Hemoglobin type identification method, learned model generation method, learned model, hemoglobin analysis system, and program

Abstract

To provide a hemoglobin type identification method, learned model generation method, learned model, hemoglobin analysis system and program that can identify a type of hemoglobin in a blood sample even when using separation data in which a separation time of separation processing to a blood sample is short.SOLUTION: A hemoglobin type identification method is configured to: a prepare a learned model created by using a data set including learning-purpose frequency analysis data created by analyzing a frequency of separation data created by separation processing to a learning-purpose blood sample having a type of hemoglobin known, and the known type of the hemoglobin as a learning-purpose data set of machine learning; analyze a frequency of separation data created by separation processing to a learning-purpose blood sample having the type of hemoglobin unknown, and prepare inspection-purpose frequency analysis data; an input the inspection-purpose frequency analysis data to the learned model, and thereby identify a type of the hemoglobin included in the inspection-purpose blood sample.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a method for discriminating hemoglobin types, a method for generating a trained model, a trained model, a hemoglobin analysis system, and a program. [Background technology]

[0002] Conventionally, methods have been proposed for measuring glycohemoglobin (eg, HbA1a, HbA1b, HbA1c, etc.) and variant hemoglobin (eg, HbF, HbS, etc.) in a measurement sample such as blood by liquid chromatography (see, for example, Patent Documents 1, 2, 3, and 4). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6856434 [Patent Document 2] Patent No. 5948727 [Patent Document 3] Patent No. 6492856 [Patent Document 4] JP 2017-203677 A Summary of the Invention [Problem to be solved by the invention]

[0004] In the separation and fractionation methods described in Prior Art Documents 1 to 4, the separation time is extended in order to identify the type of hemoglobin in a blood sample, and the determination is based on the peak pattern at the specific time, so that the measurement takes a long time.

[0005] The present disclosure aims to provide a method for discriminating the type of hemoglobin that can identify the type of hemoglobin in a blood sample with higher discrimination accuracy for the blood sample, a method for generating a trained model, a trained model, a hemoglobin analysis system, and a program. [Means for solving the problem]

[0006] One aspect of the method for discriminating hemoglobin types disclosed herein includes preparing a trained model by using a dataset including training frequency analysis data generated by frequency analyzing separation data generated by a separation process on a training blood sample whose hemoglobin type is known and the known hemoglobin type as a training dataset for machine learning, performing frequency analysis on separation data generated by a separation process on a test blood sample whose hemoglobin type is unknown, preparing test frequency analysis data, and inputting the test frequency analysis data into the trained model to discriminate the type of hemoglobin contained in the test blood sample.

[0007] According to this, in one aspect, there is provided a method for determining the type of hemoglobin that can identify the type of hemoglobin in a blood sample even when separation data in which the separation time of a separation process for a blood sample is short is used. Effect of the Invention

[0008] According to the present disclosure, there is provided a method for determining the type of hemoglobin capable of identifying the type of hemoglobin in a blood sample with higher accuracy. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram of an HPLC device. [Diagram 2] FIG. 1 is a configuration diagram of a control system of an HPLC device, which is a trained model generating device. [Diagram 3] 1 is a flowchart illustrating a method for generating a trained model according to an embodiment of the present disclosure. [Figure 4] FIG. 1 shows data used in machine learning to discriminate hemoglobin types according to the present disclosure, and is a chromatogram of a blood sample measured by an HPLC device. [Diagram 5]5 is a diagram showing data used in machine learning to discriminate hemoglobin types according to the present disclosure, following FIG. 4, and is a diagram showing a chromatogram resulting from standardization processing of the chromatogram in FIG. [Figure 6] 6 is a diagram showing data used in machine learning to discriminate hemoglobin types according to the present disclosure, following FIG. 5, and is a diagram showing a wavelet-transformed chromatogram resulting from the standardization process of FIG. 5. [Figure 7] FIG. 1 is a block diagram of a control system of an HPLC device which is a hemoglobin discrimination device. [Figure 8] 1 is a flowchart illustrating a method for determining the type of hemoglobin contained in an unknown sample using a trained model according to an embodiment of the present disclosure. [Figure 9] FIG. 1 is a diagram illustrating the peak shapes when a blood sample is measured using an HPLC system at different separation times, where (A) is a chromatogram when hemoglobin E is measured, and (B) is a chromatogram when abnormal hemoglobin A is measured. [Figure 10] 10A and 10B are diagrams illustrating the peak shapes when blood samples are measured at different separation times using an HPLC apparatus different from the HPLC apparatus used in the measurement in FIG. 9, in which (A) is a chromatogram when hemoglobin E is measured, and (B) is a chromatogram when abnormal hemoglobin A is measured. [Figure 11] FIG. 1 illustrates the peak shapes when the same blood sample is measured using different HPLC instruments, where (A) is a chromatogram when the measurement time is long, and (B) is a chromatogram when the measurement time is short. [Figure 12] 5A and 5B show data used in machine learning to discriminate hemoglobin types according to another embodiment of the present disclosure, where (A) is an amplitude spectrum obtained by Fourier transforming the chromatogram in FIG. 4, and (B) is a power spectrum obtained by Fourier transforming the chromatogram in FIG. 4. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, one embodiment of the present disclosure will be described.

[0011] (Configuration of learning model generation device) 1 shows a schematic configuration diagram of a High Performance Liquid Chromatography (HPLC) device 10 as an example of the trained model generation device in this embodiment. The trained model generation device is an example of a hemoglobin analysis system.

[0012] The HPLC device 10 is configured to automatically measure the concentrations of glycated hemoglobin (e.g., HbA1a, HbA1b, HbA1c, etc.) and variant Hb (e.g., HbF, HbS, etc.) in whole blood by setting a blood collection tube 18, and has a sample preparation unit 12, an analysis unit 58, and a photometry unit 16.

[0013] The HPLC device 10 includes a plurality of eluent bottles 22A, 22B, 22C, 22D, 22E (five in FIG. 1), etc. The eluent bottles 22A to 22E respectively hold eluents A to E to be supplied to an analytical column 60, which will be described later. The eluents differ in composition, component ratio, pH, osmotic pressure, etc., depending on the application.

[0014] The blood collection tubes 18 are stored, for example, in a rack (not shown) and are configured to move to a position where they can be sampled by a nozzle 12N in the sample preparation unit 12, which will be described later.

[0015] The sample preparation unit 12 is for preparing a sample to be introduced into the analytical column 60 from blood collected from a blood collection tube 18. The sample preparation unit 12 has a nozzle 12N and a dilution tank .

[0016] The nozzle 12N is for collecting various liquids including the blood sample 20 in the blood collection tube 18, and is capable of aspirating and discharging liquids and is movable in the vertical and horizontal directions. The operation of the nozzle 12N is controlled by the control unit 100, which will be described later.

[0017] The analytical unit 58 controls the adsorption and desorption of components contained in the blood sample, such as hemoglobin, which is a biological component, to the packing material of the analytical column 60, and provides various biological components to the photometric unit 16. The set temperature in the analytical unit 58 is, for example, about 40°C. The analytical column 60 holds a packing material for selectively adsorbing components contained in the sample. For example, a methacrylic acid-methacrylic acid ester copolymer is used as the packing material.

[0018] In addition to the analytical column 60 , the analytical unit 58 includes a manifold 61 , a liquid delivery pump 62 , and an injection valve 63 .

[0019] The manifold 61 is for selectively supplying an eluent from a specific eluent bottle among the plurality of eluent bottles 22A to 22E to the analytical column 60. The manifold 61 is connected to the eluent bottles 22A, 22B, 22C, 22D, and 22E via pipes 80A to 80E, respectively, and is connected to an injection valve 63 via a pipe 84.

[0020] The liquid sending pump 62 is provided to provide power for moving the eluent to the injection valve 63 , and is provided midway through the pipe 84 .

[0021] The injection valve 63 is capable of collecting a certain amount of sample to be introduced and introducing the sample to the analytical column 60, and is provided with a plurality of introduction ports and discharge ports (not shown). The injection valve 63 is connected to an injection loop 64. The injection loop 64 is capable of holding a certain amount of liquid (e.g., several μL). By appropriately switching the injection valve 63, it is possible to select a state in which the injection loop 64 communicates with the dilution tank 24 so that the sample to be introduced is supplied from the dilution tank 24 to the injection loop 64, and a state in which the injection loop 64 communicates with the analytical column 60 via the prefilter PF and the piping 85 so that the sample to be introduced is introduced from the injection loop 64 to the analytical column 60. For example, a six-way valve can be used as the injection valve 63. The prefilter PF is a filter for filtering the sample and the eluent.

[0022] The photometry unit 16 is for optically detecting the hemoglobin contained in the sample supplied from the analytical column 60, and is connected via piping 87 to a waste liquid tank 88 for discharging the sample supplied from the analytical column 60.

[0023] Here, when a sample is introduced into the analytical column 60, the amount of each component contained in the sample that is supplied to the photometric unit 16 varies over time because the time that the analytical column 60 retains each component differs for each component. In other words, the components contained in the sample are separated by the analytical column 60.

[0024] The photometric unit 16 is controlled by the control unit 100 described later to continuously measure the absorbance of components contained in the sample supplied from the analytical column 60 and transmits the absorbance measurement results to the control unit 100.

[0025] Therefore, separation data of the sample is obtained by recording the absorbance measurement results transmitted from the photometric unit 16. That is, the analysis unit 58 is a member that performs the separation process in this embodiment, and the analysis unit 58 and the photometric unit 16 are an example of a separation section in this embodiment.

[0026] Fig. 2 shows a block diagram of a control system of the HPLC apparatus 10. As shown in Fig. 2, the HPLC apparatus 10 includes a control unit 100. The control unit 100 includes a central processing unit (CPU) 100A, a read only memory (ROM) 100B, a random access memory (RAM) 100C, a non-volatile memory 100D, and an input / output interface (I / F) 100E, all of which are connected via a bus 100F. The HPLC apparatus 10 also includes an operation unit (not shown) that accepts input from an operator, and a display unit 26.

[0027] The sample preparation unit 12, the analysis unit 58, the photometry unit 16, and the display unit 26 are connected to the input / output interface 100E.

[0028] In this embodiment, the display unit 26 is, for example, a liquid crystal display, and is controlled by the CPU 100A to display information from the control section 100 to the operator.

[0029] In the present embodiment, as an example, a basic program (not shown) for controlling the operation of the HPLC device 10 is recorded in advance in the non-volatile memory 100D of the CPU 100A. In the present embodiment, as an example, a first program 120A for generating a trained model 120C is further stored in the non-volatile memory 100D. The CPU 100A reads and executes various programs stored in the non-volatile memory 100D. Alternatively, the first program 120A may be recorded in a recording medium such as a CD-ROM, and the CPU 100A may execute the program by reading it using a CD-ROM drive or the like.

[0030] Furthermore, CPU 100A executes a basic program to control manifold 61 and injection valve 63, and changes the type of eluent introduced into analytical column 60 or the flow rate of the eluent per unit time. As a result, CPU 100A determines the flow rate of components in the sample flowing out of analytical column 60 per unit time, and sets the measurement time of the components contained in the sample (the time for measuring absorbance by photometry unit 16).

[0031] Next, a procedure for generating a trained model used in the method for determining the type of hemoglobin according to this embodiment will be described with reference to FIGS. 3 to 6 as appropriate.

[0032] (Procedure for generating a trained model according to the method of this embodiment) In the trained model generating device of this embodiment, the control unit 100 executes various programs according to this embodiment to generate a trained model for discriminating the type of hemoglobin contained in a sample. Specifically, when the CPU 100A reads the first program 120A, the CPU 100A executes the following processes.

[0033] (Procedure that the first program 120A causes the CPU 100A to execute) In this embodiment, the first program 120A causes the CPU 100A to execute the processing shown in FIG.

[0034] In step S12, the CPU 100A stores in the RAM 100C the absorbances corresponding to the components of a sample whose hemoglobin type is known (hereinafter referred to as a known sample) measured by the photometric unit 16, and creates a chromatogram as shown in Fig. 4. Then, the CPU 100A stores the created chromatogram in the RAM 100C. After that, the CPU 100A proceeds to step S14.

[0035] In step S14, the CPU 100A performs a standardization process on the chromatogram stored in the RAM 100C to generate standardized data as shown in Fig. 5. The CPU 100A then stores the generated standardized data in the RAM 100C. After that, the CPU 100A proceeds to step S16.

[0036] The specific conversion method for the standardization process is as follows and shown in Equation 1. 1. Standardization is performed on each data point of the chromatogram (scaling is performed by subtracting the average value from the measured value for each measured data point and then dividing the result by the standard deviation). 2. Add 0.001 plus the absolute value of the minimum value in the standardized data to each standardized data point. 3. For each data point, take the logarithm to the base e (Napier's constant).

[0037]

number

[0038] In the above formula 1, x' i is the value at each data point i after standardization, and x i is the value at each data point i in the chromatogram. μ is the average value of each data point in the chromatogram, and σ is the standard deviation of the values ​​of each data point in the chromatogram. Min is a function that indicates the minimum value among multiple values. In addition, ln is a logarithmic function with Napier's constant as the base.

[0039] It should be noted that the above-described conversion method and parameters are merely an example of the standardization process according to this embodiment, and the standardization process according to this embodiment is not limited to the above-described conversion method and parameters.

[0040] In step S16, CPU 100A performs wavelet analysis on the standardized data stored in RAM 100C to generate wavelet analysis data as shown in Fig. 6. CPU 100A then records the generated wavelet analysis data in ROM 100B. After that, CPU 100A proceeds to step S18.

[0041] In step S18, CPU 100A creates a learning data set in which the types of hemoglobin contained in the sample from which the wavelet analysis data was created correspond to the wavelet analysis data. Then, CPU 100A records the created learning data set in ROM 100B. After that, CPU 100A proceeds to step S20.

[0042] In step S20, CPU 100A checks the number of recorded learning data sets. If the number of learning data sets is less than the predetermined number, an affirmative decision is made in step S20, and the process proceeds to step S12. On the other hand, if the number of learning data sets has reached the predetermined number, a negative decision is made in step S20, and the process proceeds to step S22.

[0043] In step S22, CPU 100A constructs a convolutional neural network and inputs a plurality of learning data sets recorded in ROM 100B. Then, CPU 100A learns the type and amount of hemoglobin contained in the sample and the characteristics of the wavelet analysis data based on the plurality of learning data sets input to the convolutional neural network. After that, CPU 100A proceeds to step S24.

[0044] In step S24, CPU 100A records the learned parameters in non-volatile memory 100D. After that, CPU 100A ends the processing by first program 120A.

[0045] The trained parameters are combined with a subprogram (not shown) and are part of a trained model 120C described later. In other words, the first program 120A causes the CPU 100A to execute a process of generating the trained model 120C.

[0046] As described above, the first program 120A performs wavelet analysis on the standardized data. Thus, in this embodiment, the CPU 100A that executes the first program 120A is an example of a frequency analysis unit.

[0047] Next, a procedure for determining the type of hemoglobin contained in a sample whose type of hemoglobin is unknown (hereinafter referred to as an unknown sample) using the trained model according to this embodiment will be described with reference to Figures 1, 7, and 8. Note that in the HPLC device 11 which is the hemoglobin discrimination device according to this embodiment, the configuration of the HPLC device 10 is the same as that of the HPLC device 10 except for the control unit 100.

[0048] (Configuration of hemoglobin discrimination device) 7 is a configuration diagram of the control unit 101 of the HPLC device 11, which is a hemoglobin discrimination device according to this embodiment. The hemoglobin discrimination device is an example of a hemoglobin analysis system.

[0049] The HPLC device 11, which is a hemoglobin discrimination device, is different from the HPLC device 10, which is the trained model generation device described above, in the configuration of the programs held in the non-volatile memory 100D. Specifically, the hemoglobin discrimination device has a trained model 120C and a second program 120B, which discriminates hemoglobin in a blood sample using the trained model 120C, stored in the non-volatile memory.

[0050] The CPU 100A in the hemoglobin discrimination device reads and executes various programs stored in the non-volatile memory 100D. The trained model 120C includes trained parameters generated by the trained model generation device described above. The trained model 120C may be transferred in any manner from the non-volatile memory 100D of the trained model generation device, and is recorded in the non-volatile memory 100D of the hemoglobin discrimination device via a network, for example. The second program 120B may be recorded in a recording medium such as a CD-ROM, and the CPU 100A may execute the program by reading it using a CD-ROM drive or the like.

[0051] Other configurations of the control unit 101 are the same as those of the control unit 100 of the HPLC device 10 of the trained model generation device described above. In the following description, when there is no need to distinguish between the HPLC device 10 of the trained model generation device and the HPLC device 11 which is a hemoglobin discrimination device, they will be referred to as HPLC devices 10 and 11. Similarly, in the following description, when there is no need to distinguish between the control unit 100 of the HPLC device 10 and the control unit 101 of the HPLC device 11, they will be referred to as control units 100 and 101.

[0052] (Procedure for determining hemoglobin type by the method of this embodiment) In the hemoglobin discrimination device of this embodiment, the control unit 101 executes various programs according to this embodiment to discriminate the type of hemoglobin contained in a sample. Specifically, when the CPU 100A reads the second program 120B, it executes the following processes.

[0053] (Processing that the second program 120B causes the CPU 100A to execute) In this embodiment, the second program 120B causes the CPU 100A to execute the processing shown in FIG.

[0054] In step S42, the CPU 100A reads the trained model 120C (trained parameters and subprograms) recorded in the ROM 100B. After that, the CPU 100A proceeds to step S44.

[0055] In step S44, CPU 100A stores in RAM 100C the absorbances measured by photometric unit 16 and corresponding to the components of the unknown sample, and creates a chromatogram as shown in Fig. 4. CPU 100A then stores the created chromatogram in RAM 100C. After that, CPU 100A proceeds to step S46.

[0056] In step S46, CPU 100A performs standardization processing on the chromatogram stored in RAM 100C to generate standardized data as shown in Fig. 5. CPU 100A then stores the generated standardized data in RAM 100C. After that, CPU 100A proceeds to step S48.

[0057] In step S48, CPU 100A performs wavelet analysis on the standardized data stored in RAM 100C to generate wavelet analysis data as shown in Fig. 6. Then, CPU 100A records the generated wavelet analysis data in ROM 100B. After that, CPU 100A ends the processing by second program 120B.

[0058] The hemoglobin discrimination device according to this embodiment may further execute the following process.

[0059] For example, in step S48, when the type of hemoglobin contained in the unknown sample has not been determined, CPU 100A may cause display unit 26 to display that the type of hemoglobin has not been determined.

[0060] On the other hand, when the type of hemoglobin contained in the unknown sample is determined in step S48, another process may be performed. For example, CPU 100A may calculate the amount of hemoglobin contained in the unknown sample for each hemoglobin type from the value of the peak corresponding to the target hemoglobin type in the chromatogram of the unknown sample, and display the calculated value on display unit 26.

[0061] As described above, the second program 120B performs wavelet analysis on each of the standardized data. Thus, in this embodiment, the CPU 100A that executes the second program 120B is an example of a frequency analysis unit.

[0062] Next, the results of discriminating hemoglobin contained in an unknown sample using the method for generating the trained model 120C and the method for discriminating the type of hemoglobin according to this embodiment will be described as Example 1 and Example 2. Note that in each example, the "unknown sample" refers to a sample used for testing discrimination in the trained model 120C, and is a sample that has actually had the type of hemoglobin contained in the sample discriminated and that was not used to generate the trained model 120C.

[0063] In each example, a diabetes testing device HA-8180V or HA-8190V manufactured by Arkray, Inc. was used. In each measuring device, the measurement time was set to 90 seconds for the HA-8180V and 24 seconds for the HA-8190V. The sampling frequency was set to 2048 Hz (sampling interval 488.28125 microseconds).

[0064] In each example, blood samples containing hemoglobin S, hemoglobin C, hemoglobin D, or hemoglobin E were prepared, as well as blood samples in which hemoglobin S, hemoglobin C, hemoglobin D, and hemoglobin E were not observed. Hereinafter, samples in which hemoglobin S, hemoglobin C, hemoglobin D, and hemoglobin E were not observed will be referred to as "healthy subject samples."

[0065] Example 1 In Example 1, a HA-8180V was used as a measurement device to create a trained model 120C and to identify the type of hemoglobin contained in an unknown sample using the trained model 120C.

[0066] In this embodiment, in the learning stage, learning was performed using the procedure shown in Fig. 3. More specifically, in this embodiment, a chromatogram of a known sample was created, and the created chromatogram was standardized to create wavelet analysis data. Then, a learning data set including the wavelet analysis data was created, and learning was performed using a convolutional neural network to create a trained model 120C.

[0067] In this embodiment, in the estimation stage of the unknown sample, the determination was performed using the procedure shown in Fig. 8. More specifically, in this embodiment, a chromatogram of the unknown sample was created, and the created chromatogram was standardized to create wavelet analysis data. Then, the type of hemoglobin contained in the unknown sample was determined based on the trained model 120C for the wavelet analysis data.

[0068] Table 1 shows the number of training data (the number of training data sets, that is, the number of known samples) used in the training of this embodiment, the number of test data (the number of samples used in the discrimination test, that is, the number of unknown samples), the number of correct answers, and the accuracy rate.

[0069] [Table 1]

[0070] Example 2 In Example 2, a HA-8190V was used as a measurement device to create a trained model 120C and to identify the type of hemoglobin contained in an unknown sample using the trained model 120C.

[0071] In this embodiment, in the learning stage, learning was performed using the procedure shown in Fig. 3. More specifically, in this embodiment, a chromatogram of a known sample was created, and the created chromatogram was standardized to create wavelet analysis data. Then, a learning data set including the wavelet analysis data was created, and learning was performed using a convolutional neural network to create a learned model.

[0072] In this embodiment, in the estimation stage of the unknown sample, the determination was performed using the procedure shown in Fig. 8. More specifically, in this embodiment, a chromatogram of the unknown sample was created, and the created chromatogram was standardized to create wavelet analysis data. Then, the type of hemoglobin contained in the unknown sample was determined based on the trained model for the wavelet analysis data.

[0073] Table 2 shows the number of learning data and the number of test data used in the learning according to this embodiment, the number of correct answers, and the correct answer rate.

[0074] [Table 2]

[0075] (summary) As described above, the HPLC apparatuses 10 and 11 of the present embodiment have demonstrated the following.

[0076] As shown in Tables 1 and 2, it was shown that the type of hemoglobin can be discriminated using the method for creating a trained model and the method for discriminating the type of hemoglobin in Examples 1 and 2. In addition, this result suggests that the accuracy of identifying the type of hemoglobin in a blood sample is improved based on the trained model created using wavelet analysis data. In other words, it was shown that the method for creating a trained model and the method for discriminating the type of hemoglobin in a blood sample can be identified even when the separation time is short, using the method for creating a trained model and the method for discriminating the type of hemoglobin in this embodiment.

[0077] (Reference example: How to distinguish hemoglobin types) Incidentally, one method for identifying the type of hemoglobin contained in an unknown sample is to identify it from the shape of the peaks appearing in a chromatogram (the positional relationship between the peaks appearing in the chromatogram).

[0078] Figure 9(A) is a chromatogram obtained for a blood sample containing hemoglobin E using the HA-8180V as the measurement device, with the flow rate set to a measurement time of 200 seconds. As shown in Figure 9(A), two peaks were confirmed in this sample before and after the largest peak, which was the hemoglobin A0 peak.

[0079] Similarly, Fig. 9(B) is a chromatogram obtained for a blood sample containing abnormal hemoglobin A under conditions where the measurement time was set to 200 seconds. As shown in Fig. 9(B), this chromatogram shows that peak A appeared around 30 seconds as a fractionated region, but peak B did not appear, unlike Fig. 9(A). From this, it is considered that, under these measurement conditions, it is possible to distinguish between hemoglobin E and abnormal hemoglobin A by observing the presence or absence of peak B in a sample having peak A. Thus, by setting the measurement time to 200 seconds, hemoglobin can be distinguished, but this measurement time is longer than the time set in the above-mentioned Example 1.

[0080] Here, Fig. 10(A) is a chromatogram obtained using the same sample and the same device as Fig. 9(A) under the condition that the measurement time was set to 90 seconds as in Example 1. As shown in Fig. 10(A), this chromatogram shows that peak A appears around 25 seconds as a fractionated region, but peak B does not appear.

[0081] Similarly, Fig. 10(B) is a chromatogram obtained using the same sample as Fig. 9(B) but with a different measuring device, the HA-8190V, and with a measurement time set to 90 seconds. As shown in Fig. 10(B), this chromatogram shows that peak A appears around 25 seconds as a fractionated region, but peak B does not appear.

[0082] In this way, peak B that appears in FIG. 9(A) is not fractionated in FIG. 10(A), so if the measurement time is shortened, the type of hemoglobin contained in the sample is likely to be erroneously identified.

[0083] For example, Fig. 11(A) is a chromatogram obtained under conditions in which the measurement time was set to 90 seconds for a blood sample containing hemoglobin S. As shown in Fig. 11(A), in the sample containing hemoglobin S, peak C appeared at around 28 seconds, following the largest peak (the peak of hemoglobin A0) that appeared at around 20 seconds.

[0084] On the other hand, Fig. 11(B) is a chromatogram obtained using the same sample and the same device as Fig. 11(A) under the same conditions with the measurement time set to 24 seconds as in Example 2. As shown in Fig. 11(B), this chromatogram shows that peak C did not appear after the largest peak that appeared around 18 seconds.

[0085] Thus, even when the same HPLC apparatuses 10 and 11 are used, if the measurement time is shortened, the type of hemoglobin contained in the sample is likely to be erroneously determined.

[0086] On the other hand, it has been shown that the HA-8180V, which is an example of the HPLC devices 10 and 11 in this embodiment, can identify the type of hemoglobin in a blood sample in a measurement time of 90 seconds. Also, it has been shown that the HA-8190V, which is an example of the HPLC devices 10 and 11 in this embodiment, can identify the type of hemoglobin in a blood sample in a measurement time of 24 seconds. Therefore, it has been shown that by creating the trained model 120C using wavelet analysis data, the type of hemoglobin in a blood sample can be identified even when separation data in which the separation time of the separation process for the blood sample is short is used.

[0087] In addition, even if the separation time of the separation process for the blood sample is long, the peak value of the chromatogram may not be clear, and the accuracy of the discrimination of the hemoglobin type based on the chromatogram may be poor. On the other hand, in the method for generating a trained model and the method for discriminating the hemoglobin type according to the present embodiment, since frequency analysis is performed on the components contained in the blood sample, even if the peaks overlap and the peaks appearing in the chromatogram are somewhat unclear, the overlapped peaks are separated as the frequency analysis result. Therefore, according to the method for generating a trained model and the method for discriminating the hemoglobin type according to the present embodiment, it is expected that the effect of improving the discrimination accuracy of the hemoglobin type can be expected even in a chromatogram with an unclear peak value.

[0088] In addition, for hemoglobin F, hemoglobin A2 (HbA2), and other variant hemoglobin types not implemented in the above-mentioned Examples 1 and 2, a trained model can be generated and a method for discriminating the type of hemoglobin can be applied in a manner similar to that of this embodiment.

[0089] (Other embodiments) In the above description, the control units 100 and 101 are incorporated in the HPLC devices 10 and 11, but this is not limiting. For example, the main components of the control units 100 and 101 may be arranged on a network, and the HPLC devices 10 and 11 may operate as the control units 100 and 101 by communicating with the components arranged on the network. For example, the first program 120A and the second program 120B may be acquired by the control units 100 and 101 of the HPLC device 10 as appropriate by downloading them from a server arranged on the network. For example, the HPLC devices 10 and 11 may communicate with a server arranged on the network, and the HPLC devices 10 and 11 may transmit a chromatogram to the server, and then the server may transmit frequency analysis data to the HPLC devices 10 and 11, thereby operating the HPLC devices 10 and 11 as frequency analysis units.

[0090] Furthermore, the first program 120A, the second program 120B, and the trained model 120C described above are executed by the control units 100 and 101, but this is not limiting and they may be executed by a server located on a network. For example, the HPLC devices 10 and 11 may be configured to transmit a chromatogram of an unknown sample to a server located on a network and receive the amount of hemoglobin contained in the unknown sample from the server. Any of these configurations is included in the hemoglobin analysis system of this embodiment.

[0091] In addition, in the hemoglobin type discrimination method according to the present disclosure, as long as the trained model 120C is prepared, there is no limitation on the method of preparing the trained model 120C. In other words, as long as the trained model 120C is prepared, there is no limitation on the equipment that creates the chromatogram, the equipment that performs the frequency analysis, the equipment that generates the trained model 120C, and the equipment that performs the discrimination.

[0092] More specifically, the device for creating a chromatogram may be a device separate from the HPLC devices 10 and 11, which has only the function of creating a chromatogram from a blood sample. The device for performing frequency analysis may be a device separate from the HPLC devices 10 and 11, which has only the function of performing wavelet analysis from a chromatogram. In the above description, the learning model generation device generates the learned model 120C, and the hemoglobin discrimination device discriminates the type of hemoglobin based on the learned model 120C, but these devices may be the same device. These devices may all be incorporated into the same HPLC devices 10 and 11, or some of the devices may be incorporated into a common device, or each may be a separate device. For this reason, according to the technology of this embodiment, if a learned model 120C created for each model of the HPLC device 11 is prepared, it is not necessary to create the learned model 120C in each individual HPLC device 10.

[0093] In the above description, the control units 100 and 101 execute various programs to realize the processes according to the embodiments by using a software configuration, but the present invention is not limited to this. For example, the processes according to the above embodiments may be realized by using a hardware configuration or a combination of a hardware configuration and a software configuration.

[0094] In the above description, wavelet analysis data is used as frequency analysis data, but the technology according to the present disclosure is not limited to this.

[0095] For example, in the frequency analysis, by performing a fast Fourier transform instead of the wavelet analysis, amplitude data as shown in Fig. 12(A) or power spectrum data as shown in Fig. 12(B) may be used. Also, instead of the wavelet analysis, data obtained by performing a short-time Fourier transform may be used.

[0096] In the above description, the hemoglobin concentration and wavelet analysis data of the known sample are used in the learning data set, but the learning data set may include a chromatogram. In this case, the chromatogram of the unknown sample is included in the input data to the trained model 120C in the discrimination procedure.

[0097] In the above description, the HPLC devices 10 and 11 are used as an example of a hemoglobin analysis system, but the technology according to the present disclosure is not limited to this. In other words, as long as chromatograms of known samples and unknown samples can be obtained, an electrophoretic analysis device may be used instead of the HPLC devices 10 and 11. In the electrophoretic analysis device, standardization processing can be optionally performed based on the created pherogram, wavelet analysis can be performed, and wavelet analysis data can be created. According to the technology according to the present disclosure, even in this case, the trained model 120C can be generated and the type of hemoglobin can be determined.

[0098] In the above description, the trained model 120C generating method constructs a convolutional neural network and then generates the trained model 120C, but the technology according to the present disclosure is not limited to this. That is, if the trained model 120C can be obtained, for example, a recurrent neural network may be constructed and trained. In this case, the trained model 120C can also be generated.

[0099] Furthermore, preferred aspects of the present disclosure will be described below.

[0100] (Appendix 1) A trained model is prepared by using a data set including learning frequency analysis data generated by frequency analyzing separation data generated by a separation process for a training blood sample having a known hemoglobin type and the known hemoglobin type as a training data set for machine learning; preparing frequency analysis data for testing by subjecting separated data generated by a separation process for a blood sample for testing, the type of hemoglobin being unknown, to frequency analysis; inputting the test frequency analysis data into a trained model to identify the type of hemoglobin contained in the test blood sample; How to distinguish between different types of hemoglobin.

[0101] (Appendix 2) the separated data for the learning frequency analysis data and the separated data for the inspection frequency analysis data are chromatograms; A method for distinguishing types of hemoglobin as described in Appendix 1.

[0102] (Appendix 3) The learning frequency analysis data and the inspection frequency analysis data are generated by performing frequency analysis on the separated data for the learning frequency analysis data and the separated data for the inspection frequency analysis data, which have been subjected to standardization processing, respectively. A method for determining types of hemoglobin according to appendix 1 or 2.

[0103] (Appendix 4) the learning frequency analysis data and the inspection frequency analysis data are generated by performing wavelet analysis; A method for determining the type of hemoglobin according to any one of claims 1 to 3.

[0104] (Appendix 5) The machine learning includes a convolutional neural network. A method for determining the type of hemoglobin according to any one of claims 1 to 4.

[0105] (Appendix 6) The hemoglobin is hemoglobin S, hemoglobin C, hemoglobin D, or hemoglobin E. A method for determining the type of hemoglobin according to any one of claims 1 to 5.

[0106] (Appendix 7) generating learned parameters by using a data set including learning frequency analysis data generated by frequency analyzing separation data generated by a separation process for a learning blood sample having a known hemoglobin type and the known hemoglobin type as a learning data set for machine learning; How to generate a trained model.

[0107] (Appendix 8) The separation data is a chromatogram. A method for generating a trained model as described in Appendix 7.

[0108] (Appendix 9) The frequency analysis for generating the learning frequency analysis data is performed on the separated data that has been standardized. A method for generating a trained model according to appendix 7 or appendix 8.

[0109] (Appendix 10) The frequency analysis for generating the learning frequency analysis data is wavelet analysis. A method for generating a trained model according to any one of Supplementary Note 7 to Supplementary Note 9.

[0110] (Appendix 11) The machine learning includes a convolutional neural network. A method for generating a trained model according to any one of Supplementary Note 7 to Supplementary Note 10.

[0111] (Appendix 12) The hemoglobin is hemoglobin S, hemoglobin C, hemoglobin D, or hemoglobin E. A method for generating a trained model according to any one of Supplementary Note 7 to Supplementary Note 11.

[0112] (Appendix 13) a trained model that, when frequency analysis data generated by frequency analysis of separation data generated by a separation process for a blood sample is input, outputs a type of hemoglobin contained in the blood sample from the frequency analysis data, The trained model is generated by using training data including training frequency analysis data generated by frequency analysis of separation data generated by a separation process for a training blood sample having a known hemoglobin type, and the known hemoglobin type as a training dataset for machine learning. Trained model.

[0113] (Appendix 14) a separation unit that performs a separation process on a blood sample for testing, the type of hemoglobin being unknown, and generates separation data for testing; a frequency analysis unit that performs frequency analysis on the inspection separated data to generate inspection frequency analysis data; A discrimination unit that discriminates the type of hemoglobin contained in the test blood sample using the trained model described in Appendix 13 measured using the test frequency analysis data; an output unit that outputs the amount of each type of hemoglobin identified by the identification unit contained in the test blood sample; A hemoglobin analysis system comprising:

[0114] (Appendix 15) inputting a trained model generated by using training data including training frequency analysis data generated by frequency analyzing separation data generated by a separation process for a training blood sample having a known hemoglobin type and the known hemoglobin type as a training dataset for machine learning; inputting test frequency analysis data generated by frequency analysis of separation data generated by a separation process for a test blood sample having an unknown hemoglobin type; outputting the type of hemoglobin contained in the test blood sample based on the trained model; A program that causes a computer to carry out processing. [Explanation of symbols]

[0115] 10 HPLC equipment 11 HPLC equipment 12 Sample Preparation Unit 12N nozzle 16 Photometric Units 18 Blood collection tube 20 Blood Samples 22 Eluent bottle 24 Dilution tank 26 Display unit 58 Analysis Units 60 Analytical Columns 61 Manifold 62 Liquid delivery pump 63 Injection valve 64 Injection Loop 80 Piping 83 Piping 84 Piping 85 Piping 86 Piping 87 Piping 88 Waste liquid tank 100 Control section 100A CPU 100B ROM 100C RAM 100D Non-volatile Memory 100E Input / Output Interface 100F Bus 101 Control section

Claims

1. A trained model is prepared by using a data set including learning frequency analysis data generated by frequency analyzing separation data generated by a separation process for a training blood sample having a known hemoglobin type and the known hemoglobin type as a training data set for machine learning; preparing frequency analysis data for testing by subjecting separated data generated by a separation process for a blood sample for testing, the type of hemoglobin being unknown, to frequency analysis; inputting the test frequency analysis data into a trained model to identify the type of hemoglobin contained in the test blood sample; How to distinguish between different types of hemoglobin.

2. the separated data for the learning frequency analysis data and the separated data for the inspection frequency analysis data are chromatograms; The method for determining the type of hemoglobin according to claim 1 .

3. The learning frequency analysis data and the inspection frequency analysis data are generated by performing frequency analysis on the separated data for the learning frequency analysis data and the separated data for the inspection frequency analysis data, which have been subjected to standardization processing, respectively. The method for determining the type of hemoglobin according to claim 1 .

4. the learning frequency analysis data and the inspection frequency analysis data are generated by performing wavelet analysis; The method for determining the type of hemoglobin according to claim 1 .

5. The machine learning includes a convolutional neural network. The method for determining the type of hemoglobin according to claim 1 .

6. The hemoglobin is hemoglobin S, hemoglobin C, hemoglobin D, or hemoglobin E. The method for determining the type of hemoglobin according to claim 1 .

7. generating learned parameters by using a data set including learning frequency analysis data generated by frequency analyzing separation data generated by a separation process for a learning blood sample having a known hemoglobin type and the known hemoglobin type as a learning data set for machine learning; How to generate a trained model.

8. The separation data is a chromatogram. The method for generating a trained model according to claim 7.

9. The frequency analysis for generating the learning frequency analysis data is performed on the separated data that has been standardized. The method for generating a trained model according to claim 7.

10. The frequency analysis for generating the learning frequency analysis data is wavelet analysis. The method for generating a trained model according to claim 7.

11. The machine learning includes a convolutional neural network. The method for generating a trained model according to claim 7.

12. The hemoglobin is hemoglobin S, hemoglobin C, hemoglobin D, or hemoglobin E. The method for generating a trained model according to claim 7.

13. a trained model that, when frequency analysis data generated by frequency analysis of separation data generated by a separation process for a blood sample is input, outputs a type of hemoglobin contained in the blood sample from the frequency analysis data, The trained model is generated by using training data including training frequency analysis data generated by frequency analysis of separation data generated by a separation process for a training blood sample having a known hemoglobin type, and the known hemoglobin type as a training dataset for machine learning. Trained model.

14. a separation unit that performs a separation process on a blood sample for testing, the type of hemoglobin being unknown, and generates separation data for testing; a frequency analysis unit that performs frequency analysis on the inspection separated data to generate inspection frequency analysis data; a discrimination unit that discriminates the type of hemoglobin contained in the test blood sample by the trained model according to claim 13, which is measured using the test frequency analysis data; an output unit that outputs the type of hemoglobin determined by the determination unit; A hemoglobin analysis system comprising:

15. inputting a trained model generated by using training data including training frequency analysis data generated by frequency analyzing separation data generated by a separation process for a training blood sample having a known hemoglobin type and the known hemoglobin type as a training dataset for machine learning; inputting test frequency analysis data generated by frequency analysis of separation data generated by a separation process for a test blood sample having an unknown hemoglobin type; outputting the type of hemoglobin contained in the test blood sample based on the trained model; A program that causes a computer to carry out processing.

Images