12 in association with the cell tumor information stored in the process shown in S502.

[0070] Next, the estimation unit 25 determines whether the processes shown in S501 to S502 have been performed on all cell images isolated by the process shown in S403 (S503). If the estimation unit 25 determines that the processes shown in S501 to S502 have not been performed on all cell images isolated by the process shown in S403 (S503-NO), the process returns to S501. Thereafter, the processes shown in S501 to S503 are repeated until the estimation unit 25 determines that the processes shown in S501 to S502 have been performed on all cell images isolated by the process shown in S403 (S503-YES).

[0071] When the estimation unit 25 determines that the processes shown in S501 and S502 have been performed on all cell images isolated in the process shown in S403 (S503-YES), the estimation unit 25 extracts a total distribution, which is a histogram showing the distribution of cell features (S504). The total distribution is a histogram obtained by summing the histogram of CAR-T cells and the histogram of T cells. The estimation unit 25 sequentially reads out features corresponding to the feature information stored in the storage unit 12, and arranges the read features according to size to extract the total distribution, and stores total distribution information showing the extracted total distribution in the storage unit 12.

[0072] Next, the estimation unit 25 generates a virtual T cell distribution, which is a histogram that hypothesizes the distribution of the feature amounts of T cells (S505). The estimation unit 25 acquires a UTD histogram that corresponds to the UTD histogram information stored in the storage unit 12.

[0073] Next, the estimation unit 25 extracts the respective distributions of CAR-T cells and T cells (S506). The estimation unit 25 acquires the virtual T-cell distribution information, the first virtual CAR-T-cell distribution information, and the second virtual CAR-T-cell distribution information stored in the storage unit 12. The estimation unit 25 extracts the respective distributions of CAR-T cells and T cells based on the Gaussian distributions corresponding to the acquired virtual T-cell distribution information, the first virtual CAR-T-cell distribution information, and the second virtual CAR-T-cell distribution information, respectively. Specifically, the first coefficient w １ , the second coefficient w ２ and the third coefficient w ３In equation (1), D represents the total distribution, UTD represents the hypothetical T cell distribution, and G Ｃ１ (μ Ｃ１ , σ Ｃ１ ) indicates the hypothetical CAR-T cell primary distribution, and G Ｃ２ (μ Ｃ２ , σ Ｃ２ ) indicates the hypothetical CAR-T cell second distribution. ε = D - (w １ ×UTD distribution +w ２ ×G Ｃ１ (μ Ｃ１ , σ Ｃ１ ) + w ３ ×G Ｃ２ (μ Ｃ２ , σ Ｃ２ )) (1) The estimation unit 25 estimates the UTD distribution, G Ｃ１ (μ Ｃ１ , σ Ｃ１ ) and G Ｃ２ (μ Ｃ２ , σ Ｃ２ ) as Gaussian basis functions to find the first coefficient w that minimizes ε. １ , the second coefficient w ２ and the third coefficient w ３ Determine.

[0074] The estimation unit 25 determines the first coefficient w １ , the second coefficient w ２ and the third coefficient w ３ and UTD distribution, G Ｃ１ (μ Ｃ１ , σ Ｃ１ ) and G Ｃ２ (μ Ｃ２ , σ Ｃ２ ) The distribution of CAR-T cells and T cells is extracted from the data.

[0075] Next, the estimation unit 25 counts the number of T cells (S507). From the distribution of cell features estimated in the process shown in S506, the estimation unit 25 counts the number of cells estimated to have T cell features for each of a plurality of feature amounts forming a histogram showing the distribution of T cell features. The estimation unit 25 calculates the number of T cells by summing the numbers of T cells counted for each of the plurality of feature amounts forming the histogram showing the distribution of T cell features, and stores T cell count information showing the calculated number of T cells in the storage unit 12.

[0076] Next, the estimation unit 25 counts the number of CAR-T cells (S508). From the distribution of CAR-T cell feature amounts estimated in the process shown in S506, the estimation unit 25 counts the number of cells estimated to have CAR-T cell feature amounts for each of a plurality of feature amounts forming a histogram showing the distribution of CAR-T cell feature amounts. The estimation unit 25 calculates the number of CAR-T cells by summing the numbers of CAR-T cells counted for each of the plurality of feature amounts forming the histogram showing the distribution of CAR-T cell feature amounts, and stores CAR-T cell count information showing the calculated number of CAR-T cells in the memory unit 12.

[0077] Then, the estimation unit 25 calculates the content rate of CAR-T cells (S509). The estimation unit 25 calculates the content rate of CAR-T cells contained in the turbid solution of T cells and CAR-T cells by dividing the number of CAR-T cells counted in the process shown in S508 by the sum of the number of T cells counted in the process shown in S507 and the number of CAR-T cells counted in the process shown in S508.

[0078] (Action and effect of the content rate estimation device according to the embodiment) The content rate estimation device 1 estimates the content rate of cell images related to CAR-T cells in multiple cell images by utilizing the distribution of feature amounts, and can estimate the content rate with higher accuracy than when the probability output from the second activation function layer 226 of the trained model 220 is used.

[0079] 12A to 12D are histograms showing the CAR gene expression rate of donor A, FIGS. 13A to 13D are histograms showing the CAR gene expression rate of donor B, FIGS. 14A to 14D are histograms showing the CAR gene expression rate of donor C, and FIGS. 15A to 15D are histograms showing the CAR gene expression rate of donor D. FIGS. 12A, 13A, 14A, and 15A show the distribution of feature amounts obtained by summing the histograms of CAR-T cells and T cells after CAR gene transfer into T cells from donors A to D, respectively. FIGS. 12B, 13B, 14B, and 15B show the distributions of actually measured values ​​of feature amounts of CAR-T cells and T cells after CAR gene transfer into T cells from donors A to D, respectively (UTD distribution). Figures 12C, 13C, 14C, and 15C show the distribution of T cell feature amounts before CAR gene introduction into T cells from each of donors A to D. Figures 12D, 13D, 14D, and 15D show the distribution of actual measured values ​​and estimated values ​​of feature amounts of CAR-T cells and T cells after introduction of CAR-T cells from each of donors A to D. In each of Figures 12A to 12D, 13A to 13D, 14A to 14D, and 15A to 15D, the horizontal axis represents feature amount, and the vertical axis represents probability density.

[0080] 12B, 13B, 14B, and 15B, the single peak in the distribution of T cell feature quantities and the two peaks in the distribution of CAR-T cell feature quantities have substantially the same peak value and half-width without any individual difference between donors. Because the two peaks in the distribution of CAR-T cell feature quantities have substantially the same peak value and half-width, the distribution of CAR-T cell feature quantities can be approximated by a Gaussian distribution.

[0081] Furthermore, as shown in Figures 12B and 12C, 13B and 13C, 14B and 14C, and 15B and 15C, the peak values ​​of the feature amounts before and after CAR gene introduction are substantially the same. Therefore, the distribution of the feature amounts of T cells after CAR gene introduction can be estimated from the distribution of the feature amounts of T cells before CAR gene introduction.

[0082] Furthermore, as shown in Figures 12D, 13D, 14D, and 15D, the distributions of the actual measured values ​​and estimated values ​​of the feature amounts of the CAR-T cells and T cells after the introduction of the CAR-T cells are approximately the same.

[0083] Table 1 shows a comparison between the results of CAR-T cell content estimation by the content estimation device 1 and the results of CAR-T cell content estimation based on the probabilities output from the second activation function layer 226 of the trained model 220. In Table 1, the "Donor" column indicates the donor who provided the cells, and the "Measured Value" column indicates the measured value [%] of the CAR-T cell content using fluorescent labeling. The "Estimated Value" column in the "Content Estimation Device 1" column indicates the estimated value [%] of the CAR-T cell content using the content estimation device 1, and the "Estimated Value" column in the "Trained Model 220" column indicates the estimated value [%] of the CAR-T cell content using the trained model 220. The "Error" column in the "Content Estimation Device 1" column shows the difference [%] between the estimated value of the CAR-T cell content using the content estimation device 1 and the actual measured value, and the "Error" column in the "Trained Model 220" column shows the difference [%] between the measured value of the CAR-T cell content using the trained model 220 and the actual measured value.

[0084]

[0085] The estimated values ​​of the CAR-T cell content using the trained model 220 differ from the actual measured values ​​by an absolute value of 5% or less for donors C and D, but by an absolute value of 5% or more for donors A and B. In particular, the absolute value of the difference from the actual measured value for donor B is extremely large, at 17.0%. On the other hand, the estimated values ​​of the CAR-T cell content using the content estimation device 1 differ from the actual measured value by an absolute value of 5% or less for all donors, enabling the CAR-T cell content to be estimated with high accuracy.

[0086] (Variation of Content Estimation Device According to the Embodiment) In the content estimation device 1, the estimation unit 25 estimates the content of CAR-T cells using UTD histogram information, which is a histogram showing the distribution of T cell features in cells of a donor before CAR introduction. However, the content estimation device according to the embodiment may estimate the content of CAR-T cells without using UTD histogram information. For example, the content estimation device according to the embodiment may estimate the content of CAR-T cells by considering the peak of the total distribution, which is a histogram showing the distribution of cell features, as the peak of the distribution of T cell features.

[0087] 16A to 16D are diagrams illustrating the content rate estimation process according to a modified example. Fig. 16A shows the total distribution of cell feature amounts for donor A, Fig. 16B shows the total distribution of cell feature amounts for donor B, Fig. 16C shows the total distribution of cell feature amounts for donor C, and Fig. 16D shows the total distribution of cell feature amounts for donor D.

[0088] For all of donors A to D, the region above the peak of the total distribution, which is the distribution of the sum of the feature amounts of CAR-T cells and T cells, approximately matches a Gaussian distribution with the peak as the mean value. Furthermore, because the feature amounts of T cells are dominant in the region above the peak of the total distribution, the distribution of the principal component of the feature amounts of T cells can be a Gaussian distribution with the peak of the total distribution as the mean value and with a half width approximately matching the half width of the region above the peak of the total distribution.

[0089] Furthermore, in the region smaller than the peak of the total distribution, the features of CAR-T cells and T cells are mixed, but the T cell features have a main component of the cell features whose average value is the peak of the total distribution, as well as a subcomponent whose peak is in a region smaller than the peak of the total distribution. Because the subcomponent of the distribution of T cell features is very small compared to the main component, it can be made into a Gaussian distribution with a predetermined mean value and variance determined from actual measured values ​​obtained from multiple donors. Information indicating the subcomponent of the distribution of T cell features is stored in the memory unit 12 as virtual T cell subcomponent information including the mean value and variance that define the Gaussian distribution.

[0090] 17 is a flowchart showing in more detail the process shown in S405 of the content rate estimation process according to the modified example, which is executed by the content rate estimation device 1. The process shown in FIG. 17 is executed mainly by the processing unit 20 in cooperation with each element of the content rate estimation system 100, based on a content rate estimation program stored in advance in the storage unit 12.

[0091] The content rate estimation process according to the modified example, steps S401 to S404 and S406, have been described with reference to Fig. 10, and therefore detailed description thereof will be omitted here. Also, steps S601 to S604 are similar to steps S501 to S504, and therefore detailed description thereof will be omitted here.

[0092] Following the process of S604, the estimation unit 25 generates a virtual T-cell principal component distribution, which is a histogram imagining the principal components of the distribution of T-cell features (S605). The estimation unit 25 generates the virtual T-cell principal component distribution from the peak of the total distribution, which is a histogram showing the distribution of the cell features extracted in the process of S604, and the half-width of the region larger than the peak, and stores virtual T-cell principal component distribution information showing the generated virtual T-cell principal component distribution in the storage unit 12. The virtual T-cell principal component distribution information includes the mean value and variance of the Gaussian distribution corresponding to the virtual T-cell principal component distribution information.

[0093] Next, the estimation unit 25 extracts the respective distributions of CAR-T cells and T cells (S606). The estimation unit 25 acquires the virtual T-cell principal component distribution information, virtual T-cell sub-component distribution information, virtual CAR-T-cell first distribution information, and virtual CAR-T-cell second distribution information stored in the storage unit 12. The estimation unit 25 extracts the respective distributions of CAR-T cells and T cells based on the Gaussian distributions corresponding to the acquired virtual T-cell principal component distribution information, virtual T-cell sub-component distribution information, virtual CAR-T-cell first distribution information, and virtual CAR-T-cell second distribution information. Specifically, the first coefficient w １ , the second coefficient w ２ , the third coefficient w ３ and the fourth coefficient w ４ In equation (2), D represents the total distribution, and G Ｔ１ (μ Ｔ１ , σＴ１ ) indicates the virtual T cell principal component distribution, and G Ｔ２ (μ Ｔ２ , σ Ｔ２ ) indicates the hypothetical T cell subcomponent distribution, and G Ｃ１ (μ Ｃ１ , σ Ｃ１ ) indicates the hypothetical CAR-T cell primary distribution, and G Ｃ２ (μ Ｃ２ , σ Ｃ２ ) indicates the hypothetical CAR-T cell second distribution. ε = D - (w １ ×G Ｔ１ (μ Ｔ１ , σ Ｔ１ ) + w ２ ×G Ｔ２ (μ Ｔ２ , σ Ｔ２ ) +w ３ ×G Ｃ１ (μ Ｃ１ , σ Ｃ１ ) + w ４ ×G Ｃ２ (μ Ｃ２ , σ Ｃ２ )) (2) The estimation unit 25 Ｔ１ (μ Ｔ１ , σ Ｔ１ ), G Ｔ２ (μ Ｔ２ , σ Ｔ２ ), G Ｃ１ (μ Ｃ１ , σ Ｃ１ ) and G Ｃ２ (μ Ｃ２ , σ Ｃ２ ) as Gaussian basis functions to find the first coefficient w that minimizes ε. １ , the second coefficient w ２ , the third coefficient w ３ and the fourth coefficient w ４ Determine.

[0094] The estimation unit 25 determines the first coefficient w １ , the second coefficient w ２ , the third coefficient w ３ and the fourth coefficient w ４ and G Ｔ１ (μ Ｔ１ , σ Ｔ１ ), G Ｔ２ (μ Ｔ２ , σ Ｔ２ ), GＣ１ (μ Ｃ１ , σ Ｃ１ ) and G Ｃ２ (μ Ｃ２ , σ Ｃ２ ) and extract the distributions of CAR-T cells and T cells. The processes shown in S607 to S609 are the same as the processes shown in S507 to S509, and therefore detailed description thereof will be omitted here.

[0095] The content estimation process according to the modified example does not use a histogram showing the distribution of T cell features in cells before introduction of the donor's CAR gene, and therefore can estimate the CAR-T cell content more easily than the content estimation process according to the embodiment described with reference to Figure 10.

[0096] Furthermore, the content estimation process of CAR-T cells by the content estimation device 1 is used in the process of estimating the gene expression rate of the CAR gene. However, the content estimation device according to the embodiment may be used for processes other than the process of estimating the gene expression rate of the CAR gene, as long as it executes the process of estimating the content of CAR-T cells. The content estimation device according to the embodiment may use the results of estimating the content of CAR-T cells to calculate the amount of protein produced by gene introduction or the amount of change in a structure, or may use the results of estimating the CAR expression rate in a cell population to execute a process of estimating an appropriate culture period for CAR-T cells.

[0097] Furthermore, in the inclusion rate estimation system 100, the trained model 220 is stored in a server storage unit 202 included in the server 200, and the inclusion rate estimation process is executed by the inclusion rate estimation device 1. However, in the inclusion rate estimation system according to the embodiment, the inclusion rate estimation process may be executed by a server having a server storage unit 202 that stores the trained model 220. Furthermore, in the inclusion rate estimation system according to the embodiment, the trained model 220 may be stored in a storage unit 12 included in the inclusion rate estimation device 1.

[0098] FIG. 18 is a conceptual diagram showing the configuration of a content rate estimation system according to a modified example, and FIG. 19 is a block diagram of the content rate estimation device shown in FIG.

[0099] The inclusion rate estimation system 100a differs from the inclusion rate estimation system 100 in that it has an inclusion rate estimation device 2 instead of the inclusion rate estimation device 1 and the server 200. The inclusion rate estimation device 2 differs from the inclusion rate estimation device 1 in that it has a memory unit 17 instead of the memory unit 12. The configurations and functions of the components of the inclusion rate estimation system 100a other than the memory unit 17 are the same as the configurations and functions of the components of the inclusion rate estimation system 100 that are assigned the same reference numerals, and therefore detailed description thereof will be omitted here. Like the inclusion rate estimation system 100, the inclusion rate estimation system 100a executes a learning image capture process, a learning process, an inclusion rate estimation image capture process, and an inclusion rate estimation process.

[0100] The storage unit 17 differs from the storage unit 12 in that it stores the trained model 220. The configuration and functions of the storage unit 17 other than storing the trained model 220 are the same as the configuration and functions of the storage unit 12, and therefore detailed description thereof will be omitted here.

[0101] Furthermore, the content estimation device 1 stores UTD histogram information generated for each donor from which cells are collected in the storage unit 12, but the content estimation device according to the embodiment may store UTD histogram information generated from cells collected from a single donor in the storage unit 12. The content estimation device according to the embodiment performs the content estimation process using UTD histogram information generated from cells collected from a single donor, thereby reducing the cost of generating UTD histogram information.

[0102] Although the content estimation device 1 estimates the content of CAR-T cells, the content estimation device according to the embodiment may estimate the content of T cells in which the CAR gene has not been expressed.Furthermore, the content estimation device according to the embodiment may estimate the content of cells into which an exogenous gene other than the CAR gene has been introduced.

[0103] Furthermore, the content estimation device 1 estimates the content of CAR-T cells by approximating the distribution of feature amounts of CAR-T cells and T cells to a Gaussian distribution. However, the content estimation device according to the embodiment may estimate the content of cells into which an exogenous gene other than the CAR gene has been introduced by statistical processing other than approximating to a Gaussian distribution.

[0104] Furthermore, the content rate estimation device 1 calculates the first coefficient w １ , the second coefficient w ２ and the third coefficient w ３ However, the content rate estimation device according to the embodiment uses the first coefficient w １ , the second coefficient w ２ and the third coefficient w ３ At least one of the mean value and variance of the virtual T cell distribution, the virtual CAR-T cell first distribution, and the virtual CAR-T cell second distribution may be used as a parameter. Furthermore, the content rate estimation device according to the embodiment may use any one of the coefficient, mean value, and variance shown in formula (1) as a parameter.

[0105] The content estimation device according to the embodiment may be a device that estimates the content of cells into which a gene has actually been introduced after a gene introduction operation has been performed on an entire cell population. The gene to be introduced is not particularly limited, and examples thereof include CAR (chimeric antigen receptor). The cells are also not particularly limited, and animal cells, for example, can be used. Examples of animal cells that may be the subject of estimation include spleen cells, nerve cells, glial cells, pancreatic beta cells, bone marrow cells, mesangial cells, Langerhans cells, epidermal cells, epithelial cells, endothelial cells, fibroblasts, fibrocytes, muscle cells (e.g., skeletal muscle cells, cardiac muscle cells, myoblasts, and muscle satellite cells), adipocytes, immune cells (e.g., macrophages, T cells, B cells, natural killer cells (NK cells), mast cells, neutrophils, basophils, eosinophils, monocytes, and megakaryocytes), synoviocytes, chondrocytes, osteocytes, osteoblasts, osteoclasts, mammary gland cells, hepatocytes, stromal cells, egg cells, and sperm cells, as well as stem cells that can be induced to differentiate into these cells (including pluripotent stem cells such as neural stem cells, hematopoietic stem cells, mesenchymal stem cells, dental pulp stem cells, iPS cells, and ES cells), progenitor cells, blood cells, oocytes, and fertilized eggs. T cells include αβ T cells, γδ T cells, helper T cells, cytotoxic T cells, regulatory T cells, suppressor T cells, tumor-infiltrating T cells, memory T cells, naive T cells, NK T cells, TCR-T cells, STAR receptor T cells, CAR-T cells, and the like. Furthermore, animal cells also include the above-mentioned cells produced by in vitro differentiation induction of primary cells, the above-mentioned stem cells (e.g., iPS cells), and the like. Animal cells also include various cancer cells. Animal cells may be of one type only, or may contain two or more types. Furthermore, the organism from which the animal cells are derived is not particularly limited, and examples of such organisms include mammals, such as humans, mice, rats, cows, horses, pigs, rabbits, dogs, cats, goats, monkeys, and chimpanzees. The organism from which the animal cells are derived is preferably humans, and cells that undergo changes around their outline or change in morphology upon gene introduction, such as CAR-T cells, are preferred. Cells that undergo changes around their contours or morphological changes include cells that express receptors on the cell surface due to gene introduction, and cells that express intracellular domains within the cell near the cell surface.Also included are cells that secrete proteins produced by gene introduction, such as antibody-producing cells.

[0106] The content estimation device according to the embodiment may be used in a manufacturing process for producing a gene-introduced cell preparation, in a process for estimating whether a cell has been introduced with an exogenous gene or has not been introduced with an exogenous gene based on an acquired bright-field image.

[0107] The manufacturing method of a cell preparation according to the embodiment includes: (I) a step of acquiring a bright-field image of a cell photographed at a position shifted from the focal position; and (II) a step of estimating, based on the bright-field image, whether the cell is a cell into which an exogenous gene has been introduced or a cell into which an exogenous gene has not been introduced.

[0108] In the method for producing a cell preparation according to the embodiment, steps (I) and (II) can be carried out by the method described above. The cell preparation according to the embodiment is preferably produced as a parenteral preparation by mixing an effective amount of transfected animal cells with a pharmaceutically acceptable carrier according to known procedures, such as the method described in the Japanese Pharmacopoeia. The cell preparation according to the embodiment is preferably produced as a parenteral preparation such as an injection, suspension, or infusion. Parenteral administration methods include intravenous, intraarterial, intramuscular, intraperitoneal, and subcutaneous administration. Pharmaceutically acceptable carriers include solvents, bases, diluents, excipients, soothing agents, buffers, preservatives, stabilizers, suspending agents, isotonic agents, surfactants, and solubilizers.

[0109] REFERENCE SIGNS LIST 1 Content estimation device 21 Instruction unit 22 Acquisition unit 23 Image processing unit 24 Learning unit 25 Estimation unit 26 Output unit 100 Content estimation system 101 Microscope main body 200 Server 220 Trained model

Claims

1. A content estimation device comprising: a memory unit that stores a trained model that has been trained to output, when a cell image is input, a feature that indicates whether a cell represented in the cell image is a specific cell; an acquisition unit that acquires a plurality of the cell images; an estimation unit that estimates the content of the specific cell in the cells represented in the plurality of cell images using the distribution of the feature values ​​obtained by inputting the plurality of cell images to the trained model; and an output unit that outputs a signal indicating the content estimated by the estimation unit.

2. The content estimation device described in claim 1, wherein the cell image is an image representing either a cell not expressing an exogenous gene or a cell expressing an exogenous gene, and the specific cell is either a cell not expressing an exogenous gene or a cell expressing an exogenous gene.

3. The content rate estimation device described in claim 1, wherein the estimation unit extracts a distribution of features of a first cell, which is the specific cell, and a distribution of features of a second cell different from the first cell from the distribution of features acquired by the acquisition unit, and estimates the content rate from the distribution of features of the first cell and the second cell.

4. The content estimation device described in claim 3, wherein the estimation unit estimates that a first peak, which is the largest peak in the distribution of the feature acquired by the acquisition unit, is the peak of the second cell, estimates that a second peak, which is the peak next to the first peak in the distribution of the feature acquired by the acquisition unit, is the peak of the first cell, and estimates the content based on the first peak and the second peak.

5. The content rate estimation device described in claim 4, wherein the estimation unit extracts the distribution of the features of the first cell and the distribution of the features of the second cell from the distribution of the features acquired by the acquisition unit through statistical processing.

6. The content estimation device described in claim 5, wherein the estimation unit extracts the distribution of the features of the first cell and the distribution of the features of the second cell from the distribution of the features acquired by the acquisition unit by approximating them with a Gaussian distribution.

7. A content estimation device described in any one of claims 4 to 6, wherein the first cell is either a cell that does not express an exogenous gene or a cell that expresses an exogenous gene, and the second cell is the other cell of the cell that does not express an exogenous gene or the cell that expresses an exogenous gene.

8. The content estimation device described in claim 7, wherein the cells not expressing the foreign gene are T cells not expressing the CAR gene, and the cells expressing the foreign gene are CAR-T cells expressing the CAR gene.

9. The content estimation device according to claim 8, wherein the estimation unit estimates the content by estimating that the distribution of the feature amounts of the CAR-T cells is formed by superimposing two Gaussian distributions.

10. The content estimation device according to claim 9, wherein the estimation unit extracts the distribution of the features of the first cells and the distribution of the features of the second cells using as parameters at least one of the coefficients and the mean value and standard deviation of the Gaussian distributions, so as to minimize the difference between a distribution calculated by multiplying each of two Gaussian distributions forming the distribution of the features of the T cells and the distribution of the features of the CAR-T cells by a coefficient, and the distribution of the multiple features acquired by the acquisition unit.

11. The content rate estimation device described in claim 10, wherein the estimation unit extracts the distribution of the features of the first cell and the distribution of the features of the second cell using any one of the coefficient, the mean value, and the standard deviation as a parameter.

12. A content estimation device according to claim 10 or 11, wherein the estimation unit estimates the distribution of the T cell features based on the distribution of the T cell features collected from the subject.

13. The content estimation device according to claim 10 or 11, wherein the estimation unit estimates the distribution of the T cells based on the first peak.

14. A content estimation device according to any one of claims 3 to 13, wherein the feature is a numerical value that can be converted into a probability distribution indicating whether the cell is the first cell or the second cell.

15. A content estimation device described in any one of claims 3 to 14, wherein the trained model includes a convolutional layer that performs convolution processing on the cell image, and a feature generation layer that generates features of the first cell or the second cell from the output of the convolutional layer.

16. A content estimation device according to any one of claims 1 to 15, wherein the cell image is captured by bright-field observation.

17. A content estimation device according to any one of claims 1 to 16, wherein the acquisition unit generates the cell images so that one cell is represented per image.

18. The content estimation device described in claim 17, wherein the acquisition unit generates the cell image by placing the one cell at the center of the cell image and changing the brightness of cells other than the one cell to the brightness of the background.

19. A content estimation method including the steps of: inputting a plurality of cell images into a trained model that has been trained to output features indicating whether a cell represented in a cell image is a specific cell; acquiring a plurality of the features corresponding to the cells represented in each of the plurality of cell images; estimating the content rate of the specific cells represented in the plurality of cell images from the distribution of the acquired plurality of features; and outputting a signal indicating the estimated content rate.

20. A content estimation device comprising: a memory unit that stores a trained model that has been trained to output, when a cell image is input, a signal indicating a feature that indicates whether the cell shown in the cell image is a CAR-T cell; an acquisition unit that acquires a plurality of the cell images derived from the same subject; an estimation unit that estimates the content of CAR-T cells in the cells shown in the plurality of cell images using a feature distribution obtained by inputting the plurality of cell images to the trained model; and an output unit that outputs a signal indicating the content estimated by the estimation unit.

21. A content estimation device comprising: an acquisition unit that acquires a plurality of cell images; an estimation unit that estimates the content of a specific cell among cells represented in a plurality of cell images by utilizing a feature distribution obtained by inputting a cell image into a trained model that is trained to output features indicating whether the cell represented in the cell image is a specific cell; and an output unit that outputs the content estimated by the estimation unit.

22. A content estimation system comprising: a server having a memory unit that stores a trained model that has been trained to output features indicating whether a cell represented in a cell image is a specific cell when the cell image is input; an acquisition unit that acquires a plurality of the cell images; an estimation unit that estimates the content of the specific cell in the cells represented in the plurality of cell images using a feature distribution obtained by inputting a number of the cell images into the trained model; and a terminal device having an output unit that outputs a signal indicating the content estimated by the estimation unit.

Images