Disease differentiation support methods, disease differentiation support devices, programs

Abstract

We propose a new method to support the differential diagnosis of myelodysplastic syndromes, such as differentiating between myelodysplastic syndromes and other conditions. [Solution] The computer-based disease differentiation support method obtains a first parameter regarding the classification and abnormal morphology of cells obtained by analyzing images of cells contained in a sample taken from a subject, and a second parameter regarding the number of each type of cell contained in the sample. Using a pre-trained computer algorithm, it generates differentiation support information to assist in differentiating myelodysplastic syndrome from other conditions based on the first and second parameters.

Description

[Technical Field]

[0001] The present invention relates to a disease differentiation support method and the like. [Background Art]

[0002] Myelodysplastic syndromes (hereinafter appropriately referred to as "MDS (Myelo Dysplastic Syndromes)") is a disease that often progresses to acute myeloid leukemia (hereinafter appropriately referred to as "AML (Acute Myeloid Leukemia)") and the like, and early detection is considered important. Patent Document 1 discloses a genetic diagnosis technique for MDS using a next-generation sequencer. Patent Document 2 discloses a technique for differentiating between MDS and aplastic anemia (hereinafter appropriately referred to as "AA (Aplastic Anemia)") based on information on cell types and abnormal findings obtained by analyzing blood cell images. Patent Document 3 discloses a technique for differentiating between polycythemia vera (hereinafter appropriately referred to as "PV (Polycythemia Vera)"), essential thrombocythemia (hereinafter appropriately referred to as "ET (Essential Thrombocythemia)"), and primary myelofibrosis (hereinafter appropriately referred to as "PMF (Primary Myelo Fibrosis)") by inputting parameters obtained by analyzing blood cell images and parameters obtained from a blood cell counter into a machine learning algorithm. [Prior Art Documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent No. 6241735 [Patent Document 2] Japanese Patent No. 7381003 [Patent Document 3] Japanese Unexamined Patent Publication No. 2021-162323 [Summary of the Invention] [Problems to be Solved by the Invention]

[0004] To detect MDS early, it is necessary to differentiate it at the screening stage. The technology in Patent Document 1 is not suitable for screening because it uses a next-generation sequencer. The technology in Patent Document 2 is a technology specialized in differentiating specimens that are highly likely to be either MDS or AA, as is evident from the fact that it is a machine learning algorithm trained on 89 MDS cases and 43 AA cases, and cannot be applied to tests that screen for MDS specimens from blood specimens where the presence or absence of the disease is unknown. Patent Document 3 discloses a technology specialized in differentiating PV, ET, and PMF, but does not disclose a screening test for MDS.

[0005] This invention has been made in view of the above-mentioned problems, and one of its objectives is to propose a new method for supporting the differential diagnosis of myelodysplastic syndromes, such as differentiating between myelodysplastic syndromes and other conditions, which can be applied to screening tests. [Means for solving the problem]

[0006] According to a first aspect of the present invention, a computer-based disease differentiation support method obtains a first parameter relating to the classification and abnormal morphology of cells obtained by analyzing images of cells contained in a sample taken from a subject, obtains a second parameter relating to the number of each type of cell contained in the sample, and uses a pre-trained computer algorithm to generate differentiation support information for assisting in the differentiation between myelodysplastic syndrome and other conditions based on the first and second parameters. According to a second aspect of the present invention, a disease differentiation support device for assisting in the differentiation of diseases obtains a first parameter relating to the number of abnormally shaped cells obtained by analyzing images of cells contained in a sample taken from a subject, obtains a second parameter relating to the number of each type of cell contained in the sample, and uses a pre-trained computer algorithm to generate differentiation support information for assisting in the differentiation of myelodysplastic syndrome from other conditions based on the first parameter and the second parameter. According to a third aspect of the present invention, a program to be executed by a computer obtains a first parameter relating to the number of abnormally shaped cells obtained by analyzing images of cells contained in a sample taken from a subject, obtains a second parameter relating to the number of each type of cell contained in the sample, and uses a pre-trained computer algorithm to generate differential diagnosis support information to assist in the differentiation of myelodysplastic syndromes based on the first and second parameters. [Effects of the Invention]

[0007] According to the present invention, the ability to differentiate MDS in screening tests can be improved. [Brief explanation of the drawing]

[0008] [Figure 1] A diagram outlining methods to support the differential diagnosis of diseases. [Figure 2] A figure showing an example of a group of cell classification parameters. [Figure 3] A figure showing an example of a group of abnormal morphological parameters. [Figure 4] An example of the second parameter group is shown below. [Figure 5] A diagram showing an example of training data. [Figure 6] A diagram showing an example configuration of the disease differentiation support system 1 in the first embodiment. [Figure 7] A diagram showing an example configuration of a cell image analysis device. [Figure 8] A diagram showing an example configuration of a blood cell counter equipped with an optical detection unit. [Figure 9] A diagram showing an example configuration of a blood cell counter equipped with an electrical resistance detection unit. [Figure 10] A diagram showing an example of the hardware configuration of the training device 100A and the disease differentiation support device 100B. [Figure 11] A diagram showing an example of the functional configuration of training device 100A. [Figure 12] A diagram illustrating an example of the training process flow. [Figure 13]A diagram showing the hardware configuration of disease differential diagnosis support devices 200A, 200B and a terminal device 200C. [Figure 14] A diagram showing an example of the functional configuration of the disease differential diagnosis support device 200A. [Figure 15] A diagram showing an example of the flow of differential diagnosis support processing. [Figure 16] A diagram showing a configuration example of a disease differential diagnosis support system 2 according to a second embodiment. [Figure 17] A diagram showing an example of the functional configuration of the disease differential diagnosis support device 200B. [Figure 18] A diagram showing a configuration example of a disease differential diagnosis support system 3 according to a third embodiment. [Figure 19] A diagram showing an example of the functional configuration of the disease differential diagnosis support device 100B. [Figure 20] A diagram showing the correspondence between parameters and Shap values. [Figure 21] A diagram showing a comparison between machine-based disease prediction results and doctor's diagnosis results. [Figure 22] A diagram showing the accuracy and the like of prediction results obtained by a machine method. [Figure 23] A diagram showing an ROC curve of prediction results obtained by a machine method. [Figure 24] An example of a graph in which probabilities are plotted. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, the outline and embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description and the drawings, the same reference numerals denote the same or similar components, and thus descriptions of the same or similar components will be omitted. Further, although there are portions in the text that include descriptions such as "for example" and similar expressions, it should be noted that the present invention is not limited to the content of those descriptions, applies to the entire embodiments described below.

[0010] 1. Method for supporting differential diagnosis of diseases The present embodiment is an embodiment that supports differential diagnosis related to myelodysplastic syndrome (hereinafter referred to as "MDS").

[0011] In this embodiment, a computer algorithm is used to support the identification of MDS, specifically, for example, the identification of MDS from other conditions. In differentiating MDS from other conditions, "other conditions" may include, for example, other hematopoietic disorders that are not MDS (other hematopoietic disorders that are not myelodysplastic syndrome), and the method of this embodiment may include supporting the differentiation between MDS and other hematopoietic disorders (for example, non-MDS, as described later).

[0012] As will be explained in more detail later, in this embodiment, the diagnostic tool is generated using information from healthy individuals (HC (healthy) as described later). In other words, it may also include supporting the differentiation between non-MDS, MDS, and HC.

[0013] The method for supporting the identification of MDS in this embodiment involves obtaining multiple types of first parameters and multiple types of second parameters related to cells from a sample containing cells taken from a subject, and using a computer algorithm to generate identification support information for supporting the identification of MDS based on the obtained multiple types of first parameters and multiple types of second parameters.

[0014] In this embodiment, the first parameter may be a parameter relating to abnormal cell morphology obtained by analyzing images of cells contained in a sample taken from a subject, for example, by analyzing images of cells contained in the sample. Details will be described later.

[0015] The second parameter may be a parameter relating to the number of each type of cell contained in the sample, and may be obtained, for example, from the results of analysis obtained by analyzing the optical or electrical signals from cells contained in a sample taken from a subject. Details will be described later.

[0016] Parameters relating to the number of each type of cell contained in the sample may include, for example, the number of cells of each cell type, the concentration of cells per predetermined amount of sample calculated based on this number (e.g., the concentration of red blood cells per 1 μL), and the ratio of a particular cell to a certain cell (e.g., the ratio of eosinophils to 100 white blood cells).

[0017] Figure 1 is a diagram illustrating the overview of the method for supporting the identification of MDS in this embodiment. As shown in Figure 1, the sample S collected from the subject is divided into, for example, sample S1 used for analysis in step A and sample S2 used for analysis in step B.

[0018] In step A, for example, cells contained in sample S1 are imaged using a cell image analysis device 400, and the obtained cell images are analyzed to acquire a group of first parameters consisting of multiple types of first parameters, including parameters related to cell classification and abnormal morphology.

[0019] In step B, for example, an optical or electrical signal is obtained from the cells contained in sample S2 using a blood cell counter 450, and a group of second parameters consisting of multiple types of second parameters, including the number or ratio of each type of cell contained in sample S2, is obtained by analyzing the obtained optical or electrical signal. The group of second parameters may also include the hemoglobin concentration in the blood. In that case, the hemoglobin concentration is obtained by irradiating the measurement sample prepared with sample S2 and SLS hemolytic agent with light of a wavelength of 555 nm, which is the absorption wavelength of SLS hemoglobin, and measuring the absorbance.

[0020] In process C, a pre-trained computer algorithm CA (identifier) ​​is input with a first parameter group consisting of multiple first parameters and a second parameter group consisting of multiple second parameters to generate identification support information to assist in the identification of MDS.

[0021] In this embodiment, the cells contained in the sample belong to a predetermined cell group, for example, the cell groups that make up each organ of a human being. A given group of cells typically includes several cell types that can be morphologically classified by histological or cytological microscopy. Morphological classification (also called "morphological classification") may include classification of cell types and classification of cell morphological characteristics. For example, the cells under analysis belong to a specific cell lineage belonging to a specific cell group. A specified cell lineage is a group of cells belonging to the same lineage that have differentiated from a certain type of tissue stem cell. Preferably, the specified cell lineage is a hematopoietic lineage, and more preferably, the hematopoietic cells are peripheral blood cells or bone marrow cells.

[0022] Generally, in a blood test, a hematopoietic sample S taken from the subject is used in a blood cell counter to measure red blood cell count, white blood cell count, platelet count, hemoglobin concentration, hematocrit value, red blood cell indices, and white blood cell classification values. In addition, especially when a blood disorder is suspected, in addition to blood cell counting using a blood cell counter, a smear is prepared from the sample S and the morphology of the blood cells is actually observed to check for any morphological abnormalities in the cells.

[0023] (1) First parameter group In this embodiment, in step A shown in Figure 1, in order to obtain the first parameter (first parameter group), morphological features are extracted from each individual cell on the smear prepared from the sample S1, either under a microscope or from the cell in the image captured by a slide scanner.

[0024] For extracting morphological features, specimens stained with bright-field staining may be used, for example. Bright-field staining may be selected from, for example, Wright staining, Giemsa staining, Wright-Giemsa staining, and May-Giemsa staining. The specimen is not limited as long as the morphology of each cell belonging to a given cell group can be observed individually. Examples include smear specimens and imprint specimens. For example, a smear specimen using peripheral blood or bone marrow as the sample may be used.

[0025] For example, in extracting morphological features from cells, the morphological classification of individual cells on a specimen is performed. Also, for example, if there are abnormal findings in the cells, the abnormal findings are classified. Morphological classification allows us to obtain, for example, at least one of the cell types and the proportion of identical cells in the sample as parameters related to the morphological classification of cells. Furthermore, by classifying the abnormal findings, at least one of the types of abnormal findings and the ratio of cells exhibiting the same type of abnormal finding is obtained as a parameter related to the abnormal findings.

[0026] In this embodiment, abnormal morphology (abnormal morphological features) may be classified for cells of the type classified by morphological classification. In this case, information on the number and proportion of classified cells may be obtained in order to calculate the number of cells with abnormal morphology (comparison display of normal number and abnormal number) and the proportion of abnormal cells in the total. In this sense, the cell classification parameters exemplified below may be included as the first parameter related to the abnormal morphology of cells. The classification of abnormal morphology may be performed, for example, using a computer algorithm that is pre-designed to assign abnormal findings (abnormal comments) corresponding to the type of cell.

[0027] Figure 2 shows an example of a group of cell classification parameters (cell classification parameter group), which is one of the first parameters used in this embodiment. In reality, classification of many types of cells can be obtained by cell image analysis, but here we show examples of items that are considered to be particularly relevant to the differentiation of MDS. The following 20 items from (1A) to (1T) fall into this category. (1A) Segmented neutrophils (hereinafter referred to as "SN") (%) (1B) Band Neutrophils (%) (hereinafter referred to as "BN") (%) (1C) Postmyelocytes (%) (1D) Myelocytes (%) (1E) Promyelocytes (%) (1F) Blasts (%) (1G) Eosinophils (%) (1H) Basophil (%) (1I) Lymphocytes (%) (1J) Reactive lymphocytes (%) (1K) Monocyte (%) (1L) Smudge (%) (1M) Nucleated red blood cells (%) (1N) Large platelets (%) (1O) Megakaryocytes (%) (1P) Artifact (%) (1Q) Platelet aggregation (%) (1R) Nucleated red blood cells ( / 100WBC) (1S) Megakaryocyte ( / 100WBC) (1T) Large platelet ( / 100WBC)

[0028] The 17 classifications (1A) to (1Q) may be calculated, for example, as "the number of cells identified by the AI ​​per sample / the total number of AI-detected images per sample" (%). The three classifications (1R) to (1T) may be calculated, for example, as the number of non-leukocytes such as nucleated red blood cells, megakaryocytes, and large platelets, using the same counting method as in an actual laboratory.

[0029] Figure 3 shows an example of the categories of abnormal findings (abnormal finding parameter group) included in the first parameter. In reality, various classifications of abnormal morphologies can be obtained through cell image analysis, but here we exemplify items that are considered to have a particularly high correlation with the differentiation of MDS. Specifically, for example, by selecting BN and SN, the number of cells validated at the time of setting the abnormal finding threshold was ≥ 50 cells, the AUC value was ≥ 0.8, and the sensitivity, specificity, and accuracy at the time of evaluation were all 80% or higher, and the output performance of abnormal findings was extracted, resulting in the following 10 items (1a) to (1j). • BN (Banded Neutrophils) (1a) BN - Spherical / Elliptic nucleus (%) (1b) BN-degranulation (%) (1c)BN-toxic granules (%) (1d) BN-Smudge (%) ·SN (segmental neutrophil) (1e)SN-Pseudopelgel (%) (1f)SN-degranulation (%) (1g)SN-toxic granules (%) (1h)SN-Döhle bodies (%) (1i) SN-Smudge (%) (1j)SN-apoptosis (%)

[0030] (%) may represent the proportion of cells classified into a specific cell type that exhibit abnormal findings. For example, (1a) BN-spherical / elliptic nucleus (%) may be calculated as (number of BN cells with abnormal findings of spherical / elliptic nucleus per sample) / (number of cells classified as BN per sample). Furthermore, for example, (1e)SN-Pseudo-Pelgel (%) may be calculated as (number of SN cells with abnormal findings of pseudo-Pelgel per sample) / (number of cells classified as SN per sample).

[0031] The above is just one example, but the first parameter in this embodiment may include, for example, a parameter relating to at least one of the number and ratio of atypical lymphocytes; or a parameter relating to at least one of the number of each type of cell having an abnormal morphology selected from neutrophils, eosinophils, platelets, lymphocytes, monocytes, basophils, metamyelocytes, myelocytes, promyelocytes, blasts, plasma cells, immature eosinophils, immature basophils, erythroblasts, and megakaryocytes, and at least one of the ratio of each type of cell having an abnormal morphology.

[0032] Furthermore, the abnormal cell morphology in this embodiment may include, for example, at least one selected from nuclear morphological abnormalities, granular abnormalities, abnormal cell size, cellular malformations, cell destruction, vacuoles, immature cells, presence of inclusion bodies, Döhle bodies, satellite phenomena, nuclear chromatin abnormalities, petal-like nuclei, high N / C ratio, bleb-like morphology, smudge, and hairy cell-like morphology. Furthermore, the above-mentioned nuclear morphological abnormalities may include, for example, at least one selected from hyperlobulation, hypolobulation, pseudopelger nucleus abnormalities, ring-shaped nuclei, spheroidal nuclei, elliptic nuclei, apoptosis, polynucleation, nuclear disintegration, enucleation, naked nuclei, irregular nuclear margins, nuclear fragmentation, internuclear bridges, multiple nuclei, notched nuclei, nuclear fission, and nucleolar abnormalities. Furthermore, the granular abnormalities described above may include, for example, at least one selected from degranulation, granular distribution abnormalities, toxic granules, Auer bodies, Fagott cells, and pseudo-Chediak-Higashi granule-like granules. Furthermore, the above-mentioned abnormalities in cell size may include, for example, giant platelets.

[0033] Furthermore, the method for obtaining the first parameter is not particularly limited. For example, the first parameter may be obtained by the examiner, or it may be obtained using the cell image analysis device 300 or cell image analysis device 400 described later.

[0034] Furthermore, the first parameter may be obtained using a deep learning algorithm, for example, as described in U.S. Patent Publication No. 2019-0347467. U.S. Patent Publication No. 2019-0347467 is incorporated herein. The discriminator used in the method for obtaining the first parameter may include, for example, multiple deep learning algorithms having a neural network structure, as shown in Figure 1. The identification device includes a first deep learning algorithm DL1 and a second deep learning algorithm DL2, wherein the first deep learning algorithm DL1 extracts cell features, and the second deep learning algorithm DL2 identifies the cells to be analyzed based on the features extracted by the first deep learning algorithm. The second deep learning algorithm, DL2, may output morphological classification results and the probability of corresponding to that classification, or classification of abnormal findings and the probability of corresponding to that classification, for each cell. The first deep learning algorithm DL1 may be, for example, a convolution-connected neural network, and the second deep learning algorithm DL2, located downstream of the first deep learning algorithm, may be, for example, a fully-connected neural network.

[0035] (2) Second parameter In this embodiment, in step B shown in Figure 1, the sample S2 is measured using a blood cell counter to obtain the second parameter.

[0036] The second parameter may be the Complete Blood Count (CBC) parameter in the blood cell count test. The CBC parameter may include the following eight items, for example: ·White blood cell count (WBC) • Red blood cell count (RBC) • Hemoglobin (Hb) • Hematocrit (HCT) ·Mean corpuscular volume (MCV) • Mean corpuscular hemoglobin (MCH) • Mean corpuscular hemoglobin concentration (MCHC) ·Platelet count (PLT)

[0037] The second parameter input to the computer algorithm includes at least one of the eight CBC items described above. Preferably, the second parameter includes at least two of the eight CBC items. More preferably, the second parameter may include at least WBC, RBC, Hb, and PLT from the eight CBC items. Even more preferably, the second parameter may include all eight of the eight CBC items described above.

[0038] In addition to the eight items mentioned above, CBC parameters may include other parameters that can be obtained simultaneously with CBC measurement. For example, CBC parameters may include the following parameters: • Red blood cell distribution width (RDW) ·Platelet distribution width (PDW)

[0039] Figure 4 shows an example of the second parameter group (CBC parameter group). The second parameter group includes, for example, the following items (2A) to (2I). (2A) White blood cell count (WBC) (2B) Red blood cell count (RBC) (2C) Hemoglobin (Hb) (2D) Hematocrit (HCT) (2E) Mean corpuscular volume (MCV) (2F) Mean corpuscular hemoglobin (MCH) (2G) Mean corpuscular hemoglobin concentration (MCHC) (2H) Red blood cell distribution width (RDW) (2I) Platelet count (PLT)

[0040] In a preferred example, the second parameter group does not include items other than CBC parameters. In other words, the second parameter group includes the eight CBC items mentioned above and other parameters that can be measured simultaneously by CBC measurement, but does not include other parameters obtained by measurements other than CBC measurement. Other measurements include, for example, leukocyte classification (DIFF), reticulocytes (RET), and platelet optical measurement (PLT-F). By constructing the second parameter group with only CBC parameters, there is an advantage in that, for example, the dependence on the type of hematology counter can be reduced. There are various models of hematology counters. For example, there are low-end models that only support CBC measurement (e.g., Sysmex XQ series), mid-end models that support CBC / DIFF / RET measurement (e.g., Sysmex XN-L series), and high-end models that support CBC / DIFF / RET / PLT-F measurement (e.g., Sysmex XR series). As listed here, the measurable items differ depending on the model. Therefore, if parameters that can only be measured by a limited number of models, such as PLT-F, are included in the second parameter group, MDS screening can only be achieved when measured by a specific model. In this respect, if the second parameter group consists only of CBC parameters, MDS screening can be achieved even with a low-end model, for example.

[0041] The above is just one example, but the second parameter in this embodiment may include, for example, a value related to at least one selected from red blood cells, nucleated red blood cells, small red blood cells, platelets, hemoglobin, reticulocytes, immature granulocytes, neutrophils, eosinophils, basophils, lymphocytes, monocytes, hematocrit, mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), mean platelet volume (MPV), and erythrocyte distribution width (RDW).

[0042] (3) Differential diagnosis of diseases In this embodiment, identification support information for assisting in the identification of MDS is generated based on the first parameter group consisting of the first parameter and the second parameter group consisting of the second parameter. The generation of this identification support information may be performed by the processing unit 20, described later, using a computer algorithm. Computer algorithms may include artificial intelligence algorithms, such as machine learning algorithms and deep learning algorithms. While deep learning may be included within machine learning, for convenience, these will be explained separately.

[0043] Machine learning algorithms may include algorithms such as trees, regression, support vector machines, Bayesian algorithms, clustering algorithms, and random forests. A gradient boosting tree algorithm may also be used as a machine learning algorithm. Furthermore, a Multimodal Deep Neuronal Network (Multimodal DNN) may be used as the gradient boosting tree.

[0044] Deep learning algorithms may have a neural network structure.

[0045] The computer algorithm is trained according to the following method and functions as a discriminator for MDS.

[0046] 1-2. Creation of a discriminator In this embodiment, the training data for the artificial intelligence algorithm may be generated, for example, by arranging a first training parameter group and a second training parameter group as a matrix at the same hierarchical level, and further associating them with labels indicating disease names (hereinafter referred to as "disease labels"). The first set of training parameters may be obtained from training smears.

[0047] Figure 5 shows an example of training data, illustrating a training dataset consisting of multiple training data points. The training data may be linked to data identification information, such as the identification information of the sample used to generate the training data, or the identification information of the patient from whom the sample was collected.

[0048] The training data may be, for example, data in which parameter type, row number, parameter name, parameter value, and disease label are associated. The parameter type (first column) may be set to include a first parameter (labeled "1st" in this figure) and a second parameter (labeled "2nd" in this figure). The row numbers (second column) may be set to numbers in ascending order, for example, starting from 1. The parameter name (third column) may be set to the name of the corresponding parameter.

[0049] The parameter values ​​(fourth column) may be set to the values ​​of the corresponding parameters, for example, the unit values ​​shown in Figures 2 to 4 may be set.

[0050] The first parameter may be obtained, for example, by using the second deep learning algorithm DL2 to count the morphological classification results and / or the classification results of abnormal findings, and then weighting the obtained count results by the probability of belonging to that classification. For example, if 1 out of 100 cells is classified as BN, and the probability of a cell being classified as BN is 90%, then 1 can be multiplied by 90% to calculate "0.9".

[0051] The disease label (5th column) may be configured to store a label corresponding to the disease.

[0052] The first training parameter group and the second training parameter group may be generated for each sample from a sample taken from a patient who has received a confirmed diagnosis of a disease by a physician (hereinafter referred to as the "training sample"). Then, for each sample, a matrix consisting of the first training parameter group and the second training parameter group may be associated with the disease label and used as training data.

[0053] In this embodiment, the disease label may store one of the following based on the patient's confirmed diagnosis: "MDS", "non-MDS", or "HC" (healthy).

[0054] The disease label "MDS" may be set, for example, for training data generated from samples taken from patients with a confirmed diagnosis of MDS.

[0055] The disease label "non-MDS" may be applied to training data generated from samples taken from patients who have received a definitive diagnosis of a hematopoietic disorder other than MDS, such as at least one of the following, rather than meaning healthy. • ALL (Acute Lymphoblastic Leukemia): Acute lymphoblastic leukemia ·AML(Acute Myeloid Leukemia): Acute myeloid leukemia ML (Malignant Lymphoma): Malignant Lymphoma ⇒DLBCL (Diffuse Large B-Cell Lymphoma): Diffuse large B-cell lymphoma ⇒FL (Follicular Lymphoma): Follicular Lymphoma ⇒MCL (Mantle Cell Lymphoma): Mantle cell lymphoma ⇒ATL (Adult T cell Leukemia): Adult T cell leukemia / lymphoma • MPN (Myeloproliferative Neoplasms): Myeloproliferative neoplasms Other hematopoietic tumors

[0056] The disease label "HC" may be set, for example, for training data generated from samples taken from patients who have been definitively diagnosed as neither MDS nor non-MDS.

[0057] In the example shown in Figure 5, the first training parameter group and the second training parameter group are arranged in rows, but they can also be arranged in columns. Furthermore, as shown in Figure 5, each parameter may be represented by an abbreviation or by a label value. Furthermore, disease labels may also be represented by label values.

[0058] Furthermore, the first and second training parameter groups may be created based on the results of selecting items that have a high correlation with differential diagnosis of MDS after performing a predetermined statistical analysis. Examples of predetermined statistical analyses include one-dimensional analysis of variance (ANOVA), Pearson correlation, and Spearman's rank correlation. By selecting statistical parameters, the accuracy of differential diagnosis can be improved.

[0059] The above training data is input into a computer algorithm, which is then trained to generate a discriminator. If we use machine learning algorithms here, we train one algorithm for each disease. On the other hand, when using deep learning algorithms, a single algorithm can be trained on multiple diseases.

[0060] Computer algorithms can be trained using software such as Python.

[0061] Trained computer algorithms are used as discriminators for MDS, assisting in its identification.

[0062] 1-3. Generation of analysis data and generation of differential diagnosis support information The analysis data (data for analysis) is generated by obtaining a first set of analysis parameters and a second set of analysis parameters from the sample to be analyzed collected from the subject, and then combining these. The first set of parameters for analysis and the second set of parameters for analysis may be generated in the same way as the first set of parameters for training and the second set of parameters for training, respectively. Furthermore, the parameters included in the analysis data may be the same type as, for example, the parameters included in the training data. The analysis data may be generated by matrixing the first set of analysis parameters and the second set of analysis parameters at the same hierarchical level, for example, in the same order as the training data.

[0063] Next, the analysis data is input into the identification device trained in "1-2. Generation of Identification Device" to generate identification support information. This differential information is, for example, information that helps determine whether a subject predicted from the analysis data has MDS or not. For example, the differential information may be the probability that the subject belongs to the MDS group, the non-MDS group, or the HC group (the probability of belonging to each group / class). Furthermore, the differential can be configured to output the patient's disease label based on this probability. The differential diagnosis information generated in this way can be described as information about which group the subject is most likely to belong to: the first group of patients who have or are likely to have myelodysplastic syndrome (e.g., the MDS group), the second group of patients who have or are likely to have other hematopoietic disorders other than myelodysplastic syndrome (e.g., the non-MDS group), or the group of healthy individuals (e.g., the HC group).

[0064] Note that the first parameter for analysis and the first parameter for training may be of the same type, and the second parameter for analysis and the second parameter for training may be of the same type. Furthermore, the first parameter for analysis and the first parameter for training may be obtained based on analysis using the same cell image analysis device, or they may be obtained based on analysis using a different cell image analysis device of the same type. Similarly, the second parameter for analysis and the second parameter for training may be obtained based on analysis from the same hematology counter, or they may be obtained based on analysis from a different hematology counter of the same type.

[0065] [Examples] The following describes an example of a disease differentiation support system, which is a system for supporting the differentiation of MDS in this embodiment.

[0066] [First Embodiment] 2. Disease Differentiation Support System 1 Figure 6 shows an example of the configuration of the disease differentiation support system 1 in this embodiment. The disease differentiation support system 1 includes, for example, a training device 100A and a disease differentiation support device 200A. The vendor-side device 100 functions as the training device 100A, and the user-side device 200 operates as the disease differentiation support device 200A.

[0067] The training device 100A is connected, for example, to a cell image analysis device 300 and a blood cell counter 350. The training device 100A acquires, for example, a first set of training parameters from the cell image analysis device 300 and a second set of training parameters from the blood cell counter 350.

[0068] The disease differentiation support device 200A is connected, for example, to a cell image analysis device 400 and a blood cell counter 450. The disease differentiation support device 200A, for example, acquires a first set of analysis parameters from the cell image analysis device 400 and a second set of analysis parameters from the blood cell counter 450.

[0069] The recording medium 98 is a computer-readable and non-volatile recording medium, such as a DVD-ROM or a USB memory stick. The following describes each component.

[0070] 2-1. Cell image analysis device Figure 7 shows an example of the configuration of the cell image analysis device 300. The cell image analysis device 300 comprises at least an imaging unit 304, which includes a stage 309 for placing specimens, a magnifying glass unit 302 such as a microscope, and an image sensor 301 for capturing microscopic images. The imaging unit 304 acquires images of each cell on a training specimen 308 set on the stage 309. The cell image analysis device 300 acquires a first set of parameters from the acquired images. The cell image analysis device 300 may be equipped with an information processing unit 305, which may acquire and write out the first set of parameters and communicate with the training device 100A.

[0071] Next, we will explain the configuration of the cell image analysis device 400. The configuration of the cell image analysis device 400 is basically the same as that of the cell image analysis device 300, and includes an imaging unit 404, which comprises a stage 409 for placing specimens, a magnifying glass unit 402 such as a microscope, and an image sensor 401 for capturing microscopic images. The imaging unit 404 acquires images of each cell on the analysis specimen 408 set on the stage 409. The cell image analysis device 400 acquires a first set of analysis parameters from the acquired images. The cell image analysis device 400 may be equipped with an information processing unit 405, which acquires and writes the first set of parameters and communicates with the disease differentiation support device 200A.

[0072] For cell image analysis devices 300 and 400, for example, the Automated Digital Cell Morphology Analyzer DI-60 manufactured by Sysmex Corporation may be used.

[0073] 2-2. Blood cell counter Figures 8 and 9 show examples of the configuration of the blood cell counter 350. The blood cell counter 350 is a flow cytometer, etc., equipped with an optical detection unit 411 for detecting an optical signal with a flow cell as shown in Figure 8. In Figure 8, the light emitted from the laser diode, which is the light source 4111, is irradiated onto cells passing through the flow cell 4113 via the irradiation lens system 4112.

[0074] In this embodiment, the light source 4111 of the flow cytometer is not particularly limited, and a light source 4111 with a wavelength suitable for exciting fluorescent dyes is selected. Examples of such light sources 4111 include semiconductor lasers including red semiconductor lasers and / or blue semiconductor lasers, gas lasers such as argon lasers and helium-neon lasers, and mercury arc lamps. Semiconductor lasers are particularly preferred because they are much cheaper than gas lasers.

[0075] As shown in Figure 8, forward scattered light emitted from particles passing through the flow cell 4113 is received by a forward scattered light receiving element 4116 via a focusing lens 4114 and a pinhole section 4115. The forward scattered light receiving element 4116 may be a photodiode or the like. Side scattered light is received by a side scattered light receiving element 4121 via a focusing lens 4117, a dichroic mirror 4118, a bandpass filter 4119, and a pinhole section 4120. The side scattered light receiving element 4121 may be a photodiode, a photomultiplier, or the like. Side fluorescence is received by a side fluorescence receiving element 4122 via a focusing lens 4117 and a dichroic mirror 4118. The side fluorescence receiving element 4122 may be an avalanche photodiode, a photomultiplier, or the like.

[0076] The received signals output from each of the light receiving units 4116, 4121, and 4122 are subjected to analog processing such as amplification and waveform processing by analog processing units having amplifiers 4151, 4152, and 4153, respectively, and then sent to the measurement unit control unit 480.

[0077] The measurement unit control unit 480 is connected to the information processing unit 351, which acquires a second parameter based on the optical signal acquired by the optical detection unit 411.

[0078] Furthermore, the blood cell counter 350 may also include, for example, an electrical resistance detection unit 412 as shown in Figure 9. Figure 9 shows an example configuration where the electrical resistance detection unit 412 is a sheath-flow type electrical resistance detection unit. The electrical resistance detection unit 412 (sheath flow type electrical resistance detection unit) comprises a chamber wall 412a, an aperture section 412b for measuring the electrical resistance of cells, a sample nozzle 412c for supplying the sample, and a recovery tube 412d for collecting cells that have passed through the aperture section 412b. The area around the sample nozzle 412c and recovery tube 412d within the chamber wall 412a is filled with sheath fluid. The dashed arrow, represented by the symbol 412s, indicates the direction of the sheath fluid flow. Red blood cells 412e and platelets 412f discharged from the sample nozzle pass through the aperture section 412b while being enveloped by the flow of sheath fluid 412s. A constant DC voltage is applied to the aperture section 412b, and it is controlled so that a constant current flows while only sheath fluid is flowing. Cells are poor conductors of electricity, meaning they have high electrical resistance. Therefore, when a cell passes through the aperture 412b, its electrical resistance changes, and the number of times a cell has passed through the aperture 412b and the corresponding electrical resistance can be detected. Since electrical resistance increases in proportion to the cell volume, the measurement unit information processing unit 481 calculates the volume of cells that have passed through the aperture 412b from the signal intensity related to the electrical resistance value and can send the cell count for each volume as a histogram to the information processing unit 351. The blood cell counter 350 measures a training sample and acquires a second set of training parameters.

[0079] Furthermore, the blood cell counter 350 may also include, for example, an HGB detection unit. The HGB detection unit measures the hemoglobin concentration in the blood using the SLS hemoglobin method. The HGB detection unit measures the hemoglobin concentration in the blood by irradiating a measurement sample, prepared from blood and an SLS hemolytic agent, with light at a wavelength of 555 nm, which is the absorption wavelength of SLS hemoglobin, and measuring the absorbance. The blood cell counter 350 measures a training sample and acquires a second training parameter group.

[0080] The configuration of the hemocytometer 450 may be the same as that of the hemocytometer 350. The hemocytometer 450 measures a training sample and acquires a second set of parameters for analysis. Examples of hemocytometers 350 and 450 include the XN-2000 hemocytometer manufactured by Sysmex Corporation.

[0081] 2-3.Training device The training device 100A trains a computer algorithm using a first training parameter group, a second training parameter group, and the disease names associated with them as training data to generate a differential identifyr. The training device 100A acquires the first parameters from the cell image analysis device 300 via the recording medium 98 or the network 99. The training device 100A acquires the second parameters from the blood cell counter 350 via the recording medium 98 or the network 99. The training device 100A provides the generated differential identifyr to the disease differential identifyr support device 200A. The differential identifyr is provided via the recording medium 98 or the network 99. The disease differential identifyr support device 200A uses the differential identifyr to generate differential identifyr support information to assist in disease differential identification.

[0082] (1) Hardware configuration of the training device Figure 10 shows an example of the hardware configuration of the training device 100A. The training device 100A comprises, for example, a processing unit 10 (10A), an input unit 16, and an output unit 17. The training device 100A is composed of, for example, a general-purpose computer.

[0083] The processing unit 10 includes, for example, a CPU (Central Processing Unit) 11 that performs data processing as described later, a memory 12 used as a work area for data processing, a storage unit 13 that records programs and processing data as described later, a bus 14 that transmits data between each unit, an interface unit 15 that performs data input and output with external devices, and a GPU (Graphics Processing Unit) 19.

[0084] The input unit 16 and the output unit 17 are connected to the processing unit 10. For example, the input unit 16 is an input device such as a keyboard, touch panel, or mouse, and the output unit 17 is a display device such as a liquid crystal display. The output unit 17 may also include an audio output device such as a speaker.

[0085] GPU19 functions as an accelerator to assist the computational processing (e.g., parallel processing) performed by CPU11. In other words, in the following explanation, the processing performed by CPU11 includes the processing that CPU11 performs using GPU19 as an accelerator.

[0086] Furthermore, the processing unit 10 pre-records a computer program and a computer algorithm for performing the training process described in Figure 12 below in the storage unit 13, for example, in executable format. The executable format is, for example, a format generated by a compiler converting from a programming language. The processing unit 10 works in cooperation with the operating system stored in the storage unit 13 to perform the training process of the computer algorithm using the computer program for performing the training process (hereinafter sometimes simply referred to as the "training program").

[0087] In the following description, unless otherwise specified, the processing performed by the processing unit 10 refers to the processing performed by the CPU 11 based on the computer program and computer algorithms for training processing stored in the storage unit 13 or memory 12. The CPU 11 uses memory 12 as a working area to volatilically temporarily store necessary data (such as intermediate data during processing) and nonvolatilically records data to be stored long-term, such as calculation results, in the storage unit 13 as appropriate.

[0088] (2) Functional configuration of the training device Referring to Figure 11, the processing unit 10A of the training device 100A functions, for example, as a training data generation unit 101, a training data input unit 102, and an algorithm update unit 103. These functions are realized by installing a training program (e.g., Python) that causes the computer to perform training processing into the storage unit 13 or memory 12 of the processing unit 10A, and then having the CPU 11 execute this program. The training data database (DB) 104 stores the first training parameter group acquired by the processing unit 10A from the cell image analysis device 300 and the second training parameter group acquired from the blood cell counter 350. The training data DB 104 also stores labels of disease names corresponding to the parameters. The algorithm database (DB) 105 may store the computer algorithm before training and the computer algorithm after training.

[0089] The training data generation unit 101 may be a functional unit corresponding to step S11 described later, the training data input unit 102 may be a functional unit corresponding to step S12 described later, and the algorithm update unit 103 may be a functional unit corresponding to step S15 described later.

[0090] (3) Processing of the training program The processing unit 10A of the training device 100A performs training processing as shown in Figure 12, for example, according to a training program (not shown) stored in the memory unit 13.

[0091] Figure 12 is a flowchart showing an example of the training process flow. Note that this process is merely an example, and some steps may be omitted or additional steps may be added. Furthermore, this process is only one example of a training process and is not limited to it.

[0092] The processing unit 10A receives, for example, a command to start training data acquisition input from the operator via the input unit 16. Then, the processing unit 10A acquires, for example, a first training parameter group, a second training parameter group, and a disease label (step S11). Specifically, for example, it acquires the first training parameter group from the cell image analysis device 300 and stores it in the training data DB 104 in the storage unit 13. Also, for example, it acquires the second training parameter group from the blood cell counter 350 and stores it in the training data DB 104 in the storage unit 13. Also, for example, it acquires a disease label and stores it in the training data DB 104 in the storage unit 13.

[0093] The processing unit 10A generates training data by associating the first training parameter group and the second training parameter group with disease labels, for example, in accordance with the method described in "1-2. Generation of Identifiers" above, and stores it in the training data database 104 in the storage unit 13.

[0094] The above processing may be performed on multiple samples (patients) to obtain a training dataset, for example, as shown in Figure 5.

[0095] Furthermore, disease labels corresponding to each parameter group may be received, for example, from input by the operator via the input unit 16 for each parameter group, and linked to the first training parameter group and the second training parameter group. Alternatively, when each parameter is acquired by the cell image analyzer 300 or the blood cell counter 350, patient information may be linked to each parameter, and the processing unit 10A may read that information.

[0096] The processing unit 10A receives, for example, a training process start command input by the operator from the input unit 16. Then, the processing unit 10A reads, for example, a computer algorithm stored in the algorithm DB 105 in the memory unit 13. The processing unit 10A then reads training data from the training data DB 10 and inputs the training data into the computer algorithm to train it (step S12).

[0097] Next, the processing unit 10A determines whether or not to terminate the training (step S13). Specifically, for example, it determines whether the computer algorithm has been trained using all the training data. If it is determined that training should not be terminated (step S13: NO), the processing unit 10A reads, for example, other training data (next training data) from the training data DB 10 (step S14), and returns to step S12. On the other hand, if it is determined that training should be terminated (step S13: YES), the processing unit 10A records, for example, the trained computer algorithm (discriminator) in the algorithm DB 105 in the storage unit 13 (step S15).

[0098] A trained computer algorithm functions as a diagnostic tool, generating diagnostic support information to assist in the identification of MDS.

[0099] Then, the processing unit 10A terminates the training process.

[0100] 2-4. Disease differentiation support device The disease differentiation support device 200A acquires a first set of analytical parameters and a second set of analytical parameters, as well as a differential, and generates differential support information to assist in disease differentiation. The disease differentiation support device 200A acquires the first set of analytical parameters from the cell image analyzer 400 via the recording medium 98 or network 99. The disease differentiation support device 200A acquires the second set of analytical parameters from the blood cell counter 450 via the recording medium 98 or network 99.

[0101] (1) Hardware configuration of disease differentiation support device Figure 13 will be used to explain the hardware configuration of the disease differentiation support device 200A. The configuration of the disease differentiation support device 200A may be basically the same as that of the training device 100A. However, the processing unit 10 (10A), input unit 16, and output unit 17 of the training device 100A may be replaced with the processing unit 20 (20A), input unit 26, and output unit 27 of the disease differentiation support device 200A.

[0102] Furthermore, the CPU 11, memory 12, storage unit 13, bus 14, interface unit 15, and GPU 19 of the training device 100A may be replaced with the CPU 21, memory 22, storage unit 23, bus 24, interface unit 25, and GPU 29 of the disease differentiation support device 200A.

[0103] Furthermore, the processing unit 20 pre-records, for example, a computer program for performing the differential diagnosis support processing described in Figure 15 below, in the storage unit 13, for example, in executable format. The executable format is, for example, a format generated by a compiler that converts from a programming language. The processing unit 10 uses the computer program for supporting disease differentiation recorded in the storage unit 13 and the differential diagnoses generated by the training device 100A to generate differential diagnosis support information to support disease differentiation.

[0104] In the following description, unless otherwise specified, the processing performed by the processing unit 20 refers to the processing actually performed by the CPU 21 of the processing unit 20, based on the computer program and diagnostic tool stored in the storage unit 23 or memory 22 to assist in disease differentiation. The CPU 21 uses memory 22 as a working area to volatilically temporarily store necessary data (intermediate data during processing, etc.) and nonvolatilically records data to be stored long-term, such as calculation results, in the storage unit 23 as appropriate.

[0105] (2) Functional configuration of the disease differentiation support device Referring to Figure 14, the processing unit 20A of the disease differentiation support device 200A functions, for example, as an analysis data acquisition unit 201, an analysis data input unit 202, an analysis unit 203, an analysis data database (DB) 204, and a differential diagnostic database (DB) 205. These functions are realized by installing a computer program (e.g., Python) that causes the computer to perform the process of generating differentiation support information into the storage unit 23 or memory 22 of the processing unit 20A, and by having the CPU 21 execute this computer program and a computer program to support disease differentiation, including differential diagnostics. The analysis data database (DB) 204 stores the first set of analysis parameters acquired by the processing unit 10A from the cell image analysis device 400 and the second set of analysis parameters acquired from the blood cell counter 450. The differential diagnostic database (DB) 205 stores differential diagnostics acquired, for example, from the training device 100A.

[0106] The data acquisition unit for analysis 201 may, for example, be a functional unit corresponding to step S21 described later, the data input unit for analysis 202 may, for example, be a functional unit corresponding to steps S22 and S23 described later, and the analysis unit 203 may, for example, be a functional unit corresponding to step S24 described later.

[0107] (3) Processing of computer programs to assist in the differential diagnosis of diseases The processing unit 20A of the disease differentiation support device 200A performs differentiation support processing, for example, as shown in Figure 15, according to a differentiation support program stored in the memory unit 23.

[0108] Figure 15 is a flowchart illustrating an example of the diagnostic support process. Note that this process is merely an example; some steps may be omitted, or additional steps may be added. Furthermore, this process is only one example of a diagnostic support process and is not limited to it.

[0109] The processing unit 20A receives, for example, a command to start the analysis data acquisition process input by the operator from the input unit 26. Then, the processing unit 20A acquires, for example, a first set of analysis parameters and a second set of analysis parameters (step S21). Specifically, for example, it acquires the first set of analysis parameters from the cell image analysis device 400 and stores them in the analysis data DB 204 in the storage unit 23. Also, for example, it acquires the second set of analysis parameters from the blood cell counter 450 and stores them in the analysis data DB 204 in the storage unit 23.

[0110] The processing unit 20A receives, for example, a command to start the identification device acquisition process input by the operator from the input unit 26. Then, the processing unit 20A acquires an identifier (step S22). Specifically, for example, it acquires an identifier from the training device 100A. Alternatively, if the identifier is already stored in the identifier database 205 in the storage unit 23, the stored identifier may be read.

[0111] The processing unit 20A receives, for example, an analysis processing start command input by the operator from the input unit 26. Then, the processing unit 20A inputs, for example, the first set of analytical parameters and the second set of analytical parameters acquired in step S21 to the identification device acquired in step S22 (step S23).

[0112] The processing unit 20A generates values ​​indicating the probability of each group (each class) using a discriminator as discrimination support information and records them in the storage unit 23 (step S24).

[0113] The processing unit 20A outputs the value generated in step S24 to the output unit 27 (step S25). Specifically, for example, the generated value is displayed on a display device, which is a type of output unit 27. Alternatively, the generated value may be output as sound by a sound output unit, which is a type of output unit 27.

[0114] Then, the processing unit 20A terminates the identification support processing.

[0115] [Second Example] 3. Disease Differentiation Support System 2 Figures 16 and 17 illustrate another aspect of the disease differentiation support system.

[0116] Figure 16 shows an example of the configuration of the disease differential diagnosis support system 2. The disease differentiation support system 2 comprises, for example, a user-side device 200, a cell image analysis device 400, and a blood cell counter 450, with the user-side device 200 operating as a disease differentiation support device 200B that performs both training and disease differentiation support. The disease differentiation support device 200B performs the functions of both the training device 100A and the disease differentiation support device 200A. The disease differentiation support device 200B is connected to the cell image analysis device 400 and the blood cell counter 450.

[0117] (1) Hardware configuration of disease differentiation support device 200B The hardware configuration of the disease differentiation support device 200B may be the same as the hardware configuration of the user-side device 200 shown in Figure 13. In Figure 13, the processing unit 20A may be read as processing unit 20B in the case of the disease differentiation support device 200B.

[0118] (2) Functional configuration of the disease differentiation support device 200B Figure 17 shows an example of the functional configuration of the disease differentiation support device 200B. The processing unit 20B of the disease differentiation support device 200B functions as a training data generation unit 101, a training data input unit 102, an algorithm update unit 103, an analysis data acquisition unit 201, an analysis data input unit 202, an analysis unit 203, a parameter database (DB) 304, and an algorithm database (DB) 305. The parameter database (DB) 304 combines the functions of the training data DB 104 described in 2-3.(2) above and the analysis data DB 204 described in 2-4.(2) above. The algorithm database (DB) 305 combines the functions of the algorithm DB 105 described in 2-3.(2) above and the differential diagnose DB 205 described in 2-4.(2) above. Specifically, the parameter DB304 stores the first training parameter group and the first analysis parameter group acquired by the processing unit 20B from the cell image analysis device 400, and the second training parameter group and the second analysis parameter group acquired from the blood cell counter 450. The algorithm DB305 stores the computer algorithm before training and the computer algorithm after training, which are the discriminators.

[0119] The processing unit 20B of the disease differentiation support device 200B may, during training, perform, for example, the training processing shown in 2-3(3) above and in Figure 12. The training data generation unit 101 corresponds to step S11 described in 2-3.(3) above, the training data input unit 102 corresponds to step S12, and the algorithm update unit 103 corresponds to step S15. Here, the "storage unit 13 of the processing unit 10A", "training data DB 104", and "algorithm DB 105" described in 2-3.(3) above may be read as "storage unit 23 of the processing unit 20B", "parameter DB 304", and "algorithm DB 305", respectively.

[0120] The processing unit 20B of the disease differentiation support device 200B may perform the differentiation support processing shown in 2-4.(3) above and in Figure 15 when generating differentiation support information. The data acquisition unit 201 for analysis corresponds to step S21 described in 2-4.(3) above, the data input unit 202 for analysis corresponds to steps S22 and S23, and the analysis unit 203 corresponds to step S24. Here, "analysis data DB204" and "identifier DB205" described in 2-4.(3) above may be read as "parameter DB304" and "algorithm DB305," respectively.

[0121] [Third Embodiment] 4. Disease Differentiation Support System 3 Figures 18 and 19 illustrate another aspect of the disease differentiation support system.

[0122] Figure 18 shows an example of the configuration of the disease differential diagnosis support system 3. The disease differentiation support system 3 comprises, for example, a vendor-side device 100 and a user-side device 200. The vendor-side device 100 comprises a processing unit 10 (10B), an input unit 16, and an output unit 17. The vendor-side device 100 operates as a disease differentiation support device 100B, similar to the disease differentiation support device 200B described above, performing both training processing and differentiation support information generation processing. On the other hand, the user-side device 200 operates as a terminal device 200C.

[0123] Here, the disease differentiation support device 100B is a cloud server-side device, for example, composed of a general-purpose computer. The disease differentiation support device 100B is connected to the cell image analysis device 300 and the blood cell counter 350 in a communicative manner. The disease differentiation support device 100B is also connected to the terminal device 200C in a communicative manner via the network 99. The terminal device 200C is a general-purpose computer, etc., and is connected to the cell image analysis device 400 and the blood cell counter 450 in a communicative manner.

[0124] (1) Hardware configuration of disease differentiation support device 100B The hardware configuration of the disease differentiation support device 100B may be the same as the hardware configuration of the vendor-side device 100 shown in Figure 11. In Figure 11, the processing unit 10A may be read as processing unit 10B in the case of the disease differentiation support device 100B.

[0125] (2) Functional configuration of disease differential diagnosis support system 3 Figure 19 shows an example of the functional configuration of the disease differential diagnosis support system 3. The functional configuration of the disease differentiation support device 100B may be the same as that described in 3.(2) above and in Figure 17.

[0126] Terminal device 200C acquires a first analysis parameter from cell image analysis device 400 and a second analysis parameter from blood cell counter 450, and transmits these analysis parameters to disease differentiation support device 100B via network 99. Disease differentiation support device 100B generates differentiation support information from the analysis parameters transmitted from terminal device 200C and transmits it to terminal device 200C.

[0127] 5. Computer Programs Another embodiment of this disclosure relates to a computer program that causes a computer to perform the training process described in 2-3.(3) above and in Figure 12.

[0128] Another embodiment of this disclosure relates to a computer program that causes a computer to perform the identification support processing shown in 2-4.(3) above and Figure 15.

[0129] The above computer program may be provided as a program product such as a recording medium that stores the above computer program. The above computer program may be stored on a recording medium such as a hard disk, a semiconductor memory element such as flash memory, or an optical disc. The format in which the program is stored on the above recording medium is not limited as long as the above processing unit can read the program. The storage on the above recording medium may be non-volatile.

[0130] 6. Verification of effectiveness The following shows the results of verifying the effectiveness of the diagnostic support methods for MDS disclosed herein.

[0131] 6-1. Sample (1) Training samples / Verification samples In this experiment, we used routine smears from the following patients who had visited Juntendo University Hospital and had already been diagnosed by a physician, as training and validation samples. ·45 patients with MDS (MDS group: N1=45) ·279 non-MDS patients (MDS group: N2=279) ·213 patients with HC (HC group: N3=213)

[0132] 6-2. Acquisition and Selection of Each Parameter (1) Acquisition of the first set of parameters Peripheral blood smears were prepared, and images of the smears were acquired using an Automated Digital Cell Morphology Analyzer DI-60 (Sysmex Corporation). From the acquired images, cell classification parameters and abnormality parameters were obtained using a computer algorithm described in Patent No. 7381003, etc. Cell classification parameters, as described above, are statistical information for each type obtained by classifying individual cells contained in the acquired image into one of several types (e.g., segmental neutrophils, band neutrophils, etc.) and aggregating the classification results of multiple cells contained in a single sample, such as the number or proportion of cells of each type. Parameters related to abnormalities, as described above, are the number or proportion of cells that have been assigned a predetermined abnormality among cells classified into a specific type. As a result, approximately 150 items were obtained for cell classification parameters and abnormality parameter.

[0133] (2) Acquisition of the second set of parameters Using the XN series hemocytometer, we obtained measurements of parameters related to the number of each cell type in the sample as a second parameter. As a result, approximately 15 second parameters were obtained.

[0134] (3) Selection of the first and second parameters For the first and second parameter groups described above, one-dimensional analysis of variance (ANOVA) was used to select items that were particularly highly correlated with differential diagnosis. Parameter selection was carried out only under the condition that there were no missing values ​​in any sample and that all parameters had sufficient accuracy to construct a computer algorithm for differential diagnosis. As a result, 30 items were selected as the first parameter from approximately 150 items, for example, 20 items shown in Figure 2 and 10 items shown in Figure 3. For the second parameter, 9 items were selected from approximately 15 items, for example, 9 items shown in Figure 4.

[0135] 6-3. Training For each training sample, the first and second parameter groups selected in 6-2.(3) above were obtained, and a matrix was generated for each training sample by arranging these parameters at the same hierarchical level. The disease labels of each patient from whom the training sample was collected were linked to each matrix, and training data as shown in Figure 5 was created. Then, the created training data was input into a gradient boosting tree, the algorithm was trained, and a classifier was generated. Python was used as the software.

[0136] 6-4. Verification For each validation sample, the first and second parameter groups selected in 6-2.(3) above were obtained, and a matrix was generated for each validation sample by arranging these parameters at the same hierarchical level. This matrix was then used as the analysis data for each validation sample. The generated analysis data was input into the identification device, and the identification results were obtained.

[0137] Furthermore, in this embodiment, the performance of machine learning was evaluated using the following learning and evaluation method (which can also be considered a cross-validation method). The dataset of all samples (MDS group (N1=45), non-MDS group (N2=279), HC group (N3=213), totaling 537 samples) was divided into five groups so that each group had an equal proportion. Four of these groups were used as training data, and the remaining group was used as test data. This was then performed for all five patterns in all five groups, and the best-performing result was adopted. This method differs from general methods (methods that use data independent of training for evaluation), but it is applicable even when sufficient sample data (specimens) is not available, making it a useful technique.

[0138] Figure 20 is a heatmap of the Shap values ​​showing the contribution of a total of 39 parameters: the first parameter set shown in Figures 2 and 3 (20 cell classification parameters and 10 abnormal findings parameters, totaling 30 items), and the second parameter set shown in Figure 4 (9 CBC parameters). In this diagram, the area on the right side of the diagram corresponds to the MDS group (N1=45), the area in the center of the diagram corresponds to the non-MDS group (N2=279), and the area on the left side of the diagram corresponds to the HC group (N3=213). Due to the limitations of the diagram, it can be difficult to determine which part has the greatest contribution. Therefore, to make it easier to understand, black circles have been placed around the parameters that are particularly effective in identification.

[0139] Looking at the MDS group, we can see that (2E) mean corpuscular volume (MCV), one of the second parameters (CBC parameters) in Figure 4, is particularly effective in differential diagnosis, meaning it is especially effective in differentiating MDS. Furthermore, among the first parameters, the large platelets ((1N) large platelets (%), (1T) large platelets ( / 100 WBC)) from the cell classification parameters in Figure 2, and the (1e) SN-pseudopelger (%) from the abnormal findings parameters in Figure 3, are particularly effective in differential diagnosis, meaning they are especially effective in differentiating MDS.

[0140] Based on this finding, it is possible to use at least all or some of the above parameters as essential parameters for training and differentiation, or not. In other words, for example, one or both of large platelets and SN-pseudopelger may be used as essential parameters, and mean corpuscular volume (MCV) may be used as essential parameters, or not.

[0141] Furthermore, the device may be configured to allow users to set the parameters used for training and identification based on user input.

[0142] Figures 21 to 24 are diagrams that support the usefulness of the mechanical method using the identification device in this embodiment. Figure 21 is a table showing the degree of discriminative ability with respect to the physician's definitive diagnosis in each case group. Of the 45 patients diagnosed with MDS by a physician (MDS group), 32 were predicted to have MDS using the automated method. The AUC value was 0.99. Of the 279 patients diagnosed with non-MDS by a physician (non-MDS group), 278 were predicted to have non-MDS by the automated method. The AUC value was 1.00. Of the 213 patients diagnosed with HC by a physician (HC group), 212 were predicted to have HC using the automated method. The AUC value was 1.00.

[0143] Figure 22 is a table showing the results of calculating accuracy, precision, sensitivity, and specificity based on the results in Figure 21. These results show that the accuracy for each of the MDS group, non-MDS group, and HC group was 0.974 or higher (at least 0.974), which is good.

[0144] Figure 23 shows the ROC curves for MDS, non-MDS, and HC predicted by the machine method. In all cases, the AUC value was 0.99 or higher (at least 0.99), indicating favorable results.

[0145] Figure 24 shows the probabilities output by the diagnostic tool for all patients plotted in a three-dimensional space (MDS, non-MDS, HC). Using a cutoff value of "0.7" for probability, the only cases where prediction errors occurred when "probability ≥ 0.7" were three instances where the confirmed diagnosis was MDS but the patient was predicted to be non-MDS. However, these three patients were actually cases that had progressed from MDS to ALL (acute lymphoblastic leukemia). In other words, they were actually non-MDS cases, so the prediction can be considered correct.

[0146] 7. Effects of the Embodiment In this embodiment, a disease differentiation support device (e.g., disease differentiation support devices 200A, 200B, 100B) obtains a first parameter (e.g., abnormal findings parameter in Figure 3, cell classification parameter in Figure 2) regarding abnormal cell morphology obtained by analyzing images of cells contained in a sample taken from a subject, and a second parameter (e.g., CBC parameter in Figure 4) regarding the number of each type of cell contained in the sample. Using a pre-trained computer algorithm (e.g., a pre-trained differentiater), the device generates differentiation support information to assist in differentiating between myelodysplastic syndrome (MDS) and other conditions (e.g., non-MDS, healthy individuals (HC)) based on the first and second parameters. This allows for effective differentiation between myelodysplastic syndrome and other conditions based on a first parameter related to abnormal cell morphology and a second parameter related to the number of each type of cell in the sample, obtained by analyzing images of cells contained in a sample taken from a subject using a pre-trained computer algorithm.

[0147] Furthermore, obtaining the first parameter may include obtaining a first parameter (e.g., a first parameter for analysis) based on the analysis of a first cell image analysis device (e.g., cell image analysis device 400) that analyzes images, and obtaining the second parameter may include obtaining a second parameter (e.g., a second parameter for analysis) based on the analysis of an optical or electrical signal obtained from cells contained in the sample by a first hematological counter (e.g., hematological counter 450). This makes it easy to obtain the first and second parameters.

[0148] Furthermore, the computer algorithm may be pre-trained based on a first training parameter selected from multiple types of parameters obtained based on analysis by a second cell image analyzer (e.g., cell image analyzer 300), a second training parameter selected from multiple types of parameters obtained based on analysis by a second blood cell counter (e.g., blood cell counter 350), and a disease label representing the disease. The first training parameter may be selected based on the results of significance testing (e.g., ANOVA) performed between disease groups for each of the multiple types of parameters obtained based on analysis by the second cell image analyzer, and the second training parameter may be selected based on the results of significance testing between disease groups for each of the multiple types of parameters obtained based on analysis by the second blood cell counter. This allows us to obtain appropriate parameters for the first and second training parameters.

[0149] Furthermore, the first parameter and the first training parameter may be the same type of parameter, the second parameter and the second training parameter may be the same type of parameter, the first cell image analysis device and the second cell image analysis device may be the same or different devices of the same type, and the first blood cell counter and the second blood cell counter may be the same or different devices of the same type.

[0150] Furthermore, generating differential diagnosis support information may include generating primary information regarding which of the following groups the subject is most likely to belong to: a first patient group (e.g., MDS group) that has or is likely to have myelodysplastic syndrome (MDS), a second patient group (e.g., non-MDS group) that has or is likely to have another hematopoietic disorder other than myelodysplastic syndrome, and a group of healthy individuals (e.g., HC group). This makes it possible to generate primary information, which is likely to be used as differential diagnostic support information, regarding which group a subject is most likely to belong to: Group 1 of patients who have or are likely to have myelodysplastic syndrome; Group 2 of patients who have or are likely to have other hematopoietic disorders other than myelodysplastic syndrome; or Group 3 of healthy individuals.

[0151] Furthermore, other hematopoietic disorders may include acute lymphoblastic leukemia (ALL), acute myeloid leukemia (AML), malignant lymphoma (ML), myeloproliferative neoplasms (MPN), and other hematopoietic malignancies.

[0152] Furthermore, the disease differentiation support device may be configured to output the first information to an output device (for example, a display device, an audio output device). This allows the system to present users with information on which group a subject is most likely to belong to: Group 1, who have or are likely to have myelodysplastic syndrome; Group 2, who have or are likely to have other hematopoietic disorders other than myelodysplastic syndrome; or Group 3, who are healthy.

[0153] 8. Other Embodiments (1) The disease differentiation support device may display various types of information, such as differentiation support information, on a display device which is a type of output unit, and may also output sound from a sound output device which is a type of output unit.

[0154] (2) In the above embodiment, MDS, non-MDS, and HC were used as disease labels, but the embodiment is not limited to this, and for example, non-MDS and HC may be used as a common disease label (e.g., "Others") and the same processing as in the above embodiment may be performed.

[0155] (3) For example, the probabilities obtained by the diagnostic tool, as shown in Figure 24, may be plotted on a plane or in space corresponding to the number of diseases to be classified (including HC), and this graph may be displayed on the display unit of the disease differentiation support device or on an external display device. In other words, the information that visualizes the prediction results by the machine method may be displayed on the display device. The cutoff value for probability can be set based on known methods, and this cutoff value may also be displayed along with the graph.

[0156] Furthermore, in the example of the above embodiment, for example, regions where there is a high probability of non-MDS but also a possibility of MDS may be displayed in a different manner from other regions, such as a probability of non-MDS being "less than 0.8", a probability of MDS being "0.2 or higher", and a probability of HC being "less than 0.1". This allows physicians and others to use plots in this region as an indicator for carefully determining whether MDS is present. Furthermore, to make the plots included in this region identifiable, objects such as arrows as shown in Figure 24 may be added to their display, or they may be displayed in a different manner from other plots (for example, by changing the color, changing the shape, or making them blink).

[0157] Furthermore, in cases where a diagnosis (not necessarily a definitive diagnosis) has already been made by a physician, plots corresponding to subjects whose predicted results are considered incorrect to the physician's diagnosis may be displayed in a different manner from other plots (for example, by changing the color, changing the shape, or making them flash).

[0158] In this case, for example, the processing unit of the disease differentiation support device may generate a graph plotting the probability values ​​for each disease generated in step S24 after (or in) step S25 of the process shown in Figure 15, and display it on the display unit.

[0159] Thus, the first information regarding which group a subject is most likely to belong to—the first patient group having or being likely to have myelodysplastic syndrome, the second patient group having or being likely to have another hematopoietic disorder other than myelodysplastic syndrome, or the healthy control group—includes the probability that the subject belongs to the first patient group, the second patient group, or the healthy control group. The disease differentiation support device may display second information based on this probability on its display device. This allows the user to be informed, through a probability-based display, of which group the subject is most likely to belong to: Group 1 of patients who have or are likely to have myelodysplastic syndrome; Group 2 of patients who have or are likely to have other hematopoietic disorders other than myelodysplastic syndrome; or Group 3 of healthy individuals. [Explanation of Symbols]

[0160] 100A training device 200A,200B,100B Disease differentiation support device 300,400 Cell image analysis equipment 350,450 blood cell counter

Claims

1. A computer-based method for supporting disease differentiation, By analyzing images of cells contained in a sample taken from the subject, we obtain a first parameter related to cell classification and abnormal morphology. A second parameter is obtained regarding the number of each type of cell contained in the aforementioned sample. Using a pre-trained computer algorithm, differential information is generated to assist in differentiating between myelodysplastic syndrome and other conditions, based on the first and second parameters. Disease differentiation support method.

2. Obtaining the first parameter includes obtaining the first parameter based on the analysis of the image by a first cell image analysis device that analyzes the image, Obtaining the second parameter includes obtaining the second parameter based on the analysis of an optical or electrical signal obtained from cells contained in the sample using a first hemocytometer. The disease differentiation support method according to claim 1.

3. The aforementioned computer algorithm is pre-trained based on a first training parameter selected from multiple parameters obtained based on the analysis of the second cell image analysis device, a second training parameter selected from multiple parameters obtained based on the analysis of the second blood cell counter, and a disease label representing the disease. The first training parameter is a parameter selected based on the results of significance testing between disease groups for each of the multiple parameters obtained based on the analysis of the second cell image analysis device. The second training parameter is a parameter selected based on the results of significance testing between disease groups for each of the multiple parameters obtained based on the analysis of the second blood cell counter. The disease differentiation support method according to claim 2.

4. The first parameter and the first training parameter are of the same type. The aforementioned second parameter and the aforementioned training second parameter are of the same type. The first cell image analysis device and the second cell image analysis device are the same or different devices of the same type. The first blood cell counter and the second blood cell counter are the same or different devices of the same type. The disease differentiation support method according to claim 3.

5. The first cell image analysis device analyzes the image captured by the imaging unit. The disease differentiation support method according to claim 2.

6. The first blood cell counter analyzes the optical signal or electrical signal obtained by the detection unit. The disease differentiation support method according to claim 2.

7. The detection unit includes a flow cytometer, The first blood cell counter analyzes the optical signal obtained by the flow cytometer. The disease differentiation support method according to claim 6.

8. The abnormal morphology includes at least one selected from nuclear morphological abnormalities, granular abnormalities, cell size abnormalities, cellular malformations, cell destruction, vacuoles, immature cells, presence of inclusion bodies, Döhle bodies, satellite phenomena, nuclear chromatin abnormalities, petal-like nuclei, high N / C ratio, bleb-like morphology, smudge, and hairy cell-like morphology. The disease differentiation support method according to claim 1.

9. The nuclear morphological abnormality includes at least one selected from hyperlobulation, hypolobulation, pseudopelger nucleus abnormality, ring-shaped nucleus, spherical nucleus, elliptic nucleus, apoptosis, multinucleus, nuclear decay, denucleation, naked nucleus, irregular nuclear margin, nuclear fragmentation, internuclear bridge, multiple nuclei, notched nucleus, nuclear fission, and nucleolar abnormality. The granular abnormality includes at least one selected from degranulation, granular distribution abnormality, toxic granules, Auer bodies, Fagott cells, and pseudoChediak-Higashi granule-like granules. The aforementioned abnormalities in cell size include, The disease differentiation support method according to claim 8.

10. The first parameter includes a parameter relating to at least one of the number and proportion of atypical lymphocytes; or a parameter relating to at least one of the number of each type of cell having an abnormal morphology selected from neutrophils, eosinophils, platelets, lymphocytes, monocytes, basophils, metamyelocytes, myelocytes, promyelocytes, blasts, plasma cells, immature eosinophils, immature basophils, erythroblasts, and megakaryocytes, and the proportion of each type of cell having an abnormal morphology. The disease differentiation support method according to claim 1.

11. The second parameter is composed of multiple parameters obtained by measuring CBC (Complete Blood Count). The disease differentiation support method according to claim 1.

12. The second parameter includes a value related to at least one selected from red blood cells, nucleated red blood cells, small red blood cells, platelets, hemoglobin, reticulocytes, immature granulocytes, neutrophils, eosinophils, basophils, lymphocytes, monocytes, hematocrit, mean corpuscular volume (MCV), mean corpuscular hemoglobin level (MCH), mean corpuscular hemoglobin concentration (MCHC), mean platelet volume (MPV), and erythrocyte distribution width (RDW). The disease differentiation support method according to claim 1.

13. The aforementioned computer algorithm includes a machine learning algorithm. The disease differentiation support method according to claim 1.

14. The machine learning algorithm includes an algorithm selected from trees, regression, support vector machines, Bayesian methods, clustering, and random forests. The disease differentiation support method according to claim 13.

15. The aforementioned machine learning algorithm includes a gradient boosting tree algorithm, The disease differentiation support method according to claim 14.

16. Obtaining the first parameter includes analyzing the image using a deep learning algorithm having a neural network structure. The disease differentiation support method according to claim 1.

17. Obtaining the first parameter includes obtaining the first parameter based on the cell classification result output from the deep learning algorithm and the probability of the cell being classified as such. The disease differentiation support method according to claim 16.

18. Generating the aforementioned differential diagnosis support information includes generating first information regarding which of the following groups the subject is most likely to belong to: a first patient group suffering from or potentially suffering from myelodysplastic syndrome; a second patient group suffering from or potentially suffering from another hematopoietic disorder other than myelodysplastic syndrome; and a healthy control group. The disease differentiation support method according to claim 1.

19. The aforementioned other hematopoietic disorders include acute lymphoblastic leukemia, acute myeloid leukemia, malignant lymphoma, myeloproliferative neoplasms, and other hematopoietic malignancies. The disease differentiation support method according to claim 18.

20. This includes outputting the first information to an output device. The disease differentiation support method according to claim 18 or 19.

21. The first information includes the probability that the subject belongs to the first patient group, the second patient group, or the healthy control group. This includes displaying second information based on the aforementioned probability on a display device. The disease differentiation support method according to claim 18 or 19.

22. A disease differentiation support device for assisting in the differentiation of diseases, A first parameter is obtained regarding the number of abnormally shaped cells by analyzing images of cells contained in a sample taken from a subject, and a second parameter is obtained regarding the number of each type of cell contained in the sample. A pre-trained computer algorithm is used to generate differential information to assist in differentiating between myelodysplastic syndrome and other conditions based on the first and second parameters. Disease differentiation support device.

23. Generating the aforementioned differential diagnosis support information includes generating first information regarding which of the following groups the subject is most likely to belong to: a first patient group suffering from or potentially suffering from myelodysplastic syndrome; a second patient group suffering from or potentially suffering from another hematopoietic disorder other than myelodysplastic syndrome; and a healthy control group. The disease differentiation support device according to claim 22.

24. Control is performed to output the first information to the output device. The disease differentiation support device according to claim 23.

25. The first information includes the probability that the subject belongs to the first patient group, the second patient group, or the healthy control group. The system controls the display device to show the second piece of information based on the aforementioned probability. The disease differentiation support device according to claim 23.

26. A program to be executed by a computer, By analyzing images of cells contained in samples taken from the subject, we obtained a first parameter related to the number of cells with abnormal morphology. A second parameter is obtained regarding the number of each type of cell contained in the aforementioned sample. Using a pre-trained computer algorithm, differential diagnosis support information is generated based on the first and second parameters to assist in the differentiation of myelodysplastic syndromes. A program to cause the computer to perform the above action.

27. Generating the aforementioned differential diagnosis support information includes generating first information regarding which of the following groups the subject is most likely to belong to: a first patient group suffering from or potentially suffering from myelodysplastic syndrome; a second patient group suffering from or potentially suffering from another hematopoietic disorder other than myelodysplastic syndrome; and a healthy control group. The program according to claim 26.

28. The program according to claim 27, which causes the computer to output the aforementioned identification support information to an output device.

29. The first information includes the probability that the subject belongs to the first patient group, the second patient group, or the healthy control group. The program according to claim 27, which causes the computer to perform the action of displaying second information based on the aforementioned probability on a display device.

Images