RU2839419C1 - Cell analysis method and cell analyzer - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: group of inventions relates to computer engineering and can be used for cell analysis. Method is carried out using a cell analyser comprising a first processor, a second processor and a flow cytometer, and comprises: obtaining, based on the control of the first processor, matrix data on each of the plurality of cells in the sample by flowing the sample in the channel irradiated with the light beam and detecting signals from the irradiated cells in the sample, wherein the matrix data have, as matrix data elements, values, each of which corresponds to the signal intensity obtained at the time point while the cell is irradiated in the duct; execution by a second processor of parallel processing for processing matrix data in accordance with an artificial intelligence algorithm containing a plurality of matrix operations; and classification by the first processor based on the result of parallel processing of the cell type for each of the plurality of cells.

EFFECT: providing the required efficiency of processing data having an increased amount of information on the cell.

30 cl, 53 dwg

Description

AREA OF TECHNOLOGY

[0001] The present invention relates to a method for analyzing cells and a cell analyzer.

BACKGROUND OF THE INVENTION

[0002] Patent Document 1 describes a method in which: data obtained by measuring blood cells with a flow cytometer is analyzed in a data processing system with a processor installed therein; and the cells are classified by type. Patent Document 1 states that information on the volume and electrical conductivity of the cells is further used when the cells cannot be classified according to the optical information obtained by measuring with the flow cytometer.

LIST OF REFERENCES

[PATENT DOCUMENT]

[0003] Patent Document 1 Laid-open Publication of Japanese Patent Application No. 2012-519848 (Translation of PCT International Application)

ESSENCE OF THE INVENTION

PROBLEMS SOLVED BY THE INVENTION

[0004] If the amount of information used for classifying cells is increased to improve the accuracy of classifying cells, in the case of a sample such as blood or urine containing many cells, the amount of data per sample becomes huge due to the increase in the amount of information obtained from an individual cell. For example, to classify individual cells using a deep learning algorithm, the amount of information obtained from individual cells must be significantly increased to extract the features of each cell. Patent Document 1 does not describe a system that can process a significantly increased amount of information in the range of the required performance.

[0005] One aspect of the present invention relates to providing a method for analyzing cells and a cell analyzer that, in a configuration in which data obtained from a plurality of cells contained in a sample is analyzed, can satisfy the required performance with respect to cell data having a significantly increased amount of information.

PROBLEM SOLVING

[0006] In order to solve the above problem, a method for analyzing cells according to one aspect of the present invention using a cell analyzer (4000, 4000', 4000'') comprising a host processor (3001, 4831, 6001, 8111) and a parallel processing processor (3002, 4833, 6002, 8112) includes: obtaining, based on control of the host processor (3001, 4831, 6001, 8111), data on each of a plurality of cells in a sample; executing parallel processing on the data by the parallel processing processor (3002, 4833, 6002, 8112); and generating cell type information on each of the plurality of cells based on the result of the parallel processing.

[0007] In order to solve the above problem, a method for analyzing cells according to another aspect of the present invention includes: performing a measurement of a cell by a measuring unit (400, 400a, 500, 500a, 700); executing, by a parallel processing processor (3002, 4833, 6002, 8112), connected to the measuring unit (400, 400a, 500, 500a, 700) not via the Internet or an intranet, parallel processing of data related to the cell based on the measurement; and generating cell type information based on the result of the parallel processing.

[0008] In order to solve the above problem, a method for analyzing cells according to another aspect of the present invention includes: aspirating a sample with a measuring unit (400, 400a, 500, 500a, 700) installed in a cell analyzer (4000, 4000', 4000''); generating data about a cell in the sample subjected to aspiration; executing parallel processing with respect to the data by a parallel processing processor (3002, 4833, 6002, 8112) installed in the cell analyzer (4000, 4000', 4000''); and generating cell type information based on the result of the parallel processing executed by the parallel processing processor (3002, 4833, 6002, 8112).

[0009] In order to solve the above problem, a cell analyzer according to one aspect of the present invention includes: a measuring unit (400, 400a, 500, 500a, 700) configured to measure a plurality of cells contained in a sample; a processor (3001, 4831, 6001, 8111) configured to process information related to the analysis of the plurality of cells; and a processor (3002, 4833, 6002, 8112) for parallel processing, wherein the measuring unit (400, 400a, 500, 500a, 700) receives data about each of the plurality of cells, the processor (3002, 4833, 6002, 8112) for parallel processing performs parallel processing with respect to this data, and the processor (3001, 4831, 6001, 8111) processes information related to the cell type about each of the plurality of cells that was created based on the result of the parallel processing.

ADVANTAGEOUS EFFECTS OF THE INVENTION

[0010] While many cells are analyzed based on large volume data, it is possible to meet the performance requirements.

BRIEF DESCRIPTION OF DRAWINGS

[0011] [FIG. 1] FIG. 1(a) shows an example of classifying leukocytes according to a conventional method. FIG. 1(b) shows an example of classifying leukocytes according to the proposed method.

[FIG. 2] FIG. 2(a) shows an example of supplying light to a cell flowing in a flow cell. FIG. 2(b) shows an example of sampling a forward scattered light signal, a side scattered light signal, and a fluorescence signal. FIG. 2(c) shows an example of waveform data obtained by sampling.

[FIG. 3] FIG. 3 shows an example of a method for creating training data.

[FIG. 4] FIG. 4 shows an example of a label value.

[FIG. 5] FIG. 5 shows an example of an analysis method for analyzing data.

[FIG. 6] FIG. 6 shows an example of the external appearance of the cell analyzer.

[FIG. 7] FIG. 7 shows an example of a block diagram of a measuring unit.

[FIG. 8] FIG. 8 shows an example of a sample suction unit and a sample preparation unit.

[FIG. 9] FIG. 9 shows an example of the configuration of an optical system of a flow cytometry detector (FCD).

[FIG. 10] FIG. 10 shows an example of the configuration of the processing unit.

[FIG. 11] FIG. 11 shows an example of a processor configuration for parallel processing.

[FIG. 12] FIG. 12 shows an example of the implementation of a parallel processing processor in a measuring unit.

[FIG. 13] FIG. 13 shows another example of implementing a parallel processing processor in a measuring unit.

[FIG. 14] FIG. 14 shows another example of implementing a parallel processing processor in a measuring unit.

[FIG. 15] FIG. 15 shows another example of implementing a parallel processing processor in a measuring unit.

[FIG. 16] FIG. 16 shows a general flow chart of an operation in which a processor executes arithmetic processes for matrix data using a parallel processing processor.

[FIG. 17] FIG. 17(a) shows a formula for calculating a matrix product. FIG. 17(b) shows an example of arithmetic processes executed in parallel in a parallel processing processor.

[FIG. 18] FIG. 18 shows how arithmetic processes are executed in a processor for parallel processing.

[FIG. 19] FIG. 19(a) shows an example of waveform data of a forward scattered light signal as waveform data input to a deep learning algorithm. FIG. 19(b) shows a general diagram of a matrix operation between the waveform data and a filter.

[FIG. 20] FIG. 20 shows an example of a sample analysis operation performed by the cell analyzer.

[FIG. 21] FIG. 21 shows an example of the cell analysis process.

[FIG. 22] FIG. 22 shows an example of parallel processing.

[FIG. 23] FIG. 23 shows another example of a block diagram of a measuring unit.

[FIG. 24] FIG. 24 shows an example of a block diagram of a processing unit.

[FIG. 25] FIG. 25 shows a general flow chart of an operation in which a processor executes arithmetic processes for matrix data using a parallel processing processor.

[FIG. 26] FIG. 26 shows another example of a block diagram of a measuring unit.

[FIG. 27] FIG. 27 shows another example of a block diagram of a processing unit.

[FIG. 28] FIG. 28 shows an example of the configuration of the measuring unit, the processing unit, and the analysis unit.

[FIG. 29] FIG. 29 shows another example of a block diagram of a measuring unit.

[FIG. 30] FIG. 30 shows an example of a block diagram of an analysis unit.

[FIG. 31] FIG. 31 shows a general flow chart of an operation in which a processor executes arithmetic processes for matrix data using a parallel processing processor.

[FIG. 32] FIG. 32 shows another example of a block diagram of a processing unit.

[FIG. 33] FIG. 33 shows another example of a block diagram of a measuring unit.

[FIG. 34] FIG. 34 shows another example of a block diagram of an analysis unit.

[FIG. 35] FIG. 35 shows an example of a block diagram of a measuring unit.

[FIG. 36] FIG. 36 shows a schematic example of an optical system of a flow cytometer.

[FIG. 37] FIG. 37 shows a schematic example of a sample preparation unit of the measuring unit.

[FIG. 38] FIG. 38 shows a schematic example of a waveform data analysis system.

[FIG. 39] FIG. 39 shows an example of a block diagram of a device on the supplier side.

[FIG. 40] FIG. 40 shows an example of a block diagram of a measuring unit.

[FIG. 41] FIG. 41 shows another example of a block diagram of an analysis unit.

[FIG. 42] FIG. 42 shows an example of a functional block diagram of a deep learning device.

[FIG. 43] FIG. 43 shows an example of a flow chart of an operation of a processing node for creating training data.

[FIG. 44] FIG. 44 is a schematic diagram for describing a neural network. FIG. 44(a) is a schematic diagram showing the overall layout of a neural network. FIG. 44(b) is a schematic diagram showing an arithmetic operation at each node. FIG. 44(c) is a schematic diagram showing arithmetic operations between nodes.

[FIG. 45] FIG. 45 shows a mixed matrix of a determination result according to a reference method and a determination result using a deep learning algorithm.

[FIG. 46] FIG. 46(a) shows the ROC curve of a neutrophil. FIG. 46(b) shows the ROC curve of a lymphocyte. FIG. 46(c) shows the ROC curve of a monocyte.

[FIG. 47] FIG. 47(a) shows the ROC curve of eosinophil. FIG. 47(b) shows the ROC curve of basophil. FIG. 47(c) shows the ROC curve of control blood (CONT).

[FIG. 48] FIG. 48 shows an example of a configuration of a cell analyzer as an image analyzer.

[FIG. 49] FIG. 49 shows an example of a configuration of a processing unit.

[FIG. 50] FIG. 50 shows an example of a method for creating training data.

[FIG. 51] FIG. 51 shows an example of a label value.

[FIG. 52] FIG. 52 shows an example of a method for analyzing an image.

[FIG. 53] FIG. 53 shows a variant of the analysis result.

DESCRIPTION OF IMPLEMENTATION OPTIONS

[0012] Hereinafter, in the present description, general arrangements and embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description and in the drawings, the same reference numerals denote the same or similar components, and therefore the description of the same or similar components is omitted.

[1. Method of cell analysis]

[0013] In the present embodiment, a method for analyzing cells is disclosed, including: in a cell analyzer including a master processor and a parallel processing processor, obtaining data about each of the cells based on control of the master processor; executing parallel processing with respect to the data by the parallel processing processor; and creating cell type information with respect to each of the cells based on a result of the parallel processing.

[0014] According to the proposed analysis method, even when analyzing data having a huge volume of several hundred megabytes to several gigabytes per sample, the processing with respect to the cell data can be performed in parallel by a parallel processing processor provided separately from the master processor. Therefore, for example, even when cells are classified by a deep learning algorithm that uses data having a huge volume, the data processing is contained only in the cell analyzer. For example, there is no need to transmit the cell data via the Internet or an intranet to the server for analysis with the deep learning algorithm stored therein. Therefore, according to the proposed analysis method, there is no need to transmit a large volume of data from the cell analyzer to the server for analysis and to receive the analysis result returned from the server for analysis. Thus, the performance of the cell analyzer can be maintained while the accuracy of cell classification is improved.

[0015] An example of the general scheme of the present embodiment will be described with reference to FIG. 1. FIG. 1(a) is a drawing schematically showing the classification of leukocytes according to a conventional method, and FIG. 1(b) is a drawing schematically showing the classification of leukocytes according to the present method. In FIG. 1(a) and FIG. 1(b), FSC represents an analog signal indicating the signal intensity of forward scattered light, SSC represents an analog signal of side scattered light, and SFL represents an analog signal indicating the signal intensity of side fluorescence. As shown in FIG. 1(a), in the conventional method, each individual cell contained in a sample is measured by a flow cytometer, and the peak heights of the pulses of the analog signals of the corresponding forward scattered light, side scattered light, and side fluorescence are obtained as the forward scattered light intensity, the side scattered light intensity, and the side fluorescence intensity. Then, based on the forward scattered light intensity, the side scattered light intensity, and the side fluorescence intensity, each cell is classified into a specific type. The results of the cell classification are displayed as a scatter diagram, as shown in FIG. 1(a). In the scatter diagram in FIG. 1(a), the horizontal axis represents the side scattered light intensity, and the vertical axis represents the side fluorescence intensity.

[0016] As shown in FIG. 1(a), in the conventional classification of leukocytes, the type of each blood cell is determined based on only the peak height information of the analog signal wave. In contrast, in the method of the present embodiment, as data regarding cells in a sample, as shown in FIG. 1(b), the entire waveform of an analog signal obtained from a single cell by a flow cytometer is analyzed as data on the analysis target, whereby the cell is classified. FIG. 1(b) shows a waveform obtained by plotting the analog signal obtained by the flow cytometer. However, as described later, the data of a cell in a sample in the present embodiment means digital data (waveform data described later) in which values indicating the signal intensity at a plurality of time points obtained by performing analog-to-digital (A/D) conversion of the analog signal are used as elements. This group of digital data represents matrix data, and in the present embodiment, matrix data consisting of, for example, one row x multiple columns (i.e., a one-dimensional data array).

[0017] In the present embodiment, the deep learning algorithm 50 shown in FIG. 1(b) is made to learn the waveform data of a signal for each cell type before it is trained. Then, the waveform data of a signal for each cell of which the cell type is unknown and which is contained in a sample is input to the deep learning algorithm 60 being trained, whereby a cell type determination result is output from the deep learning algorithm 60 with respect to each cell. The deep learning algorithm 50, 60 is one of the artificial intelligence algorithms and is implemented as a neural network that includes a middle layer composed of multiple layers. In the present embodiment, when processing in connection with the analysis of the waveform data is to be executed according to the trained deep learning algorithm 60, a plurality of matrix operations included in the deep learning algorithm 60 are executed by parallel processing using a parallel processing processor installed in the cell analyzer. The cell analyzer includes: a parallel processing processor capable of performing parallel processing; and an instruction execution processor (hereinafter simply referred to as the "processor") that causes the parallel processing processor to perform parallel processing.

[0018] Further, each individual cell in a biological sample that is analyzed to determine its type may also be referred to as an "analysis target cell." In other words, a biological sample may contain a plurality of analysis target cells. The plurality of cells may include cells of a plurality of types that are to be analyzed.

[0019] An example of a biological sample is a biological sample collected from a subject. For example, a biological sample may include peripheral blood such as venous blood and arterial blood, urine, and other biological fluid other than blood and urine. Examples of a biological fluid other than blood and urine may include bone marrow aspirate, ascites, pleural effusion, cerebrospinal fluid, and the like. Hereinafter, a biological fluid other than blood and urine may be simply referred to as a "biological fluid." A blood sample may be any blood sample that is in a state where the number of cells can be counted and the cell types can be determined. Preferably, the blood is peripheral blood. Examples of blood include peripheral blood collected using an anticoagulant such as ethylenediaminetetraacetate (sodium salt or potassium salt), sodium heparin, or the like. The peripheral blood may be collected from an artery or from a vein.

[0020] The cell types determined in the present embodiment are types according to cell types based on morphological classification, and are different depending on the kind of the biological sample. When the biological sample is blood and the blood is taken from a healthy individual, the cell types determined in the present embodiment include, for example, a red blood cell, a nucleated cell such as a leukocyte, a platelet, and the like. The nucleated cells include, for example, neutrophils, lymphocytes, monocytes, eosinophils, and basophils. Neutrophils include, for example, segmented neutrophils and band neutrophils. Meanwhile, when the blood is taken from an unhealthy individual, the nucleated cells may include, for example, at least one type selected from the group consisting of an immature granulocyte and an abnormal cell. Such cells are also included in the cell types determined in the present embodiment. Immature granulocytes may include, for example, cells such as metamyelocytes, myelocytes, promyelocytes, and myeloblasts.

[0021] Nucleated cells may include, in addition to normal cells, abnormal cells that are not found in the peripheral blood of a healthy individual. Examples of abnormal cells are cells that appear when an individual has a certain disease, and such abnormal cells are, for example, tumor cells. In the case of the hematopoietic system, the certain disease may be a disease selected from the group consisting of, for example, myelodysplastic syndrome; leukemia such as acute myeloid leukemia, acute myeloblastic leukemia, acute promyelocytic leukemia, acute myelomonocytic leukemia, acute monocytic leukemia, erythroleukemia, acute megakaryoblastic leukemia, acute myeloid leukemia, acute lymphocytic leukemia, lymphoblastic leukemia, chronic myelogenous leukemia, or chronic lymphocytic leukemia; malignant lymphoma, such as Hodgkin's lymphoma or non-Hodgkin's lymphoma; and multiple myeloma.

[0022] Additionally, abnormal cells may include, for example, cells that are not normally observed in the peripheral blood of a healthy individual, such as: lymphoblasts; plasma cells; atypical lymphocytes; reactive lymphocytes; erythroblasts, which are nucleated red blood cells such as proerythroblasts, basophilic erythroblasts, polychromatic erythroblasts, orthochromatic erythroblasts, promegaloblasts, basophilic megaloblasts, polychromatic megaloblasts, and orthochromatic megaloblasts; megakaryocytes, including micromegakaryocytes; and the like.

[0023] When the biological sample is urine, the cell types determined in the present embodiment may include, for example, a red blood cell, a white blood cell, an epithelial cell such as a transitional epithelial cell, a squamous epithelial cell, and the like. Examples of abnormal cells include, for example, bacteria, fungi such as filamentous fungi and yeast, tumor cells, and the like.

[0024] When the biological sample is a biological fluid that does not normally contain blood components, such as ascites, pleural effusion, or cerebrospinal fluid, the cell types may include, for example, a red blood cell, a white blood cell, and a large cell. Here, a "large cell" means a cell that is separated from the inner lining of a body cavity or a cavity of an internal organ and that is larger than a white blood cell. For example, mesothelial cells, histiocytes, tumor cells, etc. correspond to a "large cell."

[0025] When the biological sample is a bone marrow aspirate, the cell types determined in the present embodiment may include, as normal cells, mature blood cells and immature hematopoietic cells. Mature blood cells include, for example, red blood cells, nucleated cells such as leukocytes, platelets, and the like. Nucleated cells such as leukocytes include, for example, neutrophils, lymphocytes, plasma cells, monocytes, eosinophils, and basophils. Neutrophils include, for example, segmented neutrophils and band neutrophils. Immature hematopoietic cells include, for example, hematopoietic stem cells, immature granulocyte cells, immature lymphoid cells, immature monocytic cells, immature erythroid cells, megakaryocytic cells, mesenchymal cells, etc. Immature granulocytes may include, for example, cells such as metamyelocytes, myelocytes, promyelocytes, myeloblasts, etc. Immature lymphoid cells include, for example, lymphoblasts, etc. Immature monocytic cells include monoblasts, etc. Immature erythroid cells include, for example, nucleated erythrocytes such as proerythroblasts, basophilic erythroblasts, polychromatic erythroblasts, orthochromatic erythroblasts, promegaloblasts, basophilic megaloblasts, polychromatic megaloblasts, and orthochromatic megaloblasts. Megakaryocytic cells include, for example, megakaryoblasts, etc.

[0026] Examples of abnormal cells that may be part of the bone marrow include, for example, cells from hematopoietic tumors in a disease selected from the group consisting of: myelodysplastic syndrome; leukemia such as acute myeloid leukemia, acute myeloblastic leukemia, acute promyelocytic leukemia, acute myelomonocytic leukemia, acute monocytic leukemia, erythroleukemia, acute megakaryoblastic leukemia, acute myeloid leukemia, acute lymphocytic leukemia, lymphoblastic leukemia, chronic myelogenous leukemia, or chronic lymphocytic leukemia; a malignant lymphoma such as Hodgkin's lymphoma or non-Hodgkin's lymphoma; and multiple myeloma, described above, and metastatic tumor cells from a malignant tumor that developed in an organ other than the bone marrow.

[0027] FIG. 1 shows an example of using, as a signal obtained from each cell, a forward scattered light signal, a side scattered light signal, and a side fluorescence signal, which are optical signals obtained by applying light to a cell flowing in a flow cell. However, the signal is not particularly limited as long as the signal indicates a feature of each cell and enables classification of cells for each type.

[0028] The signal obtained from each cell may be any of a signal indicating a morphological feature of the cell, a signal indicating its chemical feature, a signal indicating its physical feature, and a signal indicating its genetic feature, but is preferably a signal indicating a morphological feature of the cell. The signal indicating a morphological feature of the cell is preferably an optical signal obtained from the cell.

[0029] Preferably, the optical signal is a light signal obtained as an optical response as a result of applying light to the cell. The light signal may include at least one type selected from a signal based on light scattering, a signal based on light absorption, a signal based on light transmission, and a signal based on fluorescence.

[0030] The signal based on light scattering may include a scattered light signal caused by the application of light and a light loss signal caused by the application of light. The scattered light signal serves as a parameter that indicates a cell feature and is different according to the angle of reception of the scattered light relative to the direction of passage of the applied light. The forward scattered light signal is used as a parameter that indicates the cell size. The side scattered light signal is used as a parameter that indicates the complexity of the cell nucleus.

[0031] The term "forward" in forward scattered light means the direction of travel of light emitted from the light source. When the angle of light emission is taken as 0 degrees, "forward" may include a forward low-angle light scattering angle in which the light receiving angle is approximately 0 to 5 degrees, and/or a forward high-angle light scattering angle in which the light receiving angle is approximately 5 to 20 degrees. The term "sideways" ("lateral") is not limited, provided that "sideways" does not overlap with "forward". When the angle of light emission is taken as 0 degrees, "sideways" may include a light receiving angle of approximately 25 degrees to 155 degrees, preferably approximately 45 degrees to 135 degrees, and more preferably approximately 90 degrees.

[0032] A signal based on light scattering may include polarized light or depolarized light as a component of the signal. For example, scattered light caused by applying light to a cell is received through a polarizing plate, as a result of which only scattered light polarized at a specific angle can be received. At the same time, when light is applied to a cell through a polarizing plate and the resulting scattered light is received through a polarizing plate that allows only polarized light having an angle different from the angle of the polarizing plate for applying light to pass through it, only depolarized scattered light can be received.

[0033] The light loss signal indicates the lost amount of received light based on the decrease in the amount of received light at the light receiving portion, which is caused by the light being supplied to the cell and the light being scattered. Preferably, the light loss signal is obtained as the light loss in the direction of the optical axis of the supplied light (axial light loss). The light loss signal can be expressed as a proportion of the amount of received light during the cell flow in the flow cell, when the amount of received light at the light receiving portion in the state in which the cell does not flow in the flow cell is taken as 100%. Like the forward scattered light signal, the axial light loss is used as a parameter that indicates the cell size, but the obtained signal is different depending on whether the cell has translucency or not.

[0034] The fluorescence-based signal may be fluorescence excited by applying light to a cell labeled with a fluorescent substance, or may be intrinsic fluorescence originating from an unstained cell. The fluorescent substance may be a fluorescent dye that binds to a nucleic acid or membrane protein, or may be a labeled antibody obtained by modifying an antibody with a fluorescent dye that binds to a specific cell protein.

[0035] The optical signal can be obtained in the form of image data obtained by applying light to a cell and capturing an image of the cell to which the light was applied. Image data can be obtained by capturing, using an imaging element such as a TDI (time delay and integration) camera or a CCD (charge-coupled device) camera, an image of each individual cell flowing in a flow path, using a so-called imaging flow cytometer. Alternatively, a sample or measurement probe containing cells is applied, sprayed or dropped onto a glass slide and an image of the glass slide is captured by the imaging element, whereby cell image data can be obtained.

[0036] The signal obtained from the cell is not limited to an optical signal, and may be an electrical signal obtained from the cell. As for the electrical signal, for example, a constant current is applied to the flow cell, and the change in impedance caused by the cell flowing in the flow cell can be used as the electrical signal. The electrical signal thus obtained serves as a parameter reflecting the volume of the cell. Alternatively, as for the electrical signal, the change in impedance at the time of applying radio frequency to the cell flowing in the flow cell can be used as the electrical signal. The electrical signal thus obtained serves as a parameter reflecting the electrical conductivity of the cell.

[0037] The signal obtained from the cell may be a combination of at least two kinds of signals from the above-described signals obtained from the cell. By combining a plurality of signals, the cell features can be subjected to pleiotropic analysis, and thus, cell classification with higher accuracy is possible. Regarding such a combination, it is possible to combine, for example, at least two of a plurality of optical signals, for example, a forward scattered light signal, a side scattered light signal and a fluorescence signal. Alternatively, it is possible to combine scattered light signals with different angles, for example, a low-angle scattered light signal and a high-angle scattered light signal. In addition, it is alternatively possible to combine an optical signal and an electrical signal. The type and number of signals to be combined are not particularly limited.

[0038] In the cell analysis method of the present embodiment, the cell type determination method is not limited to the method in which a deep learning algorithm is used. Signal intensity is obtained from individual cells passing through a predetermined position in the flow path, for each of the cells, at a plurality of time points during the time period in which the cell passes through the predetermined position, and based on a result in which the obtained signal intensities at a plurality of time points with respect to each individual cell are recognized as a pattern, the cell type can be determined. The pattern can be recognized as a numerical pattern of signal intensities at a plurality of time points, or can be recognized as a shape pattern obtained by plotting the signal intensities at a plurality of time points. When the pattern is recognized as a numerical pattern, if the numerical pattern of the analysis target cell and the numerical pattern for which the cell type is already known are compared with each other, the cell type can be determined. For comparison between the numerical pattern of the target cell and the reference numerical pattern, one can use, for example, the Spearman rank correlation coefficient, z-score, etc. When comparing the graph shape pattern of the target cell and the graph shape pattern for which the cell type is already known, the cell type can be determined. For comparison between the graph shape pattern of the target cell and the graph shape pattern for which the cell type is already known, one can use the coincidence of the geometric shape of the patterns, or one can use, for example, a feature descriptor represented by a SIFT descriptor.

<General scheme of the cell analysis method>

[0039] Next, with reference to the examples shown in FIG. 2 and FIG. 3 to FIG. 5, a method for creating training data 75 and a method for analyzing waveform data will be described.

<Waveform data>

[0040] FIG. 2 is a schematic diagram for describing waveform data used in the proposed analysis method. As shown in FIG. 2(a), when a sample containing a cell C is caused to flow in a flow cell FC and light is supplied to the cell C flowing in the flow cell FC, a forward scattered light FSC is generated in a forward direction with respect to the passing direction of the light. Similarly, a side scattered light SSC and a side fluorescence SFL are generated in a side direction with respect to the passing direction of the light. The forward scattered light is received by a first light receiving portion D1, and a signal corresponding to an amount of the received light is output. The side scattered light is received by a second light receiving portion D2, and a signal corresponding to an amount of the received light is output. The side fluorescence is received by a third light receiving portion D3, and a signal corresponding to an amount of the received light is output. Accordingly, an analog signal representing the change in signal associated with the passage of time is output from each of the light-receiving parts D1-D3. The analog signal corresponding to forward-scattered light will be called the "forward-scattered light signal", the analog signal corresponding to side-scattered light will be called the "side-scattered light signal", and the analog signal corresponding to side-scattered fluorescence will be called the "fluorescence signal". One pulse of each analog signal corresponds to one cell.

[0041] Analog signals are input to an A/D converter to be converted into digital signals. FIG. 2(b) schematically shows the conversion into a digital signal performed by the A/D converter. Here, for simplicity of description, the analog signal is shown as being directly input to the A/D converter. The analog signal can be converted directly without changing its level into a digital signal. However, processing such as noise removal, background correction, and normalization can be performed as necessary. As shown in FIG. 2(b), from a start point representing a time point when the level of the forward scattered light signal among the analog signals input from the light receiving portions D1 to D3 has reached a level set as a predetermined threshold, the A/D converter samples the forward scattered light signal, the side scattered light signal, and the fluorescence signal. The A/D converter samples the corresponding analog signals at a given sampling rate (e.g. 1024-point sampling with an interval of 10 nanoseconds, 128-point sampling with an interval of 80 nanoseconds, 64-point sampling with an interval of 160 nanoseconds, etc.).

[0042] FIG. 2(c) schematically shows waveform data obtained during sampling. During the sampling, matrix data having values digitally indicating the level of an analog signal at a plurality of time points as elements is obtained as waveform data corresponding to one cell. In this way, the A/D converter creates a digital forward scattered light signal, a digital side scattered light signal and a digital side fluorescence signal corresponding to one cell. The A/D conversion is repeated until the number of cells in the digital signal reaches a predetermined number or until a predetermined period of time has elapsed from the start of flowing the sample in the flow cell. Accordingly, as shown in FIG. 2(c), a digital signal obtained by combining waveform data for N cells contained in one sample is obtained. The waveform data will be referred to as the set of sampling data for each cell (in the example in FIG. 2, the set of 1024 digital values obtained every 10 nanoseconds from t=0 ns to t=10240 ns), and the set of waveform data obtained from a single sample will be referred to as the digital signal.

[0043] Each piece of waveform data generated by the A/D converter can be assigned an index for identifying a corresponding cell. As indices, for example, integers from 1 to N are assigned to the generated pieces of waveform data in sequential order, and the forward scattered light waveform data, the side scattered light waveform data, and the side fluorescence waveform data obtained from the same cell are each assigned the same index.

[0044] Since one piece of waveform data corresponds to one cell, the index corresponds to the cell that was measured. Since the same index is assigned to the pieces of waveform data corresponding to the same cell, the deep learning algorithm described below can analyze, as one set, the forward scattered light waveform data, the side scattered light waveform data, and the fluorescence waveform data corresponding to an individual cell, and can classify the cell type.

<Creating training data>

[0045] FIG. 3 is a schematic diagram showing an example of a method for creating training data used to train a deep learning algorithm to determine a cell type. The training data 75 is waveform data created based on a forward scattered light (FSC) analog signal 70a, a side scattered light (SSC) analog signal 70b, and a side fluorescence (SFL) analog signal 70c that have been obtained with respect to a cell contained in a sample by measuring the sample with a flow cytometer. The method for obtaining the waveform data has been described above.

[0046] As for the training data 75, for example, a sample is measured by a flow cytometer, and waveform data of a cell determined to have a high probability of being a specific cell type by analyzing the cells contained in the sample based on a scatter diagram in accordance with a conventional method may be used. An example using a blood cell counter will be described. First, a blood sample is measured by a flow cytometer, and waveform data of forward scattered light, side scattered light, and fluorescence of each individual cell contained in the sample are collected. Based on the intensity of the side scattered light (pulse height of the side scattered light signal) and the intensity of fluorescence (pulse height of the fluorescence signal), each cell is classified into a group of a neutrophil, a lymphocyte, a monocyte, an eosinophil, a basophil, an immature granulocyte, or an abnormal cell. A label value corresponding to the classified cell type is assigned to the waveform data of the cell, thereby obtaining training data. For example, the mode, mean or median of the side scattered light intensity and the side fluorescence intensity of the cells included in the neutrophil group are obtained, representative cells are identified based on this value, and the waveform data of these cells are assigned a label value of "1" corresponding to a neutrophil, so that training data can be obtained. The method for creating training data is not limited to this. For example, only specific cells are selected by a cell sorter, each cell is measured by a flow cytometer, and the obtained waveform data are assigned a label value for the cell, so that training data can be obtained.

[0047] The analog signals 70a, 70b, 70c respectively represent a forward scattered light signal, a side scattered light signal and a side fluorescence signal at the time the neutrophil was measured by the flow cytometer. When these analog signals are subjected to A/D conversion as described above, forward scattered light signal waveform data 72a, side scattered light signal waveform data 72b and side fluorescence signal waveform data 72c are obtained. Each of the cells adjacent to each other in each of the waveform data 72a, 72b, 72c stores a signal level at an interval corresponding to a sampling frequency, for example at an interval of 10 nanoseconds. Each of the pieces of waveform data 72a, 72b, 72c is combined with a label value 77 indicating the type of the cell which is the source of the data, and these three pieces of waveform data corresponding to the cell, in other words, the intensity data of the three signals (the intensity of the forward scattered light signal, the intensity of the side scattered light signal and the intensity of the side fluorescence signal), are input so as to form a set as training data 75 in the deep learning algorithm 50. Since in the example of FIG. 3 the cell which is the source of the training data is a neutrophil, the waveform data 72a, 72b, 72c are assigned "1" as a label value 77 indicating that the cell is a neutrophil, thereby obtaining the training data 75. In FIG. 4 shows an example of the label value 77. Since training data 75 is created for each cell type, different label value 77 is assigned as the label value according to the cell type.

<General Framework for Deep Learning>

[0048] Using FIG. 3 as an example, a general scheme for training a neural network will be described. Preferably, the neural network 50 is a convolutional neural network having a convolutional layer. The number of nodes of the input layer 50a in the neural network 50 corresponds to the number of elements of an array included in the waveform data of the input training data 75. The number of elements of the array is equal to the sum of the number of elements of the waveform data 72a, 72b, 72c of forward scattered light, side scattered light, and side fluorescence corresponding to one cell. In the example of FIG. 3, each of the waveform data 72a, 72b, 72c includes 1024 elements, and therefore the number of nodes of the input layer 50a is 1024×3=3072. The waveform data 72a, 72b, 72c are input into the input layer 50a of the neural network 50. The label value 77 of each piece of waveform data of the training data 75 is input into the output layer 50b of the neural network, as a result of which the neural network 50 is trained. The reference numeral 50c in FIG. 3 represents the middle layer.

<Waveform data analysis method>

[0049] FIG. 5 shows an example of a method for analyzing waveform data of a cell that is an analysis target. In the method for analyzing waveform data, from an analog signal 80a of forward scattered light, an analog signal 80b of side scattered light, and an analog signal 80c of side fluorescence obtained from a cell that is an analysis target by a flow cytometer, analysis data 85 is created consisting of the waveform data obtained in the above-described manner.

[0050] Preferably, the analysis data 85 and the training data 75 have at least the same acquisition condition. The acquisition condition includes conditions for measuring cells contained in a sample by a flow cytometer, such as a condition for preparing a sample for measurement, a flow rate at which a sample for measurement is made to flow in a flow cell, an intensity of light supplied to the flow cell, a gain at the light-receiving parts receiving scattered light and fluorescence, and the like. The acquisition condition further includes a sampling frequency during the execution of A/D conversion of an analog signal.

[0051] When the target cell of the analysis flows in the flow cell, an analog signal 80a of forward scattered light, an analog signal 80b of side scattered light and an analog signal 80c of side fluorescence are obtained. When these analog signals 80a, 80b, 80c are subjected to A/D conversion as described above, the time points when the signal intensities were obtained are synchronized for each cell and data 82a of the waveform of the forward scattered light signal, data 82b of the waveform of the side scattered light signal and data 82c of the waveform of the side fluorescence signal are obtained. The waveform data fragments 82a, 82b, 82c are combined so that the data fragments on the intensity of three signals (the forward scattered light signal intensity, the side scattered light signal intensity, and the side fluorescence signal intensity) of each cell form a set, and the resulting set is input as analysis data 85 into the deep learning algorithm 60.

[0052] When the analysis data 85 is input into the input layer 60a of the neural network 60 forming the trained deep learning algorithm 60, the analysis result 83 is output from the output layer 60b as classification information related to a cell type and corresponding to the analysis data 85. Reference numeral 60c in FIG. 5 represents a middle layer. The classification information related to a cell type is, for example, a probability with which a cell belongs to each of a plurality of cell types. In addition, it can be determined that the analysis target cell for which the analysis data 85 is obtained belongs to the classification that has the highest value among the probabilities, and the analysis result 83 can include a label value 82 or the like that is an identifier representing its cell type. The analysis result 83 can be the label value itself, or can be data obtained by replacing the label value with information (for example, a character sequence) that indicates the cell type. In the example of FIG. 5, based on the analysis data 85, the deep learning algorithm 60 outputs the label value "1" that has the highest probability that the target cell of the analysis for which the analysis data 85 is obtained belongs to this classification. Then, as the analysis result 83, the character data "neutrophil" corresponding to this label value is output. The output of the label value may be performed by the deep learning algorithm 60, but another computer program may output the most preferred label value based on the probability calculated by the deep learning algorithm 60.

[2. Cell analyzer and measurement of biological sample in cell analyzer]

[0053] The waveform data of the cell according to the present embodiment and the analog signal of the cell that is the source of the waveform data may be obtained in the first cell analyzer 4000 or the second cell analyzer 4000'. In (a) of FIG. 6, an example of the appearance of the cell analyzer 4000 is shown. In (b) of FIG. 6, an example of the cell analyzer 4000' is shown. In (a) of FIG. 6, the cell analyzer 4000 includes: a measuring unit 400 and a processing unit 300 for controlling the setting of the measurement conditions of the sample and its measurement in the measuring unit 400, and for analyzing the result of the data analysis of each cell by the deep learning algorithm 60. In (b) of FIG. 6 cell analyzer 4000' includes: a measuring unit 500 and a processing unit 300 for controlling the setting of conditions for measuring a sample and measuring it in the measuring unit 500, and for analyzing the result of analyzing the data of each cell by means of the deep learning algorithm 60. The measuring unit 400, 500 and the processing unit 300 can be connected so as to be communicative with each other by wire or wirelessly. Next, an example of the configuration of the measuring unit 400, 500 will be shown below, but the present embodiment should not be considered as limited to the example below. The processing unit 300 can be used together with the device 100 described below on the supplier side.

<First cell analyzer and sample preparation for measurement>

(Configuration of the measuring unit and processing unit)

[0054] An example of a configuration will be described in which the measuring unit 400 is a blood analyzer including a FCM detector that is a flow cytometer for detecting each cell in a blood sample, more specifically, the measuring unit 400 is a blood cell counter.

(Configuration example 1)

[0055] With reference to FIG. 7 to FIG. 19, configuration examples of the measuring unit 400 and the processing unit 300 will be described. FIG. 7 shows an example of a block diagram of the measuring unit 400. As shown in FIG. 7, the measuring unit 400 includes: a PCM detector 410 for detecting blood cells; an analog processing unit 420 for processing an analog signal output from the PCM detector 410; a measuring unit controller 480; a sample preparation unit 440; a hardware mechanism unit 430; and a sample suction unit 450.

[0056] FIG. 8 is a schematic diagram for describing a sample suction unit 450 and a sample preparation unit 440. The sample suction unit 450 includes: a tip 451 for suctioning a blood sample (whole blood) from a blood collection tube T; and a pump 452 for providing a negative pressure/positive pressure at the tip. The tip 451 moves up and down by a hardware mechanism unit 430 to be inserted into the blood collection tube T. When the pump 452 provides a negative pressure in a state in which the tip 451 is inserted into the blood collection tube T, the blood sample is suctioned through the tip 451. The hardware mechanism unit 430 may include a manipulator for inverting and shaking the blood collection tube T before suctioning the blood from the blood collection tube T.

[0057] The sample preparation unit 440 includes five reaction chambers 440a-440e. The reaction chambers 440a-440e are used in the DIFF, RET, WPC, PLT-F and WNR measurement channels, respectively. Each reaction chamber has a hemolytic agent container containing a hemolytic agent and a coloring liquid container containing a coloring liquid, which serve as reagents for the corresponding measurement channel, connected thereto via flow paths. One reaction chamber and the reagents (hemolytic agent and coloring liquid) associated with it form a measurement channel. For example, the DIFF measurement channel consists of: a DIFF hemolytic agent and a DIFF coloring liquid, which serve as reagents for measuring DIFF; and the DIFF reaction chamber 440a. The other measurement channels are configured in a similar manner. An example of a configuration is shown here in which one measurement channel includes one hemolytic agent and one coloring liquid. However, one measurement channel does not necessarily have to include both the hemolytic agent and the staining liquid, and multiple measurement channels may share a single common reagent.

[0058] By means of the horizontal movement and the up-and-down movement of the hardware mechanism unit 430, the tip 451 with the aspirated blood sample is accessed from above into the reaction chamber among the reaction chambers 440a to 440e which corresponds to the measurement channel corresponding to the order, and the tip 451 discharges the aspirated blood sample. The sample preparation unit 440 supplies the corresponding hemolytic agent and the corresponding coloring liquid into the reaction chamber with the blood sample discharged therein to mix the blood sample, the hemolytic agent and the coloring liquid in the reaction chamber, thereby preparing the sample for measurement. The prepared sample for measurement is supplied from the reaction chamber to the FCM detector 410 through the flow path to subject the cells to measurement by flow cytometry.

[0059] FIG. 9 shows an example of a configuration of an optical system of the FCM detector 410. As shown in FIG. 9, in measurement by a flow cytometer, when each cell contained in a sample to be measured passes through a flow cell (a cell with a pinched flow) 4113 provided in the flow cytometer, a light source 4111 supplies light to the flow cell 4113, and scattered light and fluorescence emitted from a cell in the flow cell 4113 due to this light are detected.

[0060] In FIG. 9, light emitted from a laser diode, which is a light source 4111, is supplied through a light delivery lens system 4112 to each cell passing through a flow cell 4113.

[0061] In the present embodiment, the light source 4111 of the flow cytometer is not particularly limited, and the light source 4111 is selected to have a wavelength suitable for exciting the fluorescent dye. As such a light source 4111, for example, a semiconductor laser light source including a red semiconductor laser light source and/or a blue semiconductor laser light source, a gas laser light source such as an argon laser light source or a helium-neon laser light source, a mercury arc lamp, or the like is used. In particular, the semiconductor laser light source is suitable because it is very inexpensive compared with the gas laser light source.

[0062] As shown in FIG. 9, the forward scattered light emitted from the particle passing through the flow cell 4113 is received by the forward scattered light receiving element 4116 through the condenser lens 4114 and the pinhole member 4115. The forward scattered light receiving element 4116 is a photodiode. The side scattered light is received by the side scattered light receiving element 4121 through the condenser lens 4117, the dichroic mirror 4118, the band pass filter 4119 and the pinhole member 4120. The side scattered light receiving element 4121 is a photodiode. The lateral fluorescence is received by the lateral fluorescence receiving element 4122 through the condenser lens 4117 and the dichroic mirror 4118. The lateral fluorescence receiving element 4122 is an avalanche photodiode. A photomultiplier can be used as the forward scattered light receiving element 4116, the side scattered light receiving element 4121 and the lateral fluorescence receiving element 4122.

[0063] The received light signals output from the corresponding light receiving elements 4116, 4121, 4122 are input into the analog processing unit 420 through the amplifiers 4151, 4152, 4153, respectively.

[0064] Referring again to FIG. 7, the analog processing unit 420 performs processes such as noise removal and smoothing of the analog signals input from the PCM detector 410 and outputs the processed analog signals to the measuring unit controller 480.

[0065] The controller 480 of the measuring unit includes an A/D converter 482, a processor 4831, a RAM 4834, a storage 4835, a bus controller 4850, a processor 4833 for parallel processing and an interface unit 489 connected to the processing unit 300. In addition, the controller 480 of the measuring unit includes an interface unit 484 connected to the A/D converter 482 and an interface unit 488 connected to the hardware engine unit 430. As shown in FIG. 7, in this example of the configuration, the processor 4833 for parallel processing is installed in the cell analyzer 4000 in a built-in form in the measuring unit 400.

[0066] The processor 4831 is connected to the interface node 489, the interface node 488, the interface node 484, the RAM 4834 and the storage 4835 via the bus 485. The processor 4831 is connected to the processor 4833 for parallel processing via the bus 485. The processing unit 300 is connected to the components of the measuring unit 400 via the interface node 489 and the bus 485. The bus 485 is a transmission line with a data transfer rate of, for example, at least several hundred Mbps. The bus 485 can be a transmission line with a data transfer rate of, for example, at least 1 Gbps. The bus 485 performs data transfer based on, for example, the PCI-Express or PCI-X standard.

[0067] The A/D converter 482 converts the analog signal output from the analog processing unit 420 into a digital signal. The A/D converter 482 converts the analog signal, which goes from the start of the sample measurement to the end of the measurement, into a digital signal. When a plurality of analog signals (for example, analog signals respectively corresponding to the forward scattered light intensity, the side scattered light intensity, and the fluorescence intensity) are obtained by measuring in a specific measurement channel, the A/D converter 482 converts each analog signal, which goes from the start of the measurement to the end of the measurement, into a digital signal. As described with reference to FIG. 9, for example, three kinds of analog signals (i.e., the forward scattered light signal, the side scattered light signal, and the fluorescence signal) are input through a plurality of corresponding signal transmission paths 421 to the A/D converter 482. The A/D converter 482 converts each of the analog signals input from the plurality of signal transmission paths 421 into a digital signal. Each signal transmission path 421 is configured to transmit an analog signal, for example, in the form of a differential signal.

[0068] As described with reference to FIG. 2, the A/D converter 482 samples analog signals at a given sampling frequency (for example, sampling at 1024 points with an interval of 10 nanoseconds, sampling at 128 points with an interval of 80 nanoseconds, sampling at 64 points with an interval of 160 nanoseconds, or the like). By executing the processes of sampling three kinds of analog signals corresponding to each cell, the A/D converter 482 generates forward scattered light signal waveform data, side scattered light signal waveform data, and fluorescence signal waveform data for each cell. The A/D converter 482 assigns an index to each piece of the generated waveform data. As shown in FIG. 2(c), the pieces of the generated waveform data form a digital signal, so that the pieces of waveform data of N cells contained in one sample are adjacent to each other. Accordingly, three digital signals are generated corresponding to three types of analog signals (forward scattered light signal, side scattered light signal, and fluorescence signal) obtained from N cells.

[0069] In addition to creating waveform data from each analog signal, the A/D converter 482 can calculate a peak value from a pulse of the analog signal.

[0070] The A/D converter 482 inputs the generated digital signals to the bus 485. The bus controller 4850 transfers the digital signals output from the A/D converter 482 to the RAM 4834, such as via a DMA (direct memory access) transfer. The RAM 4834 stores the digital signals.

[0071] Using the processor 4833 for parallel processing, the processor 4831 executes a process of analyzing the waveform data included in the generated digital signal in accordance with the deep learning algorithm 60. That is, the processor 4831 is programmed to execute a process of analyzing the waveform data included in the digital signal in accordance with the deep learning algorithm 60. The analysis software 4832 for performing the analysis of the cell data based on the deep learning algorithm 60 may be stored in the storage 4835. In this case, the processor 4831 executes the analysis software 4832 stored in the storage 4835, thereby performing the data analysis process based on the deep learning algorithm 60. The processor 4831 is, for example, a central processing unit (CPU). For example, the 4831 processor may be a Core i9, Core i7, or Core i5 manufactured by Intel Corporation, a Ryzen 9, Ryzen 7, Ryzen 5, or Ryzen 3 manufactured by AMD, or the like.

[0072] The processor 4831 controls the parallel processing processor 4833. The parallel processing processor 4833 performs parallel processing, such as matrix operations, under the control of the processor 4831. That is, the processor 4831 is a master processor with respect to the parallel processing processor 4833, and the parallel processing processor 4833 is a subordinate (slave) processor to the processor 4831. The processor 4831 is also called a master processor or a primary processor.

[0073] The parallel processing processor 4833 executes in parallel a plurality of arithmetic processes that represent at least a part of the processing related to the analysis of waveform data. The parallel processing processor 4833 is, for example, a graphics processor (GPU), a field programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). When the parallel processing processor 4833 is an FPGA, the parallel processing processor 4833 may have an arithmetic process related to, for example, the trained deep learning algorithm 60 programmed into it in advance. When the parallel processing processor 4833 is an ASIC, the parallel processing processor 4833 may have a circuit for executing an arithmetic process related to the trained deep learning algorithm 60 built into it in advance, or may have, for example, a programmable module built into it in addition to such a built-in circuit. As a 4833 parallel processing processor, you can use, for example, GeForce, Quadro, TITAN, Jetson, etc., manufactured by NVIDIA Corporation. In the case of Jetson series, you can use, for example, Jetson Nano, Jetson Tx2, Jetson Xavier, or Jetson AGX Xavier.

[0074] The processor 4831 executes a computing process related to, for example, controlling the measuring unit 400. The processor 4831 executes a computing process related to, for example, control signals transmitted/received between the hardware mechanism unit 430, the sample preparation unit 440, and the sample suction unit 450. The processor 4831 executes a computing process related to, for example, transmitting/receiving information with respect to the processing unit 300. The processor 4831 executes processes related to, for example, reading program data from the storage 4835, deploying the program in the RAM 4834, and transmitting/receiving data with respect to the RAM 4834. The above-described processes executed by the processor 4831 must be executed, for example, in a predetermined sequential order. For example, if the processes required to control the hardware mechanism unit 430, the sample preparation unit 440, and the sample suction unit 450 are taken as A, B, and C, then in some cases these processes need to be executed in the sequential order of B, A, and C. Since the processor 4831 often executes such continuous processes that depend on the sequential order, even when the number of arithmetic units (each may be called a “processor core,” “core,” or the like) increases, the processing speed does not always increase.

[0075] At the same time, the parallel processing processor 4833 executes a large number of conventional calculation processes, such as, for example, arithmetic operations on matrix data including a large number of elements. In the present embodiment, the parallel processing processor 4833 executes parallel processing in which at least part of the processes of analyzing waveform data according to the deep learning algorithm 60 are parallelized. The deep learning algorithm 60 includes, for example, a large number of matrix operations. For example, the deep learning algorithm 60 may include at least 100 matrix operations, or may include at least 1000 matrix operations. The parallel processing processor 4833 has a plurality of arithmetic units, and the respective arithmetic units may simultaneously execute matrix operations. That is, the parallel processing processor 4833 may parallelize matrix operations by a plurality of respective arithmetic units as parallel processing. For example, the matrix operation included in the deep learning algorithm 60 can be divided into a plurality of arithmetic processes that do not depend on the sequential order with respect to each other. The arithmetic processes divided in this way can be executed in parallel by a plurality of arithmetic units respectively. Each of these arithmetic units can be called a "processor core", a "core", or the like.

[0076] As a result of executing such parallel processing, acceleration of arithmetic processing in the measuring unit 400 as a whole can be realized. A process such as a matrix operation included in the deep learning algorithm 60 can be called, for example, "single instruction stream and multiple data streams (SIMD) processing". The processor 4833 for parallel processing is suitable, for example, for such a SIMD operation. Such a processor 4833 for parallel processing can be called a vector processor.

[0077] As described above, the processor 4831 is suitable for executing various and complex processes. At the same time, the processor 4833 for parallel processing is suitable for executing a large number of normal processes in parallel. By executing a large number of normal processes in parallel, the turnaround time (TAT) required for the computing process is shortened.

[0078] The parallel processing executed by the parallel processing processor 4833 is not limited to matrix operations. For example, when the parallel processing processor 4833 executes the learning process according to the deep learning algorithm 50, the target of the parallel processing may be differential operations or the like related to the learning process.

[0079] As for the number of arithmetic units of the processor 4831, for example, a dual-core (number of cores: 2), quad-core (number of cores: 4) or eight-core (number of cores: 8) are used. At the same time, the parallel processing processor 4833 has at least ten arithmetic units (number of cores: 10) and can execute, for example, ten matrix operations in parallel. There is also a parallel processing processor 4833 that has, for example, several tens of arithmetic units. There is also a parallel processing processor 4833 that has, for example, at least 100 arithmetic units (number of cores: 100) and which can execute 100 matrix operations in parallel. There is also a parallel processing processor 4833 that has, for example, several hundred arithmetic units. There is also a 4833 parallel processor which has, for example, at least 1000 arithmetic units (number of cores: 1000) and which can execute 1000 matrix operations in parallel. There is also a 4833 parallel processor which has, for example, several thousand arithmetic units.

[0080] FIG. 11 shows an example of a configuration of a parallel processing processor 4833. The parallel processing processor 4833 includes a plurality of arithmetic units 4836 and a RAM 4837. The respective arithmetic units 4836 execute arithmetic processes on matrix data in parallel. The RAM 4837 stores data related to the arithmetic processes executed by the arithmetic units 4836. The RAM 4837 is a memory device with a capacity of at least 1 gigabyte. The RAM 4837 may be a memory device having a capacity of 2 gigabytes, 4 gigabytes, 6 gigabytes, 8 gigabytes, 10 gigabytes or more. Each arithmetic unit 4836 receives data from the RAM 4837 and executes an arithmetic process. The 4836 arithmetic unit may be referred to as the "processor core", "core", or the like.

[0081] Each of FIGS. 12 to FIG. 14 shows an example of installing the parallel processing processor 4833 in the measuring unit 400. Each of FIGS. 12 to FIG. 14 shows an example in which the parallel processing processor 4833 is installed in the cell analyzer 4000 in an integrated manner in the measuring unit 400. Each of FIGS. 12 and FIG. 13 shows an example of installing in which the processor 4831 and the parallel processing processor 4833 are represented as separate items. As shown in FIG. 12, the processor 4831 is mounted, for example, on the substrate 4838. The processor 4833 for parallel processing is mounted on the graphics board 4830, and the graphics board 4830 is connected to the substrate 4838, for example, via the connector 4839. The processor 4831 is connected to the processor 4833 for parallel processing via the bus 485. As shown in FIG. 13, the processor 4833 for parallel processing can be mounted directly on the substrate 4838 and connected to the processor 4831, for example, via the bus 485. FIG. 14 shows an example of an installation in which the processor 4831 and the processor 4833 for parallel processing are represented as a single whole. As shown in FIG. 14, the processor 4833 for parallel processing can be, for example, integrated into the processor 4831 mounted on the substrate 4838.

[0082] FIG. 15 shows another example of installing the parallel processing processor 4833 in the measuring unit 400. FIG. 15 shows an example in which the parallel processing processor 4833 is installed in the measuring unit 400 via an external device 4800 connected to the measuring unit 400. For example, the parallel processing processor 4833 is mounted on the external device 4800, which is a USB (Universal Serial Bus) device, and this USB device is connected to the bus 485 via an interface unit 487, as a result of which the parallel processing processor 4833 is installed in the cell analyzer 4000. The USB device may be a small device, such as a USB adapter, for example. The interface unit 487 is a USB interface with a transfer rate of several hundred Mbps, for example, and more preferably is a USB interface with a transfer rate of several Gbps to several tens of Gbps or higher. As an external device 4800 with a processor 4833 mounted on it for parallel processing, it is possible to use, for example, a neural miniature computer 2 manufactured by Intel Corporation.

[0083] A plurality of USB devices, each with a parallel processing processor 4833 mounted thereon, may be connected to the interface node 487, whereby a plurality of parallel processing processors 4833 may be installed in the cell analyzer 4000. The parallel processing processor 4833 mounted on one USB device may, in some cases, have a smaller number of arithmetic units 4836 than a GPU or the like. Thus, if a plurality of USB devices are connected to the measuring unit 400, scaling of the number of cores may be realized.

[0084] Each of FIG. 16, FIG. 17, and FIG. 18 shows a general diagram of arithmetic processes executed by the parallel processing processor 4833 based on the control of the analytical software 4832 running on the processor 4831. FIG. 16 shows an example of a configuration of the parallel processing processor 4833 executing arithmetic processes. The parallel processing processor 4833 includes a plurality of arithmetic units 4836 and a RAM 4837. The processor 4831 executing the analytical software 4832 can send a command to the parallel processing processor 4833, causing the parallel processing processor 4833 to execute at least a portion of the arithmetic processes necessary for analyzing the waveform data in accordance with the deep learning algorithm 60. The processor 4831 causes the processor 4833 to execute arithmetic processes related to the analysis of waveform data based on a deep learning algorithm for parallel processing. All or at least part of the waveform data corresponding to the signals detected by the DMD detector 410 are stored in the RAM 4834. The data stored in the RAM 4834 are transferred to the RAM 4837 of the processor 4833 for parallel processing. The data stored in the RAM 4834 are transferred to the RAM 4837, for example, by a DMA (direct memory access) method. A plurality of arithmetic units 4836 of the processor 4833 for parallel processing respectively execute arithmetic processes in parallel with respect to data stored in the RAM 4837. Each of the plurality of arithmetic units 4836 receives necessary data from the RAM 4837 for executing the arithmetic process. Data corresponding to the arithmetic result is stored in the RAM 4837 of the processor 4833 for parallel processing. Data corresponding to the arithmetic result is transferred from the RAM 4837 to the RAM 4834, for example, by the DMA method.

[0085] FIG. 17 shows a general diagram of a matrix operation executed by the parallel processing processor 4833. Before analyzing the waveform data in accordance with the deep learning algorithm 60, a matrix product calculation (matrix operation) is executed. The parallel processing processor 4833 executes a plurality of arithmetic processes, such as those related to the matrix operation, in parallel. In (a) of FIG. 17, a matrix product calculation formula is shown. In the calculation formula shown in (a), a matrix c is obtained by product of an n-row × n-column matrix a and an n-row × n-column matrix b. As shown as an example in FIG. 17, the calculation formula is described in the syntax of a hierarchical loop. In (b) of FIG. 17, an example of arithmetic processes executed in parallel in the parallel processing processor 4833 is shown. The calculation formula shown as an example in FIG. 17(a) can be divided into n × n arithmetic processes, where n × n represents, for example, the number of combinations of loop variable i for the first hierarchical level and loop variable j for the second hierarchical level. Such divided arithmetic processes are arithmetic processes that are independent of each other and can therefore be executed in parallel.

[0086] FIG. 18 is a conceptual diagram showing that a plurality of arithmetic processes, shown as an example in (b) of FIG. 17, are executed in parallel in the processor 4833 for parallel processing. As shown in FIG. 18, each of the plurality of arithmetic processes is assigned to one of the plurality of arithmetic units 4836 of the processor 4833 for parallel processing. The respective arithmetic units 4836 execute the assigned arithmetic processes in parallel. That is, the respective arithmetic units 4836 simultaneously execute the divided arithmetic processes.

[0087] By means of arithmetic operations shown as an example in FIG. 17 and FIG. 18, performed by the processor 4833 for parallel processing, information on the probability with which a cell corresponding to the waveform data belongs to each of a plurality of cell types is obtained, for example. Based on the results of the arithmetic operations, the processor 4831, executing the analysis software 4832, performs an analysis related to the cell type of the cell corresponding to the waveform data. The arithmetic results are stored in the RAM 4837 of the processor 4833 for parallel processing, to be transmitted from the RAM 4837 to the RAM 4834. The processor 4831 transmits the result of the analysis performed on the basis of the arithmetic results stored in the RAM 4834 to the processing unit 300 via the bus 485 and the interface unit 489.

[0088] The arithmetic operations with respect to the probability with which a cell belongs to each of the plurality of cell types may be performed by a processor other than the parallel processing processor 4833. For example, the arithmetic results from the parallel processing processor 4833 may be transferred from the RAM 4837 to the RAM 4834, and based on the arithmetic results read from the RAM 4834, the processor 4831 may perform arithmetic operations with information related to the probability with which a cell corresponding to each piece of waveform data belongs to each of the plurality of cell types. Alternatively, the arithmetic results from the processor 4833 for parallel processing may be transmitted from the RAM 4837 to the processing unit 300, and the processor installed in the processing unit 300 may perform arithmetic operations with information related to the probability with which a cell corresponding to each piece of waveform data belongs to each of the plurality of cell types.

[0089] In the present embodiment, the processes shown in FIG. 17 and FIG. 18 are applicable, for example, to an arithmetic process (also called a filtering process) related to a convolutional layer in the deep learning algorithm 60.

[0090] FIG. 19 shows a general diagram of an arithmetic process related to a convolutional layer. In (a) of FIG. 19, an example of forward scattered light (FSC) waveform data is shown as the waveform data to be input to the deep learning algorithm 60. As shown in FIG. 2, the waveform data of the present embodiment is data in the form of a one-dimensional matrix. To explain this more simply, the waveform data is an array in which elements are arranged in a row. Here, for convenience of description, the number of elements of the waveform data is taken to be n (n is an integer of 1 or more). In (a) of FIG. 19, a plurality of filters are shown. Each filter is created through the training process of the deep learning algorithm 50. Each of the plurality of filters is one-dimensional matrix data indicating features of the waveform data. Although each filter shown in (a) of FIG. 19, is a matrix data of 1 row × 3 columns, the number of columns is not limited to three. A matrix operation is performed on each filter and the waveform data, which are input to the deep learning algorithm 60, whereby features corresponding to a cell type of the waveform data are calculated. (b) of FIG. 19 shows an overall diagram of a matrix operation between the waveform data and a filter. As shown in (b) of FIG. 19, a matrix operation is performed while each filter is shifted relative to the elements of the waveform data, one by one. The matrix operation calculation is performed in accordance with the following formula 1.

[Mathematical expression 1]

In Formula 1, the x suffixes are variables indicating the row number and the column number of the waveform data. The h suffixes are variables indicating the row number and the column number of the filter. In the example shown in FIG. 19, the waveform data is one-dimensional matrix data, and the filter is a 1 row × 3 column matrix data, and thus, L=1, M=3, p=0, q=0, 1, 2, i=0, and j=0, 1, …, n-1.

[0091] The parallel processing processor 4833 parallelly executes the matrix operation represented by formula 1 by means of a plurality of corresponding arithmetic units 4836. Based on the arithmetic processes executed by the parallel processing processor 4833, classification information related to a cell type is generated about each cell. The generated classification information is transmitted to the processing unit 300, which is used to generate and display a test result of a sample based on the classification information.

[0092] The measuring unit 400 may process the waveform data and the identification information so that they are related to each other. Specifically, the measuring unit 400 may create an analysis result (i.e., classification information related to a cell type about each cell) from the waveform data and the identification information so that they are related to each other. The measuring unit 400 transmits, for example, the cell type classification information about each cell and the identification information in relation to each other to the processing unit 300. Examples of the identification information include: (1) identification information of a biological sample corresponding to the waveform data; (2) identification information of a cell corresponding to the waveform data (for example, the index described above); (3) identification information of a patient corresponding to the waveform data; (4) identification information of the test corresponding to the waveform data; (5) identification information of the cell analyzer by which the waveform data was measured; and (6) identification information of an entity (hereinafter referred to as a “test-related entity”), such as a hospital or a testing department, where the waveform data was measured. It should be noted that (1) the identification information of the biological sample corresponding to the waveform data may include information for determining the priority of parallel processing, such as: information on the time point at which an order for measuring the biological sample was registered; information on the time point at which the analyzer identified the biological sample; information on the time point at which the analyzer started measuring the biological sample; information for identifying whether the biological sample is an urgent sample or a scheduled sample; and information for identifying whether the measurement of the biological sample is a repeat measurement or a new measurement. When the measuring unit 400 receives a test order, for example, from a LIS (laboratory information system) or the processing unit 300, the measuring unit 400 can receive at least one of the above-mentioned identification information (1) to (6) or a combination thereof from the LIS or the processing unit 300. For example, at least one of (1) to (6), shown as examples, is transmitted to the processing unit 300 in connection with the classification information and is provided as a test result to the user via the processing unit 300. A plurality of combinations (1) to (6), shown as examples, can be transmitted to the processing unit 300 in connection with the classification information. The measuring unit 400 itself can also create, for example, at least one of the above-mentioned identification information (1) to (6) or a combination thereof.

[0093] The association between the waveform data and the above-mentioned identification information (at least one of the above (1) to (6) or a combination thereof) can be performed by the processing unit 300. When the processing unit 300 receives a test order, for example, from the LIS, the processing unit 300 receives the identification information of the biological sample or the identification information of the patient. The processing unit 300 instructs the measuring unit 400 to perform a measurement corresponding to the above-mentioned test order. When the processing unit 300 has received from the measuring unit 400 a result (i.e., the waveform data) corresponding to the measurement instruction, the processing unit 300 associates the result (i.e., the waveform data) with the above-mentioned identification information.

[0094] Referring again to FIG. 6, the processing unit 300 is connected to the processor 4831 via the interface node 489 and the bus 485 and can receive the analysis results from the processor 4831 and the processor 4833 for parallel processing. The interface node 489 is, for example, a USB interface.

[0095] FIG. 10 shows the configuration of the processing unit 300. The processing unit 300 includes a processor 3001, a bus 3003, a storage 3004, an interface unit 3006, a display unit 3015, and an operation unit 3016. The processing unit 300 is implemented as hardware by a general-purpose personal computer and functions as a processing unit of the cell analyzer 4000 by executing a specialized program stored in the storage 3004.

[0096] Processor 3001 is a CPU and can execute a program stored in storage 3004.

[0097] The storage 3004 includes a hard disk device. The storage 3004 stores at least a program 60 for processing classification information about each cell transmitted from the measuring unit 400 and for generating a sample test result. The sample test result means, as described below, a result of counting the blood cells contained in the sample based on the analysis result 83 and the classification information 82 about each individual cell obtained by the measuring unit 400.

[0098] The display unit 3015 includes a computer screen. The display unit 3015 is connected to the processor 3001 via the interface unit 3006 and the bus 3003. The display unit 3015 can receive an image signal input from the processor 3001 and can display the analysis result 83 received from the measuring unit 400 and the test result obtained by the processor 3001 analyzing the analysis result 83.

[0099] The operating unit 3016 includes a pointing device, including a keyboard, a mouse, or a touch pad. A user, such as a physician or a laboratory technician, operates the operating unit 3016 to enter an order (e.g., a single order) for a measurement into the cell analyzer 4000, while being able to enter a measurement instruction in accordance with the measurement order. The operating unit 3016 can also receive an instruction to display a test result from the user. By operating the operating unit 3016, the user can see various types of information related to the test result, such as a graph, a table, or signal information assigned to a sample.

[0100] The measuring unit 400 described above is connected to the processing unit 300 via the interface node 3006. The processor 4831 of the measuring unit 400 can transmit to the processor 3001 of the processing unit 300 the classification information about each individual cell created by the deep learning algorithm 60 in connection with the identification information of the sample. The processor 3001 stores in the storage 3004 the analysis result 83 of each cell received from the measuring unit 400 in connection with the identification information of the sample.

<Functioning of the cell analyzer>

[0101] With reference to FIG. 20 to FIG. 22, a sample analysis operation performed by the 4000 cell analyzer will be described.

[0102] When the processor 3001 of the processing unit 300 receives a measurement order and a measurement instruction from the user via the operation unit 3016, the processor 3001 transmits a measurement command to the measuring unit 400 (step S1).

[0103] After receiving the measurement command, the processor 4831 of the measuring unit 400 starts measuring the sample. The processor 4831 causes the sample aspiration unit 450 to aspirate the sample from the blood collection tube T (step S10). Then, the processor 4831 causes the sample aspiration unit 450 to distribute the aspirated sample into one of the reaction chambers 440a-440e of the sample preparation unit 440. The measurement command transmitted from the processing unit 300 in step S1 includes information about the measurement channel for which the measurement of the measurement order is requested. Based on the information about the measurement channel included in the measurement command, the processor 4831 controls the sample aspiration unit 450 so as to discharge the sample into the reaction chamber of the corresponding measurement channel.

[0104] The processor 4831 causes the sample preparation unit 440 to prepare a sample for measurement (step S11). Specifically, after receiving an order from the processor 4831, the sample preparation unit 440 supplies reagents (a hemolytic agent and a coloring liquid) to a reaction chamber with a sample discharged therein to mix the sample with the reagents. Accordingly, a sample for measurement is prepared in the reaction chamber in which red blood cells are hemolyzed by a hemolytic agent and in which cells such as leukocytes or reticulocytes serving as a target for the measurement channel are colored with a coloring liquid.

[0105] The processor 4831 causes the PCM detector 410 to measure the prepared measurement sample (step S12). Specifically, the processor 4831 controls the hardware mechanism unit 430 to send the measurement sample located in the reaction chamber of the sample preparation unit 440 to the PCM detector 410. The reaction chamber and the PCM detector 410 are connected to each other by a flow path, and the measurement sample sent from the reaction chamber flows in the flow cell 4113 and is irradiated with laser light by the light source 4111 (see FIG. 9). When a cell contained in the measurement sample passes through the flow cell 4113, the cell is exposed to light. The forward scattered light, the side scattered light and the side fluorescence generated by the cell are detected by the light receiving elements 4116, 4121, 4122, respectively, and analog signals corresponding to the intensity of the received light types are output. The analog signals are output to the A/D converter 482 via the analog processing unit 420.

[0106] The A/D converter 482 samples each analog signal at a given frequency, creating a digital signal including waveform data of each individual cell (step S13). The methods for creating the waveform data and the digital signal are described above. The processor 4831 receives each digital signal generated by the A/D converter 482 into the RAM 4834. For example, the processor 4831 controls the bus controller 4850, causing the digital signal generated by the A/D converter 482 to be received into the RAM 4834 via DMA transmission. By means of the DMA transmission, the digital signal is directly transmitted to the RAM 4834, not through the processor 4831. Specifically, the processor 4831 receives in the RAM 4834 the digital signal of the forward scattered light signal, the digital signal of the side scattered light and the digital signal of the fluorescence signal, which were obtained from each cell contained in the target test sample. Each digital signal is stored in the RAM 4834.

[0107] The processor 4831 performs cell classification based on the waveform data included in each generated digital signal using the deep learning algorithm 60 (step S14). The process of step S14 will be described later.

[0108] The processor 4831 transmits the analysis result 83 including the classification information 82 about each individual cell obtained as the result of step S14 to the processing unit 300 in association with the identification information of the sample (step S15). Regarding the analysis results 83, for example, the analysis result 83 of each of a plurality of cells contained in one sample is sent in association with the identification information of the sample to the processing unit 300.

[0109] After receiving the analysis results 83 from the measuring unit (step S2), the processor 3001 of the processing unit 300 analyzes the analysis results 83 using the program stored in the storage 3004 and creates a sample test result (step S3). During step S3, for example, based on the label values included in the analysis results 83 of individual cells, the number of cells for each cell type is counted. For example, with respect to one sample, when there are N pieces of classification information with an assigned label value of "1" that indicates a neutrophil, the count result that the number of neutrophils=N is obtained as the sample test result.

[0110] The processor 3001 obtains a count result for a measurement parameter corresponding to a measurement channel based on the analysis results 83 and stores the count result, together with the sample identification information, in the storage 3004. The measurement parameter corresponding to the measurement channel is the parameter whose count result is requested in the measurement order. For example, the measurement parameter corresponding to the DIFF channel is the counts of five classifications of white blood cells, i.e., monocytes, neutrophils, lymphocytes, eosinophils, and basophils. The measurement parameter corresponding to the RET channel is the count of reticulocytes. The measurement parameter corresponding to the PLT-F channel is the count of platelets. The measurement parameter corresponding to the WPC is the count of hematopoietic progenitor cells. The measurement parameter corresponding to the WNR is the counts of white blood cells and nucleated red blood cells. The count result is not limited to the result of the parameter (also called the "recorded parameter") for which the measurement is requested, as listed above, and may include the count result of another cell that can be measured in the same measurement channel. For example, when the measurement channel is DIFF, as shown in FIG. 4, the count result also includes immature granulocytes (IG) and abnormal cells in addition to the five classifications of leukocytes. In addition, the processor 3001 analyzes the obtained count result to generate a test result of the sample and stores this result in the storage 3004. Analyzing the count result includes making a determination, for example, whether the count result is within the normal range, whether any abnormal cells are detected, whether the difference from the previous test result is within the acceptable range, and the like.

[0111] The processor 3001 displays the generated test result on the display unit 3015 (step S4).

[0112] Next, with reference to FIG. 21, the cell classification process in step S14 will be described. The cell classification process in step S14 is a process executed by the processor 4831 in accordance with the operation of the analysis software 4832. The processor 4831 causes each digital signal received in the RAM 4834 in step S13 to be transmitted to the processor 4833 for parallel processing (S101). As shown in FIG. 16, the processor 4831 causes the digital signal to be transmitted in a DMA manner from the RAM 4834 to the RAM 4837. For example, the processor 4831 controls the bus controller 4850 to DMA transmit the digital signal from the RAM 4834 to the RAM 4837.

[0113] The processor 4831 instructs the parallel processing processor 4833 to execute parallel processing of the waveform data included in the digital signal (S102). The processor 4831 instructs the execution of the parallel processing by, for example, calling a function of the core of the parallel processing processor 4833. The process executed by the parallel processing processor 4833 will be described below with reference to a flow chart shown as an example in FIG. 22. The processor 4831 instructs the parallel processing processor 4833 to execute a matrix operation, for example, related to the deep learning algorithm 60. The digital signal is divided into a plurality of pieces of waveform data for sequentially inputting them to the deep learning algorithm 60. An index corresponding to each cell and included in the digital signal is not input to the deep learning algorithm 60. The waveform data input to the deep learning algorithm 60 is subjected to arithmetic operations performed by the processor 4833 for parallel processing.

[0114] The processor 4831 receives the results of arithmetic operations executed by the processor 4833 for parallel processing (S103). The arithmetic results are transferred by a DMA method from the RAM 4837 to the RAM 4834, for example, as shown in FIG. 16.

[0115] Based on the arithmetic results from the parallel processing processor 4833, the processor 4831 generates a cell type analysis result for each measured cell (S104).

[0116] FIG. 22 shows an example of the operation of the arithmetic processes of the parallel processing processor 4833, executed based on an instruction from the processor 4831, in accordance with the operation of the analytical software 4832.

[0117] The processor 4831 executing the analytical software 4832 causes the parallel processing processor 4833 to execute the assignment of arithmetic processes to the arithmetic units 4836 (S110). For example, the processor 4831 causes the parallel processing processor 4833 to execute the assignment of arithmetic processes to the arithmetic units 4836 by calling a function of the core of the parallel processing processor 4833. As shown in FIG. 18, for example, a matrix operation related to the deep learning algorithm 60 is divided into a plurality of arithmetic processes, and the respective divided arithmetic processes are assigned to the arithmetic units 4836. A plurality of pieces of waveform data are sequentially input to the deep learning algorithm 60. The matrix operation corresponding to the waveform data is divided into a plurality of arithmetic processes to be assigned to the arithmetic units 4836.

[0118] Arithmetic processes are processed in parallel by a plurality of arithmetic units 4836 (S111). The arithmetic processes are executed on a plurality of pieces of waveform data.

[0119] The arithmetic results generated by parallel processing by the plurality of arithmetic units 4836 are transferred from the RAM 4837 to the RAM 4834 (S112). For example, the arithmetic results are transferred by a DMA method from the RAM 4837 to the RAM 4834.

(Configuration example 2)

[0120] With reference to FIG. 23 and FIG. 24, another configuration example of a cell analyzer 4000 consisting of a measuring unit 400 and a processing unit 300 will be described. In the present configuration example, a processor for parallel processing is provided in the processing unit 300.

[0121] FIG. 23 shows another example of a block diagram of the measuring unit 400. Except that the measuring unit 400 shown in FIG. 23 is not provided with the A/D converter 482, the processor 4831, the RAM 4834, the storage 4835 and the processor 4833 for parallel processing, but is provided with the connection port 4201, the measuring unit 400 shown in FIG. 23 has configurations and functions similar to the measuring unit 400 described with reference to FIG. 6 and FIG. 7 and in the description related thereto. The connection cable 4202 is connected to the connection port 4201.

[0122] FIG. 24 shows an example of a block diagram of a processing unit 300. As shown in FIG. 24, the processing unit 300 includes a processor 3001, a parallel processing processor 3002, a storage 3004, a RAM 3005, an interface unit 3006, an A/D converter 3008, a bus controller 4850, an interface unit 3009, and they are connected to a bus 3003. That is, in the example of FIG. 24, the parallel processing processor 3002 is installed in the cell analyzer 4000 in an integrated manner in the processing unit 300. The bus 3003 is a transmission line with a data transfer rate of, for example, not less than several hundred Mbps. The bus 3003 can be a transmission line with a data transfer rate of not less than 1 Gbps. The bus 3003 performs data transfer, for example, based on the PCI-Express or PCI-X standard. The configurations of the processor 3001, the processor 3002 for parallel processing, the storage 3004 and the RAM 3005 and the processes executed by them are similar to the configurations and processes of the processor 4831, the processor 4833 for parallel processing, the storage 4835 and the RAM 4834 described with reference to FIG. 11 - FIG. 19 above. The A/D converter 3008 samples each analog signal output from the measuring unit 400, as described above, and creates a digital signal including data on the waveform of the cells. The method for creating the digital signal was described above. In the example of FIG. 23 and FIG. 24 connecting cable 4202 is provided with transmission paths, the number of which corresponds to the types of analog signals transmitted, for example, from measuring unit 400 to processing unit 300. For example, connecting cable 4202 is implemented by means of cables with twisted pairs and has pairs of wires, wherein the number of pairs corresponds to the types of analog signals transmitted to processing unit 300. The transmission path from connecting port 3007 to A/D converter 3008 can also have wires, the number of which corresponds to the types of analog signals transmitted to processing unit 300. In the transmission path from connecting port 3007 to A/D converter 3008, analog signals are transmitted, for example, in the form of differential signals.

[0123] As shown in FIG. 25, the processor 3001 and the parallel processing processor 3002 have configurations and functions similar to the processor 4831 and the parallel processing processor 4833 described above. The parallel processing processor 3002 includes a plurality of arithmetic units 3200 and a RAM 3201. The analysis software 3100 for analyzing the cell type of each measured cell is executed in the processor 3001. The parallel processing processor 3002 does not necessarily have to be directly connected to the bus 3003. For example, the parallel processing processor 3002 can be mounted on a USB device, and this USB device can be connected to the bus 3003 via an interface node (not shown), whereby the parallel processing processor 3002 can be installed as part of the processing unit 300 in the cell analyzer 4000. This USB device may be a small device such as a USB adapter.

[0124] As shown in FIG. 23 and FIG. 24, the processing unit 300 is connected to the interface node 489 of the measuring unit 400 via the interface node 3006. The processing unit 300 transmits control signals for the hardware mechanism node 430 and the sample preparation node 440 to the measuring unit 400 via the interface node 3006. The interface node 3006 is, for example, a USB interface.

[0125] In the example shown in FIG. 23 and FIG. 24, the processing unit 300 is connected to the connection port 4201 of the measuring unit 400 via the connection port 3007 connected to the A/D converter 3008 and the connection cable 4202 connected to the connection port 3007, in addition to the interface unit 3006. The connection port 4201 is connected to the analog processing unit 420. Each analog signal output from the connection port 4201 to the processing unit 300 is a signal obtained as a result of the output of the DAC detector 410 of the measuring unit 400, processed by the analog processing unit 420, as described above. The analog processing unit 420 performs a process including noise elimination on the analog signal input from the DMD detector 410. The analog signal processed by the analog processing unit 420 is transmitted to the processing unit 300 via the connection port 4201 and the connection cable 4202. The connection cable 4202 is made having a length of, for example, 1 meter or less, to reduce noise during signal transmission. The analog signal is transmitted in the form of, for example, a differential signal to the processing unit 300 via the connection cable 4202. The processing unit 300 can include a plurality of connection ports 3007. The processing unit 300 can receive analog signals from a plurality of measuring units 400 via a plurality of connection ports 3007.

[0126] Each analog signal transmitted from the measuring unit 400 via the connecting cable 4202 is converted into a digital signal by the A/D converter 3008 of the processing unit 300. As described with reference to FIG. 2, for example, the A/D converter 3008 samples the transmitted analog signal at a given sampling frequency (for example, sampling at 1024 points with an interval of 10 nanoseconds, sampling at 128 points with an interval of 80 nanoseconds, sampling at 64 points with an interval of 160 nanoseconds, or the like) and creates waveform data for each cell. The waveform data is stored in the storage 3004 or RAM 3005 via the bus 3003. The waveform data is transmitted by the DMA method, for example, to the RAM 3005. The processor 3001 and the processor 3002 for parallel processing perform arithmetic processes on the waveform data stored in the storage 3004 or RAM 3005.

[0127] The analytical software 3100 running in the processor 3001 has functions similar to the functions of the analytical software 4832 shown in FIG. 16. The processor 3001 executes the analytical software 3100, thereby creating cell type-related classification information about each measured cell, through operations similar to the operations described with reference to FIG. 16, FIG. 17, FIG. 18, FIG. 21 and FIG. 22 and in the associated description.

[0128] In the case of the cell analyzer 4000 of the configuration example 2 shown in FIG. 23 and FIG. 24, in the flow chart shown in FIG. 20, step S13 (creating a digital signal) and step S14 (classifying cells) are performed in the processing unit 300. Step S15 (transmitting classification information) is excluded. The processes in FIG. 21 and FIG. 22 are performed by the processor 3001 and the processor 3002 for parallel processing from the processing unit 300.

[0129] The processor 3001 may process the waveform data and the identification information so that they are associated with each other. Specifically, the processor 3001 may associate with each other an analysis result (i.e., classification information related to a cell type about each cell) of the waveform data and the identification information. Examples of the identification information include: (1) the identification information of a biological sample corresponding to the waveform data; (2) the identification information of a cell corresponding to the waveform data; (3) the identification information of a patient corresponding to the waveform data; (4) the identification information of a test corresponding to the waveform data; (5) the identification information of a cell analyzer by which the waveform data was measured; and (6) the identification information of a testing facility where the waveform data was measured. When the processor 3001 receives a test order, for example, from the LIS, the processor 3001 can receive at least one of the above identification information (1) - (6) or a combination thereof from the LIS. For example, at least one of (1) - (6), shown as examples, is provided to the user as a test result in connection with the classification information. A plurality of combinations of (1) - (6), shown as examples, can be provided to the user as a test result in connection with the classification information.

(Configuration example 3)

[0130] With reference to FIG. 26 and FIG. 27, another configuration example of the cell analyzer 4000, consisting of the measuring unit 400 and the processing unit 300, will be described. In the present configuration example, also the processor 3002 for parallel processing is installed in the cell analyzer 4000 in a built-in form in the processing unit 300.

[0131] FIG. 26 shows another example of a block diagram of the measuring unit 400. Except that the measuring unit 400 shown in FIG. 26 is not provided with the processor 4831, the RAM 4834, the storage 4835 and the processor 4833 for parallel processing and is provided with an interface unit 4851 and a transmission line 4852 for transmitting digital signals generated in the A/D converter 482 to the processing unit 300, the measuring unit 400 shown in FIG. 26 has configurations and functions similar to the measuring unit 400 described with reference to FIG. 6 and FIG. 7 and in the description related thereto. The interface unit 4851 is an interface serving as a dedicated line having a communication channel bandwidth of, for example, not less than 1 gigabit/second. For example, the interface node 4851 is an interface in accordance with Gigabit Ethernet, USB 3.0 or Thunderbolt 3. When the interface node 4851 corresponds to Gigabit Ethernet, the transmission line 4852 is, for example, a LAN cable. When the interface node 4851 corresponds to USB 3.0, the transmission line 4852 is a USB cable in accordance with USB 3.0. The transmission line 4852 is a specialized transmission line for transmitting digital signals, for example, between the measuring unit 400 and the processing unit 300.

[0132] FIG. 27 shows another example of a block diagram of the processing unit 300. Except that the processing unit 300 shown in FIG. 27 is not provided with the A/D converter 3008 and the connection port 3007 and is provided with the interface node 3010, the processing unit 300 shown in FIG. 27 has configurations and functions similar to the processing unit 300 described with reference to FIG. 24 and in the description related thereto. The processing unit 300 can be connected to a plurality of measuring units 400 through a plurality of interface nodes 3010 and a plurality of interface nodes 3006.

[0133] The processor 3001 and the parallel processing processor 3002 have configurations and functions similar to the processor 3001 and the parallel processing processor 3002 described with reference to FIG. 25 and in the description related thereto. In FIG. 27, the parallel processing processor 3002 does not necessarily have to be directly connected to the bus 3003. For example, the parallel processing processor 3002 can be mounted on a USB device, and this USB device can be connected to the bus 3003 via an interface node (not shown). This USB device can be a small device, such as a USB adapter, for example.

[0134] The analysis software 3100 running on the processor 3001 has functions similar to the functions of the analysis software 4832 shown in FIG. 16. The analysis software 3100 analyzes the cell type for each measured cell by operations similar to those described with reference to FIG. 16, FIG. 17, FIG. 18, FIG. 21 and FIG. 22 and in the description related thereto.

[0135] In the case of the cell analyzer 4000 of the configuration example 3 shown in FIG. 26 and FIG. 27, in the flow chart shown in FIG. 20, step S14 (cell classification) is performed in the processing unit 300. Step S15 (transmission of classification information) is omitted. The processes in FIG. 21 and FIG. 22 are performed by the processor 3001 and the processor 3002 for parallel processing of the processing unit 300.

[0136] In the configuration of FIG. 26 and FIG. 27, the analog signals (forward scattered light signal, side scattered light signal, side fluorescence signal) of each cell generated in the PCM detector 410 are converted into digital signals in the A/D converter 482 in the measuring unit 400. The digital signals are sent to the processing unit 300 via the interface unit 484, the bus 485, the interface unit 4851, and the transmission line 4852. The transmission line 4852 is a dedicated transmission line for transmitting digital signals between the measuring unit 400 and the processing unit 300, as described above. For example, the measuring unit 400 and the processing unit 300 are connected on a one-to-one basis via the transmission line 4852. In other words, the transmission line 4852 is a transmission line that does not transmit data, for example, related to a device other than the components constituting the cell analyzer 4000 (for example, the measuring unit 400 and the processing unit 300). The transmission line 4852 is a transmission line other than an intranet or the Internet, for example. Accordingly, even when the digital signals created by the A/D conversion in the measuring unit 400 are transmitted to the processing unit 300, a bottleneck in the communication speed in transmitting the digital signals can be avoided.

(Configuration example 4)

[0137] With reference to FIG. 28, FIG. 29, FIG. 30, FIG. 31 and FIG. 32, another example of the configuration of the cell analyzer 4000 will be described.

[0138] In the present configuration example, as shown as an example in FIG. 28, an analysis unit 600 is provided between the measuring unit 400 and the processing unit 300. That is, in the configuration in FIG. 28, FIG. 29, FIG. 30, FIG. 31 and FIG. 32, the cell analyzer 4000 is composed of the measuring unit 400, the processing unit 300 and the analysis unit 600. The analysis unit 600 analyzes the cell type of each measured cell. As described later, in the present configuration example, the processor 6002 for parallel processing is installed in the cell analyzer 4000 in a built-in manner in the analysis unit 600.

[0139] The measuring unit 400 shown as an example in FIG. 29 has configurations and functions similar to the measuring unit 400 described with reference to FIG. 26 and in the description related thereto. The analysis unit 600 is provided between the measuring unit 400 and the processing unit 300. The analysis unit 600 can be connected to a plurality of measuring units 400. The analysis unit 600 can be connected to a plurality of processing units 300. The interface node 4851 is an interface having a communication channel bandwidth of, for example, not less than 1 gigabit/second. For example, the interface node 4851 is an interface corresponding to Gigabit Ethernet, USB 3.0, or Thunderbolt 3. When the interface node 4851 corresponds to Gigabit Ethernet, the transmission line 4852 is, for example, a LAN cable. When the interface unit 4851 corresponds to USB 3.0, the transmission line 4852 is a USB cable in accordance with USB 3.0. The transmission line 4852 is a specialized transmission line for carrying out the transmission of digital signals between the measuring unit 400 and the processing unit 300, as described above. For example, the measuring unit 400 and the processing unit 300 are connected on a one-to-one basis via the transmission line 4852.

[0140] FIG. 30 shows an example of a configuration of an analysis unit 600. The analysis unit 600 includes, for example, a processor 6001, a parallel processing processor 6002, a bus 6003, a storage 6004, a RAM 6005, an interface unit 6006, and an interface unit 6007, and they are connected to the bus 6003. The bus 6003 is a transmission line with a data transfer rate of, for example, not less than several hundred Mbps. The bus 3003 may be a transmission line with a data transfer rate of not less than 1 Gbps. The bus 3003 performs data transfer based on the PCI-Express or PCI-X standard, for example. The analysis unit 600 may be connected to a plurality of measuring units 400 via a plurality of interface nodes 6006. When a plurality of measuring units 400 are provided, the analysis unit 600 may be connected to each of the measuring units 400 (for example, the plurality of measuring units 400 and the plurality of analysis units 600 are connected on a one-to-one basis, respectively).

[0141] As shown in FIG. 31, the processor 6001 and the parallel processing processor 6002 have configurations and functions similar to the processor 4831 and the parallel processing processor 4833 described above. The parallel processing processor 6002 includes a plurality of arithmetic units 6200 and a RAM 6201. The analysis software 6100 that analyzes the cell type of each measured cell operates on the processor 6001. The analysis software 6100 running on the processor 6001 has functions similar to the functions of the analysis software 4832 shown in FIG. 16. The analysis software 6100 analyzes the cell type of each measured cell through operations similar to those described with reference to FIG. 16, FIG. 17, FIG. 18, FIG. 21, and FIG. 22 and in the description relating thereto. The analytical software 6100 transmits the classification information about each measured cell to the processing unit 300 via the interface node 6007. The interface node 6007 corresponds to, for example, Ethernet (registered trademark) or USB. The interface node 6007 may be an interface capable of wireless communication (for example, WiFi (registered trademark) or Bluetooth (registered trademark)).

[0142] FIG. 32 shows an example of a configuration of the processing unit 300. The processing unit 300 shown in FIG. 32 does not necessarily have to include a processor 3002 for parallel processing, unlike the processing unit 300 shown in FIG. 24 and FIG. 27. In addition, the analysis software 3100 shown in FIG. 24 and FIG. 27 does not necessarily have to operate on the processor 3001 shown in FIG. 32. The processing unit 300 receives the analysis results from the analysis unit 600 via an interface node 3006. The interface node 3006 corresponds to, for example, Ethernet or USB. The interface node 3006 can be an interface capable of wireless communication (for example, WiFi or Bluetooth).

[0143] In the configuration of FIG. 29, FIG. 30, FIG. 31 and FIG. 32, the analog signals (forward scattered light signal, side scattered light signal, side fluorescence signal) of each cell generated in the PCM detector 410 are converted into digital signals by the A/D converter 482 in the measuring unit 400. The waveform data is sent to the analyzing unit 600 via the interface node 484, the bus 485, the interface node 4851 and the transmission line 4852. The interface node 4851 is a dedicated interface connecting the measuring unit 400 and the analyzing unit 600 to each other as described above, and the interface node 4851 connects the measuring unit 400 and the analyzing unit 600 on a one-to-one basis. In other words, the transmission line 4852 is a transmission line that does not transmit data, for example, related to a device other than the components constituting the cell analyzer 4000 (for example, the measuring unit 400 and the processing unit 300). The transmission line 4852 is a transmission line other than an intranet or the Internet, for example. Accordingly, even when the digital signals created by the A/D conversion in the measuring unit 400 are transmitted to the processing unit 300, a bottleneck in the communication speed in transmitting the digital signals can be avoided.

[0144] In the case of the cell analyzer 4000 of the configuration example 4 shown in FIG. 29, FIG. 30, FIG. 31 and FIG. 32, in the flow chart shown in FIG. 20, step S14 (cell classification) and step S15 (classification information transmission) are performed in the analysis unit 600. That is, the digital signals generated in step S13 are transmitted from the measuring unit 400 to the analysis unit 600, and the classification information is transmitted to the processing unit 300 in step S15. The processes in FIG. 21 and FIG. 22 are performed by the processor 6001 and the processor 6002 for parallel processing of the analysis unit 600.

(Configuration example 5)

[0145] With reference to FIG. 28, FIG. 33 and FIG. 34, another configuration example of the cell analyzer 4000 is described. Like the configuration example 4 described above, the cell analyzer of this configuration example 5 also includes a measuring unit 400, a processing unit 300 and an analysis unit 600. The measuring unit 400 shown in FIG. 33 has functions and configurations similar to the measuring unit 400 described with reference to FIG. 23 and in the description related thereto. The measuring unit 400 shown in FIG. 33, is connected to the analysis unit 600 via the connecting cable 4202. For example, the connecting cable 4202 is implemented by means of a twisted-pair cable and has pairs of wires, wherein the number of pairs corresponds to the types of analog signals transmitted to the processing unit 300. The connecting cable 4202 is made having a length of, for example, 1 meter or less, in order to reduce noise during signal transmission. The measuring unit 400 transmits analog signals to the analysis unit 600 via the connecting cable 4202.

[0146] The analysis unit 600 shown in FIG. 34 has functions and configurations similar to the analysis unit 600 described with reference to FIG. 30 and in the description related thereto. That is, in the example of FIG. 34, the processor 6002 for parallel processing is installed in the cell analyzer 4000 in an integrated manner in the analysis unit 600. The analysis unit 600 shown in FIG. 34 further includes a connection port 6008 and an A/D converter 6009. The analog signals transmitted from the analysis unit 600 through the connection cable 4202 are input to the A/D converter 6009 through the connection port 6008. The A/D converter 6009 converts the analog signals into digital signals through a process similar to the process performed by the A/D converter 482.

[0147] The analysis unit 600 may be connected to a plurality of measuring units 400 via a plurality of connection ports 6008. When a plurality of measuring units 400 are provided, the analysis unit 600 may be connected to each of the measuring units 400 (for example, the plurality of measuring units 400 and the plurality of analysis units 600 are connected on a one-to-one basis, respectively).

[0148] As shown in FIG. 31, the processor 6001 and the parallel processing processor 6002 have configurations and functions similar to the processor 4831 and the parallel processing processor 4833 described above. The analysis software 6100 that analyzes the cell type of each measured cell operates on the processor 6001. The analysis software 6100 running on the processor 6001 has functions similar to the functions of the analysis software 4832 shown in FIG. 16. The analysis software 6100 analyzes the cell type of each measured cell by operations similar to those described with reference to FIG. 16, FIG. 17, FIG. 18, FIG. 21 and FIG. 22 and in the description related thereto. The analysis software 6100 transmits the result of the analysis of the cell type of each measured cell to the processing unit 300 via the interface node 6007. The interface node 6007 corresponds to, for example, Ethernet or USB. The interface node 6007 may be an interface capable of wireless communication (for example, WiFi or Bluetooth).

[0149] In the case of the cell analyzer 4000 of Configuration Example 5 shown in FIG. 33 and FIG. 34, in the flow chart shown in FIG. 20, step S13 (digital signal generation), step S14 (cell classification), and step S15 (classification information transmission) are performed in the analyzing unit 600. That is, digital signal generation (step S13) is performed in the A/D converter 6009 of the analyzing unit 600, cell classification (step S14) based on the digital signals is performed by the processor 6001 and the processor 6002 for parallel processing, and the classification information is transmitted from the analyzing unit 600 to the processing unit 300 (step S15). The processes in FIG. 21 and FIG. 22 are performed by the processor 6001 and the processor 6002 for parallel processing of the analysis block 600.

(Amount of waveform and digital signal data)

[0150] Next, the data volumes of the waveform data and the digital signal will be described. In the present embodiment, for example, with respect to each of the forward scattered light analog signal (FSC), the side scattered light analog signal (SSC), and the side fluorescence analog signal (SFL), sampling is performed for each cell at a plurality of time points at a certain interval. Examples of the sampling frequency include sampling at 1024 points at an interval of 10 nanoseconds, sampling at 128 points at an interval of 80 nanoseconds, sampling at 64 points at an interval of 160 nanoseconds, and the like. The amount of data is, for example, 2 bytes per sampling. With respect to each of the FSC, SSC, and SFL, data (in the case of a frequency of 1024 points, 2 bytes×1024=2048 bytes) in an amount corresponding to the sampling frequency are obtained. This amount of data is the data volume per cell. In one dimension, FSC, SSC and SFL are measured with respect to at least 100 cells, for example. Alternatively, in one dimension, FSC, SSC and SFL can be measured with respect to at least 1000 cells, for example. Alternatively, in one dimension, FSC, SSC, SFL can be measured, for example, with respect to from about 10000 to about 140000 cells. Thus, for example, when the number of cells measured in one dimension is 100000 and the sampling frequency is 1024, the data volume of the digital signal of each of FSC, SSC and SFL is 2 bytes×1024×100000=204800000 bytes, and the data volume for FSC, SSC and SFL in total is 614400000 bytes.

[0151] In addition, FSC, SSC and SFL are measured for each measurement channel. Thus, for example, when the number of cells measured in one measurement is 100000, the sampling frequency is 1024, and the number of measurement channels is 5, the data volume for each of FSC, SSC and SFL is 2 bytes×1024×100000×5=1024000000 bytes, and the data volume for FSC, SSC and SFL in total is 3072000000 bytes.

[0152] Thus, the volume of each digital signal is, for example, from several hundred megabytes to several gigabytes per sample, and is at least 1 gigabyte, depending on the number of cells, the sampling frequency and the number of measurement channels.

[0153] According to the present embodiment, when a digital signal having a huge volume of several hundred megabytes to several gigabytes per sample is analyzed, the analysis process using the deep learning algorithm 60 is contained within the cell analyzer 4000 or 4000' as described above, and the digital signal is not transmitted to a server for analysis provided outside the cell analyzer 4000 or 4000' via the Internet or an intranet. Therefore, a decrease in productivity associated with an increase in the communication load caused by transmitting the digital signal from the cell analyzer 4000 or 4000' to the server for analysis can be avoided.

<The second cell analyzer and the measurement of biological sample in the second cell analyzer>

[0154] As an example of the configuration of the second cell analyzer 4000', an example of a block diagram of a urine particle analyzer or a biological fluid analyzer is shown, in which the measuring unit 500 includes a flow cytometer for measuring a urine sample or a biological fluid sample.

[0155] FIG. 35 is an example of a block diagram of a measuring unit 500. In FIG. 35, the measuring unit 500 includes: a sample distribution unit 501; a sample preparation unit 502; an optical detector 505; an amplifying circuit 550 that amplifies an output signal (an output signal amplified by a preamplifier) of the optical detector 505; a filtering circuit 506 that performs a process of filtering the output signal from the amplifying circuit 550, an A/D converter 507 that converts the output signal (an analog signal) from the filtering circuit 506 into a digital signal; a processor 4831; a parallel processing processor 4833; a RAM 4834; a bus controller 4850; a storage device 511a; and a LAN (local area network) adapter 512. The processor 4831, the parallel processing processor 4833, the RAM 4834, the bus controller 4850, the storage device 511a and the LAN adapter 512 are connected to the bus 508. The deep learning algorithm 60 and the analytical software based on the deep learning algorithm 60 are stored in the storage device 511a.

[0156] As described based on the example of the first cell analyzer 4000, the processor 4831 performs specified arithmetic processes on the digital signal output from the A/D converter 507. The processor 4831 analyzes the waveform data using the parallel processing processor 4833. Examples of the configuration of the processor 4831 and the parallel processing processor 4833 and the operation of the parallel processing processor 4833 are similar to those described above with reference to FIG. 11 to FIG. 19 and in the description related thereto.

[0157] The processing unit 300 is connected to the measuring unit 500, for example, via a LAN cable through the LAN adapter 512, and the processing unit 300 performs the creation of a test result based on the measurement data obtained by the measuring unit 500. The optical detector 505, the amplifying circuit 550, the filtering circuit 506, the A/D converter 507, the processor 4831, the processor 4833 for parallel processing and the storage device 511a form an optical measurement unit 510, which measures the sample for measurement and creates the measurement data.

[0158] FIG. 36 shows a configuration of an optical detector 505 of a measuring unit 500. In FIG. 36, a condenser lens 552 condenses in a flow cell 551 laser light emitted from a light source 553 in the form of a semiconductor laser serving as a light source, and a condenser lens 554 condenses on a forward scattered light receiving unit 555 the forward scattered light emitted from a particle in a sample to be measured. Another condenser lens 556 condenses on a dichroic mirror 557 the side scattered light and fluorescence emitted from the particle. The dichroic mirror 557 reflects the side scattered light to the side scattered light receiving unit 558 and allows fluorescence to pass through it to the fluorescence receiving unit 559. These light signals reflect characteristics of the particle in the sample to be measured. The forward scattered light receiving unit 555, the side scattered light receiving unit 558 and the fluorescence receiving unit 559 convert light signals into electrical signals and output a forward scattered light signal, a side scattered light signal and a fluorescence signal, respectively. These output signals are amplified by a preamplifier and then subjected to subsequent processing. As for each of the forward scattered light receiving unit 555, the side scattered light receiving unit 558 and the fluorescence receiving unit 559, it is possible to perform switching of the low sensitivity output and the high sensitivity output by switching the control voltage. Switching of these sensitivities is performed by the processor 4831 described later. In the present embodiment, a photodiode may be used as the forward scattered light receiving unit 555, photomultipliers may be used as the side scattered light receiving unit 558 and the fluorescence receiving unit 559, or photodiodes may be used as the side scattered light receiving unit 558 and the fluorescence receiving unit 559. The fluorescence signal output from the fluorescence receiving unit 559 is amplified by a preamplifier and then provided to branched two signal channels. The two signal channels are connected to an amplifying circuit 550 described above with reference to FIG. 35. The fluorescence input to one of the signal channels is amplified by the amplifying circuit 550 so as to have a high sensitivity.

(Preparing the sample for measurement)

[0159] FIG. 37 is a schematic diagram showing a functional configuration of the sample preparation unit 502 and the optical detector 505 shown in FIG. 35. The sample distribution unit 501 shown in FIG. 35 and FIG. 37 includes a suction tube 517 and a syringe pump. The sample distribution unit 501 sucks a sample (urine or biological fluid) 00b through the suction tube 517 and distributes the sample to the sample preparation unit 502. The sample preparation unit 502 includes a reaction chamber 512u and a reaction chamber 512b. The sample distribution unit 501 distributes a quantified sample for measurement to each of the reaction chamber 512u and the reaction chamber 512b.

[0160] In the reaction chamber 512i, the distributed biological sample is mixed with the first reagent 519u as a diluent and the third reagent 518u containing a dye. Due to the dye contained in the third reagent 518u, the particles in the sample are colored. When the biological sample is urine, the sample prepared in the reaction chamber 512u is used as a first sample for measurement for analyzing urine particles that are relatively large, such as red blood cells, white blood cells, epithelial cells, or tumor cells. When the biological sample is a biological fluid, the sample prepared in the reaction chamber 512u is used as a third sample for measurement for analyzing red blood cells in the biological fluid.

[0161] At the same time, in the reaction chamber 512b, the distributed biological sample is mixed with the second reagent 519b as a diluent and the fourth reagent 518b containing a dye. As described later, the second reagent 519b has a hemolytic effect. Due to the dye contained in the fourth reagent 518b, the particles in the sample are colored. When the biological sample is urine, the sample prepared in the reaction chamber 512b serves as a second measurement sample for analyzing bacteria in the urine. When the biological sample is a biological fluid, the sample prepared in the reaction chamber 512b serves as a fourth measurement sample for analyzing nucleated cells (leukocytes and large cells) and bacteria in the biological fluid.

[0162] A tube extends from the reaction chamber 512u to the flow cell 551 of the optical detector 505, as a result of which the sample for measurement prepared in the reaction chamber 512u can be supplied to the flow cell 551. A solenoid valve 521u is provided at the outlet of the reaction chamber 512u. A tube also extends from the reaction chamber 512b, and this tube is connected to a part of the tube extending from the reaction chamber 512u. Accordingly, the sample for measurement prepared in the reaction chamber 512b can be supplied to the flow cell 551. A solenoid valve 521b is provided at the outlet of the reaction chamber 512u.

[0163] The tube extending from the reaction chambers 512u, 512b to the flow cell 551 branches before the flow cell 551, and the branched tube is connected to the syringe pump 520a. A solenoid valve 521c is provided between the syringe pump 520a and the branch point.

[0164] Between the connection point of the tubes passing from the respective reaction chambers 512u, 512b and the branch point, the tube is further branched, and the branched tube is connected to the syringe pump 520b. Between the branch point of the tube passing from the syringe pump 520b and the connection point, a solenoid valve 521d is provided.

[0165] The sample preparation unit 502 has a sheath fluid storage unit 522 connected thereto, which stores the sheath fluid, and the sheath fluid storage unit 522 is connected to the flow cell 551 via a tube. The sheath fluid storage unit 522 has a compressor 522a connected thereto, and when the compressor 522a is driven, compressed air is supplied to the sheath fluid storage unit 522, and the sheath fluid is supplied from the sheath fluid storage unit 522 to the flow cell 551.

[0166] Regarding the two kinds of suspensions (measurement samples) prepared in the respective reaction chambers 512u, 512b, the suspension (the first measurement sample when the biological sample is urine; the third measurement sample when the biological sample is a biological fluid) in the reaction chamber 512u is introduced into the optical detector 505, forming a thin stream surrounded by a sheathing liquid in the flow cell 551, and the thin stream is exposed to laser light. Then, similarly, the suspension (the second measurement sample when the biological sample is urine; the fourth measurement sample when the biological sample is a biological fluid) in the reaction chamber 512b is introduced into the optical detector 505, forming a thin stream in the flow cell 551, and the thin stream is exposed to laser light. Such operations are performed automatically, forcing solenoid valves 521a, 521b, 521c, 521d, drive unit 503, etc. to function under the control of processor 4831 (controller).

[0167] The first to fourth reagents will be described in detail hereinafter. The first reagent 519u is a reagent with a buffer as a main component, contains an osmotic pressure compensation agent so as to ensure the production of a stable fluorescence signal without hemolysis of erythrocytes, and is adjusted to 100-600 mOsm/kg so as to realize an osmotic pressure suitable for classification measurement. Preferably, the first reagent 519u does not have a hemolytic effect on erythrocytes in urine.

[0168] Unlike the first reagent 519u, the second reagent 519b has a hemolytic effect. This is intended to facilitate the passage of the fourth reagent 518b described below through the cell membranes of bacteria so as to promote staining. In addition, this is also intended to limit contaminants such as mucus fibers and red blood cell fragments. The second reagent 519b contains a surfactant to acquire a hemolytic effect. A variety of anionic, nonionic and cationic surfactants can be used as the surfactant, but a cationic surfactant is particularly suitable. Since the surfactant can destroy the cell membranes of bacteria, nucleic acids from bacteria can be effectively stained with the dye contained in the fourth reagent 518b. As a result, the measurement of bacteria can be performed by a short-term staining method.

[0169] As another embodiment, the second reagent 519b may acquire a hemolytic effect not by means of a surfactant, but by being adjusted to an acidic or low pH. Having a low pH means that the pH is lower than the pH of the first reagent 519u. When the first reagent 519u is in the range from neutral or slightly acidic to slightly alkaline, the second reagent 19b is acidic or strongly acidic. When the pH of the first reagent 519u is 6.0-8.0, the pH of the second reagent 519b is lower than this, and is preferably 2.0-6.0.

[0170] The second reactant 519b may comprise a surfactant and may be adjusted to a low pH.

[0171] As another embodiment, the second reagent 519b may acquire its hemolytic effect by having a lower osmotic pressure than the first reagent 19u.

[0172] At the same time, the first reagent 519u does not contain any surfactant. In another embodiment, the first reagent 519u may contain a surfactant, but its type and concentration need to be adjusted so as not to subject the red blood cells to hemolysis. Thus, it is preferable that the first reagent 519u does not contain the same surfactant as the second reagent 519b, or even if the first reagent 519u contains the same surfactant as the second reagent 519b, the concentration of the surfactant in the first reagent 519i is lower than in the second reagent 519b.

[0173] The third reagent 518u is a coloring reagent for use in measuring urine particles (erythrocytes, leukocytes, epithelial cells, casts, or the like). As the dye contained in the third reagent 518u, a dye that colors membranes is selected in order to also color particles that do not have nucleic acids. Preferably, the third reagent 518u contains an osmotic pressure compensation agent for the purpose of preventing hemolysis of erythrocytes and for the purpose of obtaining a stable fluorescence intensity, and is adjusted to 100-600 mOsm/kg so as to realize an osmotic pressure suitable for classification measurement. The cell membrane and the core (membrane) of the urine particles are colored by the third reagent 518u. As the coloring reagent containing the dye that colors membranes, a condensed benzene derivative is used, and, for example, a cyanine-based dye can be used. The third reagent 518u stains not only cell membranes but also nuclear membranes. When the third reagent 518u is used in nucleated cells such as leukocytes and epithelial cells, the staining intensity in the cytoplasm (cell membrane) and the staining intensity in the nucleus (nuclear membrane) are combined, resulting in a staining intensity higher than that of urine particles lacking nucleic acids. Accordingly, nucleated cells such as leukocytes and epithelial cells can be distinguished from urine particles lacking nucleic acids such as red blood cells. As the third reagent, reagents described in U.S. Patent Publication No. US 5,891,733 can be used. The contents of U.S. Patent Publication No. US 5,891,733 are incorporated herein by reference. The third reagent 518u is mixed with urine or a biological fluid, together with the first reagent 519u.

[0174] The fourth reagent 518b is a staining reagent that allows bacteria to be accurately measured even when the sample contains contaminants that are equivalent in size to bacteria and fungi. The fourth reagent 518b is described in detail in European Patent Application Publication No. EP 1 136 563. A dye that stains nucleic acids is suitably used as the dye contained in the fourth reagent 518b. For example, the cyanine-based dye of U.S. Patent No. US 7,309,581 can be used as the staining reagent containing the dye that stains nuclei. The fourth reagent 518b is mixed with urine or a sample, together with the second reagent 519b. The contents of European Patent Application Publication No. EP 1 136 563 and U.S. Patent No. US 7,309,581 are incorporated herein by reference.

[0175] Thus, preferably, the third reagent 518u contains a dye that stains cell membranes, while the fourth reagent 518b contains a dye that stains nucleic acids. The urine particles may include non-nucleated particles such as red blood cells. Thus, by using the third reagent 518u containing a dye that stains cell membranes, urine particles including non-nucleated particles can be detected. Since the second reagent can destroy the cell membranes of bacteria, nucleic acids of bacteria and fungi can be effectively stained with the dye contained in the fourth reagent 18b. As a result, the measurement of bacteria can be performed by a short-term staining method.

[3. The First Waveform Data Analysis System]

<Configuration of the first waveform data analysis system>

[0176] Another embodiment relates to a waveform data analysis system. With reference to FIG. 38, a first waveform data analysis system includes a deep learning device 100A. A device 100 on the provider side functions as the deep learning device 100A. The deep learning device 100A causes a neural network 50 to learn using training data and provides a deep learning algorithm 60 trained on the training data to a user. The deep learning algorithm 60, implemented as a trained neural network, is provided from the deep learning device 100A to the measuring unit 400 via a medium 98 or a communication network 99. The measuring unit 400 performs waveform data analysis for an analysis target cell using the deep learning algorithm 60, implemented as a trained neural network.

[0177] The deep learning device 100A is implemented, for example, as a general-purpose computer and performs a deep learning process based on the flow chart described below. The measuring unit 400 performs a process of analyzing waveform data based on the flow chart described below. The carrier 98 is, for example, a machine-readable non-transitory tangible storage medium such as a DVD-ROM or a USB memory.

[0178] The deep learning device 100A is connected to the measuring unit 400a or the measuring unit 500a. The configuration of the measuring unit 400a or the measuring unit 500a is similar to the configuration of the measuring unit 400 or the measuring unit 500 described above. The deep learning device 100A receives training data about a waveform obtained by the measuring unit 400a or the measuring unit 500a. The method for creating the training data about a waveform is as described above.

[0179] As shown in FIG. 38, the measuring unit 400a or the measuring unit 500a includes a flow cell 4113, 551. In the measuring unit 400a or the measuring unit 500a, a biological sample is sent to the flow cell 4113, 551. The biological sample supplied to the flow cell 4113, 551 is irradiated with light from a light source 4112, 553, and forward scattered light, side scattered light and side fluorescence emitted from each cell in the biological sample are detected by light detectors (4116, 4121, 4122, 555, 558, 559). The measuring unit 400a or the measuring unit 500a generates waveform data from the forward scattered light signal, the side scattered light signal and the side fluorescence signal obtained by detecting light by the light detectors (4116, 4121, 4122, 555, 558, 559), and transmits the waveform data to the device 100 on the supplier side.

<Deep Learning Device Hardware Configuration>

[0180] FIG. 39 shows an example of a block diagram of a device 100 on the provider side (deep learning device 100A). The device 100 on the provider side includes a processing node 10 (10A), an input node 16, and an output node 17.

[0181] The processing unit 10 includes, for example: a CPU 11 that performs the data processing described below, a memory 12 used as a working medium for data processing; a storage 13 that stores a program and the data to be processed described below; a bus 14 that transfers data between the units; an interface unit (I/F unit) 15 that performs input/output of data with respect to an external device; and a GPU 19. An input unit 16 and an output unit 17 are connected to the processing unit 10 via the interface unit 15. For example, the input unit 16 is an input device such as a keyboard or a mouse, and the output unit 17 is a display device such as a liquid crystal display. The GPU 19 functions as an accelerator that assists arithmetic processes (for example, parallel arithmetic processes) performed by the CPU 11. That is, in the description below, the processes performed by the CPU 11 also include the processes performed by the CPU 11 using the GPU 19 as an accelerator. The GPU 19 has a function equivalent to that of the above-described 4833 processor for parallel processing. In this case, instead of the GPU 19, a chip that is suitable for neural network computing may be installed. Examples of such a chip include FPGA, ASIC, and Myriad X (Intel).

[0182] To perform the process of each stage described below with reference to FIG. 43, the processing unit 10 has a program and a deep learning algorithm 50 previously stored in the storage 13, which are implemented as a neural network before being trained according to the present embodiment, for example, in an executable form. The executable form is a form created by converting a programming language by a compiler, for example. The processing unit 10 uses the program stored in the storage 13 to perform training processes of the neural network 50 before training it.

[0183] In the following description, unless otherwise specified, the processes executed by the processing unit 10 mean the processes executed by the CPU 11 based on the program and the neural network 50 stored in the storage 13 or the memory 12. The CPU 11 temporarily stores necessary data (intermediate data to be processed, etc.) using the memory 12 as a working medium and, as necessary, stores data to be stored for a long time, such as arithmetic results, in the storage 13.

<Analyzer hardware configuration>

[0184] The configuration of the measuring unit 400 or 500, which processes the waveform data based on the algorithm provided by the device 100 on the supplier side, is equivalent to the configuration described above. The measuring unit 400 or 500 may also have the function of the device 100 on the supplier side (deep learning device 100A) to force the neural network 50 to learn using the training data. In this case, the device 100 on the supplier side (deep learning device 100A) is not necessary.

[0185] In order to execute the process of each stage in the waveform data analysis process described below, the measuring unit controller 480 has a program and a deep learning algorithm 60, implemented as a trained neural network according to the present embodiment, in an executable form, previously stored in the storage 4835, for example. The executable form is a form created by converting a programming language by a compiler, for example. The measuring unit controller 480 uses the program and the deep learning algorithm 60 stored in the storage 4835 to execute the processes.

[0186] As shown in FIG. 40, the measuring unit controller 480 may update the program and the deep learning algorithm 60 stored in the storage 4835 via a communication network, for example, via the LAN adapter 481. As shown in FIG. 41, the above-described analysis unit 600 may transmit digital signals and analysis results received from the measuring unit 400 to the deep learning device 100A via the interface node 6011. For example, the analysis unit 600 transmits waveform data and classification information corresponding to the waveform data to the deep learning device 100A via the Internet. The deep learning device 100A performs a learning process based on the waveform data and the classification information corresponding to the waveform data transmitted from the analysis unit 600 to update the deep learning algorithm 60. The deep learning device 100A transmits the updated deep learning algorithm 60 to the analysis unit 600. The analysis unit 600 updates the algorithm stored in the storage 6004 by means of the deep learning algorithm 60 transmitted from the deep learning device 100A. For example, while the processor 6001 and the processor 6002 for parallel processing process the waveform data transmitted from the measuring unit 400, in parallel with this, the analysis unit 600 transmits the waveform data to the deep learning device 100A. After completing the analysis of the waveform data, i.e. after completing the creation of the classification information, the analysis unit 600 transmits the classification information to the deep learning device 100A.

[0187] In the description below, unless otherwise specified, the processes executed by the measuring unit controller 480 actually mean the processes executed by the processor 4831 and the processor 4833 for parallel processing based on the deep learning program and algorithm 60 stored in the storage 4835 or the RAM 4834. The processor 4831 temporarily stores necessary data (such as intermediate data to be processed, measurement results, analysis results, etc.) by, for example, using the RAM 4834 as a working environment, and, as necessary, stores data to be stored for a long time, such as arithmetic results, in the storage 4835.

<Functional block and processing procedure>

(Deep learning process)

[0188] FIG. 42 shows an example of a functional block of the deep learning device 100A that performs deep learning. Referring to FIG. 42, the processing unit 10A of the deep learning device 100A according to the present embodiment includes a training data creation unit 101, a training data input unit 102, and an algorithm updating unit 103. These functional blocks are implemented when: a program for causing a computer to execute a deep learning process is installed in a storage 13 or a memory 12 of the processing unit 10A shown in FIG. 39; and the program is executed by the CPU 11 and the GPU 19. The storage 13 or the memory 12 of the processing unit 10A stores a training data base (DB) 104 and an algorithm data base (DB) 105.

[0189] The training data 76a, 76b, 76c about the waveform are obtained in advance, for example, by means of the measuring unit 400, 500, and are stored in advance in the storage 13 or memory 12 of the processing node 10A. The deep learning algorithm 50 is stored in the algorithm database 105 in advance.

[0190] The processing unit 10A of the deep learning device 100A executes the processes shown in FIG. 43. With reference to the functional blocks shown in FIG. 42, the processes of steps S211, S214, and S216 shown in FIG. 43 are executed by the training data creation unit 101. The process of step S212 is executed by the training data input unit 102. The processes of steps S213 and S215 are executed by the algorithm update unit 103.

[0191] With reference to FIG. 43, an example of a deep learning process performed by the processing unit 10A will be described. First, the processing unit 10A obtains training data 72a, 72b, 72c about a waveform. The training data 72a about a waveform is data about a waveform of forward scattered light, the training data 72b about a waveform is data about a waveform of sideways scattered light, and the training data 72c about a waveform is data about a waveform of sideways fluorescence. The training data 72a, 72b, 72c about a waveform are obtained, for example, by performing an operation by an operator, from the measuring unit 400, 500, from the carrier 98, or through the I/F unit 15 via a communication network. When the training data 72a, 72b, 72c about the wave form are received, information about what type of cell the training data 72a, 72b, 72c about the wave form indicates is also received. The information about what type of cell is indicated can be associated with the training data 72a, 72b, 72c about the wave form, or can be input by the operator via the input unit 16.

[0192] In step S211, the processing unit 10A creates training data 75 from the waveform data 72a, 72b, 72c and the obtained label value 77.

[0193] In step S212, the processing unit 10A inputs the training data 75 into the neural network 50 to obtain a test result. The test result is accumulated each time each of the plurality of pieces of the training data 75 is input into the neural network 50.

[0194] In the cell type analysis method according to the present embodiment, a convolutional neural network is used and a stochastic gradient descent method is used. Thus, in step S213, the processing unit 10A determines whether or not the test results of a predetermined set number of test times have been accumulated. When the test results of the set number of test times have been accumulated (YES), the processing unit 10A proceeds to the process of step S214. When the test results of the set number of test times have not been accumulated (NO), the processing unit 10A proceeds to the process of step S215.

[0195] Then, when the test results of a predetermined number of test times have been accumulated, the processing unit 10A updates, in step S214, the link weight w of the neural network 50 using the training results accumulated in step S212. In the cell type analysis method according to the present embodiment, since a stochastic gradient descent method is used, the link weight w of the neural network 50 is updated in the step when the test results of a predetermined number of test times have been accumulated. Specifically, the process of updating the link weight w is a process of performing calculations in accordance with the gradient descent method represented by formula 12 and formula 13 described below.

[0196] In step S215, the processing unit 10A determines whether or not the neural network 50 has been trained using the predetermined number of pieces of training data 75. When training has been performed using the predetermined number of pieces of training data 75 (YES), the deep learning process ends.

[0197] When the neural network 50 has not been trained using the specified number of pieces of training data 75 (NO), the processing unit 10A proceeds from step S215 to step S216 and performs the processes from step S211 to step S215 with respect to the next training waveform data.

[0198] In accordance with the processes described above, the neural network 50 is trained, resulting in a deep learning algorithm 60.

(Neural network structure)

[0199] As described above, in the present embodiment, a convolutional neural network is used. In (a) of FIG. 44, an example of the structure of the neural network 50 is shown. The neural network 50 includes an input layer 50a, an output layer 50b, and a middle layer 50c between the input layer 50a and the output layer 50b. The middle layer 50c is composed of a plurality of layers. The number of layers forming the middle layer 50c may be, for example, 5 or more, preferably 50 or more, and more preferably 100 or more.

[0200] In the neural network 50, a plurality of nodes 89, ranked by layers, are connected between the layers. Accordingly, information propagates in only one direction, indicated by arrow D in the figure, from the input layer 50a to the output layer 50b.

(Arithmetic operation at each node)

[0201] (b) in FIG. 44 is a schematic diagram showing arithmetic operations performed in each node. Each node 89 receives a plurality of inputs and calculates one output (z). In the case of the example shown in (b) in FIG. 44, the node 89 receives four inputs. The total input (u) received by the node 89 is represented by formula 2 below as an example. In the present embodiment, one-dimensional matrix data is used as the training data 75 and the analysis data 85. Thus, when the variables of the arithmetic expression correspond to two-dimensional matrix data, a process of transforming the variables so that they correspond to one-dimensional matrix data is performed.

[Mathematical expression 2]

Each input is multiplied by a different weight. In formula 2, b is a value called the bias. The output (z) of the node serves as the output of a given function f with respect to the overall input (s) represented by formula 2, and is represented by formula 3 below. The function f is called the activation function.

[Mathematical expression 3]

(c) in FIG. 44 is a schematic diagram showing arithmetic operations between nodes. In the neural network 50, nodes, each of which outputs a result (z), represented by formula 3, with respect to a common input (u) to node 89, represented by formula 2, are ranked by layers. Outputs from nodes of a previous layer serve as inputs to nodes of the next layer. In the example shown in (c) in FIG. 44, outputs of node 89a in the left layer serve as inputs to nodes 89b in the right layer. Each node 89b receives outputs from nodes 89a. The connection between each node 89a and each node 89b is multiplied by a different weight. When the corresponding outputs from a plurality of nodes 89a are determined as x1-x4, the inputs to the corresponding three nodes 89b are represented by formulas 4-1 through 4-3 below.

[Mathematical expression 4]

Generalizing formulas 4-1 through 4-3 yields formula 4-4. Here i=1, …, I, j=1,…, J (I is the total number of inputs, and J is the total number of outputs).

[Mathematical expression 5]

When applying the 4-4 formula to the activation function, the output is given by the formula 5 below.

[Mathematical expression 6]

(Activation function)

In the cell type analysis method according to the embodiment, a linear rectification block function is used as the activation function. The linear rectification block function is represented by formula 6 below.

[Mathematical expression 7]

Formula 6 is the function obtained by setting u=0 to the u<0 portion of the linear function with z=u. In the example shown in (c) of FIG. 44, using Formula 6, the output from node j=1 is represented by the formula below.

[Mathematical expression 8]

(Neural network training)

If a function expressed using a neural network is defined as y(x:w), the function y(x:w) changes when the parameter w of the neural network changes. Correcting the function y(x:w) such that the neural network selects a more appropriate parameter w with respect to the input x is called training/educating the neural network. It is assumed that a plurality of pairs of input and output of a function expressed using a neural network are represented. If a desired output for an input x is defined as d, the input/output pairs are given as {(x1,d1), (x2,d2), ..., (xn,dn)}. A set of pairs where each is expressed as (x,d) is called training data. Specifically, a set of pieces of waveform data (forward scattered light waveform data, side scattered light waveform data, fluorescence waveform data) shown in FIG. 3 is the training data shown in FIG. 3.

[0202] Training a neural network means adjusting the weight w such that, for any input/output pair (xn,dn), the output y(xn:w) of the neural network for a given input xn becomes as close as possible to the output dn. The error function is a measure of the closeness

[Mathematical expression 9]

between the training data and the function expressed using the neural network. The error function is also called the loss function. The error function E(w) used in the cell type analysis method according to the embodiment is represented by the following formula 7. Formula 7 is called the cross entropy.

[Mathematical expression 10]

Next, a method for calculating the cross entropy of formula 7 will be described. In the output layer 50b of the neural network 50 used in the cell type analysis method according to the embodiment, i.e., in the last layer of the neural network, an activation function is used to classify the input x into a finite number of classes according to the content. The activation function is denoted as a softmax function and is represented below by formula 8. It is assumed that in the output layer 50b, the nodes are ranked by the same number as the number of classes k. It is assumed that the total input u from each node k (k=1, …, K) of the output layer L is given as uk ^(L) from the outputs from the previous layer L-1. Accordingly, the output from the k-th node in the output layer is represented below by formula 8.

[Mathematical expression 11]

Formula 8 is a softmax function. The sum of the outputs y1, …, yK, determined by formula 8, is always 1.

[0203] When each class is expressed as C1, …, CK, the output yK from node k in the output layer L (i.e., uk ^(L)) represents the probability with which a given input x belongs to class CK See Formula 9 below. The input x is classified into the class in which the probability represented by Formula 9 becomes highest.

[Mathematical expression 12]

When training a neural network, the function expressed by the neural network is regarded as a model of the posterior probability of each class, and the likelihood of the weight w with respect to the training data is estimated in such a probability model, and the weight w that maximizes the likelihood is selected.

[0204] It is assumed that the target output dn by the softmax function of formula 8 is 1 only when the output is the correct class, and otherwise, the target output dn is 0. When the target output is expressed in the vector form dn=[dn1, …, dnK], if, for example, the correct class of the input xn is C3, only the target output dn3 becomes 1, and the other target outputs become 0. When encoding in this way, the posterior distribution is represented below by formula 10.

[Mathematical expression 13]

The likelihood L(w) of weight w applied to training data {(xn, dn)}(n=1,…, N) is given below by formula 11. Taking the logarithm of the likelihood L(w) and changing the sign yields the error function of formula 7.

[Mathematical expression 14]

Training means minimizing the error function E(w) calculated based on the training data with respect to the parameter w of the neural network. In the cell type analysis method according to the embodiment, the error function E(w) is represented by formula 7.

[0205] Minimizing the error function E(w) with respect to a parameter w has the same meaning as finding a local minimum point of the function E(w). The parameter w represents the weight of the connection between nodes. The local minimum point of the weight w is obtained by iteratively calculating a repeated update of the parameter w from an arbitrary initial value used as the starting point. One example of such a calculation is the gradient descent method.

[0206] The gradient descent method uses a vector represented by formula 12 below.

[Mathematical expression 15]

In the gradient descent method, the process of advancing the value of the current parameter w in the direction of the negative gradient (i.e., -VE) is repeated many times. When the current weight is w ^(t) and the weight after advancing is w ^(t+1) , the arithmetic operation according to the gradient descent method is represented by the following formula 13. The value t means the number of times the parameter w is advancing.

[Mathematical expression 16]

Symbol

[Mathematical expression 17]

is a constant that determines the range of the updated value of the parameter w and is called the learning coefficient. As a result of repeating the arithmetic operation represented by formula 13, the error function E(w ^(t) ) decreases due to the increase in the value of t, and the parameter w reaches a local minimum point.

[0207] It should be noted that the arithmetic operation according to formula 13 can be performed for all training data (n=1, …, N) or can be performed only for part of the training data. The gradient descent method performed only for part of the training data is called a stochastic gradient descent method. In the cell type analysis method according to the embodiment, the stochastic gradient descent method is used.

[4. Designing a Deep Learning Model]

[0208] Using the Sysmex XN-1000, blood collected from a healthy individual was measured as a healthy blood sample, and XN CHECK Lv2 (control blood from Streck (subjected to treatment such as fixation)) was measured as an unhealthy blood sample. Fluorocell WDF manufactured by Sysmex Corporation was used as a fluorescence staining reagent. Lysercell WDF manufactured by Sysmex Corporation was used as a hemolytic agent. For each cell contained in each biological sample, forward scattered light, side scattered light, and side fluorescence waveform data were acquired at 1024 points at an interval of 10 nanoseconds from the start of forward scattered light measurement. For the healthy blood sample, waveform data of cells in the blood collected from 8 healthy individuals were pooled in the form of digital signals. For each cell waveform data, the classification of neutrophils (NEUTR), lymphocytes (LYMPH), monocytes (MONOC), eosinophils (EOS), basophils (BASO), and immature granulocytes (IG) was performed manually, and each piece of waveform data was annotated with a cell type label. The time point at which the forward scattered light signal intensity exceeded the threshold was defined as the measurement start time point, and the acquisition times of the forward scattered light, side scattered light, and side fluorescence waveform data pieces were synchronized with each other to create training data. In addition, the control blood was annotated with a “control blood-derived cell (CONTR)”. The training data was input into the deep learning algorithm to train it.

[0209] For blood cells from a different healthy individual than the healthy individual from which the training cell data was obtained, waveform analysis data was obtained using the Sysmex XN-1000 in a manner similar to that for the training data. Waveform data from the control blood were mixed to create analysis data. These analysis data were fed into a constructed deep learning algorithm to obtain individual cell type data.

[0210] FIG. 45 shows the result in the form of a mixed matrix. The horizontal axis represents the determination result of the designed deep learning algorithm, and the vertical axis represents the determination result obtained manually by a human (reference method). Although a slight confusion was observed between basophils and lymphocytes and between basophils and ghosts in the determination result of the designed deep learning algorithm, the determination result of the designed deep learning algorithm demonstrated a match rate of 98.8% with the determination result of the reference method.

[0211] Then, for each cell type, ROC analysis was performed and the sensitivity and specificity were evaluated. In (a) of FIG. 46, the ROC curve of neutrophil is shown, in (b) of FIG. 46, the ROC curve of lymphocyte is shown, in (c) of FIG. 46, the ROC curve of monocyte is shown, in (a) of FIG. 47, the ROC curve of neutrophil is shown, in (b) of FIG. 47, the ROC curve of basophil is shown, and in (c) of FIG. 47, the ROC curve of control blood (CONTR). The sensitivity and specificity were, respectively, 99.5% and 99.6% for neutrophils, 99.4% and 99.5% for lymphocytes, 98.5% and 99.9% for monocytes, 97.9% and 99.8% for eosinophils, 71.0% and 81.4% for basophils and 99.8% and 99.6% for control blood (CONTR). These were good results.

[0212] From the above results, it was found that cell types can be determined with high classification accuracy using a deep learning algorithm based on signals obtained from cells contained in a biological sample and based on waveform data.

[0213] Furthermore, in some cases where unhealthy blood cells such as control blood were mixed with healthy blood cells, it was difficult to determine the type of each individual cell from a conventional scatter plot. However, it was shown that by using the deep learning algorithm of the present embodiment, even when unhealthy blood cells were mixed with healthy blood cells, it was possible to determine each individual cell.

[5. Analysis system using image analyzer]

[0214] Next, an embodiment will be described in which an image analyzer is used as a cell analyzer. The third cell analyzer 4000'', which is an image analyzer, analyzes the captured image data, thereby estimating the cell type for each cell whose image was captured.

[0215] FIG. 48 shows an example of a configuration of a cell analyzer 4000''. The cell analyzer 4000'' shown in FIG. 48 includes a measuring unit 700 and a processing unit 800, measures a sample 901 prepared by pre-processing by a pre-processing device 900, and performs analysis.

[0216] The measuring unit 700 includes a flow cell 710, light sources 720-723, condenser lenses 730-733, dichroic mirrors 740-741, a condenser lens 750, an optical unit 751, a condenser lens 752, and an imaging unit 760. The sample 701 is caused to flow in the flow path 711 of the flow cell 710.

[0217] Each of the light sources 720-723 supplies light to the sample 701 flowing in the flow cell 710. Each of the light sources 720-723 is implemented, for example, by a light source in the form of a semiconductor laser. Light with wavelengths λ11-λ14 is emitted from the light sources 720-723, respectively. The condenser lenses 730-733 condense the light with wavelengths λ11-λ14 emitted from the light sources 720-723, respectively. The dichroic mirror 740 allows light with a wavelength of λ11 to pass through it and reflects light with a wavelength of λ12. The dichroic mirror 741 allows light with a wavelength of λ11 and λ12 to pass through it and reflects light with a wavelength of λ13. Thus, different light with wavelengths λ11-λ14 acts on the sample 701 flowing in the flow channel 711 of the flow cell 710. The number of light sources in the form of semiconductor lasers of the measuring unit 700 is not particularly limited, provided that their number is 1 or more. The number of light sources in the form of semiconductor lasers can be selected, for example, from 1, 2, 3, 4, 5 or 6.

[0218] In the case where the sample 701 flowing in the flow cell 710 has been stained with a fluorescent dye, when different light with wavelengths of λ11-λ13 is exposed to the sample 701, fluorescence from the fluorescent dye staining each cell is generated. For example, fluorescence with wavelengths of λ21, λ22, λ23 corresponding to wavelengths of λ11, λ12, λ13, respectively, is generated. When light with a wavelength of λ14 is exposed to the sample 701 flowing in the flow cell 710, this light passes through each cell. The transmitted light having a wavelength of λ14 and having passed through the cell is used to create a bright field image.

[0219] The condenser lens 750 condenses the fluorescence generated from the sample 701 flowing in the flow path 711 of the flow cell 710 and the transmitted light that has passed through the sample 701 flowing in the flow path 711 of the flow cell 710. The optical unit 751 has a configuration in which four dichroic mirrors are combined. The four dichroic mirrors of the optical unit 751 reflect the fluorescence and the transmitted light at angles slightly different from each other in order to separate them on the light-receiving surface of the imaging unit 760. The condenser lens 752 condenses the fluorescence and the transmitted light.

[0220] The imaging unit 760 is implemented by a TDI (time delay and accumulation) camera. The imaging unit 760 can capture images of fluorescences and transmitted light and output to the processing unit 800 a fluorescence image corresponding to the fluorescences and a bright field image corresponding to the transmitted light as imaging signals.

[0221] The processing unit 800 includes a processing unit 811, a storage 812, an interface unit 816, and a bus 815 as a hardware configuration. The processing unit 811, the storage 812, and the interface unit 816 are connected to the bus 815. Image data (for example, fluorescence images, bright field images) generated by imaging signals captured by the imaging unit 760 of the measuring unit 700 are stored in the storage 812 via the interface unit 816. The processing unit 811 reads the image data from the storage 812 and analyzes this image data.

[0222] FIG. 49 shows an example of a configuration of the processing unit 811. The processing unit 811 includes, for example, a processor 8111, a parallel processing processor 8112, and a RAM 8113. The processor 8111, the parallel processing processor 8112, and the RAM 8113 have configurations and functions similar to the above-described processor 4831, the parallel processing processor 4833, and the RAM 4834, respectively. The image data captured by the imaging unit 760 is analyzed by the processor 8111 and the parallel processing processor 8112. The parallel processing processor 8112 executes, for example, the processes shown as an example in FIG. 16, FIG. 17, and FIG. 18, arithmetic processes for matrix data related to image data in parallel by a plurality of arithmetic units.

[0223] The processing unit 800 does not necessarily have to have the functions (processor 8111, processor 8112 for parallel processing and RAM 8113) of the processing unit 811, and these functions may be provided by the measuring unit 700.

<Creating training data>

[0224] Next, an example of creating training data in the present embodiment will be described.

[0225] Preferably, training images to be used for training the deep learning algorithm are captured in RGB colors, CMY colors, or the like. Preferably, with respect to the color image, the darkness/paleness or brightness of each of the primary colors such as red, green, and blue, or cyan, magenta, and yellow is expressed by a 24-bit value (8 bits x 3 colors). It is sufficient for each training image to include at least one hue and the darkness/paleness or brightness of such a hue, but more preferably, it includes at least two hues and the darkness/paleness or brightness of each hue. Information including a hue and the darkness/paleness or brightness of a hue is also called a tone.

[0226] The tone information of each pixel in the training image is converted, for example, from RGB colors to a format that includes brightness information and hue information. Examples of a format that includes brightness information and hue information include YUV (YCbCr, YPbPr, YIQ, etc.). An example of conversion to the YCbCr format will be described here. The training image captured in RGB colors is converted to image data based on brightness, image data based on a first hue (for example, a bluish color), and image data based on a second hue (for example, a reddish color). The conversion from RGB to YCbCr can be performed in a known manner. For example, the conversion from RGB to YCbCr can be performed in accordance with the international standard ITU-R BT.601. The image data based on the brightness, the image data based on the first hue, and the image data based on the second hue can be expressed in the form of gradation value matrix data as shown in FIG. 50 (hereinafter also referred to as tone matrix data 72y, 72cb, 72cr). The image data based on the brightness, the image data based on the first hue, and the image data based on the second hue can be expressed in 256 gradations consisting of, for example, gradations from 0 to 255. In this case, instead of the brightness, the first hue, and the second hue, the training image can be converted into three primary colors of red R, green G, and blue B, or into three primary colors of pigment cyan C, magenta M, and yellow Y.

[0227] Then, based on the tone matrix data 72y, 72cb, 72cr, tone vector data 74 is created for each pixel by combining three gradation values 72y, the first hue 72cb, and the second hue 72cr.

[0228] Then, for example, assuming that an image of a segmented neutrophil is captured in the training image, each piece of tone vector data 74 created from the training image is assigned "1" as a label value 77 indicating a segmented neutrophil, thereby obtaining training data 75. In FIG. 50, for convenience, the training data 75 is expressed as 3 pixels × 3 pixels. However, in fact, the tone vector data exists according to the number of pixels with which the training image was captured.

[0229] FIG. 51 shows an example of a label value 77. Regarding the label value, a label value 77 is assigned that is different according to the cell type and the presence or absence of a feature of each cell.

<General Framework for Deep Learning>

[0230] With reference to FIG. 50, used as an example, a general scheme for training a neural network will be described. The neural network 50 is preferably a convolutional neural network. The number of nodes of the input layer 50a in the neural network 50 corresponds to the product of the number of pixels in the training data 75 to be input and the number of luminances and hues included in the image (for example, in the above example, three, i.e., luminance 72y, the first hue 72cb and the second hue 72cr). The fragments of tone vector data 74 are input in the form of a set 76 thereof into the input layer 50a of the neural network 50. The neural network 50 is trained using, for the output layer 50b of the neural network, the value of the label 77 of each pixel of the training data 75.

[0231] Based on the training data 75, the neural network 50 extracts quantitative feature values related to the morphological cell types and cell features. The output layer 50b of the neural network outputs a result reflecting these quantitative feature values.

[0232] Reference numeral 50c in FIG. 50 represents a middle layer.

[0233] The deep learning algorithm 60 with the neural network 60 trained in this manner is used as a discriminator to identify which of a plurality of cell types that belong to a given group of cells and that are morphologically classified corresponds to the target cell of the analysis.

<Image Analysis Method>

[0234] FIG. 52 shows an example of a method for analyzing an image. In the method for analyzing an image, analysis data 81 is created from an image for analysis obtained by capturing an image of an analysis target cell. The image for analysis is an image obtained by capturing an image of an analysis target cell.

[0235] For example, in the present embodiment, it is preferable that the image captured by the imaging device is in RGB colors, CMY colors, and the like. Preferably, with respect to the color image, the darkness/paleness or brightness of each of the primary colors such as red, green, and blue, or cyan, magenta, and yellow is expressed by a 24-bit value (8 bits x 3 colors). It is sufficient that the image for analysis includes at least one hue and the darkness/paleness or brightness of such a hue, but more preferably, it includes at least two hue and the darkness/paleness or brightness of each hue. Information including the hue and the darkness/paleness or brightness of the hue is also called a tone.

[0236] For example, conversion is performed from RGB colors to a format that includes brightness information and hue information. Examples of a format that includes brightness information and hue information include YUV (YCbCr, YPbPr, YIQ, etc.). An example of conversion to the YCbCr format will be described here. A training image in RGB colors is converted into image data based on brightness, image data based on a first hue (for example, a bluish color), and image data based on a second hue (for example, a reddish color). Conversion from RGB to YCbCr can be performed in a known manner. For example, conversion from RGB to YCbCr can be performed in accordance with the international standard ITU-R BT.601. Portions of image data corresponding to brightness, the first hue, and the second hue, respectively, can be expressed in the form of matrix data of gradation values, as shown in FIG. 52 (hereinafter referred to as 79y, 79cb, 79cr tone matrix data). The brightness, first hue and second hue of 72Cr can be expressed in 256 gradations consisting of gradations from 0 to 255. In this case, instead of brightness, first hue and second hue, the training image can be converted into three primary colors of red R, green G and blue B, or into three primary colors of pigment cyan C, magenta M and yellow Y.

[0237] Then, based on the tone matrix 79y, 79cb, 79cr, for each pixel, tone vector data 80 is created by combining three gradation values 79y, the first hue 79cb and the second hue 79cr. As analysis data 81, a set of tone vector data 80 created from one image for analysis is created.

[0238] Preferably, the creation of the analysis data 81 and the creation of the training data 75 are performed under at least the same conditions of image capture and the same conditions of creating the vector data that are input from each image into the neural network.

[0239] The analysis data 81 are input to the input layer 60a of the neural network 60 forming the trained deep learning algorithm 60. The deep learning algorithm extracts quantitative feature values from the analysis data 81 and outputs the result from the output layer 60b of the neural network 60. The value output from the output layer 60b represents the probability with which the analysis target cell included in the image for analysis belongs to each of the morphological classifications of cells and features input as training data.

[0240] An analysis target cell included in the image for analysis is determined as belonging to a morphological classification having the highest value among probabilities, and a label value associated with the morphological type of the cell or the cell feature is output. The label value itself or data obtained by replacing the label value with information (e.g., a term) indicating the morphological type of the cell or the presence or absence of the cell feature is output as the analysis result 83 regarding the cell morphology. In FIG. 52, based on the analysis data 81, the label value "1" is output by the discriminator as the label value 82 having the highest probability, and the character data "segmented neutrophil" corresponding to this label value is output as the analysis result 83 regarding the cell morphology.

[0241] Reference numeral 60c in FIG. 52 represents a middle layer.

[6. Other embodiments]

[0242] Although general arrangements and specific embodiments of the present invention have been described, the present invention is not limited to the general arrangements and embodiments described above.

[0243] In the above embodiments, the functional blocks of the training data creation unit 101, the training data input unit 102, the algorithm update unit 103, the analysis data creation unit 201, the analysis data input unit 202, and the analysis unit 203 are executed in a single processor 4831 and a single processor 4833 for parallel processing. However, these functional blocks do not necessarily have to be executed in a single processor and a single processor for parallel processing, and they may be executed in a distributed manner by a plurality of processors and a plurality of processors for parallel processing.

[0244] In the above embodiments, the program for performing the process of the steps described with reference to FIG. 43 is previously stored in the storage 4835. Instead, the program can be installed in the controller 480 of the measuring unit from a machine-readable non-transitory tangible storage medium 98, such as, for example, a DVD-ROM or a USB memory. Alternatively, the controller 480 of the measuring unit can be connected to the communication network 99, and the program can be downloaded and installed via the communication network 99, for example, from an external server (not shown).

[0245] FIG. 53 shows an embodiment of the analysis result. FIG. 53 shows: the cell types contained in the biological sample measured by the flow cytometry method, which are assigned the label values shown in FIG. 4; and the number of cells of each type. Instead of displaying the number of cells, or together with displaying the number of cells, it is possible to output a proportion (for example, %) of each cell type relative to the total number of cells that were counted. The cell count can be obtained by counting the number of label values (the number of identical label values) corresponding to each cell type that was output. In the final result, a warning may be output indicating that abnormal cells are contained in the biological sample. FIG. 53 shows an example in which an exclamation mark is provided as a warning in the abnormal cell column, but such a warning is not limited to this. In addition, the distribution of each cell type can be plotted in the form of a scatter plot, and the scatter plot can be output. When plotting a scatterplot, for example, the highest values during signal intensities can be plotted on a graph such that the vertical axis represents the intensity of sideward fluorescence and the horizontal axis represents the intensity of sideward scattered light.

DESCRIPTION OF REFERENCE POSITIONS

[0246] 50 - deep learning algorithm before training

60 - trained deep learning algorithm

400, 400a, 500, 500a, 700 - measuring unit

410 - PCM detector

450 - sample suction unit

482, 507, 3008, 6009 - AID converter

3001, 4831, 6001, 8111 - processor (master processor)

3002, 4833, 6002, 8112 - parallel processing processor

4000, 4000', 4000'' - cell analyzer

3200, 6200, 4836 - arithmetic block

Claims (39)

1. A method for analyzing cells using a cell analyzer comprising a first processor, a second processor, and a flow cytometer, including: obtaining, based on control of the first processor, matrix data about each of a plurality of cells in the sample by flowing the sample in a flow path irradiated by a light beam and detecting signals from irradiated cells in the sample, wherein the matrix data has as matrix data elements values each of which corresponds to a signal intensity obtained at a time point while the cell is irradiated in the flow path; executing, by a second processor, parallel processing for processing matrix data in accordance with an artificial intelligence (AI) algorithm comprising a plurality of matrix operations; and classifying by the first processor based on the result of parallel processing by the second processor the cell type for each of the plurality of cells. 2. A cell analyzer equipped with a flow cytometer, containing: a measuring unit configured to measure a plurality of cells contained in a sample; a first processor configured to process information related to the analysis of a plurality of cells; and a second processor capable of performing parallel processing, wherein the measuring unit obtains matrix data about each of a plurality of cells by flowing a sample in a flow path irradiated by a light beam and detecting signals from the irradiated cells in the sample, wherein the matrix data has as matrix data elements values, each of which corresponds to a signal intensity obtained at a time point while the cell is irradiated in the flow path, the second processor performs parallel processing for processing matrix data in accordance with an artificial intelligence (AI) algorithm containing a plurality of matrix operations, and the first processor classifies, based on the result of parallel processing by the second processor, a cell type for each of the plurality of cells. 3. A cell analyzer according to claim 2, wherein the cell analyzer comprises a light source for irradiating cells passing through the flow path with light and a detector receiving an optical signal supplied from the light source. 4. The cell analyzer of claim 2, wherein the detector receives multiple types of optical signals. 5. The cell analyzer according to claim 2, including transmitting data to the second processor via a transmission line included in the cell analyzer. 6. The cell analyzer of claim 2, comprising transmitting data to the second processor via a transmission line other than the Internet or an intranet. 7. The cell analyzer according to item 2, wherein the transmission line has a communication channel bandwidth of at least 1 gigabit/second. 8. The cell analyzer of claim 2, wherein the transmission line is a bus, and the data is transmitted to the second processor via the bus. 9. A cell analyzer according to any one of claims 5-8, wherein data obtained by converting an analog signal related to each of the plurality of cells measured by the flow cytometer are transmitted to the second processor via a transmission line. 10. A cell analyzer according to any one of claims 2-6, wherein the information includes an identifier for identifying the cell type. 11. A cell analyzer according to any one of claims 2-8, wherein the information includes a probability with which the cell belongs to each of the plurality of cell types. 12. A cell analyzer according to any one of claims 2 to 9, comprising transmitting the analysis result, including an identifier for identifying the cell type, to a processing unit that performs an analysis of the analysis result. 13. A cell analyzer according to any one of claims 2 to 10, comprising transmitting an analysis result, including a probability with which a cell belongs to each of a plurality of cell types, to a processing unit that performs an analysis of the analysis result. 14. A cell analyzer according to any one of claims 2-11, wherein the second processor executes, in parallel as parallel processing, a plurality of arithmetic processes related to data analysis. 15. A cell analyzer according to any one of claims 2 to 12, wherein the second processor has a plurality of arithmetic units, each of which is capable of executing an arithmetic process related to data analysis, and the second processor executes, in parallel as parallel processing, the arithmetic processes by means of the corresponding arithmetic units. 16. A cell analyzer according to any one of paragraphs 2-14, wherein the artificial intelligence algorithm is a deep learning algorithm. 17. A cell analyzer according to any one of claims 2-14, wherein the second processor executes, in parallel as parallel processing, a filtering process for extracting a data feature. 18. A cell analyzer according to any one of claims 2 to 17, wherein the artificial intelligence algorithm is a deep learning algorithm, and the second processor executes, in parallel as parallel processing, a plurality of arithmetic processes in a convolutional layer in the deep learning algorithm. 19. A cell analyzer according to any one of paragraphs 2-16, wherein the second processor performs parallel processing in accordance with a single order. 20. A cell analyzer according to any one of claims 2-19, wherein the second processor performs a plurality of matrix operations for each fragment of matrix data. 21. A cell analyzer according to any one of claims 2-19, wherein the second processor performs at least 100 matrix operations for each fragment of matrix data. 22. A cell analyzer according to any one of claims 2-19, wherein the second processor performs at least 1000 matrix operations for each fragment of matrix data. 23. A cell analyzer according to any one of claims 2-22, wherein the second processor analyzes matrix data corresponding to each of at least 100 said cells. 24. A cell analyzer according to any one of claims 2-22, wherein the second processor analyzes matrix data corresponding to each of at least 1000 said cells. 25. A cell analyzer according to any one of claims 2-24, wherein the second processor analyzes matrix data having a volume of at least 1 gigabyte. 26. A cell analyzer according to any one of claims 2 to 25, wherein the second processor has at least 10 arithmetic units, each of which is capable of executing an arithmetic process related to the analysis of matrix data, and the second processor executes, in parallel as parallel processing, the arithmetic processes by means of the corresponding arithmetic units. 27. A cell analyzer according to any one of claims 2 to 25, wherein the second processor has at least 100 arithmetic units, each of which is capable of executing an arithmetic process related to the analysis of matrix data, and the second processor executes, in parallel as parallel processing, the arithmetic processes by means of the corresponding arithmetic units. 28. A cell analyzer according to any one of claims 2 to 25, wherein the second processor has at least 1000 arithmetic units, each of which is capable of executing an arithmetic process related to the analysis of matrix data, and the second processor executes, in parallel as parallel processing, the arithmetic processes by means of the corresponding arithmetic units. 29. A cell analyzer according to any one of claims 2 to 28, wherein the second processor performs parallel processing using, as input, matrix data read from a memory having a capacity of at least 1 gigabyte. 30. A cell analyzer according to any one of claims 2 to 29, wherein the data about a cell in a sample is a combination of data fragments based on a plurality of types of signals received from the cell.