JP6112597B2 - Diagnosis support device using CBC scattergram - Google Patents

Description

  The present invention relates to a diagnosis support apparatus that provides diagnosis support information based on blood cell distribution information of a blood sample.

  The CBC test is a screening test with the largest number of requests among clinical tests. By using the flow site method, particle size data of tens of thousands of blood cells per specimen can be obtained. These particle size data have various patterns and are considered to be observed as a result of blood cells reacting through complex networks such as cytokines related to immunity and inflammation. Therefore, it may be possible to indirectly detect unexplained diagnostic characteristics and dynamics of some cytokines by capturing these blood cell dynamic patterns objectively and in detail.

  The inventors have collected measurement data of about 500,000 specimens of the CBC test conducted as a screening test for daily medical care and conducted a comprehensive analysis on the relationship with the disease name history. As a result, it has been found that there is a feature that enables diagnosis of gastric cancer including early gastric cancer, which has been difficult to detect by routine clinical tests (see Non-Patent Document 1). Non-Patent Document 1 discloses that by comprehensively analyzing the relationship between leukocyte distribution data and diagnostic disease names, it is possible to scientifically discover features that could not be found visually.

Hiroaki Kataoka and four others, "CBC cell distribution data yield remarkable laboratory diagnostic characteristics", The Ace International Conference of・ The 8th International Conference of Clinical Laboratory Automation, April 2012, p. 49

  However, Non-Patent Document 1 mentioned above suggests that the white blood cell distribution pattern of the scattergram (frequency distribution diagram using two parameters) obtained by CBC examination can be used for diagnosis of gastric cancer such as early gastric cancer. However, how to generate diagnostic support information from the white blood cell distribution pattern is not clarified.

  The present invention has been made in view of such circumstances, and provides a diagnosis support apparatus capable of accurately providing diagnosis support information related to diseases such as stomach cancer from a blood cell distribution pattern of a scattergram obtained by blood cell analysis. The purpose is to provide.

In order to solve the above-described problem, a diagnosis support apparatus according to an aspect of the present invention provides a blood cell frequency distribution map in a two-dimensional coordinate space based on two characteristic parameters obtained by optically measuring a blood sample. based on, a diagnosis support apparatus for generating diagnosis support information, and the blood cell frequency distribution diagram data acquisition unit for acquiring blood cell frequency distribution diagram data for a particular subject, for a given disease to power in the blood cell frequency distribution diagram A storage unit that stores ROC curve information indicating a relationship between sensitivity and specificity for each position in the two-dimensional coordinate space, blood cell frequency distribution diagram data acquired by the blood cell frequency distribution diagram data acquisition unit, and the storage unit based on the ROC curve information stored in the secondary positive likelihood ratio specified by the sensitivity and the specificity corresponding to the power in the blood cell frequency distribution diagram Calculated for each position in the coordinate space, likelihood matrix distance is the total sum of the calculated positive likelihood ratio is calculated, and a calculation unit for generating diagnosis support information based on the likelihood matrix distance, the arithmetic unit And an output unit that outputs the diagnosis support information generated by the above.

In the above aspect, the calculation unit may be configured to generate diagnosis support information indicating that the specific patient is positive when the likelihood matrix distance exceeds a predetermined threshold.

In the above aspect, the predetermined threshold may be obtained by ROC analysis of a plurality of likelihood matrix distances obtained for a plurality of patients.

  In the above aspect, the predetermined disease may be cancer, and the blood cell frequency distribution map may be a white blood cell frequency distribution map.

  In the above aspect, the predetermined disease may be diabetes, acute pancreatitis, or cirrhosis, and the blood cell frequency distribution map may be a reticulocyte frequency distribution map.

  ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to provide the diagnostic assistance information regarding diseases, such as stomach cancer, accurately from the blood cell distribution pattern of the scattergram obtained by the CBC test.

The figure which shows the structure of the diagnosis assistance apparatus which concerns on embodiment. The schematic diagram which shows a DIFF scattergram. The schematic diagram which shows a WBC / BASO scattergram. The schematic diagram which shows a RET scattergram. The flowchart which shows the flow of operation | movement of the diagnosis assistance apparatus which concerns on embodiment. The histogram which shows distribution of likelihood matrix distance in an early gastric cancer patient and other disease group patients. ROC curve graph showing likelihood matrix distance and laboratory diagnostic characteristics for early gastric cancer. The schematic diagram which shows the method of exhaustive derivation | leading-out of an inspection diagnostic characteristic. The figure which shows the AUC matrix pattern of the gastric cancer in a DIFF channel. The figure which shows the AUC matrix pattern of the early gastric cancer in a WBC / BASO channel. The figure which shows the AUC matrix pattern of the gastric cancer in a WBC / BASO channel. The graph of the ROC curve which shows the examination diagnostic characteristic of the gastric cancer in 14-26 coordinates of a DIFF channel. The graph of the ROC curve which shows the diagnostic characteristic about the stomach part cancer using the likelihood matrix distance by a DIFF scattergram. The histogram which shows distribution of likelihood matrix distance in a gastric cancer patient and another disease group patient. The figure which shows the AUC matrix pattern of the prostate cancer in a WBC / BASO channel. The figure which shows the AUC matrix pattern of the prostate cancer in a DIFF channel. The figure which shows the AUC matrix pattern of the prostatic hypertrophy in a DIFF channel.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

<Configuration of diagnosis support device>
FIG. 1 is a diagram showing a configuration of a diagnosis support apparatus 1 according to the present embodiment. The diagnosis support apparatus 1 according to the present embodiment is configured by a computer 5 mainly composed of a main body unit 2, a display unit 3, and an input device 4. The main body 2 mainly includes a CPU 2a, a ROM 2b, a RAM 2c, a hard disk 2d, a reading device 2e, an input / output interface 2f, a communication interface 2g, and an image output interface 2h. The CPU 2a, ROM 2b, RAM 2c, hard disk 2d, reading device 2e, input / output interface 2f, communication interface 2g, and image output interface 2h are connected by a bus 2i.

  The CPU 2a can execute a computer program stored in the ROM 2b and a computer program loaded in the RAM 2c. The computer 5 functions as the diagnosis support apparatus 1 when the CPU 2a executes a diagnosis support program 6a described later.

  The ROM 2b is configured by a mask ROM, PROM, EPROM, EEPROM, or the like, and stores a computer program executed by the CPU 2a, data used for the same, and the like.

  The RAM 2c is configured by SRAM, DRAM, or the like. The RAM 2c is used for reading out computer programs recorded in the ROM 2b and the hard disk 2d. Further, when these computer programs are executed, they are used as a work area of the CPU 2a.

  The hard disk 2d is installed with various computer programs to be executed by the CPU 2a, such as an operating system and application programs, and data used for executing the computer programs. A diagnosis support program 6a is also installed in the hard disk 2d. By executing the diagnosis support program 6a by the CPU 2a, the diagnosis support apparatus 1 operates as described later.

  The reading device 2 e is configured by a flexible disk drive, a CD-ROM drive, a DVD-ROM drive, or the like, and can read a computer program or data recorded on the portable recording medium 6. The portable recording medium 6 stores a diagnosis support program 6a, and the computer 5 reads the diagnosis support program 6a from the portable recording medium 6 and installs the diagnosis support program 6a in the hard disk 2d. Is possible.

  The diagnostic support program 6a is not only provided by the portable recording medium 6, but also from the external device connected to be able to communicate with the computer 5 through an electric communication line (whether wired or wireless). It is also possible to provide it through a line. For example, the diagnosis support program 6a is stored in the hard disk of a server computer on the Internet, and the computer 5 accesses the server computer to download the diagnosis support program 6a and install it on the hard disk 2d. Is also possible.

  Further, for example, an operating system such as Windows (registered trademark) manufactured and sold by Microsoft Corporation is installed in the hard disk 2d. In the following description, it is assumed that the diagnosis support program 6a operates on the operating system.

  The hard disk 2d is provided with an ROC database 21. Details of the ROC database 21 will be described later.

  The input / output interface 2f includes, for example, a serial interface such as USB, IEEE 1394, and RS-232C, a parallel interface such as SCSI, IDE, and IEEE1284, and an analog interface including a D / A converter and an A / D converter. Has been. An input device 4 is connected to the input / output interface 2 f, and the user can input data to the computer 5 by using the input device 4.

  The communication interface 2g is an Ethernet (registered trademark) interface, for example. The computer 5 can transmit and receive data to and from the blood cell analyzer 7 by using the communication interface 2g using a predetermined communication protocol.

<Configuration of blood cell analyzer>
The diagnosis support apparatus 1 is connected to the blood cell analyzer 7 so that data communication is possible. The blood cell analyzer 7 sucks a blood sample from a sample container (collecting blood vessel), mixes the sucked blood sample with a reagent (hemolyzing agent and fluorescent dye), prepares a measurement sample used for measurement, and prepares the measurement The sample is optically measured. The blood sample is measured by the blood cell analyzer 7 for the DIFF channel that classifies leukocytes into four types (monocytes, lymphocytes, eosinophils, and neutrophils), and WBC for detecting basophils of leukocytes. / BASO channel, and RET channel for detecting reticulocytes.

  In the measurement of the DIFF channel (hereinafter referred to as “DIFF measurement”), a measurement sample in which a blood sample and a DIFF measurement reagent are mixed is supplied to an optical detector, and at this time, optical information (fluorescence intensity) is detected by the optical detector. , Forward scattered light intensity and side scattered light intensity). The optical information obtained by the DIFF measurement is processed by the control unit built in the blood cell analyzer 7, and the white blood cells in the blood sample are classified into four types. At this time, the control unit generates a scattergram of the DIFF channel (hereinafter referred to as “DIFF scattergram”). FIG. 2 is a schematic diagram showing a DIFF scattergram. As shown in the figure, the DIFF scattergram is a scattergram in which the fluorescence intensity and the side scattered light intensity are optical characteristic parameters.

  In measurement of a WBC / BASO channel (hereinafter referred to as “WBC / BASO measurement”), a measurement sample in which a blood sample and a WBC / BASO measurement reagent are mixed is supplied to an optical detector, and at this time, the optical detector Thus, optical information (fluorescence intensity, forward scattered light intensity, and side scattered light intensity) is detected. Optical information obtained by the WBC / BASO measurement is processed by the control unit, and basophils and other leukocytes in the blood sample are detected. At this time, the control unit generates a scattergram of the WBC / BASO channel (hereinafter referred to as “WBC / BASO scattergram”). FIG. 3 is a schematic diagram showing a WBC / BASO scattergram. As shown in the figure, the WBC / BASO scattergram is a scattergram having forward scattered light intensity and side scattered light intensity as optical characteristic parameters.

  In measurement of the RET channel (hereinafter referred to as “RET measurement”), a measurement sample in which a blood sample and a reagent for RET measurement are mixed is supplied to an optical detector, and at this time, optical information (fluorescence intensity) is detected by the optical detector. , Forward scattered light intensity and side scattered light intensity). The optical information obtained by the RET measurement is processed by the control unit, and reticulocytes in the blood sample are detected. At this time, the control unit generates a RET channel scattergram (hereinafter referred to as “RET scattergram”). FIG. 4 is a schematic diagram showing a RET scattergram. As shown in the figure, the RET scattergram is a scattergram having forward scattered light intensity and fluorescence intensity as optical characteristic parameters.

  Scattergram data obtained in this way is transmitted from the blood cell analyzer 7 to the diagnosis support apparatus 1.

<Explanation of ROC database>
The ROC database 21 stores a plurality of ROC curves. Each ROC curve is created using a scattergram obtained by measuring blood samples for a large number of patients. Hereinafter, a method for creating the ROC curve will be described.

  Scattergram data, which is the distribution data of each blood sample obtained by the blood cell analyzer, is located at each address (i, j) of n × n (n = 256) in a two-dimensional coordinate space with two characteristic parameters as orthogonal coordinate axes. It is configured as a data group to which the frequency F (i, j) is assigned. In order to create an ROC curve, first, 256 × 256 scattergram data is deteriorated and converted into 32 × 32 two-dimensional scattergram data. The frequency data of each of the 1024 cells in the scattergram after conversion is set to 1024 inspection values.

  For one disease, one or more of a DIFF scattergram, a WBC / BASO scattergram, and a RET scattergram are used. For example, the DIFF scattergram is used for gastric cancer, the WBC / BASO scattergram is used for early gastric cancer, and the RET scattergram is used for diabetes.

  A comprehensive ROC analysis is performed on the relationship between the data of a two-dimensional scattergram having a resolution of 32 × 32 and the history of a predetermined disease. This gives 1024 ROC curves for each scattergram used for one disease. For example, for gastric cancer, 1024 ROC curves corresponding to the DIFF scattergram are obtained, for early gastric cancer, 1024 ROC curves corresponding to the WBC / BASO scattergram are obtained, and for diabetes, 1024 ROC curves corresponding to the RET scattergram are obtained.

<Operation of diagnosis support device>
Next, the operation of the diagnosis support apparatus according to this embodiment will be described. In the following, the operation of the diagnosis support apparatus 1 when generating diagnosis support information for early gastric cancer will be described.

  First, the blood sample of the subject is measured by the blood cell analyzer 7. The blood cell analyzer 7 prepares a measurement sample by mixing a reagent with a blood sample of a subject, and performs each of DIFF measurement, WBC / BASO measurement, and RET measurement. Thereby, a DIFF scattergram, a WBC / BASO scattergram, and a RET scattergram of the subject are obtained.

  FIG. 5 is a flowchart showing a flow of operations of the diagnosis support apparatus according to the present embodiment. When creating the diagnosis support information of the subject, the CPU 2a of the diagnosis support apparatus 1 requests the blood cell analyzer 7 for the scattergram data of the subject. In response to this request, the blood cell analyzer 7 transmits the subject's scattergram data, which is received by the diagnosis support apparatus 1 (step S1).

  When the diagnosis support apparatus 1 acquires the scattergram data, the CPU 2a converts the scattergram having a resolution of 256 × 256 into a scattergram of 32 × 32 by degradation (Step S2). Thereby, each cell of the subject's scattergram corresponds to the ROC curve in the ROC database 21.

  Next, the CPU 2a selects a scattergram corresponding to a disease for which diagnosis support information is to be created, and applies an ROC curve corresponding to the frequency (measurement value) of each cell to each cell of the scattergram. The positive likelihood ratio is calculated (step S3). Here, since the disease for which diagnosis support information is created is early gastric cancer, the WBC / BASO scattergram corresponding to early gastric cancer is selected, and the positive likelihood ratio of each cell of the WBC / BASO scattergram is calculated. .

Here, the likelihood ratio is a value obtained by dividing the probability (sensitivity) that a test is positive for a patient with a disease by the probability (false positive rate) that the test is positive for a patient without a disease. That is, it can be calculated by the following formula.
Likelihood ratio = sensitivity / (1-specificity)

  When diagnosing a certain disease with a certain measurement value, these two relations can be obtained as examination diagnostic characteristics by ROC analysis. This characteristic is expressed on the ROC curve, and the sensitivity and specificity are determined by the measurement value (frequency) of the test. That is, the likelihood ratio when diagnosing a certain disease with a certain measurement value can also be obtained by the above calculation.

  A likelihood matrix distance is defined for the purpose of handling 1024 cells constituting a 32 × 32 two-dimensional scattergram as a comprehensive distance index. The likelihood ratio is calculated from the sensitivity and specificity for 1024 measurement values. When the likelihood ratio exceeds 100, the likelihood ratio is 100, and the sum of the 1024 likelihood ratios is used as the likelihood matrix distance. The CPU 2a calculates a likelihood matrix distance (step S4). That is, the CPU 2a calculates the sum of the positive likelihood ratios of the respective cells and sets this as the likelihood matrix distance.

  Next, the CPU 2a compares the likelihood matrix distance with a predetermined cut-off value and generates diagnosis support information (step S5). This cutoff value is a value determined for each disease and is stored in advance in the hard disk 2d.

  The creation of cut-off values for patients with early gastric cancer will be described. For a plurality of patients with early gastric cancer and other disease group patients, a likelihood matrix distance for early gastric cancer is obtained, and ROC analysis is performed. FIG. 6 is a histogram showing the distribution of likelihood matrix distance between patients with early gastric cancer and other disease group patients, and FIG. 7 is a diagnostic characteristic for early gastric cancer using the likelihood matrix distance by the WBC / BASO scattergram. It is a graph of the ROC curve which shows. As a result of ROC analysis, AUC was 0.918, which was a very good result. The optimal cutoff value is 623, and it can be seen that early gastric cancer can be diagnosed with performance of sensitivity 85% and specificity 86% by setting the likelihood matrix distance to 623.

  When the likelihood matrix distance as the evaluation value is greater than or equal to the cut-off value, the CPU 2a generates diagnosis support information indicating that there is a high possibility of being positive. On the other hand, when the likelihood matrix distance is less than the cutoff value, the CPU 2a generates diagnosis support information indicating that there is a high possibility of being negative.

  The CPU 2a displays the diagnosis support information generated as described above on the display unit 3 (step S6), and ends the process. In step S6, for example, “probability of early gastric cancer: high” is output as diagnosis support information indicating that there is a high possibility of being positive. On the other hand, for example, “probability of early gastric cancer: low” is output as diagnosis support information indicating that there is a high possibility of being negative.

<Evaluation test>
The inventors conducted an evaluation test of the diagnosis support method according to the present invention as follows and verified its performance.

Using a blood cell analyzer XE-2100 (manufactured by Sysmex), CBC examinations requested in daily clinical examinations and raw measurement data of about 500,000 specimens were collected from 2006 to 2012. In this study, measurement data of DIFF channel and WBC / BASO channel were used.
Scattergram data, which is distribution data of each specimen obtained by the blood cell analyzer, is a data group in which a frequency F (i, j) is assigned to each address (i, j) of n × n (n = 256) in the scattergram space. Configured as
These scattergram data were once converted into a two-dimensional scattergram having a resolution of 256 × 256 to 32 × 32, and the frequency data at each address was stored in the data warehouse as inspection values of 1024 items.
The breakdown of the number of cases in the gastric cancer-related disease group was 225 early gastric cancers, 186 gastric body cancers, 583 gastric cancers, and 349 esophageal cancers.

  FIG. 8 shows a method for exhaustive derivation of examination diagnostic characteristics. A comprehensive ROC analysis was performed on the relationship between the data of a two-dimensional scattergram having a resolution of 32 × 32 and the history of cases with 100 or more cases, and the results were stored in a database.

Data sampling of the positive group and negative group necessary for ROC analysis uses the latest CBC test values from the day when the definitive diagnosis disease name was first given until 14 days ago. The event date on which the disease name was recorded for the first time in the absence of the patient group was selected. Furthermore, a maximum of 5000 cases were sampled randomly from these patient groups and analyzed using bootstrap ROC analysis.
For bootstrap ROC analysis, refer to the following document.
Xavier Robin, Natacha Turck et al pROC: an open-source package for R and S + to analyze and compare ROC curves BMC Bioinformatics, 12, 77, 2011.

By displaying the AUC values obtained from the coordinates of the two-dimensional scattergram divided into 1024 as a matrix corresponding to the two-dimensional scattergram, an area having high examination diagnostic characteristics was visualized. Hereinafter, this is referred to as an AUC matrix.
When the AUC matrix is positive at a high value, it is displayed by hatching with a slanting line to the left (lower half, especially the left area). When it is positive at a low value, AUC × -1 is calculated, Displayed with hatched hatching (upper left area).
The example in the lower right of FIG. 8 shows an example of a typical RET (reticulocyte) channel and iron deficiency anemia AUC matrix pattern. Since iron deficiency anemia is microcytic hypochromic anemia, the frequency of large cell areas decreases and is displayed as a right-slanting diagonal line, and the small cell area increases, so the lower area is displayed as a downward-sloping diagonal line Pattern. By displaying as an AUC matrix pattern, the cell state can be easily read.

The relationship between the AUC matrix pattern and each disease group was evaluated using a phylogenetic tree and a network diagram.
In the evaluation method using the phylogenetic tree, clustering was performed using the Ward method using the Euclidean distance of the AUC matrix for disease groups with high diagnostic characteristics.
Furthermore, for the purpose of clarifying the dynamic model of the similarity relationship with the whole disease group, disease name pairs with high similarity to all disease names were selected and a network diagram was created and evaluated.
The network diagram used the Euclidean distance of the AUC pattern, and the Fruchterman Reingold method was used as a graph drawing algorithm.
For the Fruchterman Reingold method, refer to the following document.
Fruchterman, Thomas MJ; Reingold, Edward M. Graph Drawing by Force-Directed Placement Software-Practice & Experience (Wiley) 21 (11): pp1129-1164,1991.

  When searching for disease names in descending order of AUC of ROC analysis, the best results were obtained with febrile neutropenia with AUC = 0.93 in the DIFF channel, and febrile neutropenia with the WBC / BASO channel. As a result, chorioamnionitis was the best at AUC = 0.95, and unexpectedly, pregnancy related diseases including normal delivery were ranked high. The disease group that was known to affect blood cells in general was the highest. On the other hand, in the DIFF channel and the WBC / BASO channel, AUC values exceeding 0.8 for gastric cancer and esophageal cancer were obtained among the unexpected results among the top-ranking disease names.

  9A-9C show AUC matrix patterns of gastric cancer in DIFF and WBC / BASO channels. FIG. 9A is an AUC matrix pattern of gastric cancer in the DIFF channel, FIG. 9B is an AUC matrix pattern of early gastric cancer in the WBC / BASO channel, and FIG. 9C is an AUC matrix pattern of gastric cancer in the WBC / BASO channel. In both cases, feature points satisfying AUC ≧ 0.8 were detected. When a two-dimensional scattergram of a normal person was synthesized with these AUC matrix patterns, the tendency of the nuclei of lymphocytes to shift in a smaller direction was recognized as a major feature. Early gastric cancer, gastric cancer, gastric body cancer, and esophageal cancer showed similar AUC matrix patterns.

  FIG. 10 shows the diagnostic characteristics of gastric cancer at 14-26 coordinates of the DIFF channel. The 14-26 coordinates of the DIFF channel are feature points indicating the best AUC value in the 32 × 32 two-dimensional scattergram matrix. Regarding this coordinate, the confidence interval of the ROC curve was obtained using the bootstrap ROC. The average AUC was 0.79 and the optimum cutoff value was 27.5. That is, when the number of cells appearing on the 14-26 coordinates of the DIFF channel exceeded 27.5, gastric cancer was detectable with the performance of 62% sensitivity and 90% specificity.

In the present invention, a calculation method using the likelihood matrix distance is applied for the purpose of diagnosing individual specimens. The test likelihood of each specimen and the positive likelihood ratio for the disease name to be evaluated were calculated corresponding to the two-dimensional scattergram, and the sum of them was used as the evaluation value.
FIG. 6 shows the results of diagnostic characteristics of early gastric cancer using the likelihood matrix distance in the WBC / BASO channel. In the case of an average AUC of 0.92 and an optimum cut-off, the sensitivity was 86% and the specificity was 85%, and good diagnostic characteristics were obtained. Even when evaluated at the bottom of the 95% confidence interval, results were obtained that could be used as a test with a diagnostic ability exceeding 80%.

  In addition, cases up to October 2012 were added to evaluate the diagnostic characteristics of gastric cancer. There were 312 positive groups in the cases used. FIG. 11 shows the diagnostic characteristics of gastric cancer using the likelihood matrix distance based on the DIFF scattergram, and FIG. 12 shows the distribution of the likelihood matrix distance between the gastric cancer patient and other disease group patients. Also in this case, as a result of the ROC analysis, AUC was 0.989, which was a very good result.

  We also evaluated the diagnostic properties of prostate cancer. FIG. 13 is a diagram showing an AUC matrix pattern of prostate cancer in the WBC / BASO channel. In the figure, the X axis indicates the side scattered light intensity, and the Y axis indicates the forward scattered light intensity. The AUC at each position of X = 4, Y = 25 and X = 10, Y = 21 was 0.72, which was the best. Further, as a result of ROC analysis of the likelihood matrix distance, AUC was 0.85, which showed a performance significantly exceeding the examination diagnostic ability of PSA.

  FIG. 14 is a diagram showing an AUC matrix pattern of prostate cancer in the DIFF channel. High diagnostic characteristics were shown in the DIFF channel as well as the WBC / BASO channel. FIG. 15 is a diagram showing an AUC matrix pattern of prostatic hypertrophy in the DIFF channel. As can be seen from a comparison between FIG. 14 and FIG. 15, the AUC matrix pattern of prostate cancer in the DIFF channel gave different results from the pattern of the AUC matrix of prostatic hypertrophy.

(Other embodiments)
In the above-described embodiment, the configuration for generating diagnosis support information for early gastric cancer, gastric body cancer, and diabetes has been described. However, the present invention is not limited to this. It can also be configured to generate diagnostic support information for cancer diseases such as gastric cancer, prostate cancer, bile duct cancer, colon cancer using DIFF scattergram or WBC / BASO scattergram, or using RET scattergram It can also be set as the structure which produces | generates the diagnostic assistance information for diseases, such as acute pancreatitis and liver cirrhosis.

  In the above-described embodiment, the diagnosis support apparatus 1 provided separately from the blood cell analyzer 7 acquires scattergram data created by the blood cell analyzer 7 and uses the scattergram data to provide diagnosis support. Although the configuration for generating information has been described, the present invention is not limited to this. The blood cell analyzer may be configured to generate diagnosis support information using a scattergram created by measuring a blood sample. Moreover, although the structure which provides the ROC database 21 in the hard disk 2d of the diagnosis support apparatus 1 was described, it is not limited to this. For example, an ROC database server connected to the diagnosis support apparatus so as to be able to perform data communication can be constructed, and the diagnosis support apparatus can download a necessary ROC curve from the ROC database server.

  The diagnosis support apparatus according to the present invention is useful as a diagnosis support apparatus that provides diagnosis support information based on blood cell distribution information of a blood sample.

DESCRIPTION OF SYMBOLS 1 Diagnosis support apparatus 2 Main-body part 2a CPU
2c RAM
2d hard disk 2g communication interface 21 ROC database 3 display unit 5 computer 6a diagnosis support program 7 blood cell analyzer

Claims (5)

A diagnostic support apparatus that generates diagnostic support information based on a blood cell frequency distribution map in a two-dimensional coordinate space obtained by optical measurement of a blood sample, in a two-dimensional coordinate space,
A blood cell frequency distribution map data acquisition unit for acquiring blood cell frequency distribution map data relating to a specific subject;
A storage unit for storing the ROC curve information indicating the sensitivity and specificity of the relationship for a given disease to power in the blood cell frequency distribution diagram for each position on the two-dimensional coordinate space,
Based on the blood cell frequency distribution data acquired by the blood cell frequency distribution data acquisition unit and the ROC curve information stored in the storage unit, the sensitivity and the specificity corresponding to the frequency in the blood cell frequency distribution graph the positive likelihood ratio specified calculated for each position of the 2-dimensional coordinate space, and calculates a likelihood matrix distance is the total sum of the calculated positive likelihood ratio by the diagnosis based on the likelihood matrix distance An arithmetic unit for generating support information;
An output unit for outputting the diagnosis support information generated by the arithmetic unit;
Comprising
Diagnosis support device. The calculation unit is configured to generate diagnosis support information indicating that the specific patient is positive when the likelihood matrix distance exceeds a predetermined threshold.
The diagnosis support apparatus according to claim 1. The predetermined threshold is obtained by performing ROC analysis on a plurality of the likelihood matrix distances obtained for a plurality of patients.
The diagnosis support apparatus according to claim 2. The predetermined disease is cancer;
The blood cell frequency distribution diagram is a white blood cell frequency distribution diagram,
The diagnosis support apparatus according to any one of claims 1 to 3 . The predetermined disease is diabetes, acute pancreatitis, or cirrhosis;
The blood cell frequency distribution diagram is a reticulocyte frequency distribution diagram,
The diagnosis support apparatus according to any one of claims 1 to 3 .