Method, apparatus, and program for analyzing blood samples

A deep learning-based analysis method for blood coagulation curves enhances prediction accuracy by analyzing a broader range of data, addressing limitations in conventional neural network approaches.

JP7594404B2Active Publication Date: 2024-12-04SYSMEX CORP
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2020171206
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-10-09
Publication Date
2024-12-04
Estimated Expiration
2040-10-09

AI Technical Summary

Technical Problem

Existing blood coagulation analysis methods using neural networks are limited by input variables, leading to reduced accuracy in predicting the cause of prolonged blood clotting time when similar changes occur for multiple extension causes.

Method used

A blood sample analysis method utilizing a deep learning algorithm that analyzes a data group comprising a blood coagulation curve or its differential curve, enabling more accurate prediction of the cause of prolonged blood clotting time by inputting this data into a deep learning algorithm.

Benefits of technology

The method achieves higher accuracy in predicting the cause of prolonged blood clotting time compared to conventional methods, as demonstrated by improved ROC curves and AUC values.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007594404000001
    Figure 0007594404000001
  • Figure 0007594404000002
    Figure 0007594404000002
  • Figure 0007594404000003
    Figure 0007594404000003
Patent Text Reader

Abstract

To provide a method, a device, and a program for analyzing blood samples, which allow for predicting previously unpredictable causes of prolongation.SOLUTION: A challenge is cleared by a blood sample analysis method comprising acquiring a data group including a plurality of sets of data constituting a blood coagulation curve or a differential curve thereof, inputting the data group to a deep learning algorithm, and outputting information regarding the causes of prolonged blood coagulation time of a blood sample according to a result obtained from the deep learning algorithm.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] This specification discloses a method, an analyzer, and an analysis program for analyzing a blood sample. [Background technology]

[0002] In a blood coagulation test, when the blood clotting time is prolonged, the medical staff identifies the cause. As a method for supporting the identification of the cause of the prolonged blood clotting time, Patent Document 1 discloses a method for predicting the deficiency of coagulation factors, the contamination of a sample with heparin, and the presence of lupus anticoagulant using a neural network. In Patent Document 1, a time-dependent optical profile of PT or APTT is acquired, and one or more predictive variables of the minimum value of the first derivative of the optical profile, the time index of the minimum value of the first derivative, the minimum value of the second derivative of the optical profile, the time index of the minimum value of the second derivative, the maximum value of the second derivative, the time index of the maximum value of the second derivative, the overall change in transmittance during the reaction, the clotting time, the slope of the optical profile before the formation of a clot, and the slope of the optical profile after the formation of a clot are input to the neural network. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] U.S. Patent No. 6,321,164 Summary of the Invention [Problem to be solved by the invention]

[0004] In the method described in Patent Document 1, the input values ​​to the neural network are limited to specific predictive variables, making it difficult to make predictions when changes in predictive variables are similar for multiple causes of extension, and there was a demand for improved prediction accuracy. An object of the present invention is to provide a blood sample analysis method, an analysis device, and an analysis program that are capable of predicting the cause of prolongation with higher accuracy than ever before. [Means for solving the problem]

[0005] An embodiment disclosed herein relates to a method for analyzing a blood sample, comprising: acquiring a data set including a plurality of data constituting a blood clotting curve or a derivative curve thereof; inputting the data set into a deep learning algorithm; and outputting information regarding a cause of prolongation of a blood clotting time in the blood sample based on a result obtained from the deep learning algorithm.

[0006] Another embodiment disclosed herein relates to a blood sample analyzer (1). The analyzer (1) includes a measurement unit (2) and a control unit (201). The measurement unit (2) prepares a measurement sample containing the blood sample and a coagulation time measurement reagent, and outputs a plurality of pieces of detection information constituting a blood coagulation curve from the measurement sample. The control unit (201) acquires a data group including a plurality of pieces of data constituting a blood coagulation curve or a derivative curve thereof based on the plurality of pieces of detection information, inputs the data group to a deep learning algorithm, and outputs information regarding the cause of the prolongation of the blood coagulation time in the blood sample based on the results obtained from the deep learning algorithm.

[0007] Another embodiment disclosed herein relates to an analysis program (202b) for a blood sample, which, when executed by a computer, causes the computer to execute the steps of acquiring a data group including a plurality of data constituting a blood coagulation curve or a derivative curve thereof, inputting the data group into a deep learning algorithm, and outputting information on the cause of the prolongation of the blood coagulation time in the blood sample based on the results obtained from the deep learning algorithm. Effect of the Invention

[0008] According to the present invention, the cause of the extension can be predicted with higher accuracy than ever before. [Brief description of the drawings]

[0009] [Figure 1] 1 shows an example of the appearance of an analysis device 1. [Diagram 2] 1 shows an example of the hardware configuration of an analysis device 1. [Diagram 3] 2 shows an example of the configuration of the light irradiation unit 10. [Figure 4] 2 shows an example of the configuration of the detector 230. [Diagram 5] 1 shows a flow of measurement and analysis processing by the control unit 201 of the analysis device 1. [Figure 6] 2 shows the flow of a measurement process executed by the control unit 201 based on the measurement program 202a. [Figure 7] 2 shows the flow of analysis processing executed by the control unit 201 based on the analysis program 202b. [Figure 8] (A) shows an example of a blood clotting curve. (B) shows an example of a normalized clotting curve. [Figure 9] (A) shows an example of a first derivative coagulation curve. (B) shows an example of a second derivative coagulation curve. [Figure 10] Provides an overview of training deep learning algorithms. [Figure 11] An example of analysis using a deep algorithm is shown below. [Figure 12] 13 shows a modified example of a data group including a plurality of data. [Figure 13] (A) is an example of a pie chart showing the probability of a softmax model, and (B) is an example of a bar chart showing the probability of a binary model. [Figure 14] An example of information stored in the additional examination DB 202f is shown below. [Figure 15] 1A shows an example of the appearance of the training device 5. FIG. 1B shows an example of the hardware configuration of the training device 5. [Figure 16] An example of a training process executed by the control unit 501 based on the training program 502b is shown. [Figure 17] 1 shows an example of the hardware configuration of a modified example of the analysis device 1. [Figure 18] 2 shows the flow of processing executed by the control unit 201 of the analysis device 1' based on the analysis program 202b'. [Figure 19] 1A is a histogram showing the prediction results by the conventional method, and FIG. 1B is a histogram showing the prediction results by the present embodiment. [Figure 20] (A) shows the ROC curve of the conventional method, and (B) shows the ROC curve of this embodiment. [Figure 21] The following shows the results of PT measurement and discrimination for blood samples administered warfarin, samples administered DOACs, and samples with reduced liver function. (A) shows the ROC curve for blood samples administered warfarin. (B) shows the ROC curve for blood samples administered DOACs. (C) shows the ROC curve for samples with reduced liver function. [Figure 22] The following shows the results of APTT measurement and differentiation for FVIII inhibitor-positive blood samples, FVIII-deficient blood, LA-positive blood samples, heparinized blood samples, and DOACs-administered blood samples. (A) shows the ROC curve for the FVIII inhibitor-positive blood sample. (B) shows the ROC curve for the FVIII-deficient blood sample. (C) shows the ROC curve for the LA-positive blood sample. (D) shows the ROC curve for the heparinized blood sample. (E) shows the ROC curve for the DOACs-administered blood sample. [Diagram 23] The following shows the results of PT and APTT measurements for FVIII inhibitor-positive blood samples, LA-positive blood samples, blood samples with reduced liver function, heparinized blood samples, and DOACs-administered blood samples. (A) shows the ROC curve for the FVIII inhibitor-positive blood sample. (B) shows the ROC curve for the LA-positive blood sample. (C) shows the ROC curve for the blood sample with reduced liver function. (D) shows the ROC curve for the heparinized blood sample. (E) shows the ROC curve for the DOACs-administered blood sample.

[0010] 1.Analyzer The analysis device of this embodiment (hereinafter simply referred to as "analysis device 1") will be described with reference to Figs. 1 to 14. 1-1. Hardware configuration of the analyzer The analyzer 1 is an apparatus that irradiates light onto a measurement sample prepared by adding a coagulation measurement reagent to a blood sample, detects the transmitted light of the light irradiated onto the measurement sample, and analyzes the blood sample based on the detected light. An example of the external appearance of the analyzer 1 of this embodiment is shown in Fig. 1. The analyzer 1 includes a measurement unit 2 for acquiring detection information, and a touch panel type display 4 that allows data input. An example of the hardware configuration of the analyzer 1 is shown in Fig. 2.

[0011] The measurement unit 2 of the analytical device 1 includes a control unit 201, a memory unit 202, a light irradiation unit 10, a sample preparation unit 20, a detector 230, an input interface (I / F) 206, an output interface (I / F) 207, a communication interface (I / F) 208, and a bus 209.

[0012] The control unit 201 includes a processor such as a central processing unit (CPU) or a field-programmable gate array (FPGA).

[0013] The memory unit 202 stores a measurement program 202a for controlling the measurement operation by the measurement unit 2, an analysis program 202b, an algorithm database (DB) 202c storing one or more deep learning algorithms 60, a reference value / threshold database (DB) 202d storing reference values ​​of the blood clotting time for each blood clotting parameter and a threshold value of the probability that a candidate cause of the prolongation of the blood clotting time is the cause of the prolongation of the blood clotting time, and an additional test database (DB) 202e storing information on additional tests.

[0014] The input interface 206 accepts input information input by an operator from the input unit 411 of the display 4 , and transmits it to the control unit 201 or the storage unit 202 .

[0015] The output interface 207 transmits the output information output by the control unit 201 to the output unit 412 of the display 4 .

[0016] The communication interface 208 communicatively connects the measurement unit 2 to the network 99. The connection may be wired or wireless. Signals within the measurement unit 2 are transmitted via a bus 209 .

[0017] A configuration example of the light irradiation unit 10 is shown in Fig. 3. In the configuration example of Fig. 3, the light irradiation unit 10 includes five light sources 320, five optical fiber units 330 provided corresponding to the five light sources 320, and one holding member 340 for holding each light source 320 and an incident end 331 of each optical fiber unit 330. The light sources 320, the optical fiber units 330, and the holding member 340 are housed in a housing 310 made of, for example, a metal.

[0018] All five light sources 320 are composed of LEDs. LEDs generally have a lifespan several tens of times longer than that of halogen lamps. This makes it possible to configure a light irradiation unit 10 that is smaller in size and has a longer lifespan than a configuration using a wideband light source such as a halogen lamp and a rotary filter. In addition, since an individual LED can be provided for each wavelength, the emission spectrum and emission intensity of each light source 320 can be individually optimized. The light source 320 includes a first light source 321 , a second light source 322 and a third light source 323 .

[0019] In the configuration example of Fig. 3, the first light source 321 is a light source for measuring blood coagulation time that generates light of about 660 nm as a first wavelength, the second light source 322 is a light source that generates light of about 405 nm as a second wavelength, and the third light source 323 is a light source that generates light of about 800 nm as a third wavelength.

[0020] In the configuration example of Fig. 3, the multiple light sources 320 further include a fourth light source 324 for generating light of a fourth wavelength different from the second wavelength. The fourth wavelength is a wavelength selected from the range of 300 nm to 380 nm, similar to the second wavelength. More preferably, light in a wavelength band of 320 nm to 360 nm can be used. In the configuration example of Fig. 3, the fourth wavelength is, for example, 340 nm.

[0021] In the configuration example of Fig. 3, the multiple light sources 320 further include a fifth light source 325 for generating light of a fifth wavelength different from the third wavelength. The fifth wavelength is a wavelength selected from the range of 550 nm to 590 nm, similar to the third wavelength. More preferably, light in the wavelength band of 560 nm to 580 nm can be used. In the configuration example of Fig. 3, the fifth wavelength is, for example, 575 nm.

[0022] The optical fiber units 330 are provided corresponding to the respective light sources 320. The five optical fiber units 330 are composed of optical fiber units 330a, 330b, 330c, 330d, and 330e provided individually for each light source 320 so that light from a first light source 321, a second light source 322, a third light source 323, a fourth light source 324, and a fifth light source 325 enters from the respective entrance ends 331.

[0023] In the configuration example of FIG. 3, each of the optical fiber units 330 includes a plurality of optical fibers 333. The optical fiber units 330 are mixed and bundled so that the optical fibers 333 corresponding to each light source 320 are distributed approximately uniformly at the output end 332. Here, the term "optical fiber" means an optical fiber strand or optical fiber core having one core. Each optical fiber unit 330 is configured as a cable or a twisted wire in which a plurality of strands are bundled. With this configuration, the light of each wavelength separately incident on the input end 331 of each optical fiber unit 330 can be output from a common output end 332 instead of being individually irradiated to the container 15. Therefore, the configuration for outputting the light of each wavelength can be simplified.

[0024] Furthermore, since the light of each wavelength can be emitted with the distribution of the light being uniform at the common emission end 332, even when the light of each wavelength is emitted from the common emission end 332, it is possible to prevent the distribution of the light for each wavelength from becoming biased.

[0025] In the configuration example of FIG. 3, five optical fiber parts 330 are twisted together in the middle to be integrated, and are configured to have two output ends 332. The two output ends 332 are provided to correspond to the two detectors 230, respectively. The two output ends 332 are connected to two outlets 311 provided in the housing 310, respectively. Each output end 332 includes an approximately equal number of optical fibers 333 that constitute each optical fiber part 330. In addition, the optical fibers 333 that constitute each optical fiber part 330 are mixed so as to be distributed approximately uniformly within the end face of the output end 332. The number of optical fibers 333 that constitute each optical fiber part 330 is determined according to the number of container installation parts 231 in the detectors 230 and 240. For example, when the number of container installation parts 231 is N and each optical fiber part 330 transmits an amount of light equivalent to M optical fibers to one container installation part 231, each optical fiber part 330 includes N×M optical fibers 333. Each output end 332 is configured by collecting (N×M) / 2 optical fibers 333 from each optical fiber portion 330 .

[0026] In the configuration example of FIG. 3, the light irradiation unit 10 further includes a homogenizing member 350 that is disposed adjacent to the exit end 332 of the optical fiber unit 330 and homogenizes the intensity distribution of the light incident from the exit end 332 side and emits the light. Here, each optical fiber 333 disposed at the exit end 332 emits only light of any one of the first wavelength to the fifth wavelength. That is, at the exit end 332, the light emission points for each wavelength are uniformly distributed and disposed. Therefore, by homogenizing the emitted light by making it incident on the homogenizing member 350, the intensity distribution of each wavelength is homogenized over the entire surface at the exit surface 352 of the homogenizing member 350. This makes it possible to effectively homogenize the variation in light intensity for each wavelength.

[0027] The uniformizing members 350 are respectively disposed at two outlets 311 provided in the housing 310. The uniformizing members 350 have an incident surface 351 facing the corresponding output end 332 of the optical fiber section 330, and an output surface 352 disposed on the outlet side of the outlet 311. As a result, light having a uniform intensity distribution that has passed through the uniformizing member 350 is output from each outlet 311. The uniformizing member 350 is configured, for example, to cause multiple reflections inside the light incident from the incident surface 351 and output the light from the output surface 352. If the intensity distribution of light of each wavelength is sufficiently uniformed at the emission end 332 of the optical fiber section 330, the uniformizing member 350 does not need to be provided.

[0028] The holding member 340 of the light irradiation unit 10 holds five light sources 320. Therefore, the five light sources 320 are supported by a common holding member 340. The holding member 340 is made of a metal such as aluminum, and is formed in a prismatic shape. In the configuration example of FIG. 3, the light source holding part 341 and the incident end holding part 342 are provided at one end and the other end of the holding member 340, respectively, and are connected to each other by a passage part 344 formed of a through hole penetrating the holding member 340.

[0029] The five light source holding parts 341 are arranged in a straight line along a direction perpendicular to the light emission direction of each light source 320. In each light source 320, a fourth light source 324 is arranged in the center, a fifth light source 325 and a second light source 322 are arranged on both sides of the fourth light source 324, and a first light source 321 and a third light source 323 are arranged on the outermost sides.

[0030] 3, a plurality of light source holding parts 341 that hold the light sources 320 and a plurality of incident end holding parts 342 that hold the incident ends 331 of the optical fiber parts 330 are arranged at positions facing each other linearly in the holding member 340. This allows the optical axis of the light source 320 and the axial center of the optical fiber part 330 at the incident end 331 to be easily and accurately aligned. The light source holding part 341 and the incident end holding part 342 are arranged at positions facing each other on approximately the same axis.

[0031] In the configuration example of FIG. 3, the light source holding section 341 holds the light source 320 via the socket 343. The light source holding section 341 includes a recess 345 connected to the passage section 344, and the socket 343 is a cylindrical member fitted into the recess 345. The light source 320 is fixedly held inside the socket 343. The incident end holding section 342 is formed by the other end portion of the passage section 344 which is a through hole penetrating the holding member 340. Therefore, the incident end holding section 342 is a hole section into which the incident end 331 can be inserted, and holds a predetermined length range including the incident end 331 of the optical fiber section 330 inserted therein.

[0032] The light irradiation section 10 may be provided with a member for focusing the light from the light source 320 onto the incident end 331 of the optical fiber section 330, and a member for adjusting the spectral characteristics such as the central wavelength and half-width of the light incident on the incident end 331.

[0033] For example, the light irradiation unit 10 further includes an optical bandpass filter 360 that transmits only light of a predetermined wavelength band. The optical bandpass filter 360 has a disk-like shape, and transmits only light of a predetermined wavelength band out of the light irradiated to one surface to the other surface. The holding member 340 holds the optical bandpass filter 360 at a position between the light source 320 and the incident end 331 of the corresponding optical fiber unit 330. This allows the central wavelength, half-width, etc. of the light emitted from the light source 320 to be adjusted to have characteristics suitable for measurement, and the light can be incident on the incident end 331. As a result, the measurement accuracy is improved. In addition, there are individual differences in the light source 320, and the central wavelength, half-width, etc. may differ, but the optical bandpass filter 360 absorbs the influence of the individual differences of the light source 320, and a stable measurement result can be ensured.

[0034] Although an example of using an LED as the light source has been shown here, it is also possible to use, for example, a halogen lamp as the light source, split the light into a first wavelength to a fifth wavelength using a bandpass filter or the like, and irradiate each of the wavelengths onto the measurement sample.

[0035] A configuration example of the detector 230 is shown in FIG. 4. The detector 230 includes a container installation section 231 as a hole extending in the vertical direction, and an emission end 382 of the light distribution member 380 is disposed in a hole 233 extending laterally from the container installation section 231. A condenser lens 234 is disposed inside the hole 233. The light receiving section 11 is provided at the end of a hole 235 formed to face the hole 233 across the container installation section 231. As a result, the emission end 382 of the light distribution member 380, the condenser lens 234, the container installation section 231, and the light receiving section 11 are disposed in a linear arrangement. The light emitted from the emission end 382 passes through the condenser lens 234, passes through the container 15 in the container installation section 231 and the measurement sample in the container 15, and is detected by the light receiving section 11. The measurement sample includes a blood sample and a coagulation time measurement reagent, and is prepared in the container 15. Each of the light receiving sections 11 outputs a plurality of electric signals (digital data) according to the intensity of the received light as detection information.

[0036] Returning to FIG. 2, the sample preparation section 20 includes a dispensing mechanism for dispensing a blood sample and a coagulation time-measuring reagent into the container 15.

[0037] The light irradiation unit 10, the sample preparation unit 20 and the detection unit 230 may be, for example, the light irradiation unit, the sample preparation unit and the detection unit described in U.S. Pat. No. 10,048,249, which is incorporated herein by reference.

[0038] 1-2. Measurement and analysis processing 5 shows a flow of the measurement and analysis process by the control unit 201. In step S1, the control unit 201 executes the measurement process based on the measurement program 202a. Next, in step S2, the control unit 201 executes the analysis process based on the analysis program 202b.

[0039] 1-3. Processing of measurement programs FIG. 6 shows the flow of the measurement process executed by the control unit 201 based on the measurement program 202a. In step S11, the control unit 201 acquires information on blood coagulation parameters (analysis items) ordered for each blood sample. The information on blood coagulation parameters may be received as information input by an operator via the display 4. Alternatively, the information may be acquired via a network, for example, from an electronic medical record system of a medical facility. The blood sample and the information on blood coagulation parameters can be linked together by an identifier issued when the test is requested, for example.

[0040] The blood coagulation parameters (analysis items) targeted by the analyzer 1 may include at least one selected from the group consisting of activated partial thromboplastin time (hereinafter sometimes abbreviated as "APTT"), prothrombin time (hereinafter sometimes abbreviated as "PT"), thrombotest, fibrinogen, factor 2 activity, factor 5 activity, factor 7 activity, factor 8 activity, factor 9 activity, factor 10 activity, factor 11 activity, factor 12 activity, and whole blood clotting time. The blood coagulation parameters may preferably include at least one selected from the group consisting of APTT and PT.

[0041] In step S12, the control unit 201 controls the measurement unit 2 to dispense the blood sample into the container 15. At this time, a buffer or the like for diluting the blood sample may be dispensed into the container 15.

[0042] In step S13, the control unit 201 controls the sample preparation unit 20 of the measurement unit 2 to dispense a coagulation time measurement reagent corresponding to each blood sample and prepare a measurement sample. At this time, if the addition of a coagulation activation reagent is required, the coagulation activation reagent is also dispensed. The coagulation time measurement reagent can be appropriately selected according to each blood coagulation parameter. A commercially available reagent can be used as the coagulation time measurement reagent.

[0043] For example, when measuring APTT, a test reagent for APTT measurement can be used as the coagulation time measurement reagent, which may contain an activator such as silica, ellagic acid, celite, etc., and animal-derived, plant-derived, or artificially synthesized phospholipids, etc. Examples include Thrombocheck APTT series manufactured by Sysmex Corporation, Coagpia (registered trademark) APTT-N, etc. manufactured by Sekisui Medical Co., Ltd., and Dataphi APTT, etc. manufactured by Siemens Healthcare Diagnostics Products GmbH. The coagulation activator reagent is a reagent capable of supplying calcium ions. According to the international standard method, the coagulation activator reagent is a 20 mM calcium chloride solution.

[0044] When measuring PT, a PT measuring test reagent containing thrombin can be used as a coagulation time measuring reagent. For example, Thrombocheck PT series manufactured by Sysmex Corporation and Coagpia (registered trademark) PT series manufactured by Sekisui Medical Co., Ltd. can be exemplified. PT measuring test reagents generally contain calcium ions necessary for activating coagulation.

[0045] In step S14, the control unit 201 controls the light irradiating unit 10 to start irradiating light onto the container 15 in which the measurement sample has been prepared in step S13. The light receiving unit 11 continuously outputs an electric signal (digital data) according to the intensity of the light received through the container 15 as detection information.

[0046] 1-4. Processing of analysis programs FIG. 7 shows the flow of the analysis process executed by the control unit 201 based on the analysis program 202b. In step S21, the control unit 201 acquires a data group including a plurality of pieces of data constituting a blood coagulation curve. Specifically, the control unit 201 arranges a plurality of pieces of data (digital data) corresponding to the received light intensity output from the light receiving unit 11 in a time series, and stores the data in the storage unit 202.

[0047] The control unit 201 acquires the detection information from the light receiving unit 11 over time, for example, every 0.1 to 0.5 seconds, preferably every 0.1 seconds, and stores it in the storage unit 202. The control unit 201 stores a plurality of data in the storage unit 202, for example, from the time when the blood sample and a predetermined coagulation time measuring reagent are added. In general, the blood sample and the coagulation time measuring reagent are added, and then the coagulation activation reagent is added. In this case, the control unit 201 starts storing the plurality of data from the time when the coagulation activation reagent is added. Alternatively, in the case where the coagulation activation reagent is mixed with the coagulation time measuring reagent, the control unit 201 starts storing the plurality of data from the time when the blood sample and the coagulation time measuring reagent are mixed. The control unit 201 ends storing the plurality of data when no change is observed in the magnitude of the plurality of data acquired over time. Note that the timing of starting and ending the storage of the plurality of data is not limited to this. For example, the storage may start when the light irradiation to the container 15 is started, and the storage may end after a predetermined time (for example, 120 seconds or 180 seconds) has elapsed since the start of the light irradiation.

[0048] The data group stored in the storage unit 202 by the control unit 201 may be a set of data obtained by excluding some of the data from the plurality of data output from the light receiving unit 11. The method for excluding some of the data may involve excluding data from a certain period, such as immediately after the start or end of light irradiation, or excluding data at a predetermined frequency. Examples of the predetermined frequency include excluding the next data set after acquiring data a certain number of times, excluding even-numbered data, excluding the next data set after acquiring data a certain number of times, etc.

[0049] Furthermore, the control unit 201 may arrange each data in the data group stored in the storage unit 202 in chronological order, or in order other than chronological order, such as by size.

[0050] The plurality of data stored in the memory unit 202 constitute a blood coagulation curve. The blood coagulation curve will be described in more detail with reference to FIG. 8(A). In this example, the plurality of data indicate the intensity of light (hereinafter referred to as transmitted light intensity) transmitted through the measurement sample when the measurement sample is irradiated with light. In FIG. 8(A), the vertical axis (Y axis) indicates the transmitted light intensity, and the horizontal axis (X axis) indicates the measurement time (seconds: sec) during which the transmitted light intensity is monitored. The blood coagulation curve can be generated by plotting the change over time of the monitored transmitted light intensity on two axes, the transmitted light intensity and the measurement time. Point I in FIG. 8(A) is the time when calcium ions, which are the coagulation activation reagent, and the coagulation time measurement reagent are added to the test sample, and is also the time when the measurement starts (tI). At the start of the measurement, the fibrinogen in the measurement sample has not changed to fibrin, and the measurement sample does not yet undergo fibrin precipitation, so the transmitted light intensity shows a high value. Thereafter, as the coagulation reaction progresses and fibrin begins to precipitate, the precipitated fibrin blocks the light and the transmitted light intensity begins to decrease. This point is point II in Figure 8(A) and is the start of coagulation. The measurement time when coagulation begins is represented as (tII). As the reaction progresses and fibrin precipitation progresses, the transmitted light intensity decreases. When most of the fibrinogen in the test sample is converted to fibrin, the reaction converges and the change in transmitted light intensity plateaus. This point is point III in Figure 8(A) and is the end of coagulation. The measurement time when coagulation ends is represented as (tIII).

[0051] In the above, an example in which the measurement sample is irradiated with light of one wavelength has been described, but the measurement sample may be irradiated with light of multiple wavelengths. For example, the measurement sample may be irradiated with light of a first wavelength by the first light source 321 to obtain a first plurality of data, the same measurement sample may be irradiated with light of a second wavelength by the second light source 322 to obtain a second plurality of data, the same measurement sample may be irradiated with light of a third wavelength by the third light source 323 to obtain a third plurality of data, the same measurement sample may be irradiated with light of a fourth wavelength by the fourth light source 324 to obtain a fourth plurality of data, and the same measurement sample may be further irradiated with light of a fifth wavelength by the fifth light source 325 to obtain a fifth plurality of data. In this case, each of the first plurality of data, the second plurality of data, the third plurality of data, the fourth plurality of data, and the fifth plurality of data may constitute a blood coagulation curve.

[0052] In the above embodiment, the multiple data are obtained based on the transmitted light intensity, but the multiple data can be obtained by any of the optical measurement methods such as the scattered light method and the transmitted light method, the physical method using magnetism to measure the viscosity of fibrin during precipitation, and the dry hematology method. In the optical measurement method, the multiple data are represented by signals indicating the amount of light such as the transmitted light and scattered light irradiated on the measurement sample. In the physical method, the multiple data are represented by signals indicating the amplitude of the steel ball depending on the viscosity of the measurement sample.

[0053] 7, the control unit 201 performs pre-processing on the data group including a plurality of data items. The pre-processing includes at least one of smoothing, sharpness, caving, normalization, and differentiation.

[0054] Normalization is intended to convert each data included in the data group into a relative value so that the vertical axis of the blood coagulation curve is expressed between 0% (L1: baseline) and 100% (L2) as shown in FIG. 8(B). The normalized value of each data is also called a relative value. A blood coagulation curve plotted by relative values ​​is also called a normalized coagulation curve. Normalization can be achieved, for example, by assuming that the change in transmitted light intensity (dH), which is the difference between the transmitted light intensity at point II, which is the start of the coagulation reaction, and the transmitted light intensity at point III, which is the end of coagulation, is 100%, and converting the change in transmitted light intensity into a relative value. A plurality of data converted into relative values ​​constitutes a blood coagulation curve.

[0055] When a data group is acquired using light of a plurality of wavelengths, a data group converted into relative values ​​is generated for each wavelength.

[0056] In the normalized clotting curve, the clotting time can be set as the time when the change in transmitted light intensity (dH) is, for example, 30%, 40%, 50%, or 60%. In a preferred embodiment, the clotting time is the time when the change in transmitted light intensity (dH) is 50% (L3).

[0057] When a data group including a plurality of data is subjected to differentiation processing, the plurality of data included in the data group form a differential curve. The differential curve includes, for example, a first differential coagulation curve and a second differential coagulation curve.

[0058] The first derivative coagulation curve is shown in Figure 9(A). The first derivative coagulation curve is obtained by performing first derivative processing on the blood coagulation curve shown in Figure 8(A) or the blood coagulation curve shown in Figure 8(B). The vertical axis (Y axis) data of each coordinate that constitutes the first derivative coagulation curve is also called the first derivative value.

[0059] When multiple data sets are acquired using multiple wavelengths of light, a first derivative coagulation curve is generated for each wavelength.

[0060] The second derivative coagulation curve is shown in Fig. 9(B). The second derivative coagulation curve is obtained by performing second derivative processing on the blood coagulation curve shown in Fig. 8(A) or the blood coagulation curve shown in Fig. 8(B).

[0061] When multiple data sets are acquired using multiple wavelengths of light, a second derivative coagulation curve is generated for each wavelength.

[0062] In step S23 shown in FIG. 7, the control unit 201 acquires the coagulation time (the time when the change in transmitted light intensity (dH) becomes 50% (L3)) based on the normalized coagulation curve generated in step S22.

[0063] In this embodiment, step S22 can be omitted. In this case, the clotting time can be calculated based on the blood clotting curve of FIG. 8(A) according to clotting time (sec)=[(tIII)-(tII)] / 2. Here, "-" indicates subtraction and " / " indicates division.

[0064] In step S24 shown in FIG. 7, the control unit 201 judges whether the blood clotting time has been prolonged. The reference value of the blood clotting time is stored in advance in the reference value / threshold value 202d according to the blood clotting parameter and the clotting time measuring reagent. The control unit 201 reads out the reference value of the blood clotting time from the reference value / threshold value DB 202d and compares it with the clotting time acquired in step S23. If the clotting time exceeds the reference value ("YES" in step S24), the control unit 201 judges that the blood clotting time has been prolonged and proceeds to step S25. If the clotting time does not exceed the reference value ("NO" in step S24), the control unit 201 judges that the blood clotting time has not been prolonged and returns the process to the measurement and analysis process shown in FIG. 5.

[0065] If the clotting time exceeds the reference value ("YES" in step S24), the control unit 201 proceeds to step S25 shown in FIG. 7 and reads out the deep learning algorithm 60 stored in the algorithm DB 202c of the storage unit 202. The deep learning algorithm 60 is selected from a plurality of deep learning algorithms 60 according to the blood clotting parameters (analysis items) of each blood sample and read out. The control unit 201 inputs a data group (such as a normalized clotting curve, a first derivative clotting curve, or a second derivative clotting curve) including the plurality of data acquired in step S22 to the read deep learning algorithm 60 and acquires a result. Note that step S22 may be omitted, and the data group acquired in step S21 may be input to the deep learning algorithm 60 in step S25.

[0066] A method for generating deep learning algorithm 60 (a method for training deep learning algorithm 50) and a group of data input to deep learning algorithm 50 and deep learning algorithm 60 will be described below.

[0067] i. Training the deep algorithm The deep learning algorithm is not limited as long as it is an algorithm having a neural network structure, and may include a convolutional neural network, a fully connected neural network, and a combination thereof. Figure 10 shows an overview of the training of the deep learning algorithm.

[0068] For training data for training the deep learning algorithm 50, a data group including a plurality of data obtained from a blood sample whose cause of blood clotting time extension is known is used. These data groups are generated according to the method described in steps S21 and S22 shown in FIG. 7 and used as first training data. As shown in FIG. 10, the data group includes a first data group D1 corresponding to wavelength 1, a second data group D2 corresponding to wavelength 2, a third data group D3 corresponding to wavelength 3, a fourth data group D4 corresponding to wavelength 4, and a fifth data group D5 corresponding to wavelength 5. The data group D1 includes a plurality of data d1, d2, d3, etc. Similarly, the data groups D2 to D5 include a plurality of data. The first training data is input to the input layer 50a of the neural network 50 shown in FIG. 10, and a label indicating the cause of blood clotting time extension corresponding to the input first training data (in the example of FIG. 10, "Factor VIII deficiency") is input to the output layer 50b as second training data. With these inputs, the first training data and the second training data are linked for each blood sample, the weights of each layer in the intermediate layer 50c of the neural network 50 are calculated, and a trained deep learning algorithm 60 shown in FIG. 11 is generated.

[0069] The deep learning algorithm 50 may be trained for each blood coagulation parameter, or may be trained for a combination of multiple blood coagulation parameters, such as a first blood coagulation parameter and a second blood coagulation parameter. For example, the deep learning algorithm 50 may be trained using only a data group including multiple data obtained by APTT measurement from a blood sample whose cause of blood coagulation time extension is known. In this case, the trained deep learning algorithm 60 becomes a deep learning algorithm 60 that performs analysis based on a data group including multiple data derived from APTT. The deep learning algorithm 50 may be trained using only a data group including multiple data obtained by PT measurement from a blood sample whose cause of blood coagulation time extension is known. In this case, the trained deep learning algorithm 60 becomes a deep learning algorithm 60 that performs analysis based on a data group including multiple data derived from PT. Furthermore, the deep learning algorithm 50 may be trained using a data group including multiple data obtained by ATT measurement, which is the first blood coagulation parameter, and a data group including multiple data obtained by PT measurement, which is the second blood coagulation parameter. In this case, the trained deep learning algorithm 60 becomes a deep learning algorithm 60 that performs analysis based on a data group including multiple data obtained by APPT measurement and a data group including multiple data derived from PT.

[0070] When multiple blood coagulation parameters to be trained are combined to form the first training data, for example, a data group including multiple data derived from a first blood coagulation parameter can be arranged followed by a data group including multiple data derived from a second blood coagulation parameter, and the resulting data can be input as the first training data, as shown in Figure 12.

[0071] As shown in FIG. 12, the first training data can be input as follows: a data group including a plurality of data at a first wavelength derived from a first blood coagulation parameter (e.g., APTT) is followed by a data group including a plurality of data at a second wavelength, a third wavelength, a fourth wavelength, and a fifth wavelength, and a data group including a plurality of data at a first wavelength derived from a second blood coagulation parameter (e.g., PT) is followed by a data group including a plurality of data at the second wavelength, a third wavelength, a fourth wavelength, and a fifth wavelength.

[0072] The first training data and the second training data are each generated from a plurality of blood samples for which the causes of prolonged blood clotting time are known, and are used to train the deep learning algorithm 50.

[0073] ii. Analysis by trained deep learning algorithms A data group including a plurality of data constituting a blood clotting curve obtained from a blood sample to be analyzed is input as analysis data to the input layer 60a of the trained deep learning algorithm 60 in Fig. 11. The blood sample to be analyzed may be collected from a patient who is not known to have a prolonged blood clotting time, or may be collected from a patient who is known to have a prolonged blood clotting time.

[0074] The analysis data is preferably a data group including a plurality of data having the same blood coagulation parameters and the same configuration as the first training data used when training the input deep learning algorithm 60. For example, as shown in FIG. 12, the data group includes a first data group D1 corresponding to wavelength 1, a second data group D2 corresponding to wavelength 2, a third data group D3 corresponding to wavelength 3, a fourth data group D4 corresponding to wavelength 4, and a fifth data group D5 corresponding to wavelength 5. The data group D1 includes a plurality of data d11, d12, d13, etc. Similarly, the data groups D2 to D5 include a plurality of data. The first training data and the analysis data having the same configuration means that the analysis data is a data group including a plurality of data acquired with the same wavelength of light as the first training data. In addition, when some data is removed from the data group including a plurality of data used in the first training data, it is preferable that the analysis data is also subjected to the same processing. Furthermore, when the first training data is subjected to preprocessing, it is preferable that the analysis data is also subjected to the same preprocessing.

[0075] When there are multiple deep learning algorithms 60 corresponding to blood coagulation parameters (analysis items), the deep learning algorithm 60 to be input can be selected from the multiple deep learning algorithms depending on the type of blood coagulation parameter.

[0076] When predicting the cause of prolongation based on multiple blood coagulation parameters, for example, as shown in FIG. 12, a data group including a plurality of data of a first wavelength derived from a first blood coagulation parameter (e.g., APTT) is followed by a data group including a plurality of data of each of a second wavelength, a third wavelength, a fourth wavelength, and a fifth wavelength, and then a data group including a plurality of data of a first wavelength derived from a second blood coagulation parameter (e.g., PT) is followed by a data group including a plurality of data of each of the second wavelength, the third wavelength, the fourth wavelength, and the fifth wavelength, and the data can be input to the deep learning algorithm 60.

[0077] The deep learning algorithm 60 outputs a result ("Factor VIII deficiency; 80%" in FIG. 11) from the output layer 60c. The result may include a label indicating the cause of the prolonged blood clotting time ("Factor VIII deficiency" in the example of FIG. 11) and the probability of belonging to that label ("80%" in FIG. 5).

[0078] Here, the label indicating the cause of the prolongation of blood clotting time input as the second training data and the label indicating the cause of the prolongation of blood clotting time output as the result may be text information or a label value.

[0079] The causes of prolonged blood clotting time output from the output layer 60c of the deep learning algorithm 60 may include at least one selected from the group consisting of liver disease; disseminated intravascular coagulation (hereinafter sometimes abbreviated as "DIC"); vitamin K (hereinafter sometimes abbreviated as "VK") deficiency; heavy bleeding; reduction, deficiency or abnormal function of clotting factors; presence of clotting factor inhibitors; presence of lupus anticoagulant (hereinafter sometimes abbreviated as "LA"); use of anticoagulants; presence of abnormal proteins such as macroglobulinemia; and those resulting from blood drawing procedures.

[0080] The decrease, deficiency or functional abnormality of a coagulation factor may include a decrease, deficiency or functional abnormality of at least one selected from the group consisting of fibrinogen (hereinafter sometimes abbreviated as "Fbg"), factor 2 (hereinafter sometimes abbreviated as "FII"), factor 5 (hereinafter sometimes abbreviated as "FV"), factor 7 (hereinafter sometimes abbreviated as "FVII"), factor 8 (hereinafter sometimes abbreviated as "FVIII"), von Willebrand factor (hereinafter sometimes abbreviated as "VWF"), factor 9 (hereinafter sometimes abbreviated as "FIX"), factor 10 (hereinafter sometimes abbreviated as "FX"), factor 11 (hereinafter sometimes abbreviated as "FXI"), factor 12 (hereinafter sometimes abbreviated as "FXII"), HMWK (High Molecular Weight Kininogen), and prekallikrein.

[0081] The coagulation factor inhibitor may include at least one selected from the group consisting of a factor 5 inhibitor, a factor 8 inhibitor, a von Willebrand factor inhibitor, and a factor 9 inhibitor.

[0082] Anticoagulants, also known as antithrombotic drugs, may include coumarin-based drugs such as warfarin potassium; heparin; synthetic Xa inhibitors such as fondaparinux sodium; oral direct Xa inhibitors such as edoxaban tosilate hydrate and apixaban; oral direct thrombin inhibitors such as dabigatran etexilate methanesulfonate; and antithrombin drugs such as argatroban hydrate.

[0083] When the predetermined blood coagulation parameter is thrombotest or whole blood clotting time, the anticoagulant may include an antiplatelet drug, which may include ticlopidine hydrochloride, clopidogrel sulfate, prasugrel hydrochloride, ticagrelor, clopidogrel sulfate-aspirin combination, cilostazol, ethyl icosapentate, beraprost sodium, sulfogrelate hydrochloride, aspirin-dialuminate combination, aspirin, etc.

[0084] Examples of blood collection technique-related problems include contamination with heparin during line blood collection, blood collection after dialysis, etc.; when the amount of blood collected is small compared to the proportion of anticoagulant already loaded in the blood collection tube; when blood vessels are thin and difficult to collect blood from, resulting in contamination with tissue fluid during collection; and when too much time has passed since blood collection.

[0085] The cause of the prolongation of the clotting time output from the output layer 60c of the deep learning algorithm 60 is preferably at least one selected from the group consisting of causes related to the prolongation of the activated partial thromboplastin time and causes related to the prolongation of the prothrombin time.

[0086] 7, and outputs information regarding the cause of the prolongation of the blood coagulation time to the display 4 or the network 99 based on the result output from the deep learning algorithm 60. The output in step S26 may be performed after the control unit 201 receives an output start request input from the display 4 by the operator.

[0087] The information on the cause of the prolongation of blood clotting time may include at least a label indicating a possible cause of the prolongation of blood clotting time. Preferably, the possible cause indicated by the label may include a probability that the possible cause is the cause of the prolongation of blood clotting time. The label indicating the possible cause of the prolongation of blood clotting time may be text information or a label value. In the case of text information, it may include a notation using an abbreviation such as FIX, FVIII, etc. In the case of a label value, it may include a notation using a symbol such as 1, 2, etc. that is previously associated with a predetermined prolongation cause.

[0088] The probability that the candidate cause is the cause of the prolongation of the blood clotting time may be expressed in a softmax format or in a binary format. The softmax format is a format that indicates the probability that a specific candidate cause is the cause of the prolongation of the blood clotting time among multiple candidate causes of the prolongation of the blood clotting time. For example, when the prolongation causes output from the deep learning algorithm 60 include only deficiency of factor 8 and deficiency of factor 9 as causes with a probability greater than 0%, the probability of each prolongation cause is indicated so that the sum of the two is 100%. The binary format is a format that indicates the probability that each of multiple candidate causes of the prolongation is the cause of the prolongation of the blood clotting time, and the sum of each candidate cause of the prolongation is not necessarily 100%. The probability that the candidate cause is the cause of the prolongation of the blood clotting time may be output in a graph format as shown in, for example, FIG. 13(A) or FIG. 13(B). FIG. 13(A) is an example of the probability indicated in the softmax format being output in a pie chart. FIG. 13(B) is an example of the probability indicated in the binary format being output in a bar graph.

[0089] When outputting information on the cause of prolongation of blood clotting time, a label indicating the most probable candidate cause of prolongation of blood clotting time may be output, for example. In this case, for example, in the examples of Figures 13(A) and 13(B), a label indicating FVIII deficiency with the most probabilities is output.

[0090] When outputting information on the cause of prolongation of blood clotting time, for example, a label indicating a candidate cause of prolongation of blood clotting time whose probability is equal to or higher than a predetermined threshold may be output. The threshold may be, for example, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, or 90%. When the threshold is set to, for example, 20%, in the example of Figures 13(A) and 13(B), FIX deficiency and FVIII deficiency are output as the cause of prolongation, and when the threshold is set to, for example, 60%, FVIII deficiency is output as the cause of prolongation.

[0091] When using a threshold value when outputting information on the cause of prolongation of the blood coagulation time, the control unit 201 can call it from the reference value / threshold value DB 202d stored in the storage unit 202. The threshold value may be input by the operator via the display 4, and the control unit 201 may accept it.

[0092] In step S27 shown in FIG. 7, the control unit 201 reads information about the additional test from the additional test DB 202f stored in the storage unit 202, and outputs it from the display 4. The information about the additional test is output based on the result output from the deep learning algorithm 60 in step S25. The additional test is at least one selected from the group consisting of a test for a coagulation factor, a test for a coagulation factor inhibitor, and a retest of the measurement item performed on the measurement sample. The additional test is intended to be a confirmation test or a retest of the result output from the deep learning algorithm 60.

[0093] For example, if the deep learning algorithm 60 outputs that the cause of prolonged blood coagulation is most likely to be liver disease, additional tests may be performed, such as biochemical tests to evaluate liver function (AST, ALT, γ-GTP, LDH, ALP, etc.).

[0094] If the deep learning algorithm60 indicates that disseminated intravascular coagulation is most likely to be the cause of prolonged blood coagulation, additional tests may include measurement of the platelet count, blood fibrinogen concentration, FDP·D-dimer, blood thrombin-antithrombin complex concentration, and fibrinolysis markers such as plasmin-α2 plasmin inhibitor complex.

[0095] If the deep learning algorithm 60 outputs that vitamin K deficiency is the most likely cause of prolonged blood clotting, then measuring blood vitamin K levels may be an additional test.

[0096] When the deep learning algorithm 60 outputs that the cause of prolonged blood clotting is most likely to be heavy bleeding, additional tests may include measuring the red blood cell count, hemoglobin concentration, hematocrit value, and platelet count.

[0097] When the deep learning algorithm 60 outputs that the most common cause of prolonged blood coagulation is a decrease, deficiency, or abnormal function of a specific coagulation factor, additional tests may include measuring the activity of the specific coagulation factor (including blood coagulation method, synthetic substrate method, etc.) and measuring the protein concentration of the specific coagulation factor by enzyme immunoassay, etc.

[0098] When the deep learning algorithm 60 outputs that the causes of prolonged blood coagulation are most likely to be the presence of a specific coagulation factor inhibitor, additional tests may include detection of the specific coagulation factor inhibitor, for example, detection of anticoagulant factor antibodies by enzyme immunoassay, mixing tests, cross-mixing tests, etc.

[0099] When the deep learning algorithm 60 outputs that the presence of lupus anticoagulant is the most likely cause of prolonged blood coagulation, additional tests may include, for example, detection of antiphospholipid antibodies by enzyme immunoassay or a cross-mixing test.

[0100] If the deep learning algorithm 60 outputs that the cause of prolonged blood clotting is the presence of an abnormal protein, serum protein analysis may be an additional test.

[0101] If the deep learning algorithm 60 outputs that anticoagulants are the most likely cause of prolonged blood clotting, additional testing may involve checking the medication history of the subject from whom the blood sample was taken.

[0102] If the deep learning algorithm 60 outputs that the cause of prolonged blood clotting is most likely due to blood manipulation, additional testing may include re-examination, including re-collection of a blood sample.

[0103] The information regarding the additional examination may be text information indicating the name of the additional examination or a label value.

[0104] In addition, when outputting information on additional examinations, in addition to the additional examination corresponding to the cause of extension having the highest probability, additional examinations corresponding to the cause of extension having the second or subsequent probability may be output. In this case, the priority of each additional examination may be output. It should be noted that this step S27 can be omitted.

[0105] Furthermore, when a more detailed cause of the extension of the blood clotting time is identified by an additional test, the operator may input the identified cause of the extension from the display 4 in association with the identification information of the blood sample on which the additional test was performed. The input cause of the extension may be stored in the additional test DB 202f of the storage unit 202 in association with the identification information of the blood sample and a data group including a plurality of data on the blood sample. FIG. 14 shows an example of the identification information, data group, and cause of the extension of the blood sample stored in the additional test DB 202f. The identification information, data group, and cause of the extension of the blood sample stored in the additional test DB 202f may be transmitted to another computer on the network 99, for example, a training computer for further training the deep learning algorithm 60, via the communication I / F 208. This allows the training computer to further train the deep learning algorithm 60 based on the received information, and the analysis ability of the deep learning algorithm 60 can be further improved.

[0106] 2.Training equipment 2-1. Training device configuration An embodiment disclosed in this specification relates to a training device 5 for a deep learning algorithm 50 (hereinafter, simply referred to as "training device 5"). An example of the external appearance of the training device 5 is shown in FIG. 15(A). The training device 5 is communicatively connected to an input unit 511 and an output unit 512. The training device 5 is a so-called general-purpose computer.

[0107] 15(B) shows an example of the hardware configuration of the training device 5. The training device 5 includes a control unit 501, a storage unit 502, an input interface (I / F) 506, an output interface (I / F) 507, a communication interface (I / F) 508, a media interface (I / F) 509, and a bus 510. The control unit 501 includes a processor such as a CPU.

[0108] The memory unit 502 stores a training program 502b described below, an algorithm database (DB) 502c that stores one or more deep learning algorithms 50 and / or deep learning algorithms 60, and a training data database (DB) 502d that stores first training data and second training data in association with each other.

[0109] The input interface 506 accepts input information input by an operator from an input unit 511 constituted by a touch panel, a keyboard, or the like, and transmits the information to the control unit 501 or the storage unit 502 .

[0110] The output interface 507 transmits the output information output by the control unit 501 to an output unit 512 such as a display.

[0111] The communications interface 508 communicatively connects the training device 5 to the network 95. The connection may be wired or wireless.

[0112] The media interface 509 transmits and receives information to and from non-volatile storage media such as a CD-ROM, a DVD-ROM, and an external hard disk. The transmission of signals within the training device 5 is carried out via a bus 510 .

[0113] 2-2. Processing of training programs FIG. 16 shows an example of the training process executed by the training program 502b.

[0114] In step S51, the control unit 501 accepts a processing start request input by the operator from the input unit 511, and calls up training data from the training data DB 502d.

[0115] Next, in step S52, the control unit 501 calls the training data for the deep learning algorithm 50 or the deep learning algorithm 60 from the algorithm DB 502c in the storage unit 502, inputs the training data to these, and performs training. Details of this training are as described in "i. Training of the deep algorithm". Inputting the training data to the deep learning algorithm 60 and training it is also called retraining of the deep learning algorithm 60.

[0116] Next, in step S53, the control unit 501 determines whether or not the deep learning algorithm 50 or the deep learning algorithm 60 has been trained using all the training data to be used. If training has been performed using all the training data (if "YES" in step S53), the control unit 501 proceeds to step S54 and stores the trained deep learning algorithm 60 in the algorithm DB 502c stored in the storage unit 202.

[0117] In step S53, if training has not been performed using all the training data ("NO" in step S53), the process returns to step S51 to continue the training process.

[0118] 3. Recording medium storing analysis programs or training programs The analysis program 202b and / or the training program 502b may be provided as a program product such as a storage medium. The computer program is stored in a storage medium such as a hard disk, a semiconductor memory device such as a flash memory, an optical disk, etc. The storage format of the program in the storage medium is not limited as long as the control unit can read the program.

[0119] In this specification, the terms anticoagulant and anticoagulant drug are used separately. The anticoagulant is filled in a blood collection tube or syringe to prevent fibrin from precipitating during blood collection. For example, when blood is collected for use in a normal coagulation test, it is preferable to collect the blood using a citrate salt, such as a sodium citrate solution, as an anticoagulant. The blood sample is prepared by mixing the anticoagulant and whole blood at a volume ratio of about 1:8.5 to 1:9.5 using, for example, a 3.1% to 3.3% (weight / volume) trisodium citrate solution as the anticoagulant. Alternatively, the blood sample may be plasma separated from a mixture of the anticoagulant and whole blood. 4. Variations FIG. 17 shows an analysis device 1' as a modified example of the analysis device 1. The analysis device 1' is, for example, a computer of a cloud service provider, and is connected to a known blood analysis device 100 via a network 99. The analysis device 1' includes a control unit 201, a memory unit 202, an input interface (I / F) 206, an output interface (I / F) 207, a communication interface (I / F) 208, and a bus 209. The control unit 201, the input interface (I / F) 206, the output interface (I / F) 207, the communication interface (I / F) 208, and the bus 209 can be the same as those of the analysis device 1. The memory unit 202' can be obtained by removing the measurement program 201a from the memory unit 202 of the analysis device 1 and changing the analysis program 202b to an analysis program 202b'. 18 shows a flow of processing executed by the control unit 201 of the analyzer 1' based on the analysis program 202b'. In step S201, the control unit 201 receives a data group including a plurality of data constituting a blood coagulation curve and a blood coagulation time from the analyzer 100. The received data group is the same as the data group acquired in step S22 of FIG. 7. The received coagulation time is the same as the coagulation time acquired in step S23 of FIG. 7. Thereafter, the control unit 201 executes steps S204 to S207. Steps S204 to S207 are the same processes as steps S24 to S27 of FIG. 7, respectively, and therefore description thereof will be omitted. As a known blood analyzer 100, for example, fully automated blood coagulation measuring devices CN-6000 and CN-3000 manufactured by Sysmex Corporation can be used.

[0120] 5.Verification of effectiveness 5-1. Comparison with conventional methods The APTT was measured using a sample (plasma) in which the cause of APTT prolongation was the presence of lupus anticoagulant (LA) and a sample in which the cause of APTT prolongation was the presence of factor VIII inhibitor (FVIII inhibitor). Using the measurement results, the prediction accuracy of the conventional method (first derivative method) and the present analysis method executed by the analysis device 1 were compared. The results are shown in Figs. 19 and 20. Fig. 19(A) shows a histogram with the peak value (min1) in the first derivative curve of each sample on the horizontal axis. Fig. 19(B) shows a histogram with the probability that the cause of APTT prolongation is the presence of factor VIII inhibitor on the horizontal axis output by the present analysis method. In the conventional method, LA-positive samples and samples containing factor VIII inhibitors overlap at min1=1.5 to min1=3.5, and the two could not be separated. On the other hand, the present analysis method was able to separate LA-positive samples and samples containing factor VIII inhibitors. This demonstrated that the present analytical method can predict the cause of prolonged blood coagulation time for samples that could not be predicted by conventional methods (samples showing min1=1.5 to min1=3.5).

[0121] For the samples shown in Figure 19, Figure 20(A) shows the ROC curve for the conventional method. Figure 20(B) shows the ROC curve evaluating the predictive accuracy of this analytical method. The AUC for the conventional method was 0.752, and the AUC for this analytical method was 0.893. These results show that this analytical method has higher sensitivity and specificity.

[0122] 5-2.Evaluation of deep learning algorithms for PT Figure 21 shows ROC curves obtained when PT measurements were performed on 94 blood samples from patients with warfarin administration, 93 blood samples from patients with DOACs administration, and 39 blood samples from patients with impaired liver function, and this analysis method was performed using the analysis device 1. For the blood samples from patients with warfarin administration (Figure 21A) and the blood samples from patients with impaired liver function (Figure 21C), predictions were made with an accuracy of more than 0.80 AUC. For the blood samples from patients with DOACs administration (Figure 21B), predictions were made with an accuracy of more than 0.74 AUC, although this was lower than for the blood samples from patients with warfarin administration and the blood samples from patients with impaired liver function.

[0123] 5-3.Evaluation of deep learning algorithms for APTT 22 shows the ROC curves obtained when APTT measurements were performed on 24 factor VIII (FVIII) inhibitor-positive blood samples, 23 factor VIII (FVIII) deficient blood samples, 80 LA-positive blood samples, 70 heparin-administered blood samples, and 79 DOACs-administered blood samples, and the present analysis method was performed using the analysis device 1. In all cases, predictions were made with an accuracy of more than 0.80 AUC.

[0124] 5-4.Evaluation of deep learning algorithms for PT and APTT Figure 23 shows the ROC curves obtained when PT and APTT measurements were performed on 15 blood samples positive for factor VIII (FVIII) inhibitors, 51 blood samples positive for LA, 36 blood samples with impaired liver function, 60 blood samples administered heparin, and 56 blood samples administered DOACs, and the analysis method was performed using the analyzer 1. In all cases, predictions were made with an accuracy of more than 0.82 AUC. [Explanation of symbols]

[0125] 1 Analyzer 2 Measuring part 201 Control section

Claims

1. A method for analyzing a blood sample, comprising: A data group including a plurality of data constituting a blood coagulation curve or a derivative curve thereof is obtained, and the plurality of data included in the data group are obtained at predetermined intervals over time from the start to the end of a coagulation reaction; inputting the data set into a deep learning algorithm including a convolutional neural network; outputting information regarding the cause of the prolongation of the blood clotting time in the blood sample based on the results obtained from the deep learning algorithm; The analytical method.

2. The method according to claim 1 , wherein the plurality of data included in the data group constitutes the blood coagulation curve.

3. The method according to claim 1 , wherein the plurality of data included in the data group constitute a first derivative curve or a second derivative curve of the blood coagulation curve.

4. The analysis method according to claim 1 , wherein the plurality of data included in the data group are obtained by optical measurement.

5. 5. The analytical method according to claim 4, wherein the data group includes a plurality of data obtained by irradiating a measurement sample containing the blood specimen and a coagulation time measurement reagent with light of a first wavelength, and a plurality of data obtained by irradiating the measurement sample with light of a second wavelength different from the first wavelength.

6. 6. The method of claim 1, wherein the data sets comprise a first data set for a first blood clotting parameter and a second data set for a second blood clotting parameter.

7. The analytical method according to any one of claims 1 to 6, wherein the cause of the prolongation of the blood coagulation time is at least one selected from the group consisting of causes relating to the prolongation of activated partial thromboplastin time and causes relating to the prolongation of prothrombin time.

8. The method of claim 7, wherein the cause of the prolonged blood coagulation is at least one selected from the group consisting of liver disease; disseminated intravascular coagulation syndrome; vitamin K deficiency; massive bleeding; reduction, deficiency or abnormal function of a coagulation factor; presence of a coagulation factor inhibitor; presence of lupus anticoagulant; use of an anticoagulant; presence of abnormal proteins; and blood sampling procedures.

9. The method of claim 8, wherein the decrease, deficiency or dysfunction of the coagulation factor is a decrease, deficiency or dysfunction of at least one selected from the group consisting of fibrinogen, factor 2, factor 5, factor 7, factor 8, von Willebrand factor, factor 9, factor 10, factor 11, factor 12, HMWK (High Molecular Weight Kininogen), and prekallikrein.

10. The method of claim 8, wherein the coagulation factor inhibitor is at least one selected from the group consisting of a factor 5 inhibitor, a factor 8 inhibitor, a von Willebrand factor inhibitor, and a factor 9 inhibitor.

11. The analytical method according to claim 8 , wherein the blood sampling procedure includes contamination with heparin.

12. 12. The method of claim 1, wherein the information on the cause of the prolongation of the blood clotting time includes a label indicating a potential cause of the prolongation of the blood clotting time, and a probability that the potential cause indicated by the label is the cause of the prolongation of the blood clotting time.

13. 12. The method of claim 1, wherein the information on the cause of the prolongation of the blood clotting time includes a label indicating a potential cause of the prolongation of the blood clotting time, and a probability that, among a plurality of potential causes of the prolongation of the blood clotting time, the potential cause indicated by the label is the cause of the prolongation of the blood clotting time.

14. 14. The method of claim 12, wherein in the output of the information relating to the cause of the prolongation of the blood clotting time, the probability is displayed in a graph format.

15. The analysis method according to claim 12 , wherein in outputting the information regarding the cause of the prolongation of blood clotting time, a label indicating the candidate cause of the prolongation of blood clotting time with the highest probability is output.

16. The analysis method according to any one of claims 12 to 14, wherein in the output of the information regarding the cause of the prolongation of the blood clotting time, a label indicating a candidate cause of the prolongation of the blood clotting time for which the probability is equal to or greater than a predetermined threshold is output.

17. The analysis method according to claim 16 , further comprising accepting a setting of the predetermined threshold value.

18. The method of any one of claims 1 to 17, further comprising outputting information regarding further testing based on results obtained from the deep learning algorithm.

19. The analytical method according to claim 18, wherein the additional test is at least one selected from the group consisting of a test for a coagulation factor, a test for a coagulation factor inhibitor, and a retest of a measurement item performed on the measurement sample.

20. The analysis method according to claim 18 or 19, wherein, in outputting the information relating to the additional tests, when there are a plurality of the additional tests, a priority of each of the additional tests is output.

21. The analytical method according to any one of claims 18 to 20, further comprising storing a cause of the prolongation of blood clotting time identified in the additional test in association with the data group.

22. The analysis method according to claim 21, further comprising outputting the associated and stored data group together with a cause of the prolongation of the blood clotting time.

23. determining whether the blood sample has a prolonged blood clotting time; The output of information regarding the cause of the prolongation of the blood coagulation time is executed when it is determined that there is a prolongation, and is not executed when it is determined that there is no prolongation.

23. The method of analysis according to any one of claims 1 to 22.

24. and receiving a request for output of information regarding a cause of the prolongation of the blood clotting time. The output of the information regarding the cause of the prolongation of the blood clotting time is executed when the output request is received.

24. The method of analysis according to any one of claims 1 to 23.

25. The analysis method according to any one of claims 1 to 24, further comprising selecting the deep learning algorithm to be input from a plurality of deep learning algorithms depending on the type of blood coagulation parameter.

26. 26. The method of claim 1, wherein the blood coagulation curve is a curve for obtaining activated partial thromboplastin time or prothrombin time.

27. The method of any one of claims 1 to 26, wherein the deep learning algorithm is trained using a dataset including the data group obtained from a measurement sample containing a blood specimen for which the cause of prolongation of blood clotting time is known and a coagulation time measurement reagent, and a label indicating the cause of the prolongation of blood clotting time.

28. preparing a measurement sample containing the blood specimen and a coagulation time measurement reagent; generating a plurality of pieces of detection information constituting the blood coagulation curve from the measurement sample; The analysis method according to claim 1 , further comprising acquiring the data group based on the generated plurality of pieces of detection information.

29. 29. The analysis method according to any one of claims 1 to 28, further comprising receiving the data group via a network from an analysis device that generates the data group from a measurement sample containing the blood sample and a coagulation time measurement reagent, and inputting the received data group into the deep learning algorithm.

30. 1. A blood sample analyzer comprising: a measurement unit that prepares a measurement sample containing the blood specimen and a coagulation time measurement reagent, and outputs a plurality of pieces of detection information constituting a blood coagulation curve from the measurement sample; A control unit, The control unit is a data group including a plurality of data constituting the blood coagulation curve or a derivative curve thereof is obtained based on the plurality of pieces of detection information, and the plurality of pieces of data included in the data group are obtained at predetermined intervals over time from the start to the end of the coagulation reaction; inputting the data set into a deep learning algorithm including a convolutional neural network; outputting information regarding the cause of the prolongation of the blood clotting time in the blood sample based on the results obtained from the deep learning algorithm; The analytical device.

31. When run on a computer, On the computer, A step of acquiring a data group including a plurality of data constituting a blood coagulation curve or a derivative curve thereof, the plurality of data included in the data group being acquired at predetermined intervals over time from the start to the end of a coagulation reaction; inputting the data set into a deep learning algorithm comprising a convolutional neural network; outputting information about the cause of prolonged blood clotting time in the blood sample based on the results obtained from the deep learning algorithm; The blood sample analysis program executes the above steps.

Citation Information

Patent Citations

  • Methods and apparatus for predicting the presence of hemostatic dysfunction in patient samples

    JP2002541431A

  • Blood coagulation time measuring device

    JP2007263907A

  • Blood coagulation analyzer

    JP2020051824A

  • Method and apparatus for predicting the presence of an abnormal level of one or more proteins in the clotting cascade

    US6321164B1

  • Blood coagulation analyzer

    US6524861B1