Cell species identification method, cell species identification system, and model generation method

Abstract

To discriminate the kind of a hematopoietic cell with small mechanical stress to the hematopoietic cell at a low cost.SOLUTION: The cell-type determination method includes an acquiring step (S01) of acquiring a refraction index indicating a three dimensional refraction index including a portion of hematopoietic cells, a feature-value calculating step (S02) of extracting a refraction index of a portion where the refraction index indicated by the refraction index acquired in the acquiring step is larger than a preset refraction index as a refraction index related to the hematopoietic cells and calculating a feature value of the extracted portion, and a type determining step (S03) of determining a type of the hematopoietic cells from the feature value calculated in the feature-value calculating step based on a determination model prepared in advance.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a cell type discrimination method and a cell type discrimination system for discriminating the type of hematopoietic cells, and a model generation method for generating a discrimination model used for said discrimination. [Background technology]

[0002] Conventionally, various cell analysis and isolation techniques based on the principle of surface antigen combinations have been established, such as FACS (fluorescence-activated cell sorting) or methods using beads (see, for example, Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Beili Zhu, Shashi K Murthy, Stem Cell Separation Technologies, Curr Opin Chem Eng.2013 Feb 1;2(1):3-7. Summary of the Invention [Problem to be solved by the invention]

[0004] The method disclosed in Non-Patent Document 1 can distinguish the types of hematopoietic stem cells and hematopoietic cells that differentiate from hematopoietic stem cells. However, the method disclosed in Non-Patent Document 1 requires labeling of cells, which is costly and time-consuming. Furthermore, mechanical stress is applied to cells during the isolation process, resulting in cell damage and cell loss. Furthermore, when FACS is used, the equipment costs are high.

[0005] The present invention has been made in consideration of the above, and aims to provide a cell type discrimination method, a cell type discrimination system, and a model generation method that can discriminate the type of hematopoietic cells with little mechanical stress on the hematopoietic cells and at low cost. [Means for solving the problem]

[0006] In order to achieve the above-mentioned object, the cell type discrimination method of the present invention includes an acquisition step of acquiring refractive index data indicating three-dimensional refractive index including a portion of hematopoietic cells; a feature calculation step of extracting, as a portion of hematopoietic cells, refractive indexes of portions where the refractive index indicated by the refractive index data acquired in the acquisition step is greater than a predetermined refractive index, and calculating feature amounts of the extracted portions; and a type discrimination step of discriminating the type of hematopoietic cells from the feature amounts calculated in the feature calculation step based on a discrimination model prepared in advance.

[0007] The cell type discrimination method according to the present invention allows the type of hematopoietic cells to be discriminated by obtaining refractive index data that indicates the three-dimensional refractive index including the hematopoietic cell portion. Therefore, there is no need to label the hematopoietic cells to be discriminated or to perform an isolation process for the hematopoietic cells to be discriminated. Therefore, the cell type discrimination method according to the present invention allows the type of hematopoietic cells to be discriminated at low cost and with little mechanical stress on the hematopoietic cells.

[0008] In the feature amount calculation step, sphericity may be calculated as the feature amount. With this configuration, the type of hematopoietic cell can be more appropriately determined.

[0009] The preset refractive index may be greater than the refractive index of the background. Alternatively, the preset refractive index may be greater than the refractive index of the background by 0.005. These configurations allow for more appropriate extraction of the portion relating to hematopoietic cells from the refractive index data, thereby enabling more appropriate identification of the type of hematopoietic cells.

[0010] Incidentally, the present invention can be described not only as an invention of a cell type discrimination method as described above, but also as an invention of a cell type discrimination system as described below. These are essentially the same inventions, just in different categories, and have similar functions and effects.

[0011] That is, the cell type discrimination system of the present invention comprises an acquisition means for acquiring refractive index data indicating the three-dimensional refractive index including a portion of hematopoietic cells, a feature calculation means for extracting, as a portion of hematopoietic cells, the refractive index of a portion where the refractive index indicated by the refractive index data acquired by the acquisition means is greater than a predetermined refractive index, and calculating the feature values ​​of the extracted portion, and a type discrimination means for discriminating the type of hematopoietic cells from the feature values ​​calculated by the feature calculation means based on a discrimination model prepared in advance.

[0012] Furthermore, in order to achieve the above-mentioned object, the model generation method of the present invention includes: a model generation acquisition step of acquiring refractive index data for model generation indicating the three-dimensional refractive index including a portion of hematopoietic cells for model generation, and information indicating the type of the hematopoietic cells for model generation; a model generation feature calculation step of extracting, as a refractive index related to the hematopoietic cells for model generation, a portion where the refractive index indicated by the refractive index data for model generation acquired in the model generation acquisition step is higher than a predetermined refractive index, and calculating feature quantities of the extracted portion; and a generation step of generating a discrimination model for discriminating the type of hematopoietic cells from the information indicating the type of hematopoietic cells for model generation acquired in the model generation acquisition step and the feature quantities calculated in the model generation feature calculation step.

[0013] According to the model generation method of the present invention, it is possible to generate a discrimination model used for the above discrimination. [Effects of the Invention]

[0014] According to the present invention, it is possible to distinguish the types of hematopoietic cells with little mechanical stress on the hematopoietic cells and at low cost. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 1 is a diagram showing the configuration of a cell type discrimination system and a model generation system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a diagram showing examples of hematopoietic cells to be discriminated by type in an embodiment. [Figure 3] FIG. 4 is a diagram showing an example of refractive index data used to distinguish the type of hematopoietic cells in the embodiment. [Figure 4] FIG. 10 is a diagram (heat map) showing an example of feature amounts calculated in the embodiment. [Figure 5] 1 is a graph (scatter plot) of feature amounts for each type of hematopoietic cell. [Figure 6] 1 is a flowchart showing a cell type discrimination method, which is a process executed in a cell type discrimination system according to an embodiment of the present invention. [Figure 7] 1 is a flowchart illustrating a model generation method, which is a process executed in a model generation system according to an embodiment of the present invention. [Figure 8] FIG. 10 shows an example of discrimination of hematopoietic cells performed according to an embodiment. [Figure 9] FIG. 10 shows an example of discrimination of hematopoietic cells performed according to an embodiment. [Figure 10] FIG. 10 shows an example of discrimination of hematopoietic cells performed according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, embodiments of the cell type discrimination method, cell type discrimination system, and model generation method according to the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are designated by the same reference numerals, and duplicated explanations will be omitted.

[0017] FIG. 1 shows a cell type discrimination system 10 that executes the cell type discrimination method according to this embodiment, and a model generation system 20 that executes the model generation method according to this embodiment. The cell type discrimination system 10 is a system (apparatus) that discriminates the type of hematopoietic cell (classifies hematopoietic cells) from refractive index data that indicates the three-dimensional refractive index of the hematopoietic cell. The cell type discrimination system 10 performs discrimination using a discrimination model prepared in advance. The model generation system 20 is a system (apparatus) that generates the discrimination model.

[0018] In this embodiment, hematopoietic cells are cells involved in hematopoiesis. Examples of hematopoietic cells are shown in Figure 2. Hematopoietic cells include 12 types of cells shown in Figure 2: hematopoietic stem cells (HSCs), multipotent progenitor cells (MPPs), common myeloid progenitor cells (CMPs), multipotent lymphoid progenitor cells (MLPs), granulocyte-macrophage progenitors (GMPs), megakaryocyte-erythroid progenitors (MEPs), granulocytes (Gra), monocytes (Mono), erythrocytes (Ery), natural killer cells (NKs), T cells (T), and B cells (B).

[0019] HSCs are cells that possess both self-renewal and pluripotency, and are the origin of all hematopoietic cells. Hematopoietic progenitor cells are cells that exist at an intermediate stage during the differentiation of HSCs into blood cells such as red blood cells or white blood cells. Figure 2 shows the differentiation relationship among the above 12 types of cells. In Figure 2, differentiation occurs from top to bottom. Figure 2 also shows the replication of HSCs. Of the above 12 types of cells, six types - HSC, MPP, CMP, MLP, GMP, and MEP - are progenitor cells (undifferentiated cells), and six types - Gra, Mono, Ery, NK, T, and B - are differentiated cells.

[0020] The cell type discrimination system 10, for example, discriminates whether the hematopoietic cells to be discriminated belong to two predetermined types out of the 12 types of hematopoietic cells described above. Alternatively, the cell type discrimination system 10 discriminates whether the hematopoietic cells to be discriminated belong to three or more predetermined types out of the 12 types of hematopoietic cells described above (multi-class classification). The cell type discrimination system 10 may also calculate a numerical value indicating the likelihood that the hematopoietic cells to be discriminated belong to a predetermined type. In this case, the cell type discrimination system 10 may calculate numerical values ​​for two types, or may calculate numerical values ​​for more types (for example, the 12 types described above).

[0021] Furthermore, the cell type discrimination system 10 may perform discrimination of types of hematopoietic cells other than those described above. Furthermore, the types of hematopoietic cells discriminated by the cell type discrimination system 10 may be other than the 12 types described above. Furthermore, discrimination may be between progenitor cells and differentiated cells, rather than classification into individual types. In other words, discrimination may be performed by classifying cells into groups containing multiple types.

[0022] Classification of hematopoietic cell types is important, for example, in regenerative medicine. Traditionally, hematopoietic stem cell transplantation has been performed on patients with blood cancers that are difficult to treat with chemotherapy and radiation therapy, aiming for a complete cure. However, hematopoietic stem cell transplantation is a high-risk treatment method, with a five-year survival rate of 50% (Japan Hematopoietic Cell Transplantation Data Center 2020). The main sources of cells are umbilical cord blood, bone marrow blood, and peripheral blood. Recently, the number of umbilical cord blood transplants has been increasing. The amount of hematopoietic stem cells contained in the blood of the transplant recipient affects the effectiveness of treatment.

[0023] Furthermore, classification of hematopoietic cell types is important in blood tests. When a blood disorder such as leukemia occurs, the balance of hematopoietic cells in the bone marrow or peripheral blood is disrupted. For example, specific hematopoietic progenitor cells proliferate abnormally. In both the above-mentioned regenerative medicine and blood tests, it is essential to distinguish the types of hematopoietic cells contained in the sample (blood).

[0024] The refractive index data used to distinguish the types of hematopoietic cells by the cell-typing discrimination system 10 includes, for example, refractive index values ​​for each three-dimensional coordinate (coordinates consisting of X-axis, Y-axis, and Z-axis). FIG. 3 shows an example of refractive index data used to distinguish the types of hematopoietic cells by the cell-typing discrimination system 10. The refractive index data shown in FIG. 3 is refractive index data for each of the 12 types of hematopoietic cells mentioned above. The refractive index data shown in FIG. 3 is refractive index data in an XY cross section (xy cross section) and an image obtained by maximum intensity projection (MIP) in the xz plane (MIP (xz plane)).

[0025] The refractive index data is obtained by imaging using the imaging device 30. The imaging device 30 generates refractive index data indicating the three-dimensional refractive index including the hematopoietic cell portion by imaging the hematopoietic cell to be determined. For example, the imaging device 30 is an optical diffraction tomography (ODT). The imaging device 30 images the hematopoietic cell to be determined that is present in a liquid (e.g., culture medium, buffer solution, or physiological saline) and generates refractive index data for only the hematopoietic cell as the imaging target. Note that the refractive index data indicates the three-dimensional refractive index including the hematopoietic cell portion, and may be any other data as long as it can be used to determine the type of hematopoietic cell according to this embodiment. Furthermore, the imaging device 30 may be any other device than ODT as long as it can obtain refractive index data.

[0026] The cell type discrimination system 10 and the model generation system 20 are configured to include a conventional computer including hardware such as a processor, memory, and a communication module. The processor is, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The functions of the cell type discrimination system 10 and the model generation system 20, which will be described later, are realized by these components operating through programs or the like. In FIG. 1, the cell type discrimination system 10 and the model generation system 20 are shown as separate systems (devices), but they may also be realized by the same system (device).

[0027] The computer that constitutes the cell-type discrimination system 10 and the model generation system 20 may be a computer system including multiple computers. Furthermore, the computer may be configured using cloud computing or edge computing.

[0028] Next, we will explain the functions of the cell type discrimination system 10 and the model generation system 20 according to this embodiment. As shown in Figure 1, the cell type discrimination system 10 is configured to include an acquisition unit 11, a feature calculation unit 12, and a type discrimination unit 13.

[0029] The acquisition unit 11 is an acquisition means for acquiring refractive index data indicating three-dimensional refractive index including that of a hematopoietic cell. The refractive index data acquired by the acquisition unit 11 is obtained, for example, by imaging the hematopoietic cell to be determined using the imaging device 30. The acquisition unit 11 acquires the refractive index data by receiving the refractive index data from the imaging device 30 or by accepting an input operation of the refractive index data by a user to the cell type discrimination system 10. The acquisition unit 11 may also acquire the refractive index data by a method other than the above. The acquisition unit 11 outputs the acquired refractive index data to the feature calculation unit 12.

[0030] The feature calculation unit 12 is a feature calculation means that extracts, as a feature relating to hematopoietic cells, a refractive index of a portion where the refractive index indicated by the refractive index data acquired by the acquisition unit 11 is greater than a preset refractive index, and calculates a feature amount of the extracted portion. The feature calculation unit 12 may calculate sphericity as the feature amount. The preset refractive index may be greater than the background refractive index. The preset refractive index may be 0.005 greater than the background refractive index.

[0031] The feature amounts calculated by the feature amount calculation unit 12 are used to distinguish the type of hematopoietic cell. The feature amount calculation unit 12 calculates the feature amounts, for example, as follows: The feature amount calculation unit 12 receives refractive index data from the acquisition unit 11. The feature amount calculation unit 12 compares the refractive index value of each coordinate included in the refractive index data with a refractive index value that is a preset threshold. The feature amount calculation unit 12 extracts, from the refractive index data, portions with refractive index values ​​greater than the threshold as those related to hematopoietic cells.

[0032] The threshold is set in advance to a value that ensures that the extracted portion appropriately corresponds to hematopoietic cells. For example, the threshold is set to a value greater than the refractive index of the background, more specifically, the refractive index of the background + 0.005. As described above, when hematopoietic cells are imaged in a liquid (e.g., culture medium, buffer solution, or physiological saline), the refractive index of the background is the refractive index of the liquid. Therefore, in this case, the threshold is set to the refractive index of the liquid + 0.005, for example, 1.338 (1.333 + 0.005, the refractive index of water), so that the refractive index of the portion associated with hematopoietic cells is appropriately extracted. Note that the threshold may be set to a value other than the above, as long as the type of hematopoietic cells can be identified from the refractive index extracted using the threshold.

[0033] The feature calculation unit 12 calculates the feature of the extracted portion from the refractive index data after extraction. The type of feature to be calculated by the feature calculation unit 12 is set in advance. There may be multiple types of feature to be calculated. That is, the type of hematopoietic cell may be determined from multiple types of feature.

[0034] The feature amount is, for example, a numerical value indicating the characteristics of the extracted portion, i.e., a portion of a hematopoietic cell. The feature amount may be related to the morphology (e.g., size or shape) of the portion. For example, the feature amount may include sphericity (the degree to which the shape of the portion is a perfect sphere). The feature amount may also include circularity of the portion (the degree to which the shape of the portion is a perfect circle). The type of feature amount may also be based on the frequency distribution of refractive index values. The type of feature amount may also be based on the spatial distribution of refractive index values.

[0035] The feature extraction may be performed using, for example, conventional software, such as the Python library pyRadiomics for radiomics analysis. In this case, the types of features calculated include morphology features (SHAPE), primary features (FIRSTORDER), gray level co-occurrence matrix (GLCM), gray level dependency matrix (GLDM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM), and neighboring gray tone difference matrix (NGTDM).

[0036] SHAPE includes Elongation, Flatness, LeastAxisLength, MajorAxisLength, Maximum2DDiameterColumn, Maximum2DDiameterRow, Maximum2DDiameterSlice, Maximum3DDiameter, MeshVolume, MinorAxisLength, Sphericity, SurfaceArea, SurfaceVolumeRatio, and VoxelVolume. FIRSTORDER includes 10Percentile, 90Percentile, Energy, Entropy, InterquartileRange, Kurtosis, Maximum, MeanAbsoluteDeviation, Mean, Median, Minimum, Range, RobustMeanAbsoluteDeviation, RootMeanSquared, Skewness, TotalEnergy, Uniformity, and Variance. GLCMs include Autocorrelation, ClusterProminence, ClusterShade, ClusterTendency, Contrast, Correlation, DifferenceAverage, DifferenceEntropy, DifferenceVariance, Id, Idm, Idmn, Idn, Imc1, Imc2, InverseVariance, JointAverage, JointEnergy, JointEntropy, MCC, MaximumProbability, SumAverage, SumEntropy, and SumSquares.

[0037] As GLDMs, there are DependenceEntropy, DependenceNonUniformity, DependenceNonUniformityNormalized, DependenceVariance, GrayLevelNonUniformity, GrayLevelVariance, HighGrayLevelEmphasis, LargeDependenceEmphasis, LargeDependenceHighGrayLevelEmphasis, LargeDependenceLowGrayLevelEmphasis, LowGrayLevelEmphasis, SmallDependenceEmphasis, SmallDependenceHighGrayLevelEmphasis, and SmallDependenceLowGrayLevelEmphasis. As GLRLMs, there are GrayLevelNonUniformity, GrayLevelNonUniformityNormalized, GrayLevelVariance, HighGrayLevelRunEmphasis, LongRunEmphasis, LongRunHighGrayLevelEmphasis, LongRunLowGrayLevelEmphasis, LowGrayLevelRunEmphasis, RunEntropy, RunLengthNonUniformity, RunLengthNonUniformityNormalized, RunPercentage, RunVariance, ShortRunEmphasis, ShortRunHighGrayLevelEmphasis, and ShortRunLowGrayLevelEmphasis.

[0038] GLSZM includes GrayLevelNonUniformity, GrayLevelNonUniformityNormalized, GrayLevelVariance, HighGrayLevelZoneEmphasis, LargeAreaEmphasis, LargeAreaHighGrayLevelEmphasis, LargeAreaLowGrayLevelEmphasis, LowGrayLevelZoneEmphasis, SizeZoneNonUniformity, SizeZoneNonUniformityNormalized, SmallAreaEmphasis, SmallAreaHighGrayLevelEmphasis, SmallAreaLowGrayLevelEmphasis, ZoneEntropy, ZonePercentage, and ZoneVariance. NGTDM includes Busyness, Coarseness, Complexity, Contrast, and Strength.

[0039] Of the above feature quantities, sphericity is the sphericity of SHAPE. FIG. 4 shows heat maps that are examples of calculated feature quantities. The example shown in FIG. 4 includes feature quantities for multiple hematopoietic cells for each type of hematopoietic cell. The horizontal axis of each heat map shown in FIG. 4 indicates the type of feature quantity, and the vertical axis indicates each individual hematopoietic cell. The heat maps in FIGS. 4(a) to 4(f) are heat maps for HSC, MPP, CMP, MLP, GMP, and MEP, respectively.

[0040] It should be noted that the feature amount calculated by the feature amount calculation unit 12 may be other than those described above. The feature amount calculation unit 12 outputs the calculated feature amount to the type determination unit 13.

[0041] The type discrimination unit 13 is a type discrimination means that discriminates the type of hematopoietic cells from the feature amounts calculated by the feature amount calculation unit 12, based on a discrimination model prepared in advance.

[0042] Figure 5 shows a graph (scatter plot) of the feature values ​​for each type of hematopoietic cell. The horizontal axis of the graph shown in Figure 5 indicates the type of hematopoietic cell (HSC, MPP, CMP, MLP, GMP, MEP, Gra, Mono, Ery, NK, T (T cell), and B (B cell)), and the vertical axis indicates the feature value of the hematopoietic cell. The graph in Figure 5 shows the feature values ​​of multiple cells for each type of hematopoietic cell. These feature values ​​were obtained by morphological and biophysical quantitative analysis based on the three-dimensional refractive index data of the hematopoietic cells.

[0043] The feature quantities related to the graphs in Figures 5(a) to 5(g) are the volume (fL) of hematopoietic cells, the surface area (μm 2 ), sphericity of hematopoietic cells, circularity of hematopoietic cells, dry mass of hematopoietic cells (pg), mean refractive index of hematopoietic cells per position, and standard deviation of refractive index of hematopoietic cells per position.

[0044] As shown in the graph of Figure 5, the feature values ​​of hematopoietic cells vary depending on the type of hematopoietic cell. For example, the circularity and sphericity are high in the progenitor cell group, and the variation in these values ​​is lower than that of differentiated cells. That is, the progenitor cells are a relatively uniform group that is close to spherical. Furthermore, the differentiated cells each have their own morphological characteristics. Therefore, the type of hematopoietic cell can be identified from the feature values ​​calculated by the feature calculation unit 12.

[0045] The type discrimination unit 13 discriminates the type of hematopoietic cells, for example, as follows. The type discrimination unit 13 stores a discrimination model generated in advance by the model generation system 20. The discrimination model is a model for inferring the type of hematopoietic cells to be determined, based on the feature values ​​calculated by the feature calculation unit 12. For example, the discrimination model is a model that inputs the feature values ​​calculated by the feature calculation unit 12 and outputs information indicating the type of hematopoietic cells to be determined, based on the feature values. For example, the discrimination model is obtained by logistic regression to which an L1 norm constraint is applied. In this case, the discrimination model discriminates between two predetermined types of hematopoietic cells to be determined. For example, the discrimination model discriminates whether the hematopoietic cells to be determined are HSCs or MPPs.

[0046] Logistic regression is the most basic method for binary classification, and its results are easy to interpret. The L1 norm is used to increase the sparsity of the selected features.

[0047] The discrimination model may be one other than those described above. For example, the discrimination model may be a machine learning model generated by machine learning other than logistic regression. Furthermore, the discrimination model may simultaneously discriminate between two or more types (for example, the above-mentioned 12 types).

[0048] The type discrimination unit 13 receives the feature values ​​from the feature calculation unit 12. The type discrimination unit 13 discriminates the type of hematopoietic cell from the received feature values ​​based on the above-described discrimination model stored in advance. For example, when a discrimination model based on logistic regression is used, the discrimination may be performed in the same manner as in a conventional method using logistic regression.

[0049] The type discrimination unit 13 outputs the discrimination result. For example, the type discrimination unit 13 displays information indicating the discrimination result on a display device provided in the cell type discrimination system 10. The type discrimination unit 13 may also transmit the information indicating the discrimination result to another device. The discrimination result is referenced or used, for example, by a user of the cell type discrimination system 10. Note that the type discrimination unit 13 may also output the discrimination result in a manner other than the above.

[0050] The cell type discrimination system 10 may have functions according to the usage scenario. For example, the cell type discrimination system 10 may have functions according to the medical transplantation of hematopoietic stem cells. In this case, the acquisition unit 11 acquires refractive index data related to a plurality of hematopoietic cells contained in blood that can be a transplant target (e.g., human umbilical cord blood, bone marrow blood, and peripheral blood). The plurality of hematopoietic cells in this case are assumed to be capable of calculating the proportion of each type of hematopoietic cell contained in the blood.

[0051] The type discrimination unit 13 discriminates the type of each of the multiple hematopoietic cells. Based on the discriminated types, the type discrimination unit 13 counts the number of hematopoietic cells of each type. The type discrimination unit 13 may also calculate the proportion of hematopoietic cells of each type. The type may be progenitor cells or differentiated cells. The calculated number or proportion may be limited to a predetermined type (used for the following evaluation).

[0052] The type discrimination unit 13 evaluates the quality of blood to be used for medical transplantation of hematopoietic stem cells from the calculated number or proportion of hematopoietic cells of each type based on pre-set and stored evaluation rules. For example, the type discrimination unit 13 evaluates that the higher the number or proportion of hematopoietic stem cells contained in the blood, the better the quality. In this case, the type discrimination unit 13 may output information indicating the evaluation result of the blood in addition to or instead of the discrimination result of the type of hematopoietic cells.

[0053] The cell type discrimination system 10 may also have functions appropriate for testing for hematologic diseases. In this case, as described above, the acquisition unit 11 acquires refractive index data for multiple hematopoietic cells contained in blood that may be subject to testing for hematologic diseases (e.g., human umbilical cord blood, bone marrow blood, and peripheral blood). As described above, the type discrimination unit 13 discriminates the type of each of the multiple hematopoietic cells and counts the number or proportion of hematopoietic cells of each type. The type discrimination unit 13 tests for hematologic diseases based on the calculated number or proportion of hematopoietic cells of each type, based on pre-defined and stored testing rules. For example, the type discrimination unit 13 evaluates that the greater the number or proportion of specific progenitor cells contained in the blood, the greater the likelihood of a hematologic disease. For example, in leukemia, specific progenitor cells abnormally proliferate in the bone marrow and appear in the blood while still immature. In this case, the type discrimination unit 13 may output information indicating the test results for hematologic diseases in addition to or instead of the results of discrimination of the types of hematopoietic cells.

[0054] The type discrimination unit 13 may score the blood (cell population) from the calculated number or percentage of hematopoietic cells of each type based on a calculation rule that has been set and stored in advance. For example, the calculation rule may be: if HSC is 0.5% or more, the score is 50; if HSC is less than 0.5%, the score is 10; if MPP is 0.5% or more, the score is 10; and if MPP is less than 0.5%, the score is 1. The calculated score is used, for example, to make judgments regarding the blood, such as whether the blood is suitable for medical transplantation of hematopoietic stem cells. For example, if the score is 60 or more, the blood is judged to be suitable (usable) for medical transplantation of hematopoietic stem cells. In this case, the type discrimination unit 13 may output information indicating the calculated score or judgment result in addition to or instead of the discrimination result of the type of hematopoietic cells. The above is the function of the cell type discrimination system 10 according to this embodiment.

[0055] As shown in FIG. 1, the model generation system 20 includes a model generation acquisition unit 21, a model generation feature calculation unit 22, and a generation unit .

[0056] The model generation acquisition unit 21 is a model generation acquisition means that acquires refractive index data for model generation that indicates the three-dimensional refractive index including the part of the hematopoietic cells for model generation, and information indicating the type of the hematopoietic cells for model generation.

[0057] The refractive index data for model generation and information indicating the types of hematopoietic cells for model generation acquired by the model generation acquisition unit 21 are learning data (teaching data) used to generate a discrimination model. The model generation acquisition unit 21 acquires a sufficient number of refractive index data for model generation to generate an appropriate discrimination model, and information indicating the types of hematopoietic cells for model generation. The refractive index data for model generation includes data related to the type of hematopoietic cell to be discriminated.

[0058] The refractive index data for model generation acquired by the model generation acquisition unit 21 is obtained, for example, by imaging hematopoietic cells prepared in advance for model generation using an imaging device. Note that the refractive index data for model generation may be generated by the imaging device 30 that generated the refractive index data used in the cell-type discrimination system 10, or may be generated by another imaging device.

[0059] The acquisition unit for model generation 21 acquires the refractive index data for model generation by receiving the refractive index data for model generation from the imaging device or by accepting an input operation of the refractive index data for model generation by a user to the model generation system 20. The acquisition unit for model generation 21 may also acquire the refractive index data for model generation by a method other than the above.

[0060] The refractive index data and information indicating the type of hematopoietic cell for the same hematopoietic cell for model generation are combined and used to generate a discrimination model. The information indicating the type of hematopoietic cell for model generation is obtained, for example, by a method based on the principle of conventional surface antigen combinations. The model generation acquisition unit 21 acquires the information indicating the type of hematopoietic cell for model generation to be combined with the refractive index data for model generation by accepting an input operation from a user to the model generation system 20. The information indicating the type of hematopoietic cell for model generation acquired by the model generation acquisition unit 21 may be in a format used for generating a discrimination model (for example, the logistic regression described above).

[0061] For example, a specific example of generating information indicating the type of hematopoietic cells used to generate a model, i.e., distinguishing the type of hematopoietic cells used to generate a model, is shown below. First, CD34+ and CD34- fractions from umbilical cord blood are purified using the following procedure. Umbilical cord blood collected from healthy maternal donors is obtained from the Kanto-Koshinetsu Cord Blood Bank of the Japanese Red Cross Society. The cord blood is subjected to density gradient centrifugation (1500 rpm, 25 minutes, 4°C) using lymphocyte separation medium (Ficoll). Mononuclear cells are isolated by this centrifugation. The isolated mononuclear cells are washed with phosphate-buffered saline (PBS, manufactured by Nacalai Tesque). The washed mononuclear cells are then incubated with CD34 immunomagnetic beads (manufactured by Miltenyi Biotec). After incubation, CD34+ and CD34- target cells are selected using a quadroMACS device. The selected CD34+ and CD34- fractions are stored in liquid nitrogen.

[0062] Next, CD34+ and CD34- cells were analyzed by flow cytometry to distinguish cell types. The frozen and stored CD34+ and CD34- fractions were thawed. The thawed CD34+ fraction was labeled with CD7-BV510, CD10-PerCP / Cy5.5, CD19-PerCP / Cy5.5, CD34-APC / Cy7, CD38-PE / Cy7, CD45RA-FITC, CD49f-PE / Cf594, CD71-Biotin, CD90-APC, CD135-BV421, and Brilliant S buffer. After a 30-minute incubation, cells were washed with PBS + 2% FBS and labeled with Streptavidin-Brilliant Violet 605. After a 30-minute incubation, cells were sorted into HSC (CD34+CD38-CD45RA-CD90+CD49f+), MPP (CD34+CD38-CD45RA-CD90-CD49f-), MLP (CD34+CD38-CD45RA+), CMP (CD34+CD38+CD10-CD45RA-Flt3+CD71-), GMP (CD34+CD38+CD10-CD45RA+CD71-), and MEP (CD34+CD38+CD10-CD45RA-Flt3-CD71+).

[0063] The thawed CD34- fraction was also labeled with CD15-V450, CD11b-APC / Cy7, CD14-PE / Cy7, CD3-APC, CD19-PE, CD56-FITC, and CD11c-BV510 to identify lymphoid / myeloid populations. After 30 minutes of incubation, the following cells were sorted: Granulocytes (CD14+CD15+CD3-CD19-), NK (CD56+CD3-CD19-CD14-), DC (CD11c+CD45+), T cells (CD3+CD19-), B cells (CD19+CD3-), and Monocytes (CD14+CD15-CD3-CD19-).

[0064] The thawed CD34- fraction was also labeled with CD235ab-PE, CD71-APC, and CD45-FITC to identify the erythrocyte population. After 30 minutes of incubation, erythrocyte (CD71+GPA+) cells were sorted.

[0065] From the above-described hematopoietic cell sorting, information indicating the type of hematopoietic cells for model generation is generated. As described above, the information indicating the type of hematopoietic cells for model generation may be obtained based on fluorescence data. Alternatively, the information indicating the type of hematopoietic cells for model generation may be obtained by Giemsa staining or immunostaining the hematopoietic cells instead of fluorescence data. Furthermore, generation of information indicating the type of hematopoietic cells for model generation (determination of the type of hematopoietic cells for model generation) may be performed by a method other than those described above.

[0066] The model generation acquisition unit 21 outputs the acquired refractive index data for model generation to the model generation feature calculation unit 22. The model generation acquisition unit 21 outputs information indicating the type of the acquired hematopoietic cell for model generation to the generation unit 23.

[0067] The model generation feature calculation unit 22 is a model generation feature calculation means that extracts the refractive index of the portion of the refractive index indicated by the refractive index data for model generation acquired by the model generation acquisition unit 21 that is greater than a predetermined refractive index as that relating to hematopoietic cells for model generation, and calculates the feature of the extracted portion.

[0068] The feature quantities calculated by the model generation feature quantity calculation unit 22 are used to generate a discrimination model. The model generation feature quantity calculation unit 22 calculates the feature quantities, for example, as follows: The model generation feature quantity calculation unit 22 receives refractive index data from the model generation acquisition unit 21. The model generation feature quantity calculation unit 22 extracts the refractive index from the refractive index data in the same way as the feature quantity calculation unit 12 described above. The acquisition unit 11 and the model generation feature quantity calculation unit 22 use the same threshold value for extracting the refractive index.

[0069] The model generation feature calculation unit 22 calculates the feature of the extracted portion from the refractive index data after extraction. The feature calculation unit 12 calculates the feature of the extracted portion in accordance with calculation rules that are set and stored in advance. The type of feature to be calculated is set in advance. There may be multiple types of feature to be calculated. That is, the type of hematopoietic cell may be determined from multiple types of feature.

[0070] The types of feature quantities calculated include types of feature quantities calculated by the feature quantity calculation unit 12. Furthermore, as will be described later, the types of feature quantities calculated may include those that are not ultimately used to distinguish the types of hematopoietic cells. The feature quantity calculation unit 22 for model generation calculates feature quantities in the same manner as the feature quantity calculation unit 12 described above. The feature quantity calculation unit 22 for model generation calculates feature quantities for each input refractive index data, i.e., for each hematopoietic cell for model generation related to the refractive index data. The feature quantity calculation unit 22 for model generation outputs the calculated feature quantities to the generation unit 23.

[0071] The generation unit 23 is a generation means that generates a discrimination model for discriminating the type of hematopoietic cell from information indicating the type of hematopoietic cell for model generation acquired by the model generation acquisition unit 21 and the features calculated by the model generation feature calculation unit 22.

[0072] The generation unit 23 generates a discrimination model, for example, as follows: The generation unit 23 receives information indicating the type of hematopoietic cells for model generation. The generation unit 23 receives features from the model generation feature calculation unit 22. Note that the generation unit 23 uses a plurality of hematopoietic cells for generation, and is able to grasp the correspondence between the information indicating the type of hematopoietic cells for generation and the features. In generating the discrimination model, these are used as corresponding information.

[0073] The generation unit 23 generates a discrimination model according to a generation rule that is set in advance and stored. For example, the generation unit 23 generates a discrimination model by logistic regression to which an L1 norm constraint is applied, using the feature amount as an input to the discrimination model and information indicating the type of hematopoietic cells to be generated as an output to the discrimination model. The generation of the discrimination model by logistic regression itself may be performed in the same manner as a conventional method. Furthermore, even if the discrimination model is one other than those described above, the generation of the discrimination model may be performed in the same manner as a conventional method (e.g., a conventional machine learning method) depending on the discrimination model.

[0074] The generation unit 23 may perform validation when generating the determination model. For example, the generation unit 23 may perform validation by stratified K-fold cross validation (K=10). The generation unit 23 may actually use the generated determination model to determine the type of hematopoietic cell only if the generated determination model is found to be valid by validation.

[0075] When generating a model, the generation unit 23 may select a type of feature to be input to the discrimination model from the types of input feature. The selection of the type of feature may be performed by a conventional method. For example, the selection of the type of feature may be performed by sequential feature selection. Sequential feature selection is an algorithm that adds a type of feature one by one and performs learning and evaluation while selecting features.

[0076] For example, in a determination model for determining whether a hematopoietic cell is HSS or MPP, the determination may be made using ten types of features, namely, shape_MinorAxisLength, shape_SurfaceArea, glcm_ClusterTendency, gldm_DependenceVariance, gldm_LowGrayLevelEmphasis, glrlm_LongRunLowGrayLevelEmphasis, glszm_GrayLevelNonUniformity, glszm_LargeAreaHighGrayLevelEmphasis, glszm_SmallAreaHighGrayLevelEmphasis, and ngtdm_Strength, by selecting (reducing) features.

[0077] The generation unit 23 outputs the generated determination model to the cell-type determination system 10. In the cell-type determination system 10, the type determination unit 13 stores the determination model and uses it for the above-mentioned type determination of hematopoietic cells. The above are the functions of the model generation system 20 according to this embodiment.

[0078] Next, the processing executed by the cell type discrimination system 10 and the model generation system 20 according to this embodiment (the operating method performed by the cell type discrimination system 10 and the model generation system 20) will be described using the flowcharts of Figures 6 and 7.

[0079] First, a cell-type discrimination method, which is a process executed by the cell-type discrimination system 10 according to this embodiment, will be described using the flowchart in Fig. 6. In this process, the acquisition unit 11 acquires refractive index data indicating the three-dimensional refractive index including a portion of hematopoietic cells (S01, acquisition step). Next, the feature calculation unit 12 extracts, as a portion of the hematopoietic cells, the refractive index of a portion where the refractive index indicated by the refractive index data is greater than a preset refractive index, and calculates the feature of the extracted portion (S02, feature calculation step).

[0080] Next, the type discrimination unit 13 discriminates the type of hematopoietic cell from the feature amount based on a discrimination model prepared in advance (S03, type discrimination step). Next, the type discrimination unit 13 outputs information indicating the discrimination result (S04). The above is the cell type discrimination method, which is the processing executed by the cell type discrimination system 10 according to this embodiment.

[0081] Next, a model generation method, which is a process executed by the model generation system 20 according to this embodiment, will be described using the flowchart of Fig. 7. In this process, the model generation acquisition unit 21 acquires refractive index data for model generation indicating the three-dimensional refractive index including a portion of hematopoietic cells for model generation, and information indicating the type of the hematopoietic cells for model generation (S11, model generation acquisition step). Next, the model generation feature calculation unit 22 extracts, as the hematopoietic cells for model generation, the refractive index of a portion where the refractive index indicated by the refractive index data for model generation is greater than a preset refractive index, and calculates the feature of the extracted portion (S12, model generation feature calculation step).

[0082] Next, the generation unit 23 generates a discrimination model for discriminating the type of hematopoietic cell from the information indicating the type of hematopoietic cell for model generation and the feature quantities (S13, generation step). Next, the generation unit 23 outputs the generated discrimination model to the cell type discrimination system 10 (S14). In the cell type discrimination system 10, the discrimination model is stored by the type discrimination unit 13 and used for the above-mentioned discrimination of the type of hematopoietic cell. The above is the model generation method, which is the processing executed by the model generation system 20 according to this embodiment.

[0083] In the discrimination according to this embodiment, the type of hematopoietic cells can be discriminated by obtaining refractive index data indicating the three-dimensional refractive index including the hematopoietic cell portion. Therefore, there is no need to label the hematopoietic cells to be discriminated or to perform an isolation process for the hematopoietic cells to be discriminated. Specifically, there is no need to prepare and attach expensive antibodies to the cells as labels to discriminate the types of hematopoietic cells. Furthermore, there is no need for complex fluorescence correction in FACS. Furthermore, because the isolation process is unnecessary, there is no risk of cell damage or cell loss during the isolation process, preventing cost increases. Furthermore, there is no need to prepare a FACS for discriminating the types of hematopoietic cells, so there is no need for FACS equipment costs. Therefore, according to the discrimination according to this embodiment, the type of hematopoietic cells can be discriminated with little mechanical stress on the hematopoietic cells and at low cost.

[0084] Furthermore, the discrimination according to this embodiment allows the type of hematopoietic cells to be discriminated without labeling, which is advantageous for administering cells to a living body in regenerative medicine. Furthermore, since the type of hematopoietic cells can be discriminated without labeling, it becomes possible to culture hematopoietic cells by type.

[0085] As in the present embodiment, sphericity may be calculated as a feature quantity. As described above, sphericity is determined according to the type of hematopoietic cell. Therefore, this configuration allows for more appropriate identification of the type of hematopoietic cell. However, the feature quantity does not need to be sphericity, and may be anything that can be used to identify the type of hematopoietic cell. For example, as described above, circularity also depends on the type of hematopoietic cell, so circularity may be used as a feature quantity.

[0086] As in this embodiment, the preset threshold value of the refractive index used to extract the refractive index may be greater than the refractive index of the background. Alternatively, the preset threshold value of the refractive index may be 0.005 greater than the refractive index of the background. These configurations allow for more appropriate extraction of the portion related to hematopoietic cells from the refractive index data, thereby enabling more appropriate identification of the type of hematopoietic cells. However, the preset threshold value of the refractive index does not have to be the above, and may be any value that allows extraction of a refractive index that can be used to identify the type of hematopoietic cells.

[0087] Furthermore, by generating a discrimination model according to this embodiment, it is possible to generate a discrimination model used for discrimination according to this embodiment.

[0088] Next, an example according to this embodiment will be described. FIG. 8 shows an example of type discrimination in which the discrimination model is generated by logistic regression to which an L1 norm constraint is applied and discriminates whether the type of hematopoietic cell is HSC or MPP. In this example, three lots (lot #1, lot #2, and lot #3) containing multiple hematopoietic cells to be determined were prepared, and discrimination was performed for each lot and for all lots. Each lot contained only HSC and MPP. In this example, discrimination was performed using all features extracted using the aforementioned pyRadiomics Python library for Radiomics analysis. FIGS. 8(a) to 8(d) show the discrimination results for lot #1, lot #2, lot #3, and all lots, respectively.

[0089] Each classification result shows the number of classification results (horizontal axis) determined by this embodiment for each actual classification (vertical axis, true label). Each classification result also shows the accuracy and recall for each actual classification.

[0090] FIG. 9 shows an example of classification. In the example shown in FIG. 9, classification was performed using feature quantities selected from those extracted using the aforementioned pyRadiomics Python library for Radiomics analysis (after reducing the number of features). In the example shown in FIG. 9, feature quantities were selected for each lot to generate a classification model. The example shown in FIG. 9 is otherwise the same as the example shown in FIG. 8. FIGS. 9(a) to 9(d) show the classification results for lot #1, lot #2, lot #3, and all lots, respectively. FIG. 9 also shows the same information as FIG. 8, and also shows the 10 feature quantities used for classification classification for each lot, along with their contribution rates (Coefficients).

[0091] FIG. 10 shows an example of type discrimination in which the discrimination model is generated by logistic regression to which an L1 norm constraint is applied and discriminates whether the type of hematopoietic cell is HSC or a predetermined type of hematopoietic cell other than HSC. In this example, the hematopoietic cells (lot) to be discriminated include only HSC and a predetermined type of hematopoietic cell other than HSC. In this example, discrimination was performed using features selected from those extracted using the aforementioned pyRadiomics Python library for Radiomics analysis (after reducing the number of features).

[0092] 10(a) to 10(e) are examples in which the above-mentioned preset types of hematopoietic cells are progenitor cells. FIG. 10(a) is an example of discrimination between HSC and MPP. FIG. 10(b) is an example of discrimination between HSC and CMP. FIG. 10(c) is an example of discrimination between HSC and MLP. FIG. 10(d) is an example of discrimination between HSC and GMP. FIG. 10(e) is an example of discrimination between HSC and MEP.

[0093] Figures 10(f)-(g) are examples in which the above-mentioned preset types of hematopoietic cells are differentiated cells. Figure 10(f) is an example of distinguishing between HSC and Gra. Figure 10(g) is an example of distinguishing between HSC and Mono. Figure 10(h) is an example of distinguishing between HSC and Ery. Figure 10(i) is an example of distinguishing between HSC and NK. Figure 10(j) is an example of distinguishing between HSC and T (T cell). Figure 10(k) is an example of distinguishing between HSC and B (B cell).

[0094] As shown in the above examples, in this embodiment, the types of hematopoietic cells can be identified with a certain level of accuracy or higher. [Explanation of symbols]

[0095] 10...cell type discrimination system, 11...acquisition unit, 12...feature calculation unit, 13...type discrimination unit, 20...model generation system, 21...acquisition unit for model generation, 22...feature calculation unit for model generation, 23...generation unit, 30...refractive index data

Claims

1. an acquisition step of acquiring refractive index data indicative of three-dimensional refractive index including a portion of hematopoietic cells; a feature calculation step of extracting, as a part relating to the hematopoietic cells, a refractive index of a part where the refractive index indicated by the refractive index data acquired in the acquisition step is greater than a predetermined refractive index, and calculating a feature amount of the extracted part; a type discrimination step of discriminating the type of the hematopoietic cells from the feature amounts calculated in the feature amount calculation step based on a discrimination model prepared in advance; A method for distinguishing cell types, comprising:

2. The cell-type discrimination method according to claim 1 , wherein in the feature calculation step, sphericity is calculated as the feature.

3. 3. The method for distinguishing cell types according to claim 1, wherein the predetermined refractive index is greater than the refractive index of a background.

4. 4. The method of claim 3, wherein the predetermined refractive index is 0.005 greater than the refractive index of the background.

5. an acquisition means for acquiring refractive index data indicating a three-dimensional refractive index including a portion of hematopoietic cells; a feature calculation means for extracting, as a part of the hematopoietic cells, a refractive index of a part of the refractive index data acquired by the acquisition means that is greater than a predetermined refractive index, and calculating a feature of the extracted part; a type discrimination means for discriminating the type of the hematopoietic cells from the feature calculated by the feature calculation means based on a discrimination model prepared in advance; A cell type discrimination system comprising:

6. a model generation acquisition step of acquiring refractive index data for model generation that indicates three-dimensional refractive index including a portion of hematopoietic cells for model generation, and information indicating the type of the hematopoietic cells for model generation; a model generation feature calculation step of extracting, as a part relating to the hematopoietic cells for model generation, a refractive index of a part where the refractive index indicated by the refractive index data for model generation acquired in the model generation acquisition step is greater than a predetermined refractive index, and calculating a feature of the extracted part; a generation step of generating a discrimination model for discriminating the type of hematopoietic cell from the information indicating the type of hematopoietic cell for model generation acquired in the model generation acquisition step and the feature amount calculated in the model generation feature amount calculation step; A model generation method including:

Images