Learning data generation method and characteristic prediction method

Abstract

To generate learning data used for machine learning for generating a prediction model that can repeatedly predict the characteristics of an object to be analyzed without performing calibration.SOLUTION: A control unit 10 of an analyzer 1, when predicting a K value that is a value related to the concentration of an object to be analyzed by using a prediction model, performs processing of acquiring a prescribed number of output values representing the characteristics of an object to be analyzed for every interval a from a specific time, on a plurality of objects to be analyzed repeatedly a predetermined number of times, and associates the output value of each of the objects to be analyzed acquired in every repetition, with a reference K value that is a K value obtained from the repetition processing that is executed first, to generate learning data.SELECTED DRAWING: Figure 6

Description

[Technical field]

[0001] The present disclosure relates to a method for creating learning data used to generate a prediction model for predicting characteristics of an object to be analyzed, and a method for predicting the characteristics. [Background technology]

[0002] Patent Document 1 describes a method for measuring the glucose concentration using an enzyme electrode. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2020-507773 Summary of the Invention [Problem to be solved by the invention]

[0004] As the analytical device repeatedly analyzes the property of the analyte, the accuracy of the measurement of the property of the analyte may decrease.

[0005] Therefore, in conventional analytical devices, it is recommended to perform a calibration operation on the predicted value, i.e., a calibration, for example when the characteristics of the object to be analyzed have been analyzed a predetermined number of times, or when a predetermined amount of time has elapsed since the analysis of the characteristics of the object to be analyzed began.

[0006] However, performing calibration in the middle of an analysis operation places a burden on the analyst.

[0007] The present disclosure aims to provide a method for creating learning data to be used in machine learning to generate a predictive model that can repeatedly predict the characteristics of an analyte without calibration, and a method for predicting the characteristics of an analyte. [Means for solving the problem]

[0008] In order to achieve the above object, a method for creating learning data according to one aspect of the present disclosure provides an analysis device including: a reaction cell into which a buffer solution and an analyte are introduced; electrodes provided in the reaction cell that output a response current having a magnitude corresponding to a concentration of the analyte in the buffer solution that comes into contact with the reaction cell when a voltage of a predetermined magnitude is applied; and a control unit that controls the introduction timing of the buffer solution and the analyte into the reaction cell, respectively, and controls to obtain a magnitude of the response current output from the electrode as an output value representing a characteristic of the analyte, the method comprising the steps of: obtaining an output value D t1 and an output value D at a time t2 that is predetermined as a time after the time t1. t2 When a prediction model generated by supervised machine learning is used to predict a K value, which is a value associated with the concentration of the analyte represented by the difference between t1 and an output value D at time t3+ia (i=0 to n-1: n is a predetermined integer of 2 or more) at each predetermined interval a within the period from time t3, which is the start of the period after the time t1 and before the time t2. t3+ia and the output value D at time t2 t2 and the output value D at the time t3+ia for each of the analytes is obtained at least once at first. t3+ia an acquisition step of repeatedly executing a process of acquiring the output value D for each of the analytes acquired initially by the acquisition step up to a predetermined number of times; t1 and an output for each of the analytes. Value D t2 a K value calculation step of calculating the K value for each of the analytes as a reference K value from the difference between the K value and the output value D for each of the analytes obtained each time the acquisition step is performed; t3+ia For each output value D t3+iaand creating a learning data set to be used in supervised machine learning of the prediction model by associating the reference K value of the analyte having the same concentration as the concentration of the analyte introduced into the reaction cell to obtain the reference K value.

[0009] A method for predicting a characteristic of an analyte according to an embodiment of the present disclosure is a model generated by supervised machine learning using the learning data created by the method for creating learning data according to an embodiment of the present disclosure, the prediction model predicting a K value, which is a value related to a concentration of the analyte, is provided by: t3+ia and an input step of inputting the output value D t3+ia and a prediction step of predicting the concentration of the analyte from the predicted value of the K value output by the prediction model to which the above-mentioned value has been input. Effect of the Invention

[0010] According to the present disclosure, it is possible to create learning data to be used in machine learning for generating a predictive model that repeatedly predicts the characteristics of an analyte without performing calibration. [Brief description of the drawings]

[0011] [Figure 1] FIG. 2 is a diagram showing an example of the appearance of an analysis device. [Diagram 2] FIG. 2 is a diagram showing an example of a flow path in an analysis device. [Diagram 3] FIG. 1 is a diagram showing the principle of glucose concentration measurement by a glucose sensor. [Figure 4] FIG. 13 is a diagram showing the relationship between the output value and the glucose concentration. [Diagram 5] FIG. 2 is a diagram showing an example of the configuration of a main part of an electric system of an analyzer including a control unit. [Figure 6] 13 is a flowchart showing an example of the flow of a learning data creation process. [Figure 7] 13 is a flowchart showing an example of the flow of a data creation process in learning data creation example A. [Figure 8] 13 is an example of an evaluation result of a predicted K value predicted by a prediction model A. [Figure 9] 13 is a flowchart showing an example of the flow of a data creation process in learning data creation example B. [Figure 10] 13 is a flowchart showing an example of the flow of a data creation process in a modified example of learning data creation example B. [Figure 11] FIG. 13 is a diagram showing an example of the setting range of time t4. [Figure 12] 13 is an example of an evaluation result of a predicted K value predicted by prediction model B. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, the present embodiment will be described with reference to the drawings. Note that the same components and processes are given the same reference numerals throughout the drawings, and duplicated explanations will be omitted. The dimensional ratios in the drawings are exaggerated for the convenience of explanation, and may differ from the actual ratios.

[0013] FIG. 1 is a diagram showing an example of the appearance of an analysis device 1 according to this embodiment. As an example, the analysis device 1 measures a value correlated with the glucose concentration contained in urine and predicts the glucose concentration contained in urine, but the object to be analyzed is not limited to urine and may be other objects such as blood. Furthermore, the characteristic of the object to be analyzed measured by the analysis device 1 is not limited to the value correlated with the glucose concentration and may be other characteristics. The value correlated with the glucose concentration will be described in detail later.

[0014] 1, the analytical device 1 is equipped with a standard solution container 2 containing a standard solution, a buffer solution container 3 containing a buffer solution, a cleaning solution container 4 containing a cleaning solution, and a drain container 5 for containing unnecessary liquid used in the analytical device 1. The standard solution container 2, the buffer solution container 3, the cleaning solution container 4, and the drain container 5 are connected to the analytical device 1 by tubes.

[0015] FIG. 2 is a diagram showing an example of flow paths for various liquids in the analysis device 1. The analysis device 1 includes a cleaning liquid supply pump 6A that draws up a cleaning liquid from a cleaning liquid container 4 and supplies it to a predetermined flow path, a standard liquid supply pump 6B that draws up a standard liquid from a standard liquid container 2 and supplies it to a predetermined flow path, a buffer liquid delivery pump 6D that draws up a buffer liquid from a buffer liquid container 3 and supplies it to a predetermined flow path, and a sample pump 6E that draws up a sample, which is an example of an analysis target including blood or urine collected from a subject, from a sample container (not shown) and supplies it to a predetermined flow path. The analysis device 1 also includes a drain pump 6C that draws up various unnecessary liquids remaining in the flow path and discharges them into a drain container 5. Hereinafter, the cleaning liquid supply pump 6A, the standard liquid supply pump 6B, the drain pump 6C, the buffer liquid delivery pump 6D, and the sample pump 6E may be collectively referred to as "pump 6".

[0016] A filter 18 for removing foreign matter is provided at a specified location in the flow path. The nozzle 9A is inserted into the sample container, and the sample pump 6E is driven to suck the sample into the nozzle 9A.

[0017] Manifold 7B controls the connection destinations of the flow paths by opening and closing the flow paths with three-way electromagnetic valves 17A to 17C, thereby changing the branching pattern of the flow paths.

[0018] A sample pump 6E is connected to the manifold 7B, and a flow path for sucking the sample into the nozzle 9A is formed by the control of the three-way electromagnetic valves 17A to 17C. An air inlet and a nozzle 9A are connected to the manifold 7B, and a flow path for introducing air into the nozzle 9A is formed by the control of the three-way electromagnetic valves 17A to 17C. When air is introduced into the nozzle 9A, the sample sucked into the nozzle 9A is discharged from the tip of the nozzle 9A. A path for discharging the unnecessary sample sucked into the nozzle 9A to the drain container 5 via the manifold 7A is formed by the control of the three-way electromagnetic valves 17A to 17C. A flow path for supplying a cleaning solution is connected to the manifold 7B, and a flow path for supplying a cleaning solution to the tip of the nozzle 9A is formed by the control of the three-way electromagnetic valves 17A to 17C.

[0019] In addition to nozzle 9A, the analyzer 1 is also provided with nozzles 9B and 9C, and the liquid is sucked in and discharged by the same control as that for nozzle 9A. In the example shown in FIG. 2, nozzle 9B is attached to a position opposite to cleaning tank 11B containing cleaning liquid, and nozzle 9C is attached to a position opposite to standard liquid tank 11A containing standard liquid. Hereinafter, nozzles 9A, 9B, and 9C may be collectively referred to as "nozzles 9." There is no restriction on the number of nozzles 9 attached to the analyzer 1. Nozzle 9 can be moved up, down, left, and right by, for example, actuator 21.

[0020] Manifold 7A is connected to drain pump 6C, and by controlling the opening and closing of the flow paths with pinch valves 8A to 8E, a flow path from which unnecessary liquid is to be sucked is selected from among the multiple flow paths connected to manifold 7A, and the unnecessary liquid sucked by drain pump 6C is discharged into drain container 5. Hereinafter, pinch valves 8A to 8E may be collectively referred to as "pinch valves 8."

[0021] The three-way solenoid valve 17D controls the inflow direction of the buffer solution. The buffer solution supplied to the flow path by the buffer solution delivery pump 6D passes through the three-way solenoid valve 17D, and is then introduced into the reaction cell 14 via the heating unit 13. The heating unit 13 is a device that sets the buffer solution to a pre-specified temperature. Hereinafter, the three-way solenoid valves 17A to 17D may be collectively referred to as the "three-way solenoid valve 17."

[0022] In addition, a degassing module 15 and a degasser 16 are provided on the buffer solution flow path. The degassing module 15 is a component that separates unnecessary components such as air dissolved in the buffer solution from the buffer solution and branches the liquid containing the unnecessary components into a flow path equipped with a check valve 19 connected to the manifold 7A. The degasser 16 is a component that prevents air bubbles from accumulating in the buffer solution delivery pump 6D and achieves smooth liquid delivery by degassing the buffer solution before it enters the buffer solution delivery pump 6D.

[0023] In this configuration, the sample aspirated by the nozzle 9A is introduced into a reaction cell 14 containing a buffer solution. The reaction cell 14 is provided with a glucose sensor 12, which measures the concentration of glucose 25 contained in the sample (hereinafter referred to as "glucose concentration").

[0024] 3 is a diagram showing the principle of glucose concentration measurement by the glucose sensor 12. The glucose sensor 12 measures the glucose concentration using an enzyme electrode in which a glucose oxidase (GOD) immobilized enzyme membrane 20 and a hydrogen peroxide electrode 22 are combined.

[0025] The GOD-immobilized enzyme membrane 20 is an enzyme membrane in which GOD 27 is wrapped in a polycarbonate membrane 20A and a cellulose acetate membrane 20B. The sample introduced into the reaction cell 14 contains, for example, glucose 25 and protein 26, but the polycarbonate membrane 20A has pores of a size that allows only glucose 25 to pass through, but not protein 26, specifically pores of about 300 angstroms. As a result, only glucose 25 passes through the polycarbonate membrane 20A. Glucose 25 that has entered the GOD-immobilized enzyme membrane 20 is decomposed by the action of GOD 27 into gluconic acid 28 and hydrogen peroxide 29, as shown in formula (1).

[0026] (chemical 1) C6H 12 O6+O2+H2O → C6H 12 O7+H2O2 (1)

[0027] Cellulose acetate membrane 20B has pores of a size that allows only hydrogen peroxide 29 to pass through, specifically, pores of 5 angstroms or more and 6 angstroms or less. Therefore, decomposed hydrogen peroxide 29 passes through cellulose acetate membrane 20B and reaches hydrogen peroxide electrode 22. Hydrogen peroxide electrode 22 is an electrode in which anode 22A is made of platinum and cathode 22B is made of silver. Voltage source 24 is connected between anode 22A and cathode 22B. When a voltage of a predetermined magnitude is applied from voltage source 24, hydrogen peroxide 29 that has reached hydrogen peroxide electrode 22 undergoes an oxidation-reduction reaction due to the voltage applied by voltage source 24, and a current flows between anode 22A and cathode 22B.

[0028] As already explained, the polycarbonate membrane 20A has pores large enough to pass only hydrogen peroxide 29, and is therefore not affected by interfering reactions caused by reducing substances. Therefore, a stable current continues to flow between the anode 22A and the cathode 22B. The current flowing between the anode 22A and the cathode 22B has a magnitude related to the glucose concentration in the buffer solution. The current flowing between the anode 22A and the cathode 22B in response to the glucose concentration is called a "response current."

[0029] The reactions at the anode 22A and the cathode 22B are expressed by formulas (2) and (3), respectively.

[0030] (Case 2) H2O2 → 2H2+O2+2e - (2) O2+4H+4e - → 2H2O ···(3)

[0031] An ammeter 23 is connected between the anode 22A and the cathode 22B, and the response current is subjected to AD (Analog to Digital) conversion by the ammeter 23 to obtain an output value indicating the magnitude of the response current.

[0032] The glucose concentration is the output value D at the time t1 when the control of introducing the sample into the reaction cell 14 in which the buffer solution has already been introduced is performed. t1 and an output value D at a time t2 that is predetermined as a time after the time t1. t2 It is calculated from the difference between

[0033] Fig. 4 is a diagram showing the relationship between the time elapsed from the start of measurement of the response current and the output value. The horizontal axis of Fig. 4 represents the time [seconds] elapsed from the start of measurement of the response current, and the vertical axis represents the output value. Curve 32A represents the output value at each time.

[0034] The output value D at time t1 when the sample is introduced into the reaction cell 14 t1 is called the base value, and the output value D at time t2 t2 is called the reaction value. Time t2 is the time when the output value that has changed due to the introduction of a sample into the reaction cell 14 becomes stable. Time t0 represents the time when a predetermined amount of buffer solution is introduced into the reaction cell 14. The output value becoming stable means that the fluctuation range between adjacent output values ​​along the time series falls within a range where it can be considered that no change is observed in the output value.

[0035] In FIG. 4, the difference between the base value and the reaction value is called the K value. The K value is a characteristic value that is correlated with the glucose concentration. Therefore, the glucose concentration of a sample can be predicted from the K value. In this disclosure, predicting the glucose concentration from the K value is called "measuring the glucose concentration." Specifically, the K value of a standard sample with a known glucose concentration is obtained by the analytical device 1, and a relational equation between the glucose concentration and the K value is obtained in advance. Then, the K value of a sample with an unknown glucose concentration is measured by the analytical device 1, and the glucose concentration is predicted from the previously obtained relational equation and the measured value of the K value.

[0036] In the above example, in principle, it takes time until time t2 to measure the glucose concentration contained in the sample. However, the analysis device 1 reduces the time required to measure the glucose concentration by predicting the K value from the degree of change in the output value before time t2 using a prediction model generated in advance by supervised machine learning.

[0037] Specifically, the control unit 10 calculates n (n is an integer equal to or greater than 2) output values ​​D at intervals a, that is, at each lapse of time a, over a predetermined period from time t3, which is the start of a period after time t1 and before time t2. t3+ia where “i” is an integer index that runs from 0 to (n-1), and the output value D t3+ia represents the output value at time t3+ia.

[0038] The control unit 10 receives n output values ​​D t3+ia is input to the prediction model to predict the K value. From now on, the n output values ​​D obtained at intervals a from time t3 are t3+ia is sometimes referred to as the "specific output value."

[0039] The number n is a value that is determined in advance, for example, through experiments, before measuring the glucose concentration. The greater the value of the number n, the higher the accuracy of the glucose concentration measurement, but the longer the time required to measure the glucose concentration. The interval a and the time t3 are also values ​​that are determined in advance, for example, through experiments, before measuring the glucose concentration. The maximum value of the number n is nmax Then, at time t3+a(n max The number n, the interval a, and the time t3 are set so that −1)<time t2.

[0040] In this way, the analyzer 1 measures the glucose concentration of the sample, but in an analyzer 1 that uses an enzyme electrode to measure the glucose concentration, as the glucose concentration is repeatedly measured, the obtained K value may deviate from the K value corresponding to the actual glucose concentration, and the prediction accuracy of the glucose concentration may decrease. This phenomenon is called the "drift phenomenon."

[0041] Therefore, the analyzer 1 has a calibration function. Calibration is a process for correcting the deviation between the obtained K value and the K value corresponding to the actual glucose concentration, and has the function of making the glucose concentration measured by the analyzer 1 closer to the actual glucose concentration even when the glucose concentration measurement is performed repeatedly.

[0042] The calibration in the analytical device 1 will now be described. The analytical device 1 draws up a standard solution having a predetermined glucose concentration from the standard solution container 2, introduces it into the reaction cell 14, and predicts the K value of the standard solution by the method already described. If the predicted K value of the standard solution deviates from the standard value of the K value corresponding to the glucose concentration of the standard solution, the analytical device 1 determines a correction coefficient for the predicted K value of the standard solution so that the predicted K value of the standard solution approaches the standard value of the K value corresponding to the glucose concentration of the standard solution. In subsequent glucose concentration measurements, the analytical device 1 corrects the K value using the correction coefficient determined by calibration, thereby bringing the glucose concentration measurement result closer to the actual glucose concentration of the sample.

[0043] Such calibration is generally performed at the user's command, for example, when the glucose concentration of a sample has been measured a predetermined number of times, or when a predetermined time has elapsed since the glucose concentration measurement began.

[0044] However, calibration of the analysis device 1 of the present disclosure may be performed at a timing that the user considers particularly necessary, such as at the start of using the analysis device 1. As will be described in detail later, the analysis device 1 has a prediction model that predicts the K value reflecting the correction coefficient, so that the user can continue measuring the glucose concentration without performing calibration, even when repeatedly measuring the glucose concentration of a sample.

[0045] The control unit 10 controls the timing of introduction of the buffer solution and sample into the reaction cell 14 by controlling the pump 6, pinch valve 8, heating unit 13, three-way solenoid valve 17, and actuator 21, the supply of cleaning solution and standard solution to the analytical device 1, the discharge of unnecessary liquid into the drain container 5, the control of the voltage source 24, calibration, and the acquisition of output values ​​by the ammeter 23.

[0046] The control unit 10 of the present disclosure is configured, for example, using a computer 30. FIG. 5 is a diagram showing an example of the configuration of the main parts of an electrical system of the analysis device 1 including the control unit 10 configured using the computer 30.

[0047] The computer 30 includes a CPU (Central Processing Unit) 30A, which is an example of a processor that performs the processes executed by the control unit 10, a RAM (Random Access Memory) 30B used as a temporary working area for the CPU 30A, a non-volatile memory 30C, and an input / output interface (I / O) 30D. The CPU 30A, the RAM 30B, the non-volatile memory 30C, and the I / O 30D are each connected via a bus 30E.

[0048] The non-volatile memory 30C is an example of a storage device that maintains stored information even if the power supplied to the non-volatile memory 30C is cut off, and may be, for example, a semiconductor memory or a hard disk. The non-volatile memory 30C stores, for example, a control program that causes the computer 30 to function as the control unit 10 and various parameters that are referenced when measuring the glucose concentration.

[0049] The non-volatile memory 30C does not necessarily have to be built into the computer 30, and may be a portable storage device that is detachable from the computer 30, such as a memory card.

[0050] To the I / O 30D, components that operate upon receiving instructions from the control unit 10, such as the pump 6, pinch valve 8, heating unit 13, three-way electromagnetic valve 17, actuator 21, ammeter 23, and voltage source 24, are connected, but the components connected to the I / O 30D are not limited to these. For example, components according to the functions of the analysis device 1 can be connected, such as a communication unit (not shown) that connects to a communication line and communicates with an external device.

[0051] <Example A of creating learning data> Next, a learning data creation process for creating learning data used to generate a prediction model for predicting the K value will be described in detail.

[0052] 6 is a flowchart showing an example of the flow of a learning data creation process executed by the control unit 10 when an instruction to create learning data is received from a user. The CPU 30A of the control unit 10 reads a control program from the non-volatile memory 30C and executes the learning data creation process.

[0053] Prior to the start of the learning data creation process, a sample set is prepared in advance for the analysis device 1. The sample set refers to a plurality of samples each having a different predetermined glucose concentration, and each sample is placed in a different sample container. Hereinafter, the predetermined glucose concentration in each sample is referred to as a "predetermined concentration."

[0054] In step S10, the control unit 10 drives the buffer solution delivery pump 6D while closing the pinch valve 8C on the flow path connecting the manifold 7A No. 2 and the reaction cell 14, thereby introducing a predetermined amount of buffer solution into the reaction cell 14 in which the glucose sensor 12 is provided so that the buffer solution comes into contact with the glucose sensor 12.

[0055] In step S20, the control unit 10 controls the voltage source 24 to apply a voltage of a predetermined magnitude to the hydrogen peroxide electrode 22 in the glucose sensor 12.

[0056] In step S30, the control unit 10 starts control to acquire an output value corresponding to the magnitude of the response current measured by the glucose sensor 12 at intervals a. The time when the acquisition of the output value starts is defined as time t0. The control unit 10 manages time using, for example, a timer function included in the CPU 30A, but may acquire time information from an external source.

[0057] While acquiring the output value at intervals a, in step S40, the control unit 10 judges whether or not time t1 has been reached. Time t1 is the time at which the output value stabilizes in the state before the sample is introduced into the reaction cell 14. Time t1 is a time that is determined before the start of the learning data creation process based on the results of an experiment or the like previously performed.

[0058] The control unit 10 may refer to the fluctuation range between adjacent output values ​​along a time series and determine that the time t1 has been reached when the output value has stabilized. In this case, the time t1 may differ for each learning data creation process.

[0059] If time t1 has not been reached, the control unit 10 repeatedly executes the determination process of step S40 and continues the process of acquiring the output value at intervals a until time t1 is reached. On the other hand, if time t1 has been reached, the process proceeds to step S50.

[0060] In step S50, the control unit 10 controls the actuator 21 to move the nozzle 9A above a sample container containing any one of the samples that has not yet been selected from the sample set, and then lowers the nozzle 9A so that the nozzle 9A enters the sample container. In this state, the control unit 10 drives the sample pump 6E to aspirate the sample into the nozzle 9A.

[0061] Furthermore, the control unit 10 controls the actuator 21 to move the nozzle 9A that has aspirated the sample above the reaction cell 14, and then lowers the nozzle 9A so that the tip of the nozzle 9A enters the reaction cell 14. In this state, the control unit 10 drives the sample pump 6E to introduce a predetermined amount of the sample in the nozzle 9A into the reaction cell 14. After the sample has been introduced into the reaction cell 14, the control unit 10 controls the actuator 21 to raise the nozzle 9A to a position where the tip of the nozzle 9A protrudes from the reaction cell 14.

[0062] In this way, by introducing the sample into the reaction cell 14 after the output value has stabilized, the output value after the sample has been introduced approaches a value that correctly reflects the amount of fluctuation in the response current obtained solely by the oxidation-reduction reaction of hydrogen peroxide 29 produced from glucose 25 contained in the sample, compared to the case where the sample is introduced when the output value is not stable.

[0063] In step S60, the control unit 10 stores the output value acquired first after the sample is introduced into the reaction cell 14 as a base value in the non-volatile memory 30C.

[0064] While continuing to obtain the output value at each interval a, in step S70, the control unit 10 determines whether or not time t3 has been reached. Time t3 is a time that is determined in advance through an experiment before the learning data creation process is executed, and represents the elapsed time from time t0.

[0065] If it is determined that time t3 has not been reached, the determination process of step S70 is repeatedly executed, and the control unit 10 continues the process of acquiring the output value at the interval a until time t3 is reached.

[0066] On the other hand, if it is determined that the time t3 has been reached, the process proceeds to step S80.

[0067] In step S80, the control unit 10 acquires n output values ​​at intervals a, and converts each of the acquired output values ​​into a specific output value D t3+ia (i=0 to n-1) and stored in the non-volatile memory 30C.

[0068] n specific output values ​​D t3+ia While acquiring the time, in step S90, the control unit 10 determines whether or not time t2 has been reached.

[0069] If the time t2 has not been reached, the control unit 10 repeatedly executes the determination process of step S90 to obtain n specific output values ​​D t3+ia On the other hand, if the time t2 has been reached, the process proceeds to step S100.

[0070] In step S100, the control unit 10 calculates the output value D t2 and obtain the output value D t2 is stored as a reaction value in the non-volatile memory 30C.

[0071] In step S110, the control unit 10 controls the voltage source 24 to stop applying voltage to the hydrogen peroxide electrode 22 in the glucose sensor 12, and ends the acquisition of the output value.

[0072] In step S120, the control unit 10 reads out from the non-volatile memory 30C the base value acquired in the process of step S60 and the reaction value acquired in the process of step S100, and calculates the K value by subtracting the base value from the reaction value. The control unit 10 stores the calculated K value in the non-volatile memory 30C.

[0073] In step S130, the control unit 10 calculates n specific output values ​​D t3+ia , and the K value calculated by the process of step S120 are stored in the non-volatile memory 30C in association with each other. t3+ia The data that associates the K value with the first result data.

[0074] In step S140, the control unit 10 determines whether or not there is any sample in the sample set for which the first result data has not yet been obtained. If there is any sample for which the first result data has not yet been obtained, the process proceeds to step S150.

[0075] In order to prepare for acquiring the first result data for the next sample included in the sample set, in step S150, the control unit 10 drives the drain pump 6C with the pinch valve 8C open to discharge the buffer solution containing the sample from the reaction cell 14, empty the reaction cell 14, and proceed to step S10. As a result, the control unit 10 repeatedly executes the processes of steps S10 to S150 until it is determined in the determination process of step S140 that there is no sample for which the first result data has not been acquired. Thus, the first result data for each sample included in the sample set is obtained.

[0076] In each of the repeated processes of steps S10 to S150, the order of selection of the samples in the process of step S50 is the same in each repeated process.

[0077] On the other hand, if it is determined in the determination process of step S140 that there is no sample in the sample set for which first result data has not been obtained, the process proceeds to step S160.

[0078] In step S160, it is determined whether the acquisition of the first result data for the sample set has been repeated M times. Here, "M" is an index representing the number of times the acquisition of the first result data is repeated for each sample set, and is an integer equal to or greater than 2. There is no upper limit on the value of the number of repetitions M, and the number of repetitions M is a number that is determined in advance before the learning data creation process is executed. The number of repetitions M is stored in advance, for example, in non-volatile memory 30C.

[0079] If the acquisition of the first result data for the sample set has not been repeated M times, the process proceeds to step S150, and the control unit 10 empties the reaction cell 14 and then executes the process of step S10. As a result, the control unit 10 repeatedly executes the processes of steps S10 to S160 until it is determined in the determination process of step S160 that the acquisition of the first result data for the sample set has been repeated M times. Therefore, M pieces of first result data for each sample included in the sample set are obtained.

[0080] Hereinafter, the term "measurement cycle" may be used to distinguish the number of times the first result data of each sample included in the sample set is acquired by acquiring the first result data in units of the sample set. The measurement cycle is a value that increases by one cycle each time the control unit 10 acquires the first result data of each sample included in the sample set. In other words, the process of steps S10 to S150 is repeated the number of times the number of samples included in the sample set corresponds to one measurement cycle. Therefore, each measurement cycle corresponds to the number of executions in which the first result data is acquired in units of the sample set. In this way, the number of measurements for each sample included in the sample set is also represented by the measurement cycle, so the number of measurements for each sample constituting the first result data is the same value as the measurement cycle.

[0081] Note that the collection of the first result data of each sample included in the sample set, ie, the first result data for each sample set, is simply referred to as "first result data of the sample set".

[0082] The control unit 10 obtains M pieces of first result data of the sample set by obtaining the first result data of the sample set for each measurement cycle.

[0083] When it is determined in the determination process of step S160 that the acquisition of the first result data for the sample set has been repeated M times, the process proceeds to step S170.

[0084] In this case, M pieces of first result data of the sample set are acquired. Therefore, in step S170, the control unit 10 performs a data creation process to create learning data for generating a prediction model, and ends the learning data creation process shown in FIG.

[0085] FIG. 7 is a flowchart showing an example of the flow of a data creation process in the learning data creation example A.

[0086] First, in step S200, the control unit 10 initializes the measurement period M to "1".

[0087] In step S210, the control unit 10 calculates a specific output value D corresponding to each of the specified concentrations from the first result data of the sample set in the Mth measurement cycle. t3+ia Get the.

[0088] In step S220, the control unit 10 calculates the specific output value D t3+ia For each of the output values ​​D t3+ia To obtain the above, a reference K value of a sample having the same specified concentration as the sample introduced into the reaction cell 14 is associated with the sample. The reference K value is the K value calculated by the process of step S120 shown in FIG. 6 in the first measurement round, and exists for each specified concentration of the sample.

[0089] A specific output value D for each specified concentration of the sample set t3+ia The data group in which the K values ​​are associated with the reference K values ​​is the learning data for the prediction model, and the specific output value D t3+ia is input data for the learning data, and the reference K value is output data for the learning data. The control unit 10 stores each of the created learning data in, for example, the non-volatile memory 30C.

[0090] In step S230, the control unit 10 determines whether the measurement period M is M max Determine whether or not M max represents the upper limit of the measurement circumference M. max If not, there is still a specific output value Dt3+ia Therefore, there is first result data for the measurement period M for which no first result data has been acquired. Therefore, the process proceeds to step S240.

[0091] In step S240, the control unit 10 adds "1" to the measurement cycle M to update the measurement cycle M, and proceeds to step S210. As a result, the control unit 10 determines in the determination process of step S230 that the measurement cycle M is M max The processes of steps S210 to S240 are repeated until it is determined that the specific output value D t3+ia Learning data associating the K value with the reference K value is created for each measurement cycle and for each specified concentration of the sample included in the sample set.

[0092] On the other hand, in the determination process of step S230, if the measurement circumference M is M max If it is determined that this is the case, learning data will have been created from the first result data of the sample set at all measurement periods M, so the data creation process shown in Figure 7 is terminated, and with the termination of the data creation process, the learning data creation process shown in Figure 6 is terminated.

[0093] An example of the learning data created by the learning data creation process shown in Fig. 6 is shown in Table 1. In Table 1, the measurement period M, the specific output value, and the reference K value of the sample are shown from the top of the table in the order in which the sample was measured.

[0094] [Table 1]

[0095] In Table 1, "M" in the top row of entries represents the measurement period. The combination of the specific output value and the reference K value along the row direction in each measurement period M represents the learning data created from the first result data of the sample set of the corresponding measurement period M.

[0096] For ease of explanation, samples having different specified concentrations included in the sample set are distinguished using an index C. Index C is an integer equal to or greater than 1. For example, index C=1 represents a sample in the sample set with a specified concentration of A [mg / dL], index C=2 represents a sample in the sample set with a specified concentration of B [mg / dL], and so on. In this manner, C types of samples included in the sample set are distinguished according to the value of index C associated with each specified concentration. In addition, the upper limit of index C, i.e., the number of samples included in the sample set, is defined as C. max This is expressed as:

[0097] In addition, the specific output value D t3+ia In order to clarify which specific output value is the specific output value of the sample having which specified concentration at which measurement period, the specific output value D is calculated using the index C and the measurement period M. t3+ia Specific output value D[M:C] t3+ia represents the specific output value at time t3+ia of the first result data of the sample having the specified concentration associated with index C among the first result data of the sample set for the Mth measurement cycle.

[0098] In addition, since the standard K value exists for each specified concentration of the sample, KC (C = 1 to C max Specifically, the reference K value K1 is a reference K value calculated from a sample having a specific specified concentration from which the first result data was obtained first in the first measurement cycle, and the reference K value K2 is a reference K value calculated from a sample having another specific specified concentration from which the first result data was obtained second in the first measurement cycle.

[0099] That is, Table 1 shows the specific output value D[M:C] t3+ia :(M=1~M max , C=1~C max , i=0~(n max -1)) for the reference K value KC: (C = 1 to C max ) represents the associated learning data.

[0100] For each learning data created from the first result data of the sample set of any measurement period M, a standard K value corresponding to the specified concentration of each sample is assigned as output data, thereby obtaining learning data that is free from the effects of the drift phenomenon.

[0101] As can be seen from the data creation process shown in FIG. 7, when creating learning data using the learning data creation example A, the K value is not required for the measurement cycle M from the second cycle onward. Therefore, for the measurement cycle M from the second cycle onward, the control unit 10 calculates the specific output value D t3+ia After acquiring the specific output values ​​D, the process may proceed to step S130 without executing the processes of steps S90 to S120. In this case, the control unit 10 does not calculate the K value for each sample in the measurement cycle M from the second cycle onward. Therefore, in step S130 in the measurement cycle M from the second cycle onward, the control unit 10 acquires the n specific output values ​​D acquired in the process of step S80 for the number of measurements and the specified concentration for the sample aspirated into the nozzle 9A in the process of step S50. t3+ia are associated with each other and stored in the non-volatile memory 30C.

[0102] <Generation of prediction model A> Using the learning data created according to the learning data creation example A, supervised machine learning was performed using a Convolutional Neural Network (CNN) to generate a prediction model A. The prediction model A is a specific output value D of a sample acquired by the analysis device 1. t3+ia This is an example of a prediction model that predicts the K value of a sample based on the

[0103] As an example, CNN is used to generate the prediction model A here, but known supervised machine learning can be used to generate the prediction model A. Specifically, logistic regression analysis, decision tree, random forest, k-nearest neighbor method, support vector machine (SVM), deep learning, and the like may be used instead of CNN.

[0104] <Evaluation of prediction model A> The prediction accuracy of the K value when predicting the K value of a sample with a known glucose concentration using prediction model A is evaluated.

[0105] Specifically, the learning data creation process shown in FIG. 6 is executed for a sample set having the same combination of specified concentrations as the sample set when the learning data was created, and preliminary data having the same data structure as the learning data shown in Table 1 is created. max is the upper limit M when creating the learning data max However, the upper limit M max As an example, the upper limit M of the measurement period M when creating the advance data is max was set to “140”.

[0106] From each of the preliminary data thus created, the preliminary data of the sample having the same specified concentration was extracted for each evaluation period M and used as the evaluation data of the prediction model A. t3+ia The K value output by prediction model A when the above is input into prediction model A, i.e., the predicted K value, is obtained, and the prediction accuracy of the K value in prediction model A is evaluated from the degree of deviation between the reference K value and the predicted K value contained in the same evaluation data.

[0107] An example of evaluation data for prediction model A is shown in Table 2.

[0108] [Table 2]

[0109] Specific output value of training data D[M:C] t3+ia To distinguish it from the above, the specific output value of the evaluation data is D L [M:C] t3+ia In order to distinguish it from the reference K value KC for the training data, the reference K value for the evaluation data is expressed as K LAs can be seen from Table 2, the evaluation data was created from a sample of the specified concentration corresponding to index C=7.

[0110] The deviation rate is used as an index showing the prediction accuracy of the K value in prediction model A. The deviation rate of the predicted K value is expressed by equation (4).

[0111] (Number 1) Deviation rate of predicted K value [%] = (predicted K value – standard K value) / standard K value × 100 (4)

[0112] To compare with the deviation rate of the predicted K value, the specific output value D of the evaluation data L [M:C] t3+ia When obtaining the K value, the deviation rate of the measured K value is calculated. The deviation rate of the measured K value is expressed by equation (5).

[0113] (Number 2) Deviation rate of actual K value [%] = (actual K value – standard K value) / standard K value × 100 (5)

[0114] Fig. 8 is an example of an evaluation result showing the deviation rate of the predicted K value in the evaluation data shown in Table 2 predicted by the prediction model A and the deviation rate of the measured K value in the evaluation data shown in Table 2. The horizontal axis of the evaluation result shown in Fig. 8 represents the measurement period M of the evaluation data, and the vertical axis represents the deviation rate. Moreover, graph 33A shown in Fig. 8 represents the predicted K value, and graph 33B represents the measured K value.

[0115] As shown in Fig. 8, in the conventional method in which the actual K value is repeatedly measured without using prediction model A and without performing calibration in between, the deviation rate tends to increase as the measurement period M increases, and a maximum deviation of 3.5% from the reference K value occurred.

[0116] On the other hand, when the K value of a sample is predicted using prediction model A, the predicted K value does not show a tendency for the deviation rate to increase as the measurement period M increases, and the deviation rate tends to fall within a certain range regardless of the measurement period M. Specifically, the predicted K value deviated from the reference K value by only about 2.4%. Therefore, even if the glucose concentration is repeatedly measured by the analysis device 1 that predicts the K value using prediction model A, calibration of the analysis device 1 is not required during the repeated measurement of the glucose concentration.

[0117] That is, the analyzer 1 that measures the glucose concentration of a sample using the prediction model A can suppress the error in the measured glucose concentration within a certain range even if the glucose concentration measurements are performed repeatedly.

[0118] <Example B of creating learning data> Next, a learning data creation process for creating learning data used to generate a prediction model for predicting the K value using a method different from the learning data creation example A will be described in detail.

[0119] The learning data creation process in learning data creation example B is different from the learning data creation process in learning data creation example A shown in Fig. 6 in the data creation process of step S170. Since the other processes are the same as the learning data creation process shown in Fig. 6, hereinafter, the data creation process performed in step S170 in Fig. 6 will be described.

[0120] 9 is a flowchart showing an example of the flow of a data creation process in learning data creation example B. The CPU 30A of the control unit 10 reads a control program from the non-volatile memory 30C and executes the data creation process.

[0121] First, in step S300, the control unit 10 initializes the measurement period M to "1".

[0122] In step S310, the control unit 10 calculates a specific output value D corresponding to each of the specified concentrations from the first result data of the sample set in the Mth measurement cycle. t3+ia Get the.

[0123] In step S320, the control unit 10 acquires the K value from the first result data in the Mth measurement cycle of a sample having a predetermined specified concentration among the samples included in the sample set. Hereinafter, a sample having a predetermined specified concentration is referred to as a "specific concentration sample." The specific concentration sample is an example of a specific analyte.

[0124] The number of specific concentration samples does not necessarily have to be one, and may be multiple. In the analysis device 1, the measurement accuracy may differ depending on the glucose concentration to be measured, so the measurement accuracy for a specific glucose concentration may be higher than the measurement accuracy for other glucose concentrations. Therefore, when the number of specific concentration samples is one, it is preferable to select as the specific concentration sample a sample in which the glucose concentration estimated from, for example, the reference K value is closest to the actual glucose concentration of the sample among the samples included in the sample set. In this case, the measurement accuracy of the glucose concentration in the analysis device 1 may be improved compared to when other samples are used as the specific concentration samples.

[0125] Such specific concentration samples are determined in advance, for example, through experiments, before the learning data creation process is executed.

[0126] In addition, when the number of specific concentration samples is, for example, two, the samples included in the sample set may be selected as the specific concentration samples, for example, the sample whose glucose concentration estimated from the reference K value is closest to the actual glucose concentration of the sample and the sample whose glucose concentration is second closest to the actual glucose concentration of the sample.

[0127] In step S330, the control unit 10 calculates, for each specific concentration sample, a drift value indicating the degree of deviation between the reference K value for the specific concentration sample and the K value for the specific concentration sample acquired by the process of step S320.

[0128] There are no restrictions on the method of calculating the drift value as long as it indicates the degree of deviation between the reference K value of the specific concentration sample and the K value of the specific concentration sample in the Mth measurement cycle. For example, as shown in formula (6), the drift value is calculated as the difference between the reference K value of the specific concentration sample and the K value of the specific concentration sample in the Mth measurement cycle.

[0129] (Number 3) Drift value at Mth lap = K value of a specific concentration sample at the Mth measurement cycle - Reference K value of a specific concentration sample (6)

[0130] The terms on the right side of equation (6) may be interchanged to calculate the drift value as (reference K value-K value).

[0131] Furthermore, for example, as shown in formula (7), the drift value may be calculated by the ratio of the K value of the specific concentration sample in the Mth measurement cycle to the reference K value of the specific concentration sample.

[0132] (Number 4) Drift value at Mth lap = K value of a specific concentration sample at the Mth measurement cycle / K value of a specific concentration sample (7)

[0133] The drift value may be calculated as the reference K value / K value by switching the numerator and denominator on the right side of equation (7).

[0134] In step S340, the control unit 10 calculates the specific output value D of each sample included in the sample set in the Mth measurement round. t3+ia The control unit 10 also associates the drift value calculated by the process of step S330 with the specific output value D t3+ia , each output value D t3+ia To obtain the K value, a reference K value of a sample having the same specified concentration as the sample introduced into the reaction cell 14 is associated with the K value.

[0135] A specific output value D for each specified concentration of the sample set t3+iaThe data group in which the drift value and the reference K value are associated with each other is the learning data for the prediction model, and the specific output value D t3+ia The drift value and the reference K value are input data for the learning data, and the reference K value is output data for the learning data. The control unit 10 stores each of the created learning data in, for example, the non-volatile memory 30C.

[0136] As can be seen from the process of step S340, all learning data created from the first result data of the sample set at the same measurement period M are associated with the same drift value.

[0137] When multiple specific concentration samples exist, multiple drift values ​​are obtained, and therefore multiple drift values ​​are associated with one learning data item.

[0138] In step S350, the control unit 10 determines whether the measurement period M is M max It is determined whether the measurement period M is equal to or exceeds the upper limit M max If not, there is still a specific output value D t3+ia Therefore, there is first result data for the measurement period M for which no first result data has been acquired. Therefore, the process proceeds to step S360.

[0139] In step S360, the control unit 10 adds "1" to the measurement cycle M to update the measurement cycle M, and proceeds to step S310. As a result, the control unit 10 determines whether the measurement cycle M is equal to or lower than the upper limit value M in the determination process of step S350. max The processes of steps S310 to S360 are repeated until it is determined that the specific output value D t3+ia Learning data in which the drift value and the reference K value are associated with each other is created for each measurement cycle M and for each specified concentration of the sample included in the sample set.

[0140] On the other hand, in the determination process of step S350, the measurement circumference M is equal to or larger than the upper limit M maxIf it is determined that this is the case, learning data will have been created from the first result data of the sample set at all measurement periods M, so the data creation process shown in Figure 9 is terminated, and with the termination of the data creation process, the learning data creation process shown in Figure 6 is terminated.

[0141] Table 3 shows an example of learning data created when the process of step S170 of the learning data creation process shown in Fig. 6 is replaced with the data creation process shown in Fig. 9. The learning data shown in Table 3 is learning data for the case where there is one specific concentration sample. For ease of explanation, the drift value created from the first result data of the sample set in the Mth measurement cycle is represented as ΔK(M).

[0142] [Table 3]

[0143] That is, Table 3 shows the specific output value D[M:C] t3+ia :(M=1~M max , C=1~C max , i=0~(n max -1)) with drift value ΔK(M):(M=1~M max ) and the standard K value KC: (C = 1 to C max ) represents the associated learning data.

[0144] As can be seen from the data creation process shown in Fig. 9, when creating learning data using learning data creation example B, in the measurement cycle M from the second cycle onwards, it is sufficient to calculate the K value of only the specific concentration sample among the samples included in the sample set. Therefore, in the measurement cycle M from the second cycle onwards, the control unit 10 calculates the specific output value D t3+ia After acquiring the K value, the process may proceed to step S130 without executing the processes of steps S90 to S120. In this case, for samples for which the K value has not been calculated, the control unit 10 sets the K value to the specific output value D t3+ia Does not correspond to.

[0145] That is, in the first result data of the sample set in the second or subsequent measurement cycle M, the K value is set only for the first result data obtained from the specific concentration sample. In the second or subsequent measurement cycle M, the K value is calculated only for the specific concentration sample among the samples included in the sample set, so that the time required to create the learning data can be shortened compared to the case where the K value of each sample included in the sample set is calculated for each measurement cycle M.

[0146] <Modification of learning data creation example B> In learning data creation example B, the same drift value is set for each piece of learning data created from the first result data of the sample set in the Mth measurement cycle. That is, in learning data creation example B, in order to set the same drift value for the learning data in units of M measurement cycles, it is necessary to calculate the K value of the specific concentration sample for each M measurement cycle.

[0147] In the following, a method for creating learning data without calculating the K value of a specific concentration sample for each measurement period M will be described.

[0148] Therefore, the control unit 10 bundles each measurement cycle M into a plurality of measurement cycles M each having the same number of cycles, starting from the first measurement cycle. For example, the upper limit value M max When bundling two measurement cycles M each having a value of "6", the control unit 10 divides the measurement cycles M into three bundles: a bundle of the first and second measurement cycles M, a bundle of the third and fourth measurement cycles M, and a bundle of the fifth and sixth measurement cycles M. The control unit 10 then processes the bundle of measurement cycles M as a new unit of repetition.

[0149] In the following, to distinguish between the measurement period M and the bundle of measurement periods M, the bundle of measurement periods M is referred to as the "data set period M". p " Here, "M p " is an integer equal to or greater than 2. The data set period M p The upper limit M pmax is the upper limit M of the measurement period M max / Q, where “Q” is the number of times per data set M pThe number of measurement periods M to be bundled is an integer equal to or greater than 1. The user can specify the upper limit M max For the data set period M p The upper limit M pmax The measurement frequency Q is preset so that is an integer. If the measurement frequency Q is set to "2", two measurement frequencies M are the data set frequency M. p When the measurement frequency Q is "1", the data set frequency M p = measurement period M, which corresponds to the method of creating learning data according to the learning data creation example B already described. Therefore, the data set period M p In order to include a plurality of measurement frequencies M, the number of measurement frequencies Q is set to 2 or more.

[0150] In the learning data creation process in this modification, the data creation process executed by step S170 in the learning data creation process in the learning data creation example A shown in Fig. 6 is replaced with a data creation process different from the data creation process shown in Fig. 9. Since the other processes are the same as the learning data creation process shown in Fig. 6, hereinafter, the data creation process executed by step S170 in Fig. 6 will be described.

[0151] 10 is a flow chart showing an example of the flow of data creation processing in this modified example. The CPU 30A of the control unit 10 reads a control program from the non-volatile memory 30C and executes the data creation processing.

[0152] First, in step S400, the control unit 10 p Initialize to “1”.

[0153] In step S410, the control unit 10 p From the first result data of the sample set of each measurement period M included in the data set period of the period, a specific output value D corresponding to each specified concentration is obtained. t3+ia Get the.

[0154] Data set circumference M pSince one cycle of the measurement includes multiple measurement cycles M, the K value of the specific concentration sample is calculated multiple times within the same data set cycle. Therefore, in step S420, the control unit 10 p The K value is obtained from the first result data of the specific concentration sample in the first measurement cycle M included in the cycle data set cycle.

[0155] The number of specific concentration samples does not necessarily have to be one, but may be more than one. In this case, the control unit 10 sets M p The K value is obtained from the first result data of the specific concentration sample in the first measurement cycle M included in the cycle data set cycle.

[0156] In step S430, the control unit 10 calculates the drift value for each specific concentration sample.

[0157] In step S440, the control unit 10 p The specific output value D of each sample in the sample set of each measurement period M included in the data set period of the period t3+ia The drift value calculated by the process of step S430 is associated with M p The specific output value D of each sample in the sample set of each measurement period M included in the data set period of the period t3+ia , each output value D t3+ia To obtain the K value, a reference K value of a sample having the same specified concentration as the sample introduced into the reaction cell 14 is associated with the K value.

[0158] M p A specific output value D for each specified concentration of the sample set of each measurement period M included in the data set period of the period t3+ia The data group in which the drift value and the reference K value are associated with each other is the learning data for the prediction model, and the specific output value D t3+ia The drift value and the reference K value are input data for the learning data, and the reference K value is output data for the learning data. The control unit 10 stores each of the created learning data in, for example, the non-volatile memory 30C.

[0159] As can be seen from the processing in step S440, the same data set period M p The same drift value is set for all of the learning data created from the first result data of the sample set of each measurement period M included in the table.

[0160] When multiple specific concentration samples exist, multiple drift values ​​are obtained, and therefore multiple drift values ​​are set for one learning data set.

[0161] In step S450, the control unit 10 p is the upper limit M pmax Determine whether the data set period M p is the upper limit M pmax If not, there is still a specific output value D t3+ia Therefore, there is first result data for the measurement period M for which no first result data has been acquired. Therefore, the process proceeds to step S460.

[0162] In step S460, the control unit 10 p Add "1" to the data set period M p Then, the control unit 10 updates the data set period M in the determination process of step S450. p is the upper limit M pmax The processes of steps S410 to S460 are repeated until it is determined that the data set is equal to or larger than the predetermined value. p From the first result data of the sample set at each measurement period M included in t3+ia The learning data that corresponds to the drift value and the reference K value is p Every Mth data set p A measurement set is created for each specified concentration of the sample contained in the sample set of each measurement period M contained in the measurement set.

[0163] On the other hand, in the determination process of step S450, p is the upper limit M pmax If it is determined that pSince the learning data has been created from the first result data of the sample set at each measurement period M included in the data creation process shown in FIG. 10, the data creation process shown in FIG. 6 is terminated.

[0164] Table 4 shows an example of learning data created when the process of step S170 of the learning data creation process shown in Fig. 6 is replaced with the data creation process shown in Fig. 10. The learning data shown in Table 4 is learning data for the case where there is one specific concentration sample.

[0165] [Table 4]

[0166] That is, Table 4 shows the specific output value D[M:C] t3+ia :(M=1~M max , C=1~C max , i=0~(n max -1)) with drift value ΔK(1+Q(M p -1)):(M p =1~M pmax ) and the standard K value KC: (C = 1 to C max ) represents the associated learning data.

[0167] As can be seen from the data creation process shown in FIG. 10, when learning data is created by this modification, the first data set period M p The K value of each sample in the sample set at the first measurement cycle M in the data set and the K value of each sample in the data set at the second or subsequent cycles M p The dataset contained in the M p It is only necessary to calculate the K value of only the specific concentration sample among the samples included in the sample set at the first measurement cycle M of each measurement.

[0168] Therefore, the measurement cycle M is the second or later cycle, and 1+Q(M p -1) In a measurement cycle M that does not correspond to the 1st cycle, a specific output value Dt3+ia After acquiring the above, the process may proceed to step S130 without executing the processes of steps S90 to S120.

[0169] In addition, the measurement period M is the second or later period, and 1+Q(M p -1) In the measurement cycle M corresponding to the cycle 1, the control unit 10 calculates the specific output value D t3+ia After acquiring the above, the process may proceed to step S130 without executing the processes of steps S90 to S120.

[0170] In this case, for samples for which the K value has not been calculated, the control unit 10 sets the K value to a specific output value D t3+ia Does not correspond to.

[0171] In this manner, according to this modified example, the K value of the specific concentration sample is calculated for each data set period M without having to calculate the K value for each data set period M. p Since it is only necessary to perform the calculation for each first measurement cycle M included in the above, the time required to create the learning data can be shortened compared to the learning data creation example B.

[0172] In the learning data creation example B and the modified example of learning data creation example B shown above, the difference between the reference K value and the K value of the specific concentration sample is used as the drift value, but a value other than the difference between the reference K value and the K value may also be used as the drift value.

[0173] For example, the control unit 10 may set the specific output value D t3+ia In addition to obtaining the output value D at time t4, which is after time t3 and before time t2 as shown in FIG. t4 Get together.

[0174] In addition, in the process of S120 in FIG. 6, the control unit 10 calculates the base value, i.e., the output value D t1 and output value D t4(D t4 -D t1 ) is calculated as the output difference value.

[0175] In the following description of each process included in the learning data creation process, the reference K value is replaced with the output difference value calculated by the process of step S120 shown in Fig. 6 in the first measurement cycle (hereinafter referred to as the "reference output difference value"). Also, the K value is replaced with the output difference value calculated by the process of step S120 shown in Fig. 6 in each measurement cycle M. As a result, the values ​​shown in formulas (8) and (9) are used as the drift values ​​instead of formulas (6) and (7).

[0176] (Number 5) Data set circumference M p Drift value of = Data set circumference M p The output difference value of a specific concentration sample at the first measurement period M included in Reference output difference value for specific concentration sample (8) Data set circumference M p Drift value of = Data set circumference M p The output difference value of a specific concentration sample at the first measurement period M included in Reference output difference value for specific concentration sample (9)

[0177] As already explained, the terms on the right side of equation (8) may be swapped to calculate the drift value as (reference output difference value - output difference value), or the numerator and denominator on the right side of equation (9) may be swapped to calculate the drift value as reference output difference value / output difference value.

[0178] The time t4 is set in advance by a user and stored in, for example, the non-volatile memory 30C. When the time t4 is set to the time t2, the output difference value becomes the K value, and the reference output difference value becomes the reference K value.

[0179] <Generation of prediction model B> Using the learning data in Table 4 created according to a modified example of the learning data creation example B, supervised machine learning using CNN was performed to generate a prediction model B. The prediction model B is a specific output value D of the sample acquired by the analysis device 1. t3+ia 13 is an example of a prediction model for predicting the K value of a sample from the drift value.

[0180] Although CNN was used to generate prediction model B, publicly known supervised machine learning can be used to generate prediction model B, as with prediction model A. Specifically, logistic regression analysis, decision tree, random forest, k-nearest neighbor method, SVM, deep learning, and the like may be used instead of CNN.

[0181] Needless to say, prediction model B may be created using the learning data of Table 3 created according to learning data creation example B.

[0182] <Evaluation of prediction model B> The prediction accuracy of the K value is evaluated when predicting the K value of a sample having a known glucose concentration using prediction model B generated by performing supervised machine learning using the learning data shown in Table 4.

[0183] Specifically, for a sample set having the same combination of specified concentrations as the sample set when the learning data was created, the learning data creation process shown in FIG. 6 is executed, in which the process of step S170 is the data creation process shown in FIG. 10, to create preliminary data having the same data structure as the learning data shown in Table 4. max is the upper limit M when creating the learning data max However, the upper limit M max As an example, the upper limit M of the measurement period M when creating the advance data is max was set to "140". That is, the data set period M p The upper limit M pmax would result in “70”.

[0184] From each of the preliminary data thus created, the preliminary data of the sample having the same specified concentration was extracted for each evaluation period M and used as the evaluation data of the prediction model B. L [M:C] t3+ia When the drift value ΔK(M) was input into prediction model B, the predicted K value output by prediction model B was obtained, and the prediction accuracy of the K value in prediction model B was evaluated from the degree of deviation between the reference K value contained in the same evaluation data and the predicted K value.

[0185] An example of evaluation data for prediction model B is shown in Table 5. As can be seen from Table 5, the evaluation data was created from a sample with a specified concentration corresponding to index C=7. The number of measurement cycles Q was set to 2.

[0186] [Table 5]

[0187] In order to evaluate the prediction accuracy of the K value in prediction model B, the deviation rates shown in equations (4) and (5) were calculated.

[0188] Fig. 12 is an example of an evaluation result showing the deviation rate of the predicted K value in the evaluation data shown in Table 5 predicted by the prediction model B and the deviation rate of the measured K value in the evaluation data shown in Table 5. The horizontal axis of the evaluation result shown in Fig. 12 represents the measurement period M of the evaluation data, and the vertical axis represents the deviation rate. Moreover, graph 33C shown in Fig. 12 represents the predicted K value, and graph 33B represents the measured K value.

[0189] As shown in FIG. 12, in the conventional method of repeatedly measuring the actual K value without using the prediction model B and without performing calibration in between, a maximum deviation of 3.5% occurred from the reference K value. On the other hand, in the case of predicting the K value of the sample using the prediction model B, the predicted K value did not show a tendency for the deviation rate to increase as the measurement period M increased, and the deviation rate tended to fall within a certain range regardless of the measurement period M. Specifically, the predicted K value only deviated by about 1.05% from the reference K value. Therefore, even if the glucose concentration is repeatedly measured by the analysis device 1 that predicts the K value using the prediction model B, calibration of the analysis device 1 is not required in the middle of the repeated measurement of the glucose concentration.

[0190] That is, the analyzer 1 that measures the glucose concentration of a sample using the prediction model B can suppress the error in the measured glucose concentration within a certain range even if the glucose concentration measurements are performed repeatedly.

[0191] The embodiment has been used above to describe one form of creating learning data for a prediction model used in the analysis device 1, but the disclosed form of creating learning data is only one example, and the form of creating learning data is not limited to the scope described in the embodiment. Various changes or improvements can be made to the embodiment without departing from the gist of the present disclosure, and forms with such changes or improvements are also included in the technical scope of the disclosure.

[0192] For example, the internal processing order in the learning data creation process shown in FIG. 6 and the internal processing order in the data creation process shown in FIGS. 7, 9, and 10 may be changed without departing from the gist of the present disclosure.

[0193] In the above embodiment, as an example, the learning data creation process and the data creation process are implemented by software. However, the same processes as those shown in the flowchart of the learning data creation process in Fig. 6 and the flowcharts of the data creation process in Fig. 7, Fig. 9, and Fig. 10 may be implemented by hardware. In this case, the processing speed can be increased compared to when each process is implemented by software.

[0194] In the above embodiment, the term "processor" refers to a processor in a broad sense, including a general-purpose processor (e.g., CPU 30A) and a dedicated processor (e.g., GPU: Graphics Processing Unit, ASIC: Application Specific Integrated Circuit, FPGA: Field Programmable Gate Array, programmable logic device, etc.).

[0195] The operations of the processor in the above embodiment may be performed not only by one processor, but also by multiple processors located at physically separate locations working together. The order of the operations of the processor is not limited to the order described in the above embodiment, and may be changed as appropriate.

[0196] In the above embodiment, the control program is stored in advance in the non-volatile memory 30C. However, the storage destination of the control program is not limited to the non-volatile memory 30C. The control program of the present disclosure may be provided in a form recorded in a storage medium readable by the computer 30.

[0197] For example, the control program may be provided in a form recorded on an optical disk such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a Blu-ray disk. The control program may also be provided in a form recorded on a portable semiconductor memory such as a Universal Serial Bus (USB) memory or a memory card. The non-volatile memory 30C, a CD-ROM, a DVD-ROM, a Blu-ray disk, a USB, and a memory card are examples of non-transitory storage media.

[0198] Furthermore, the analysis device 1 may download a control program from an external device connected to a communication line through the communication unit, and store the downloaded control program in the non-volatile memory 30C of the analysis device 1. In this case, the CPU 30A of the analysis device 1 reads the control program downloaded from the external device from the non-volatile memory 30C, and executes the learning data creation process and the data creation process.

[0199] The following are notes related to this disclosure.

[0200] (Appendix 1) a reaction cell into which a buffer solution and an analyte are introduced; an electrode provided in the reaction cell, which outputs a response current having a magnitude corresponding to a concentration of the analyte in the buffer solution contacting the reaction cell when a voltage of a predetermined magnitude is applied thereto; a control unit that controls the timing of introduction of the buffer solution and the analyte into the reaction cell, and also controls the magnitude of the response current output from the electrode to be acquired as an output value representing a characteristic of the analyte; In an analytical device comprising: The output value D at time t1 when the control unit performs control to introduce the analyte into the reaction cell in which the buffer solution has already been introduced t1 and an output value D at a time t2 that is predetermined as a time after the time t1. t2When a K value, which is a value associated with the concentration of the analyte represented by the difference between For each of the plurality of analytes having different predetermined concentrations, an output value D t1 and an output value D at time t3+ia (i=0 to n-1: n is a predetermined integer of 2 or more) at each predetermined interval a within the period from time t3, which is the start of the period after the time t1 and before the time t2. t3+ia and the output value D at time t2 t2 and after performing the process of acquiring each of them at least once at first, the output value D at the time t3+ia for each of the analytes is t3+ia an acquisition step of repeatedly executing a process of acquiring the above-mentioned information up to a predetermined number of times; The output value D for each of the analytes first acquired by the acquisition step t1 and the output value D for each analyte t2 a K value calculation step of calculating the K value for each of the analytes as a reference K value from the difference between the K value and the K value; Each output value D for each of the analysis targets obtained each time the acquisition step is performed t3+ia For each output value D t3+ia a creation step of associating the reference K value of the analyte having the same concentration as the concentration of the analyte introduced into the reaction cell to obtain the reference K value, and creating learning data to be used in supervised machine learning of the prediction model; How to create training data, including:

[0201] (Appendix 2) In each execution of the acquisition step, an output value D at a time t4 that is after the time t3 and before the time t2 t4 is acquired for each of the analytes, and in each of the second and subsequent executions of the acquisition step, the output value D at the time t1 is t1 is obtained for each of the analytes, Among the analytes, for a specific analyte having a specific concentration designated in advance, an output value D of the specific analyte obtained in each execution of the obtaining step t1 and output value D t4 a calculation step of calculating a difference between the output and the specific analyte in each execution of the acquisition step as an output difference value of the specific analyte; a drift value calculation step of calculating, for each execution of the acquisition step, a drift value indicating a degree of deviation between the output difference value of the specific analyte in a first execution of the acquisition step and the output difference value of the specific analyte in each execution of the acquisition step; Including, The generating step is a step of generating output values ​​D for each of the analysis objects obtained in each execution of the acquiring step. t3+ia and each output value D t3+ia For each input data including the drift value in the same execution time as each execution time, each output value D t3+ia and generating learning data to be used in supervised machine learning of the prediction model by associating the reference K value of the analyte having the same concentration as the concentration of the analyte introduced into the reaction cell as output data. A method for creating learning data as described in Appendix 1.

[0202] (Appendix 3) The time t4 was set to the time t2. A method for creating learning data as described in Appendix 2.

[0203] (Appendix 4) The specific analyte is the analyte whose concentration estimated from the reference K value for each analyte is closest to the actual concentration of the analyte. A method for creating learning data according to appendix 2 or appendix 3.

[0204] (Appendix 5) The specific analyte is a plurality of analytes having different concentrations that are specified in advance among the analytes, In the drift value calculation step, the drift value for each of the specific analytes is calculated each time the acquisition step is performed; In the creation step, the output value D for each of the analysis objects in each execution of the acquisition step is obtained. t3+ia and each output value D t3+ia For input data including the drift values ​​in the same execution times as each execution time, each output value D t3+ia and generating learning data to be used in supervised machine learning of the prediction model by associating the reference K value of the analyte having the same concentration as the concentration of the analyte introduced into the reaction cell as output data. A method for creating learning data according to appendix 2 or appendix 3.

[0205] (Appendix 6) In the drift value calculation step, when a bundle of a plurality of consecutive executions of the acquisition step is treated as a unit of a new execution of the acquisition step, and a plurality of output difference values ​​of the specific analyte having the same concentration are included in each of the new executions, the output value D acquired by the acquisition step performed first in each of the new executions is calculated. t1 and output value D t4 calculating at least one of the drift values ​​for the specific analyte using the output difference value calculated from the difference between the In the creation step, the output values ​​D constituting the learning data obtained from each of the execution times constituting the new execution time in which the drift value is calculated are t3+ia For each of these, the output value D t3+ia and generating the learning data by associating at least one of the drift values ​​calculated from the same new execution time obtained from the new execution time. A method for creating learning data according to any one of Supplementary Note 2 to Supplementary Note 5.

[0206] (Appendix 7) The degree of deviation is represented by a difference between the output difference value of the specific analyte in the first execution of the acquisition step and the output difference value of the specific analyte in each execution of the acquisition step, or by a ratio of the output difference value of the specific analyte in each execution of the acquisition step to the output difference value of the specific analyte in the first execution of the acquisition step. A method for creating learning data according to any one of Supplementary Note 2 to Supplementary Note 6.

[0207] (Appendix 8) A model generated by supervised machine learning using the learning data created by the learning data creation method described in Supplementary Note 1, the model predicting a K value, which is a value related to a concentration of an analyte, includes an output value D representing a characteristic of the analyte at a time t3+ia (i=0 to n-1: n is a predetermined integer of 2 or more) obtained at predetermined intervals a within a period from a time t3 that is a start of a period after a time t1 in which control is performed to introduce the analyte into a reaction cell to which a buffer solution and the analyte are introduced and before a time t2 that is predetermined as a time after the time t1, t3+ia An input step in which The output value D t3+ia a prediction step of predicting the concentration of the analyte from the predicted value of the K value output by the prediction model to which the above-mentioned input is applied; The method for predicting a property of an analyte comprising:

[0208] (Appendix 9) A model generated by supervised machine learning using the learning data created by the learning data creation method described in Supplementary Note 2, the model predicting a K value, which is a value related to a concentration of an analyte, is used to predict the concentration of the analyte when the analyte is repeatedly introduced into a reaction cell to which a buffer solution and the analyte are introduced, An output value D representing a characteristic of the analyte at a time t3+ia (i=0 to n-1: n is a predetermined integer of 2 or more) obtained at a predetermined interval a within a period from a time t3 which is the start of a period after a time t1 at which control is performed to introduce the analyte into the reaction cell and before a time t2 which is a predetermined time after the time t1. t3+ia and an output value D at time t1 obtained from the specific analyte having a specific concentration designated in advance when the specific analyte is repeatedly introduced into the reaction cell the same number of times as the number of times the analyte is introduced into the reaction cell. t1 and an output value D at time t4, which is after time t3 and before time t2. t4 and an input step of inputting into the prediction model an output difference value calculated from the difference between the output difference value and the output difference value obtained when the specific analyte is first introduced into the reaction cell; The output value D t3+ia and a prediction step of predicting the concentration of the analyte from the predicted value of the K value output by the prediction model to which the drift value is input. The method for predicting a property of an analyte comprising:

[0209] According to Appendix 1, it is possible to create learning data to be used in machine learning for generating a predictive model that repeatedly predicts the characteristics of an analyte without performing calibration.

[0210] According to Supplementary Note 2, there is an effect that the prediction accuracy of the K value is improved compared to the case where the K value is predicted using a prediction model that has been subjected to machine learning using learning data that does not include drift values.

[0211] According to Appendix 3, there is an advantage that the drift value can be calculated from the K value of the specific analyte in each execution.

[0212] According to Supplementary Note 4, there is an advantage that the measurement accuracy of the concentration of the analyte is improved compared to the case where a specific analyte is selected without taking into consideration the measurement accuracy characteristics for each concentration of the analyte in the analysis device.

[0213] According to Appendix 5, the effect is that the accuracy of measuring the concentration of the analyte is improved compared to measuring the concentration of the analyte using a predictive model generated from training data created by limiting the specific analyte to one analyte having a specific concentration.

[0214] According to Supplementary Note 6, there is an advantage that the time required for creating the learning data can be shortened, compared to the case where the output difference value of the specific analysis target is calculated every time the acquisition step is performed.

[0215] According to Appendix 7, there is an advantage that a method for calculating the degree of deviation between the output differential value of a specific analyte in the first execution of the acquisition step and the output differential value of a specific analyte in each execution of the acquisition step can be selected from a plurality of calculation methods according to the characteristics of the specific analyte.

[0216] According to Supplementary Note 8 and Supplementary Note 9, there is an advantage that the characteristics of the object to be analyzed can be repeatedly measured without performing calibration. [Explanation of symbols]

[0217] 1 Analyzer 10 Control section 12 Glucose Sensor 14 Reaction Cell 22 Hydrogen peroxide electrode 23 Ammeter 24 Voltage Source 30 Computers 30A CPU a Output value acquisition interval M measurement period Mp Dataset Circumference ΔK(M) Drift value at measurement period M t0 The time when the acquisition of output values ​​started t1 Base value acquisition time t2 Response value acquisition time t3 Start time of acquiring specific output value t4 Time used to calculate the output difference value

Claims

1. a reaction cell into which a buffer solution and an analyte are introduced; an electrode provided in the reaction cell, which outputs a response current having a magnitude corresponding to a concentration of the analyte in the buffer solution contacting the reaction cell when a voltage of a predetermined magnitude is applied thereto; a control unit that controls the timing of introduction of the buffer solution and the analyte into the reaction cell, and also controls the magnitude of the response current output from the electrode to be acquired as an output value representing a characteristic of the analyte; In an analytical device comprising: The output value D at time t1 when the control unit performs control to introduce the analyte into the reaction cell in which the buffer solution has already been introduced ｔ１ and an output value D at a time t2 that is determined in advance as a time after the time t1. ｔ２ When a K value, which is a value associated with the concentration of the analyte represented by the difference between For each of the plurality of analytes having different predetermined concentrations, an output value D ｔ１ and an output value D at time t3+ia (i=0 to n-1: n is a predetermined integer of 2 or more) at each predetermined interval a within the period from time t3, which is the start of the period after time t1 and before time t2. ｔ３＋ｉａ and the output value D at time t2 ｔ２ and the output value D at the time t3+ia for each of the analytes is obtained at least once. ｔ３＋ｉａ an acquisition step of repeatedly executing a process of acquiring the above-mentioned information up to a predetermined number of times; The output value D for each of the analytes first acquired by the acquisition step ｔ１ and the output value D for each analyte ｔ２ a K value calculation step of calculating the K value for each of the analysis objects as a reference K value from the difference between the K value and the K value; Each output value D for each of the analysis targets obtained each time the obtaining step is performed ｔ３＋ｉａ For each output value D ｔ３＋ｉａ a creation step of associating the reference K value of the analyte having the same concentration as the concentration of the analyte introduced into the reaction cell to obtain the reference K value, and creating learning data to be used in supervised machine learning of the prediction model; How to create training data, including:

2. In each execution of the acquisition step, an output value D at a time t4 that is after the time t3 and before the time t2 ｔ４ is obtained for each of the analytes, and in each of the second and subsequent executions of the obtaining step, the output value D ｔ１ is obtained for each analyte, For a specific analyte having a specific concentration designated in advance among the analytes, an output value D of the specific analyte obtained in each execution of the obtaining step ｔ１ and output value D ｔ４ a calculation step of calculating a difference between the output and the specific analyte in each execution of the acquisition step as an output difference value of the specific analyte; a drift value calculation step of calculating, for each execution of the acquisition step, a drift value indicating a degree of deviation between the output difference value of the specific analyte in a first execution of the acquisition step and the output difference value of the specific analyte in each execution of the acquisition step; Including, The generating step is a step of generating output values ​​D for each of the analysis targets in each execution of the acquisition step. ｔ３＋ｉａ and each output value D ｔ３＋ｉａ For input data including the drift value in the same execution time as each execution time, each output value D ｔ３＋ｉａ The reference K value of the analyte having the same concentration as the concentration of the analyte introduced into the reaction cell is associated as output data, and learning data to be used in supervised machine learning of the prediction model is created. The method for creating learning data according to claim 1 .

3. The time t4 is set to the time t2. The method for creating learning data according to claim 2 .

4. The specific analyte is the analyte whose concentration estimated from the reference K value for each analyte is closest to the actual concentration of the analyte. The method for creating learning data according to claim 3 .

5. The specific analyte is a plurality of analytes having different concentrations that are specified in advance among the analytes, In the drift value calculation step, the drift value for each of the specific analytes is calculated each time the acquisition step is performed; In the creation step, each output value D for each of the analysis targets in each execution of the acquisition step is acquired. ｔ３＋ｉａ and each output value D ｔ３＋ｉａ For input data including a plurality of the drift values ​​in the same execution times as each execution time, each output value D ｔ３＋ｉａ The reference K value of the analyte having the same concentration as the concentration of the analyte introduced into the reaction cell is associated as output data, and learning data to be used in supervised machine learning of the prediction model is created. The method for creating learning data according to claim 3 .

6. In the drift value calculation step, when a bundle of a plurality of consecutive executions of the acquisition step is treated as a unit of a new execution of the acquisition step, and a plurality of output difference values ​​of the specific analyte having the same concentration are included in each of the new executions, the output value D obtained by the acquisition step performed first in each of the new executions is calculated as ｔ１ and output value D ｔ４ calculating at least one of the drift values ​​of the specific analyte using the output difference value calculated from the difference between the In the creating step, the output values ​​D constituting the learning data obtained from each of the execution times constituting the new execution time in which the drift value is calculated are ｔ３＋ｉａ Each of the output values ​​D ｔ３＋ｉａ and generating the learning data by associating at least one of the drift values ​​calculated from the same new execution time obtained from the new execution time. The method for creating learning data according to any one of claims 2 to 5.

7. The degree of deviation is represented by a difference between the output difference value of the specific analyte in the first execution of the acquisition step and the output difference value of the specific analyte in each execution of the acquisition step, or by a ratio of the output difference value of the specific analyte in each execution of the acquisition step to the output difference value of the specific analyte in the first execution of the acquisition step. The method for creating learning data according to claim 2 .

8. A model generated by supervised machine learning using the learning data created by the method for creating learning data according to claim 1, the model predicting a K value, which is a value related to a concentration of an analyte, includes an output value D representing a characteristic of the analyte at a time t3+ia (i=0 to n-1: n is a predetermined integer of 2 or more) obtained at predetermined intervals a within a period from a time t3 that is a start of a period after a time t1 in which control is performed to introduce the analyte into a reaction cell into which a buffer solution and the analyte are introduced and before a time t2 that is a predetermined time after the time t1, the output value D being obtained at predetermined intervals a within the period ｔ３＋ｉａ An input step in which The output value D ｔ３＋ｉａ a prediction step of predicting the concentration of the analyte from the predicted value of the K value output by the prediction model to which the above-mentioned input is applied; The method for predicting a property of an analyte comprising:

9. 3. A model generated by supervised machine learning using the learning data created by the learning data creation method according to claim 2, wherein the model predicts a K value, which is a value associated with a concentration of an analyte, and predicts the concentration of the analyte when the analyte is repeatedly introduced into a reaction cell to which a buffer solution and the analyte are introduced, An output value D representing a characteristic of the analyte at a time t3+ia (i=0 to n-1: n is a predetermined integer of 2 or more) obtained at a predetermined interval a within a period from a time t3 which is the start of a period after a time t1 at which the analyte is introduced into the reaction cell and before a time t2 which is a predetermined time after the time t1. ｔ３＋ｉａ and an output value D at time t1 obtained from the specific analyte having a specific concentration designated in advance when the specific analyte is repeatedly introduced into the reaction cell the same number of times as the number of times the analyte is introduced into the reaction cell. ｔ１ and an output value D at a time t4 that is after the time t3 and before the time t2. ｔ４ and an input step of inputting into the prediction model an output difference value calculated from the difference between the output difference value and the output difference value obtained when the specific analyte is first introduced into the reaction cell; The output value D ｔ３＋ｉａ and a prediction step of predicting the concentration of the analyte from the predicted value of the K value output by the prediction model to which the drift value is input. The method for predicting a property of an analyte comprising:

Images