Classification model generation method, particle determination method, computer program, and information processing device.

Abstract

Provided are a classification model generation method, a particle determination method, a computer program, and an information processing device for facilitating the discrimination of particles having a specific morphological feature.　With the classification model generation method, observation data is acquired, the observation data representing results obtained by observing respective particles contained in a first sample in which particles having a specific morphological feature and other particles are mixed and in a second sample comprising the other particles and not containing the particles having the specific morphological feature, and through learning using training data that includes the observation data and information indicating whether the observation data was obtained from particles contained in the first sample or from particles contained in the second sample, a classification model is generated in a case in which observation data representing results obtained by observing particles has been input, the classification model outputting discrimination information indicating whether particles have the specific morphological feature.

Description

Technical Field

[0001] The present invention relates to a classification model generation method for determining whether particles such as cells are specific particles, a particle determination method using the generated classification model, a computer program, and an information processing apparatus.

Background Art

[0002] Conventionally, flow cytometry has been used as a method for examining individual cells. Flow cytometry is a method for analyzing cells that involves flowing cells dispersed in a fluid, irradiating each cell moving through a flow path with light, and measuring light such as scattered light or fluorescence from the irradiated cells, thereby obtaining information about the irradiated cells as a captured image or the like. By using flow cytometry, it is possible to rapidly conduct one-by-one investigations of a large number of cells. Furthermore, a ghost cytometry method (hereinafter referred to as the GC method) has been developed in which special structured illumination light is irradiated onto cells moving through a flow path in a flow cytometer, waveform data containing compressed morphological information of the cells is obtained from the cells, and the cells are classified based on the waveform data. An example of the GC method is disclosed in Patent Document 1. In the GC method, a classification model is created in advance by machine learning from the waveform data of cells prepared as training samples, and the classification model is used to determine whether the cells included in an evaluation sample are target cells to be classified. A flow cytometer using the GC method enables faster and more accurate cell analysis.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Flow cytometry using the GC method allows for the identification of cells with specific morphological characteristics without the need for labeling such as fluorescent staining. For example, if a disease causes changes in the morphology of certain cells, flow cytometry using the GC method can identify cells with altered morphology from a sample taken from a person, thereby enabling the identification of patients with that disease. To create a classification model used in the GC method, it is necessary to prepare the target cells as training samples in advance and acquire waveform data for those cells. However, it can be difficult to acquire a large amount of waveform data for the target cells, such as when the target cells are rare. Therefore, it can be difficult to identify cells with specific characteristics.

[0005] The present invention has been made in view of these circumstances, and its object is to provide a classification model generation method, a particle determination method, a computer program, and an information processing device for facilitating the identification of particles having specific morphological characteristics. [Means for solving the problem]

[0006] The classification model generation method according to the present invention is characterized by acquiring observational data representing the result of observing each particle in a first sample containing a mixture of particles having specific morphological characteristics and other particles, and a second sample containing only particles other than those having the specific morphological characteristics, and generating a classification model that outputs discrimination information indicating whether or not a particle has the specific morphological characteristics when observational data representing the result of observing a particle is input. This is achieved by learning using training data that includes the observational data and information indicating which particle in the first or second sample the observational data was obtained from.

[0007] The classification model generation method according to the present invention is characterized in that the observation data is waveform data representing the time change in the intensity of light emitted from particles irradiated with light by structured illumination, or waveform data representing the time change in the intensity of light detected by structuring the light from the irradiated particles.

[0008] The classification model generation method according to the present invention is characterized in that the first sample is a specimen collected from a person having a specific disease, and the second sample is a specimen collected from a person not having the specific disease.

[0009] The particle determination method according to the present invention involves acquiring observational data representing the result of observing a particle, inputting the acquired observational data into a classification model that outputs discrimination information indicating whether or not a particle has specific morphological characteristics when the observational data representing the result of observing a particle is input, acquiring the discrimination information output by the classification model, and determining whether or not the particle relating to the observational data is a particle having the specific morphological characteristics based on the acquired discrimination information, wherein the classification model is trained by training data that includes observational data representing the result of observing a particle and information indicating whether or not the observational data was obtained from a particle in the first sample or the second sample, for each particle contained in a first sample containing a mixture of particles having the specific morphological characteristics and other particles, and a second sample containing no particles having the specific morphological characteristics and consisting of other particles.

[0010] The particle determination method according to the present invention is characterized by outputting information relating to the particle that has been determined.

[0011] The particle determination method according to the present invention is characterized in that, when the particle to be used to acquire the observation data is collected from a person, a tag having identification information to identify the person from whom the particle was collected is attached to the particle, and when the particle related to the observation data is a particle having the specific morphological characteristics, the person from whom the particle was collected is identified based on the identification information on the tag attached to the particle related to the observation data.

[0012] The computer program according to the present invention is characterized in that, for each particle contained in a first sample containing a mixture of particles having specific morphological characteristics and other particles, and a second sample containing only particles other than those having the specific morphological characteristics, the computer obtains observational data representing the result of observing the particle, and, through learning using training data that includes the observational data and information indicating which particle in the first or second sample the observational data was obtained from, the computer generates a classification model that outputs discrimination information indicating whether or not the particle has the specific morphological characteristics when observational data representing the result of observing the particle is input.

[0013] The computer program according to the present invention acquires observational data representing the result of observing a particle, inputs the acquired observational data into a classification model that outputs discrimination information indicating whether or not the particle has specific morphological characteristics when the observational data representing the result of observing a particle is input, acquires the discrimination information output by the classification model, and causes the computer to execute a process to determine whether or not the particle relating to the observational data has the specific morphological characteristics based on the acquired discrimination information, wherein the classification model is learned by training data that includes observational data representing the result of observing each particle in a first sample containing a mixture of particles having the specific morphological characteristics and other particles, and a second sample consisting of other particles that do not contain particles having the specific morphological characteristics, and information indicating whether or not the observational data was obtained from particles in the first sample or the second sample.

[0014] The information processing apparatus according to the present invention is characterized by comprising: a data acquisition unit that acquires observation data representing the result of observing each particle contained in a first sample in which particles having specific morphological characteristics and other particles are mixed, and a second sample that does not contain the particles having the specific morphological characteristics and consists of other particles; and a classification model generation unit that generates a classification model that outputs discrimination information indicating whether or not a particle has the specific morphological characteristics when observation data representing the result of observing a particle is input, by learning using training data that includes the observation data and information indicating which of the first sample and the second sample the observation data was obtained from.

[0015] The information processing device according to the present invention comprises: an observation unit that acquires observation data representing the result of observing a particle; a discrimination information acquisition unit that inputs the acquired observation data to a classification model that outputs discrimination information indicating whether or not the particle has specific morphological characteristics when the observation data representing the result of observing a particle is input, and acquires the discrimination information output by the classification model; and a determination unit that determines whether or not the particle relating to the observation data has the specific morphological characteristics based on the acquired discrimination information. The classification model is characterized in that it is learned by training data that includes observation data representing the result of observing each particle in a first sample containing a mixture of particles having the specific morphological characteristics and other particles, and a second sample consisting of other particles that do not contain particles having the specific morphological characteristics, and information indicating whether or not the observation data was obtained from particles in the first sample or the second sample.

[0016] In one embodiment of the present invention, a classification model is trained using training data that includes observational data obtained for particles contained in a first sample and a second sample. The first sample contains particles having a specific morphological feature and other negative particles. The second sample does not contain particles having a specific morphological feature and consists of other particles. The classification model outputs discrimination information indicating whether or not a particle has a specific morphological feature when observational data is input. Even when particles having a specific morphological feature are rare and it is not easy to obtain a large amount of observational data, it is possible to generate a classification model by training using observational data and training data that includes information indicating whether or not the observational data was obtained from particles contained in the first sample or the second sample. Using the classification model, discrimination information corresponding to the observational data can be obtained, and based on the discrimination information, it is possible to determine whether or not a particle has a specific morphological feature.

[0017] In one embodiment of the present invention, the observation data is waveform data representing the time change in the intensity of light emitted from particles irradiated with light by structured illumination, or waveform data representing the time change in the intensity of light detected by structuring the light from irradiated particles. The waveform data is similar to that used in the GC method and includes compressed morphological information of the particles. Therefore, it can be used for generating classification models and determining particles.

[0018] In one embodiment of the present invention, the first sample is a sample taken from a person with a specific disease such as cancer, and the second sample is a sample taken from a person without the specific disease. Particles having specific morphological characteristics are particles whose morphology has been altered by the specific disease such as cancer (positive particles). By determining whether or not the particles contained in the sample are positive particles, it is possible to determine whether or not the person from whom the particles were collected has the specific disease.

[0019] In one embodiment of the present invention, information regarding the determined particles is displayed. The user can review the information regarding the particles.

[0020] In one embodiment of the present invention, a tag having identification information for identifying a person from whom particles have been collected is attached to the particles, and based on the tag, the person from whom the particles have been collected is specified. Based on the tag attached to the particles having specific morphological features, it is possible to specify the person from whom the particles having specific morphological features have been collected. Thereby, for example, it is also possible to specify a person having a specific disease among the people from whom particles have been collected.

Advantages of the Invention

[0021] In the present invention, even when it is not easy to prepare a large number of training samples including particles having specific morphological features to obtain observation data, a classification model that outputs discrimination information indicating whether a particle has specific morphological features in response to the input of the observation data can be generated. The present invention exhibits excellent effects such as being able to discriminate particles having specific morphological features using the classification model.

Brief Description of the Drawings

[0022] [Figure 1] It is a conceptual diagram showing an overview of a classification model generation method. [Figure 2] It is a block diagram showing a configuration example of a learning device according to Embodiment 1 for generating a classification model. [Figure 3] It is a graph showing an example of waveform data. [Figure 4] It is a block diagram showing an internal configuration example of an information processing device. [Figure 5] It is a conceptual diagram showing the function of a classification model. [Figure 6] It is a flowchart showing an example of a procedure of a process for learning a classification model. [Figure 7] It is a schematic diagram showing a learning example using the concept of MIL. [Figure 8] It is a conceptual diagram showing an overview of a particle determination method. [Figure 9] It is a block diagram showing a configuration example of a determination device according to Embodiment 1 for determining cells. [Figure 10]This is a block diagram showing an example of the internal configuration of an information processing device. [Figure 11] This flowchart shows the steps of the process that the information processing device performs to identify cells. [Figure 12] This is a schematic diagram showing an example of how the judgment results are displayed. [Figure 13] This is a block diagram showing an example configuration of a learning device according to Embodiment 2. [Figure 14] This is a block diagram showing an example configuration of the determination device according to Embodiment 2. [Figure 15] This is a block diagram showing an example configuration of a learning device according to Embodiment 3. [Figure 16] This is a block diagram showing an example configuration of the determination device according to Embodiment 3. [Modes for carrying out the invention]

[0023] The present invention will be described in detail below with reference to drawings illustrating its embodiments. <Embodiment 1> In this embodiment, the GC method is used to determine whether a cell possesses specific morphological characteristics based on waveform data representing the morphological characteristics of the cell obtained by irradiating the cell with light. The waveform data corresponds to the observed data. Below, this embodiment will be explained using an example in which cells possessing specific morphological characteristics due to the effects of a specific disease, such as cancer cells, are identified from among multiple cells contained in a sample taken from a person.

[0024] In this embodiment, a classification model necessary for cell discrimination is generated. The classification model is a pre-trained model. Figure 1 is a conceptual diagram showing an overview of the classification model generation method. Multiple first samples 111 are created, which contain a mixture of cells with specific morphological characteristics and other cells, and multiple second samples 121 are created, which do not contain cells with specific morphological characteristics and consist of other cells. For example, a sample taken from a patient 11 with a specific disease such as cancer is the first sample 111, and a sample taken from a healthy person 12 without the specific disease is the second sample 121. The symbols in parentheses in Figure 1 refer to the individual patient 11 and the first sample 111, and also to the individual healthy person 12 and the second sample 121. That is, Figure 1 In the example shown, the first sample 111a was collected from patient 11a, and the first sample 111b was collected from patient 11b. Similarly, the second sample 121a was collected from healthy individual 12a, and the second sample 121b was collected from healthy individual 12b. The first sample 111 contains cells with specific morphological characteristics due to the effects of a particular disease, such as cancer cells, as well as other cells. The other cells are those that do not possess specific morphological characteristics. The second sample 121 contains only other cells. Hereafter, cells with specific morphological characteristics will be referred to as positive cells, and other cells as negative cells. In Figure 1, positive cells are indicated by double circles, and negative cells by single circles. There are many types of negative cells. Normally, positive cells are rarely present. Typically, there are fewer positive cells than negative cells in the first sample 111.

[0025] Next, light is shone onto the cells in the first sample 111 and the second sample 121, and waveform data representing the morphological characteristics of each cell is obtained. As will be described later, the waveform data represents the time change in the intensity of the light emitted from the irradiated cell, and the waveform data contains the morphological characteristics of the cell. Next, a classification model is generated by training using training data that includes the waveform data and information indicating whether the waveform data was obtained from cells in the first sample or the second sample. The classification model outputs discrimination information indicating whether or not a cell has specific morphological characteristics when given waveform data as input.

[0026] Figure 2 is a block diagram showing an example configuration of a learning device 200 according to Embodiment 1 for generating a classification model. The learning device 200 is equipped with a channel 34 through which cells flow. Cells 4 are dispersed in a fluid, and as the fluid flows through the channel 34, individual cells 4 move sequentially through the channel 34. The learning device 200 is equipped with a light source 31 that irradiates light onto the cells 4 moving through the channel 34. The light source 31 emits white light or monochromatic light. The light source 31 is, for example, a laser light source or an LED (Light Emitting Diode) light source. Cells 4 that are irradiated with light emit light. The light emitted from cells 4 is, for example, reflected light, scattered light, transmitted light, fluorescence, Raman scattered light, or diffracted light thereof. The learning device 200 is equipped with a detection unit 32 that detects light from cells 4. The detection unit 32 includes a photomultiplier tube (PMT), a line-type PMT element, a photodiode, an APD (Avalanche Photo-Diode), or a semiconductor photosensor, among other photodetection sensors. Figure 2 shows the path of light with solid arrows.

[0027] The learning device 200 includes an optical system 33. The optical system 33 guides illumination light from a light source 31 to cells 4 in a channel 34 and causes light from cells 4 to be incident on a detection unit 32. The optical system 33 includes a spatial light modulation device 331 for modulating and structuring the incident light. The learning device 200 shown in Figure 2 is configured such that illumination light from a light source 31 is irradiated onto cells 4 via the spatial light modulation device 331. The spatial light modulation device 331 is a device that modulates light by controlling the spatial distribution of light (amplitude, phase, polarization, etc.). The spatial light modulation device 331 has, for example, multiple regions on the surface to which light is incident, and the incident light is modulated differently in two or more of these multiple regions. Here, modulation means changing the properties of light (one or more properties of light, such as intensity, wavelength, phase, and polarization state). Figure 2 illustrates a configuration in which two types of regions with different light transmittances are arranged in a predetermined pattern in a two-dimensional grid. In Figure 2, the illumination light from the light source 31 passes through the spatial light modulation device 331, creating structured illumination in which two types of light with different intensities are arranged in a predetermined pattern.

[0028] The spatial light modulation device 331 is, for example, a diffractive optical element (DOE), a spatial light modulator (SLM), or a digital micromirror device (DMD). If the illumination light emitted by the light source 31 is incoherent light, the spatial light modulation device 331 is a DMD. Another example of the spatial light modulation device 331 is a film or optical filter in which multiple types of regions with different light transmittances are arranged randomly or in a predetermined pattern. Here, "multiple types of regions with different light transmittances arranged in a predetermined pattern" means, for example, that multiple types of regions with different light transmittances are arranged in a one-dimensional or two-dimensional grid. "Multiple types of regions with different light transmittances arranged randomly" means that multiple types of regions are scattered irregularly. The aforementioned film or optical filter has a configuration having at least two types of regions: a region having a first light transmittance and a region having a second transmittance different from the first light transmittance. Thus, before the illumination light from the light source 31 is irradiated onto the cells 4, it is modulated by the spatial light modulation device 331 and converted into structured illumination light in which bright spots of different light intensities are arranged randomly or in a predetermined pattern. This configuration, in which the illumination light from the light source 31 is modulated by the spatial light modulation device 331 along the optical path from the light source 31 to the cells 4, is also referred to as structured illumination.

[0029] The illumination light from structured illumination is directed to a specific region (irradiation region) in the channel 34, and as the cell 4 moves within this irradiation region, the cell 4 is illuminated by the structured illumination light. Although the pattern of the structured illumination light directed to the cell 4 is constant and does not change over time, as the cell 4 moves within the irradiation region, it receives illumination of light with different light intensities depending on its location. Upon receiving illumination from the structured illumination light, the cell 4 emits light such as transmitted light, fluorescence, scattered light, interference light, diffracted light, or polarized light, either emitted from or generated through the cell 4. Hereafter, this light emitted from or generated through the cell 4 will also be referred to as light modulated by the cell 4. The light modulated by the cell 4 is continuously detected by the detection unit 32 while the cell 4 passes through the irradiation region of the channel 34. The detection unit 32 outputs a signal to the information processing device 2 corresponding to the intensity of the detected light. In this way, the learning device 200 can acquire waveform data representing the time change in the intensity of light modulated by the cells 4 detected by the detection unit 32.

[0030] Figure 3 is a graph showing an example of waveform data. In Figure 3, the horizontal axis represents time, and the vertical axis represents the light intensity detected by the detection unit 32. The waveform data consists of multiple intensity values ​​obtained sequentially over time. Each intensity value represents the light intensity. The waveform data here is time-series data representing the temporal change of the optical signal reflecting the morphological characteristics of cell 4. The optical signal is a signal indicating the light intensity detected by the detection unit 32. The waveform data is, for example, waveform data representing the temporal change in the light intensity emitted from cell 4 acquired by the GC method. Since the optical signal from cell 4 acquired by the GC method contains compressed morphological information of the cell, the temporal change in the light intensity detected by the detection unit 32 changes according to the morphological characteristics of cell 4, such as size, shape, internal structure, density distribution, or color distribution. The light intensity from cell 4 also changes as the intensity of the structured illumination light changes over time as cell 4 moves within the irradiation area of ​​the channel 34. As a result, the intensity of light detected by the detection unit 32 changes over time, and as shown in Figure 3, the intensity of light that changes over time forms a waveform on the graph.

[0031] The waveform data obtained by structured illumination, which represents the time variation of the intensity of light modulated by cell 4, is waveform data that compresses and includes morphological information corresponding to the morphological characteristics of cell 4. For this reason, in flow cytometers using the GC method, machine learning that uses the waveform data directly as training data is used to distinguish morphologically different cells. It is also possible to generate an image of cell 4 from the waveform data obtained by structured illumination. The learning device 200 may also be configured to acquire waveform data individually for multiple types of modulated light emitted from a single cell 4.

[0032] The optical system 33 includes a lens 332 in addition to the spatial light modulation device 331. The lens 332 focuses light from the cells 4 and directs it into the detection unit 32. In addition to the spatial light modulation device 331 and the lens 332, the optical system 33 also includes optical components such as mirrors, lenses, and filters to structure the illumination light from the light source 31 and irradiate the cells 4, thereby directing the light from the cells 4 into the detection unit 32. Note that in Figure 2, optical components that may be included in the optical system 33 other than the spatial light modulation device 331 and the lens 332 are omitted. The waveform data representing the time change in the intensity of light from the cells 4 acquired by the detection unit 32 is observational data that includes morphological information of the cells.

[0033] The learning device 200 includes an information processing device 2. The information processing device 2 performs the information processing necessary for generating a classification model. The detection unit 32 is connected to the information processing device 2. The detection unit 32 outputs a signal to the information processing device 2 according to the intensity of the detected light, and the information processing device 2 receives the signal from the detection unit 32.

[0034] Figure 4 is a block diagram showing an example of the internal configuration of the information processing device 2. The information processing device 2 is a computer, such as a personal computer or a server device. The information processing device 2 comprises an arithmetic unit 21, a memory 22, a drive unit 23, a storage unit 24, an operation unit 25, a display unit 26, and an interface unit 27. The arithmetic unit 21 is configured using, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or a multi-core CPU. The arithmetic unit 21 may also be configured using a quantum computer. The memory 22 stores temporary data generated in connection with calculations. The memory 22 is, for example, RAM (Random Access Memory). The drive unit 23 reads information from a recording medium 20, such as an optical disc or portable memory.

[0035] The storage unit 24 is non-volatile and is, for example, a hard disk or a non-volatile semiconductor memory. The operation unit 25 accepts input of information such as text by receiving operations from the user. The operation unit 25 is, for example, a touch panel, keyboard, or pointing device. The display unit 26 displays an image. The display unit 26 is, for example, a liquid crystal display or an EL display (electroluminescent display). The operation unit 25 and the display unit 26 may be integrated. The interface unit 27 is connected to the detection unit 32. The interface unit 27 sends and receives signals to and from the detection unit 32.

[0036] The arithmetic unit 21 causes the drive unit 23 to read the computer program 241 recorded on the recording medium 20, and stores the read computer program 241 in the storage unit 24. The arithmetic unit 21 executes the necessary processing for the information processing device 2 according to the computer program 241. The computer program 241 may be downloaded from outside the information processing device 2. Alternatively, the computer program 241 may be pre-stored in the storage unit 24. In these cases, the information processing device 2 does not need to have a drive unit 23. The information processing device 2 may be composed of multiple computers.

[0037] The information processing device 2 includes a classification model 242. The classification model 242 is realized by the arithmetic unit 21 executing information processing according to a computer program 241. The storage unit 24 stores the data necessary to realize the classification model 242. The classification model 242 may be configured by hardware. The classification model 242 may be realized using a quantum computer. Alternatively, the classification model 242 may be located outside the information processing device 2, and the information processing device 2 may execute processing using the external classification model 242. For example, the classification model 242 may be configured as a cloud.

[0038] Figure 5 is a conceptual diagram illustrating the function of the classification model 242. The classification model 242 receives waveform data obtained from a single cell 4 as input. The classification model 242 is trained to output discrimination information indicating whether or not cell 4 is a cell with specific morphological characteristics when waveform data is input. For example, the classification model 242 is composed of a neural network.

[0039] The information processing device 2 executes a classification model generation method by performing a process to train the classification model 242. Figure 6 is a flowchart of an example of the procedure for training the classification model 242. Hereinafter, steps will be abbreviated as S. The arithmetic unit 21 performs the following processes according to the computer program 241. Before training the classification model 242, a plurality of first samples 111 and a plurality of second samples 121 are created. The information processing device 2 acquires waveform data obtained from each cell contained in the plurality of first samples 111 and waveform data obtained from each cell contained in the plurality of second samples 121 (S11).

[0040] During the S11 process, each cell 4 contained in the multiple first samples 111 is moved through the channel 34, and structured illumination light is irradiated onto the cells 4 using the light source 31 and the spatial light modulation device 331. The cells 4 emit light modulated by the cells themselves, such as scattered light, and the emitted light is detected by the detection unit 32. The detection unit 32 outputs a signal corresponding to the intensity of the detected light to the information processing device 2, and the information processing device 2 receives the signal from the detection unit 32 at the interface unit 27. The calculation unit 21 acquires waveform data by generating waveform data representing the time change in the intensity of the light detected by the detection unit 32 based on the signal from the detection unit 32. In this way, waveform data is acquired by the information processing device 2 for each cell 4 contained in the multiple first samples 111. Similarly, waveform data is acquired by the information processing device 2 for each cell 4 contained in the multiple second samples 121. The calculation unit 21 stores the acquired waveform data in the storage unit 24. The S11 process corresponds to the data acquisition unit.

[0041] The information processing device 2 then generates training data for learning (S12). The training data includes waveform data and information indicating whether the waveform data was obtained from cells contained in the first sample 111 or the second sample 121. Waveform data obtained from cells contained in the first sample 111 is associated with information indicating that the waveform data was obtained from cells contained in the first sample 111. Waveform data obtained from cells contained in the second sample 121 is associated with information indicating that the waveform data was obtained from cells contained in the second sample 121. In S12, the calculation unit 21 generates training data by associating each waveform data with information indicating whether the waveform data was obtained from cells contained in the first sample 111 or the second sample 121. Waveform data obtained from multiple cells contained in the first sample 111 may also be associated with information indicating that they were obtained from the first sample 111, regardless of whether they are positive or negative cells, and this information may be included in the training data. Similarly, waveform data obtained from multiple cells contained in the second sample 121 may be associated with information indicating that it was obtained from the second sample 121, and this information may be included in the training data. The calculation unit 21 stores the training data in the storage unit 24.

[0042] The information processing device 2 then performs training on the classification model 242 (S13). In S13, the arithmetic unit 21 performs training using the MIL (Multiple-Instance Learning) method. The MIL method is disclosed, for example, in Marc-Andre Carbonneau, et al. "Multiple instance learning: A survey of problem characteristics and applications", Pattern Recognition, Volume 77, May 2018, Pages 329-353. Figure 7 is a schematic diagram showing an example of training using the concept of MIL. Assume that the characteristics of cells are plotted on a two-dimensional coordinate system. In Figure 7, positive cells with specific morphological characteristics are shown with double circles, and negative cells are shown with single circles. Multiple cells enclosed by solid lines are multiple cells contained in the same first sample 111. On the other hand, multiple cells enclosed by dashed lines are multiple cells contained in the same second sample 121. The dashed line in Figure 7 indicates the decision boundary when positive and negative cells are classified based on certain morphological characteristics. In MIL, the classification model is trained to appropriately define boundaries. A more appropriate decision boundary is determined through learning, such that all of the multiple cells in each second sample 121 become negative cells, and some cells with certain morphological characteristics in each first sample 111 become positive cells, while the others become negative cells.

[0043] In S13, the calculation unit 21 adjusts the calculation parameters of the classification model 242 so that when waveform data obtained from cells contained in the second sample 121 is input to the classification model 242, discrimination information indicating that the cells are not positive cells possessing specific morphological characteristics is output. Furthermore, the calculation unit 21 adjusts the calculation parameters of the classification model 242 so that when some waveform data from multiple cells contained in a single first sample 111 is input to the classification model 242, discrimination information indicating that the cells are positive cells is output, and when other waveform data is input to the classification model 242, discrimination information indicating that the cells are not positive cells is output.

[0044] The calculation unit 21 learns the classification model 242 by repeatedly adjusting the calculation parameters of the classification model 242 using training data. If the classification model 242 is a neural network, the calculation parameters of each node are adjusted. The classification model 242 is learned to output discrimination information indicating that a cell is a positive cell when waveform data obtained from some cells in the first sample 111 is input, and to output discrimination information indicating that a cell is not a positive cell when waveform data obtained from other cells in the first sample 111 and cells in the second sample 121 is input. The calculation unit 21 stores the learned data, which records the final adjusted parameters, in the storage unit 24. In this way, the learned classification model 242 is generated. The process in S13 corresponds to the classification model generation unit. After S13 is completed, the information processing device 2 terminates the process of learning the classification model 242.

[0045] The particle detection method is performed by generating a trained classification model 242 and determining whether or not a cell is a positive cell that possesses specific morphological characteristics. In the particle detection method, a sample collected from a person is used as the sample, and the classification model 242 is used to determine whether or not the cells contained in the sample are positive cells. A person in whom cells determined to be positive cells are found in the sample is identified as a positive person. Figure 8 is a conceptual diagram showing an overview of the particle detection method. For example, a sample is collected from a subject 13, such as a person undergoing cancer screening, and the sample is used as the test sample 131. The specimens include blood, fractions containing specific cells derived from blood, urine, bone marrow fluid or other bodily fluids, or washing fluid used to cleanse an affected area, and all of these contain cells. A test sample 131 contains one or more cells. Multiple test samples 131 are prepared by collecting specimens from multiple subjects 13. The symbols in parentheses in Figure 8 refer to individual subjects 13 and test samples 131. Specifically, in the example shown in Figure 8, test sample 131a is collected from subject 13a, and test sample 131b is collected from subject 13b.

[0046] Next, a tag containing identification information to identify each subject 13 is attached to the cells contained in the test sample 131. For example, in the example shown in Figure 8, a tag containing identification information to identify subject 13a is attached to the cells contained in the test sample 131a. The tag containing identification information links each cell contained in the test sample 131 to the subject 13 from whom they were collected, and it is desirable that the tag attaches to or binds to the cell itself. The tag is preferably a substance containing components formed by the linking of several types of components, such as a known peptide tag or a DNA (deoxyribonucleic acid) tag. The sequence of components contained in the tag indicates the identification information. The tag is more preferably a DNA tag; for example, in a DNA tag, the base sequence indicates the identification information. The identification information of the tag attached to the cell is different for each test sample 131 and is associated with the subject 13. For example, in the example shown in Figure 8, test sample 131a is associated with subject 13a. For example, as DNA tags that confer identification information to cells, lipid- or cholesterol-modified oligonucleotides, such as those described in WO2020 / 010366, can be used.

[0047] Next, the classification model 242 is used to determine whether each cell in the test sample 131 is a positive cell or not. If the classification model 242 determines that a cell is a positive cell, then the subject 13 from whom the test sample 131 containing the positive cell was collected is identified based on the identification information of the tag attached to that positive cell, and the identified subject 13 is determined to be positive. For example, in the example shown in Figure 8, the subject 13a from whom the test sample 131a containing the positive cell was collected is determined to be positive.

[0048] Figure 9 is a block diagram showing an example configuration of a determination device 500 according to Embodiment 1 for determining cells. The determination device 500 is equipped with a cell channel 64. Cells 4 move sequentially through the channel 64. The determination device 500 is equipped with a light source 61, a detection unit 62, and an optical system 63. The configuration of the light source 61, detection unit 62, and optical system 63 in Figure 9 is the same as the configuration of the light source 31, detection unit 32, and optical system 33 in Figure 2. The light source 61 is, for example, a laser light source or an LED light source, and the detection unit 62 has a photodetection sensor such as a photomultiplier tube, a line-type PMT element, a photodiode, an APD, or a semiconductor photosensor. In Figure 9, the path of light is shown by solid arrows.

[0049] The optical system 63 guides light from the light source 61 to the cells 4 in the channel 64 and causes the light from the cells 4 to enter the detection unit 62. The optical system 63 includes a spatial light modulation device 631 and a lens 632. Light from the light source 61 passes through the spatial light modulation device 631 before being irradiated onto the cells 4. This constitutes structured illumination. The determination device 500 can acquire waveform data representing the time change in the intensity of the light emitted from and modulated by the cells 4. The waveform data is, for example, used in the GC method and represents the morphological characteristics of the cells 4. The determination device 500 may be configured to acquire multiple waveform data for a single cell 4.

[0050] In addition to the spatial light modulation device 631 and lens 632, the optical system 63 also includes optical components such as mirrors, lenses, and filters to irradiate the cells 4 with light from the light source 61 and to cause the light from the cells 4 to enter the detection unit 62. In Figure 9, optical components other than the spatial light modulation device 631 and lens 632 are omitted. The waveform data representing the time change in the intensity of light from the cells 4 acquired by the detection unit 62 is observational data that includes morphological information of the cells.

[0051] A sorter 65 is connected to the channel 64. The sorter 65 separates specific cells from the cells 4 that have moved through the channel 64. For example, if the sorter 65 determines that the cells 4 moving through the channel 64 are specific cells, it applies a charge and voltage to the moving cells 4, thereby separating the cells 4 by changing their movement path. Alternatively, the sorter 65 may be configured to generate a pulsed flow when the cells 4 have flowed to the sorter 65, thereby separating and acquiring specific cells 4 by changing their movement path.

[0052] The determination device 500 is equipped with an information processing device 5. The information processing device 5 performs the information processing necessary for determining the cells 4. The detection unit 62 is connected to the information processing device 5. The detection unit 62 outputs a signal to the information processing device 5 according to the intensity of the detected light, and the information processing device 5 receives the signal from the detection unit 62. The sorter 65 is connected to the information processing device 5 and is controlled by the information processing device 5. The sorter 65 sorts the cells according to the control of the information processing device 5.

[0053] Figure 10 is a block diagram showing an example of the internal configuration of the information processing device 5. The information processing device 5 is a computer such as a personal computer or a server device. The information processing device 5 comprises an arithmetic unit 51, a memory 52, a drive unit 53, a storage unit 54, an operation unit 55, a display unit 56, and an interface unit 57. The arithmetic unit 51 is configured using, for example, a CPU, a GPU, or a multi-core CPU. The arithmetic unit 51 may also be configured using a quantum computer. The memory 52 stores temporary data generated in connection with calculations. .example For example, the drive unit 53 reads information from the recording medium 50, such as an optical disc.

[0054] The storage unit 54 is non-volatile and is, for example, a hard disk or a non-volatile semiconductor memory. The operation unit 55 accepts input of information such as text by receiving operations from the user. The operation unit 55 is, for example, a touch panel, keyboard, or pointing device. The display unit 56 displays an image. The display unit 56 is, for example, a liquid crystal display or an EL display. The operation unit 55 and the display unit 56 may be integrated. The interface unit 57 is connected to the detection unit 62 and the sorter 65. The interface unit 57 sends and receives signals to and from the detection unit 62 and the sorter 65.

[0055] The arithmetic unit 51 causes the drive unit 53 to read the computer program 541 recorded on the recording medium 50, and stores the read computer program 541 in the storage unit 54. The arithmetic unit 51 executes the necessary processing for the information processing device 5 according to the computer program 541. The computer program 541 may be downloaded from outside the information processing device 5. Alternatively, the computer program 541 may be pre-stored in the storage unit 54. In these cases, the information processing device 5 does not need to have a drive unit 53. The information processing device 5 may be composed of multiple computers.

[0056] The information processing device 5 includes a classification model 242. The classification model 242 is realized by the arithmetic unit 51 executing information processing according to a computer program 541. The classification model 242 is a classification model learned by the learning device 200. The information processing device 5 includes the classification model 242 by storing learned data, which records the parameters of the classification model 242 learned by the learning device 200, in the storage unit 54. For example, the learned data is read from the recording medium 50 by the drive unit 53 or downloaded. The classification model 242 may be configured by hardware. The classification model 242 may be realized using a quantum computer. Alternatively, the classification model 242 may be located outside the information processing device 5, and the information processing device 5 may execute processing using the external classification model 242. For example, the classification model 242 may be configured in the cloud.

[0057] In addition, the classification model 242 may be implemented using an FPGA (Field Programmable Gate Array) in the information processing device 5. The FPGA circuit is configured based on the parameters of the classification model 242 learned by the classification model generation method, and the FPGA executes the processing of the classification model 242.

[0058] Figure 11 is a flowchart showing the steps of the process performed by the information processing device 5 to determine the cells. The calculation unit 51 performs the following processes according to the computer program 541. As described above, a test sample 131 is prepared, and tags are attached to the cells contained in the test sample 131. The tagged cells 4 are moved through the channel 64. The information processing device 5 acquires waveform data obtained from the cells 4 moving through the channel 64 (S21). Structured illumination light is shone on the cells 4 moving through the channel 64, and the light emitted from the cells 4 is detected by the detection unit 62. The detection unit 62 outputs a signal corresponding to the detection to the information processing device 5, and the information processing device 5 receives the signal. In S21, the calculation unit 51 generates waveform data representing the time change in the intensity of the light detected by the detection unit 62 based on the signal from the detection unit 62, and stores the waveform data in the storage unit 54. The process in S21 corresponds to the observation unit.

[0059] The information processing device 5 inputs the acquired waveform data to the classification model 242 (S22). In S22, the calculation unit 51 inputs the waveform data to the classification model 242 and causes the classification model 242 to perform processing. The calculation unit 51 does not input information indicating whether the waveform data was obtained from cells contained in the first sample or the second sample. In response to the input of waveform data, the classification model 242 performs a process to output discrimination information indicating whether or not cell 4 is a positive cell having specific morphological characteristics. The calculation unit 51 acquires the discrimination information output by the classification model 242. The processing in S22 corresponds to the discrimination information acquisition unit. Based on the discrimination information output by the classification model 242, the information processing device 5 determines whether or not cell 4 is a positive cell (S23). In S23, the calculation unit 51 determines that cell 4 is a positive cell if the discrimination information indicates that cell 4 is a positive cell, and determines that cell 4 is not a positive cell if the discrimination information indicates that cell 4 is not a positive cell. The calculation unit 51 can also store information indicating the determination result in the storage unit 54, as needed, in association with the waveform data. The processing in S23 corresponds to the determination unit.

[0060] The information processing device 5 then displays the determination result on the display unit 56 (S24). Figure 12 is a schematic diagram showing an example of the display of the determination result. On the display unit 56, for example, the waveform data is displayed in the form of a graph, and the determination result of whether or not cell 4 is a positive cell is displayed in text. Figure 12 shows an example in which cell 4 is determined to be a positive cell. In S24, the calculation unit 51 may also read the determination result from the storage unit 54, generate an image representing the waveform data and the determination result, and display it on the display unit 56. The calculation unit 51 may generate an image from the waveform data and the determination result and display it on the display unit 56 without storing the information indicating the determination result in the storage unit 54. The calculation unit 51 may also generate an image of cell 4 based on the waveform data and display the image of cell 4 on the display unit 56. The calculation unit 51 may also display on the display unit 56 information about the test sample 131 in which cell 4 was contained, or information about the person 13 from whom the test sample 131 was taken. Information regarding the identified cell 4 is displayed, allowing the user to review the information about cell 4. Note that S24 may be omitted.

[0061] Next, if the information processing device 5 determines that cell 4 is not a positive cell (S25: NO), it terminates the process for determining the cell. If cell 4 is determined to be a positive cell (S25: YES), the information processing device 5 sorts cell 4 using the sorter 65 (S26). In S26, the calculation unit 51 sends a control signal from the interface unit 57 to the sorter 65 to cause the sorter 65 to sort cell 4. The sorter 65 sorts cell 4 according to the control signal. For example, when cell 4 has flowed through the channel 64 to the sorter 65, the sorter 65 applies a charge to cell 4 and applies a voltage to change the movement path of cell 4, thereby sorting cell 4. After S26 is completed, the information processing device 5 terminates the process for determining the cell.

[0062] Processes S21 to S26 are performed on each of the cells contained in the multiple test samples 131. The cells separated by the processes S21 to S26 are positive cells with specific morphological characteristics. The identification information of the tags attached to the separated cells is analyzed, and the subject 13 associated with the identification information is identified. At that time, by further biochemical or genetic testing of the separated cells, additional detailed information about the separated cells can be obtained. The identified subject 13 is the subject from whom the test sample 131 containing positive cells was collected. In this way, positive individuals are identified. If the positive cells are cells that have specific morphological characteristics due to the effect of a specific disease, the positive individual is determined to be a patient with the specific disease. In this way, patients with specific diseases such as cancer can be found among multiple subjects 13.

[0063] As detailed above, in this embodiment, a classification model is trained using training data that includes waveform data acquired for each of the cells contained in the first and second samples. The first sample contains positive cells with specific morphological characteristics and other negative cells. The second sample contains no positive cells and consists only of negative cells. The classification model outputs discrimination information indicating whether or not a cell is a positive cell when waveform data is input. Positive cells are rare, and it is not easy to acquire a large amount of waveform data of positive cells as training data. However, it is possible to generate a classification model by using training data that includes information indicating which of the first and second samples the waveform data was obtained from, and by learning using the MIL method. Using the classification model, discrimination information corresponding to the waveform data can be obtained, and it is possible to determine whether or not a cell is a positive cell based on the discrimination information. Therefore, it becomes possible to distinguish even the rare positive cells.

[0064] If positive cells are cells that can be collected from a person with a specific disease such as cancer, then determining whether or not the cells are positive allows for the determination of whether or not the person from whom the cells were collected has the specific disease. In this embodiment, a tag with identification information that identifies the person from whom the cells were collected is attached to the cells, and the person from whom the cells were collected is identified based on the tag. If the cells are determined to be positive, the person from whom the positive cells were collected can be identified based on the tag attached to the positive cells. Therefore, it is also possible to identify a person with a specific disease. In this embodiment, a method in which a tag with identification information is attached to the cells and the person from whom the cells were collected is identified based on the tag has been described, but the method of identifying the person from whom the cells were collected is not limited to this. As another method, for example, the cells contained in the test sample are observed sequentially for each person from whom the cells were collected, the test sample is identified by the observation order, and the person from whom the cells were collected is identified.

[0065] <Embodiment 2> Figure 13 is a block diagram showing an example configuration of the learning device 200 according to Embodiment 2. In Embodiment 2, the configuration of the optical system 33 is different from that of Embodiment 1 shown in Figure 2. The configuration of parts other than the optical system 33 is the same as in Embodiment 1. Light from the light source 31 is irradiated onto the cell 4 without passing through the spatial light modulation device 331. Light from the cell 4 passes through the spatial light modulation device 331, is focused by the lens 332, and enters the detection unit 32. The detection unit 32 detects the modulated light that has been structured by the spatial light modulation device 331 after being modulated by the light from the cell 4. This configuration, in which the modulated light from the cell 4 is structured by the spatial light modulation device 331 in the optical path from the cell 4 to the detection unit 32, is also referred to as structured detection. The modulated light from the cell 4 detected by the detection unit 32 has an intensity that changes over time due to the spatial light modulation device 331. The waveform data representing the time change in light intensity from cells 4 detected by the detection unit 32 through structured detection includes compressed morphological information of cells 4, similar to the case of structured illumination described above. In other words, the waveform data representing the time change in light intensity from cells 4 detected by the detection unit 32 is observational data that includes morphological information of the cells.

[0066] In Embodiment 2, the learning device 200 can also acquire waveform data representing the temporal change of light emitted from the cell 4. Similar to Embodiment 1, the waveform data represents the morphological characteristics of the cell 4. The optical system 33 has optical components other than the spatial light modulation device 331 and the lens 332. In structured detection, the optical component used in structured illumination in Embodiment 1 can be used similarly as the spatial light modulation device 331. In Figure 13, the description of optical components other than the spatial light modulation device 331 and the lens 332 is omitted. In Embodiment 2, as in Embodiment 1, the information processing device 2 generates a classification model 242 by executing the processes S11 to S13.

[0067] Figure 14 is a block diagram showing an example configuration of the determination device 500 according to Embodiment 2. In Embodiment 2, the configuration of the optical system 63 is different from that of Embodiment 1 shown in Figure 9. The configuration of parts other than the optical system 63 is the same as in Embodiment 1. Light from the light source 61 is irradiated onto the cell 4 without passing through the spatial light modulation device 631. Light from the cell 4 passes through the spatial light modulation device 631, is focused by the lens 632, and enters the detection unit 62. The detection unit 62 detects the modulated light, which has been structured by the light modulated by the cell 4 passing through the spatial light modulation device 631. The waveform data representing the time change in the intensity of the light from the cell 4 detected by the detection unit 62 through structured detection includes compressed morphological information of the cell 4. The intensity of the light from the cell 4 changes according to the morphological characteristics of the cell 4 and also changes by passing through the spatial light modulation device 631. That is, the waveform data representing the time change in the intensity of the light from the cell 4 detected by the detection unit 62 is observation data that includes morphological information of the cell.

[0068] In Embodiment 2, which employs a structured detection configuration, the determination device 500 can acquire waveform data representing the temporal change of light emitted from the cell 4. Similar to Embodiment 1, the waveform data represents the morphological characteristics of the cell 4. The optical system 63 has optical components other than the spatial light modulation device 631 and the lens 632. In Figure 14, the description of optical components other than the spatial light modulation device 631 and the lens 632 is omitted. In Embodiment 2, as in Embodiment 1, the information processing device 5 performs cell discrimination and sorting by executing the processes S21 to S26. Note that one of the learning device 200 and the determination device 500 may have the same configuration as in Embodiment 1.

[0069] In Embodiment 2, too, it is possible to generate a classification model using waveform data acquired for each cell contained in the first and second samples, and training data that includes information indicating which cell in the first or second sample the waveform data was obtained from. Using the classification model, discrimination information corresponding to the waveform data can be obtained, and it is possible to determine whether or not a cell is a positive cell with specific morphological characteristics. This makes it possible to distinguish positive cells, which are rare.

[0070] <Embodiment 3> Embodiments 1 and 2 show examples where the observation data representing the result of observing cells is waveform data containing morphological information of the cells, but Embodiment 3 shows an example where the observation data containing morphological information of cells is an image captured of the cells. Figure 15 is a block diagram showing an example configuration of the learning device 200 according to Embodiment 3. In Embodiment 3, the learning device 200 does not have a detection unit 32 and an optical system 33, but has an imaging unit 35 that captures the cells 4 moving in the channel 34. Light from the light source 31 illuminates the cells 4, light from the cells 4 enters the imaging unit 35, and the imaging unit 35 creates an image of the cells 4. For example, the imaging unit 35 is a camera having a semiconductor image sensor. The semiconductor image sensor is, for example, a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor. The learning device 200 may also have an optical system (not shown) that guides light for imaging to the cells 4 and efficiently causes the light to enter the imaging unit 35. The imaging unit 35 is connected to the information processing device 2. The interface unit 27 is connected to the imaging unit 35. The interface unit 27 transmits and receives signals to and from the imaging unit 35. The other configurations of the learning device 200 are the same as in Embodiment 1.

[0071] In Embodiment 3, the learning device 200 can acquire images of cell 4 using the imaging unit 35. The images are observational data representing the morphological characteristics of cell 4. In Embodiment 3, the classification model 242 is trained to output discrimination information indicating whether or not cell 4 is a cell with specific morphological characteristics when image data is input.

[0072] In Embodiment 3, the information processing device 2 performs the processes S11 to S13 in the same manner as in Embodiment 1. In S11, the information processing device 2 acquires captured images instead of waveform data. In S12, the information processing device 2 generates training data that includes the captured images and information indicating whether the captured images were obtained from cells contained in the first sample 111 or the second sample 121. Each captured image is associated with information indicating whether the captured images were obtained from cells contained in the first sample 111 or the second sample 121.

[0073] In S13, the information processing device 2 adjusts the calculation parameters of the classification model 242 so that when captured images obtained from cells contained in the second sample 121 are input to the classification model 242, discrimination information indicating that the cells are not positive cells possessing specific morphological characteristics is output. The calculation unit 21 also adjusts the calculation parameters of the classification model 242 so that when some captured images obtained from multiple cells contained in a single first sample 111 are input to the classification model 242, discrimination information indicating that the cells are positive cells is output, and when other captured images are input to the classification model 242, discrimination information indicating that the cells are not positive cells is output. As described above, the classification model 242 is generated by the processing in S11 to S13.

[0074] Figure 16 is a block diagram showing an example configuration of the determination device 500 according to Embodiment 3. In Embodiment 3, the determination device 500 does not include a detection unit 62 and an optical system 63, but includes an imaging unit 66 that photographs cells 4 moving in a channel 64. Light from a light source 61 illuminates the cells 4, light from the cells 4 enters the imaging unit 66, and the imaging unit 66 creates an image of the cells 4. The determination device 500 may include an optical system (not shown) that guides light for imaging to the cells 4 and efficiently directs the light into the imaging unit 66. The imaging unit 66 is connected to an information processing device 5. The interface unit 57 is connected to the imaging unit 66 and the sorter 65. The interface unit 27 transmits and receives signals between the imaging unit 66 and the sorter 65. The other configurations of the determination device 500 are the same as in Embodiment 1.

[0075] In Embodiment 3, the determination device 500 can acquire images of cells 4 using the imaging unit 66. In Embodiment 3, as in Embodiment 1, the information processing device 5 performs the processes S21 to S26. In S21, the information processing device 5 acquires images instead of waveform data. In S22, the information processing device 5 inputs the images instead of waveform data into the classification model as observational data representing the morphological characteristics of cells 4. In S24, the information processing device 5 displays the images and determination results on the display unit 56. The information processing device 5 performs cell discrimination and sorting by executing the processes S21 to S26.

[0076] In Embodiment 3, a classification model can be generated using training data that includes images taken of cells contained in the first and second samples, and information indicating which sample (first or second) the image was taken from. Using the classification model, discrimination information corresponding to the image can be obtained, and it can be determined whether or not a cell is a positive cell with specific morphological characteristics. This makes it possible to distinguish positive cells, which are rare.

[0077] In embodiments 1 to 3 described above, the first sample 111, the second sample 121, and the test sample 131 were assumed to be specimens collected from humans. However, the first sample 111, the second sample 121, and the test sample 131 may be specimens collected from sources other than humans. Alternatively, the first sample 111, the second sample 121, and the test sample 131 may be specimens prepared by methods other than specimen collection. In embodiments 1 to 3, the determination device 500 was shown to be equipped with a sorter 65 for separating cells. However, the determination device 500 may be in a configuration that does not include a sorter 65. In this configuration, the information processing device 5 omits the processing in S25 and S26. In embodiments 1 to 3, the learning device 200 and the determination device 500 were shown to be different configurations. However, some or all of the learning device 200 and the determination device 500 may be common. For example, the information processing device 2 may also be used as the information processing device 5.

[0078] In Embodiments 1 to 3, examples where the particles are cells were described, but the classification model generation method and particle determination method may also handle particles other than cells. The particles are not limited to biological particles. For example, the particles targeted by the classification model generation method and particle classification method may be microorganisms such as bacteria, yeast or plankton, tissues within living organisms, organs within living organisms, or fine particles such as beads, pollen or particulate matter.

[0079] The present invention is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. That is, embodiments obtained by combining technical means that have been appropriately modified within the scope of the claims are also included in the technical scope of the present invention. [Explanation of Symbols]

[0080] 11 patients 111 Sample 1 12 Healthy people 121 Sample 2 13. Person under examination 131 Sample for examination 200 Learning Devices 500 Judgment device 2.5 Information Processing Devices 20, 50 recording media 241, 541 Computer Programs 242 Classification Models 31, 61 light source 32, 62 Detection unit 33, 63 Optical system 331, 631 Spatial Light Modulation Devices 34, 64 channels 35, 66 Photography Department 4 cells

Claims

1. Observational data representing the results of observing each particle in a first sample containing particles having specific morphological characteristics and other particles, and a second sample containing particles that do not have the specific morphological characteristics and consist of other particles, is obtained. By learning using the aforementioned observation data and training data that includes information indicating whether the observed data was obtained from particles contained in the first sample or the second sample, a classification model is generated that outputs discrimination information indicating whether or not a particle has the aforementioned specific morphological characteristics when observation data representing the result of observing a particle is input. In the learning process, the classification model is trained to define a decision boundary for classifying particles such that some particles in the first sample are classified as particles having the specific morphological characteristics, particles other than those in the first sample are classified as particles not having the specific morphological characteristics, and particles in the second sample are classified as particles not having the specific morphological characteristics. A classification model generation method characterized by the following.

2. The aforementioned observation data is waveform data representing the time change in the intensity of light emitted from particles irradiated with light by structured illumination, or waveform data representing the time change in the intensity of light detected by structuring the light from irradiated particles. A method for generating a classification model according to claim 1, characterized by the above.

3. The first sample is a specimen collected from a person with a specific disease, The second sample is a specimen collected from a person who does not have the aforementioned specific disease. A method for generating a classification model according to claim 1 or 2, characterized by the above.

4. Obtain observational data representing the result of observing the particle, The acquired observational data is input to a classification model that outputs discrimination information indicating whether or not a particle has specific morphological characteristics when observational data representing the result of observing a particle is input, and the discrimination information output by the classification model is obtained. Based on the acquired discrimination information, a determination is made as to whether or not the particle relating to the observation data is a particle having the specific morphological characteristics. The classification model is trained by training data that includes observational data representing the results of observing each particle in a first sample containing particles having the specific morphological characteristics and other particles, and a second sample consisting of other particles but not containing particles having the specific morphological characteristics, and information indicating which of the first and second samples the observational data was obtained from. In the aforementioned training, the classification model is trained to define a decision boundary for classifying particles such that some particles in the first sample are classified as particles having the specific morphological characteristics, particles other than the aforementioned some particles in the first sample are classified as particles not having the specific morphological characteristics, and particles in the second sample are classified as particles not having the specific morphological characteristics. A particle detection method characterized by the following.

5. Output information regarding the particle for which the above determination was made. The particle determination method according to claim 4, characterized by the above.

6. The particles that are the target of the aforementioned observation data acquisition are collected from people, A tag having identification information to identify the person from whom the particles were collected is attached to the particles. If the particles related to the observation data are particles having the specific morphological characteristics, the person who collected the particles is identified based on the identification information of the tag attached to the particles related to the observation data. A particle determination method according to claim 4 or 5, characterized by the above.

7. Observational data representing the results of observing each particle in a first sample containing particles having specific morphological characteristics and other particles, and a second sample containing particles that do not have the specific morphological characteristics and consist of other particles, is obtained. By learning using the aforementioned observation data and training data that includes information indicating whether the observed data was obtained from particles contained in the first sample or the second sample, a classification model is generated that outputs discrimination information indicating whether or not a particle has the aforementioned specific morphological characteristics when observation data representing the result of observing a particle is input. In the learning process described above, the classification model is trained to define a decision boundary for classifying particles such that some particles in the first sample are classified as particles having the specific morphological characteristics, the particles in the first sample other than the aforementioned some particles are classified as particles not having the specific morphological characteristics, and the particles in the second sample are classified as particles not having the specific morphological characteristics. A computer program characterized by causing a computer to perform a process.

8. Obtain observational data representing the result of observing the particle, The acquired observational data is input to a classification model that outputs discrimination information indicating whether or not a particle has specific morphological characteristics when observational data representing the result of observing a particle is input, and the discrimination information output by the classification model is obtained. Based on the acquired discrimination information, it is determined whether or not the particle relating to the observation data is a particle having the specific morphological characteristics. Let the computer perform the process, The classification model is trained by training data that includes observational data representing the results of observing each particle in a first sample containing particles having the specific morphological characteristics and other particles, and a second sample consisting of other particles but not containing particles having the specific morphological characteristics, and information indicating which of the first and second samples the observational data was obtained from. In the aforementioned training, the classification model is trained to define a decision boundary for classifying particles such that some particles in the first sample are classified as particles having the specific morphological characteristics, particles other than the aforementioned some particles in the first sample are classified as particles not having the specific morphological characteristics, and particles in the second sample are classified as particles not having the specific morphological characteristics. A computer program characterized by the following.

9. A data acquisition unit acquires observational data representing the results of observing each particle in a first sample containing particles having specific morphological characteristics and other particles, and a second sample containing particles that do not have the specific morphological characteristics and consist of other particles. The system includes a classification model generation unit that generates a classification model that, when inputting observation data representing the result of observing a particle, outputs discrimination information indicating whether or not the particle has the specific morphological characteristics, by learning using training data that includes the observation data and information indicating whether or not the observation data was obtained from particles contained in the first sample or the second sample. The classification model generation unit learns the classification model in order to determine the decision boundary for classifying particles such that some particles in the first sample are classified as particles having the specific morphological characteristics, particles other than the some particles in the first sample are classified as particles that do not have the specific morphological characteristics, and particles in the second sample are classified as particles that do not have the specific morphological characteristics. An information processing device characterized by the following:

10. An observation unit that acquires observational data representing the results of observing particles, A classification information acquisition unit inputs the acquired observation data into a classification model that outputs discrimination information indicating whether or not a particle has specific morphological characteristics when observation data representing the result of observing a particle is input, and acquires the discrimination information output by the classification model. The system includes a determination unit that determines, based on the acquired discrimination information, whether or not the particles relating to the observation data are particles having the specific morphological characteristics, The classification model is trained by training data that includes observational data representing the results of observing each particle in a first sample containing particles having the specific morphological characteristics and other particles, and a second sample consisting of other particles but not containing particles having the specific morphological characteristics, and information indicating which of the first and second samples the observational data was obtained from. In the aforementioned training, the classification model is trained to define a decision boundary for classifying particles such that some particles in the first sample are classified as particles having the specific morphological characteristics, particles other than the aforementioned some particles in the first sample are classified as particles not having the specific morphological characteristics, and particles in the second sample are classified as particles not having the specific morphological characteristics. An information processing device characterized by the following.

11. The number of particles having the specific morphological features in the first sample is less than the number of other particles. A method for generating a classification model according to claim 1, characterized by the above.

12. The learning process involves learning the classification model using the Multiple-Instance Learning (MIL) method. A method for generating a classification model according to claim 1, characterized by the above.

13. The number of particles having the specific morphological characteristics in the first sample is less than the number of other particles. The particle determination method according to claim 4, characterized by the above.

14. The number of particles having the specific morphological characteristics in the first sample is less than the number of other particles. A computer program according to claim 7 or 8, characterized by the following:

15. The number of particles having the specific morphological characteristics in the first sample is less than the number of other particles. The information processing apparatus according to claim 9 or 10, characterized by the above.

Images