Method for estimating the cause of prolonged blood coagulation time, and information processing device.

Abstract

To estimate a cause of extension of coagulation time of the blood of a specimen to be inspected.SOLUTION: A method for estimating a cause of extension of blood coagulation time includes: acquiring a coagulation waveform indicating a change with time of a light amount due to a coagulation reaction of reaction liquid obtained by mixing a specimen to be inspected and an agent (S301b); acquiring a first waveform through before-and-after difference processing on the coagulation waveform (S302b); acquiring first and second fitting waveforms through fitting processing on the coagulation waveform and the first waveform (S303b, S304b); acquiring a third waveform through the before-and-after difference processing on the second fitting waveform (S305b); acquiring first to third normalized waveforms by normalizing light amount axes and time axes of the first and second fitting waveforms and the third waveform (S306b); extracting a feature quantity from each of the first to third normalized waveforms (S307b); and estimating the cause of extension of the coagulation time of the blood of the specimen to be inspected on the basis of a known feature quantity and the extracted feature quantity.SELECTED DRAWING: Figure 3B

Description

[Technical Field]

[0001] This disclosure relates to a method for estimating the cause of prolonged blood coagulation time, and to an information processing device. [Background technology]

[0002] Blood coagulation tests are performed to understand the pathology of the blood coagulation and fibrinolytic system, diagnose DIC (disseminated intravascular coagulation), confirm the effectiveness of thrombosis treatment, and diagnose hemophilia. In particular, blood coagulation time measurement involves mixing a sample with a reagent and measuring the time it takes for a fibrin clot to form (hereinafter referred to as blood coagulation time). If there is a congenital or acquired abnormality in blood coagulation ability, the blood coagulation time will be prolonged. Causes of prolonged blood coagulation time include deficiency of blood coagulation factors (deficiency type) and inhibition of the blood coagulation reaction by antibodies against blood coagulation factors or components in the blood coagulation time measurement reagent (e.g., phospholipids) (inhibitor type). Since the treatment strategy differs depending on the cause of prolonged blood coagulation time, it is necessary to identify the cause of the prolongation, but it is not possible to identify the cause of the prolongation by measuring blood coagulation time alone.

[0003] A common method for identifying the cause of prolonged blood clotting time is the cross-mixing test. In the cross-mixing test, the blood clotting time of multiple plasma samples, each mixed with the test sample and a normal sample at different ratios (e.g., 10:0, 9:1, 8:2, 5:5, 2:8, 1:9, 0:10), is measured. Measurements are performed immediately after sample preparation (immediate type) and after the sample has been incubated for two hours (delayed type). The obtained blood clotting time is plotted on the vertical axis, and the mixing ratio of the test sample on the horizontal axis, and the results are connected by a line to create a graph. The cause of the prolongation is determined from the shape of the immediate and delayed type graphs. This determination is qualitative and requires skilled experience from the person making the determination. Furthermore, the complexity of sample preparation and incubation during the test places a burden on the examiner. In addition, the need for additional blood collection for this test also places a burden on the patient.

[0004] To reduce the burden on examiners and patients and provide more quantitative results, a detection method using coagulation waveform analysis has been proposed. The target coagulation waveform is one obtained by optical measurement, recording the change in turbidity over time associated with the formation of fibrin clots.

[0005] Patent Document 1 discloses a method for differentiating and estimating whether the cause of prolonged blood coagulation time in a test specimen is deficiency-type or inhibitor-type, using features extracted from the coagulation waveform and its differential waveform obtained from a plasma sample mixed with a test specimen. Patent Document 2 discloses a method for estimating the presence or absence of blood coagulation abnormalities and the activity values ​​(concentrations) of blood coagulation factors in a test specimen by template matching using 50 parameters related to the centroid point of the waveform obtained by first-order differentiation of the coagulation waveform obtained from the test specimen. In particular, a method for estimating the concentration of blood coagulation factor VIII (FVIII) and blood coagulation factor IX (FIX) and differentiating which blood coagulation factor is deficient in the test specimen is disclosed. Qualitative and quantitative abnormalities of FVIII are called hemophilia A, and qualitative and quantitative abnormalities of FIX are called hemophilia B, and which blood coagulation factor is deficient is important because it relates to the selection of treatment strategy. Furthermore, Patent Document 3 discloses an APTT prolongation factor estimation system incorporating the above template matching algorithm. Furthermore, Non-Patent Document 1 discloses a method for detecting and estimating whether a test sample is deficient, inhibitory, or has anticoagulant added, using a flowchart that utilizes the slope and area of ​​a portion of the waveform obtained by first-order differentiation of the coagulation waveform acquired from the test sample, as well as the ratio of the time width of the waveform. In particular, for the inhibitory type, it detects and estimates whether the test sample is in the lupus anticoagulant (LA) positive group or the group expressing an inhibitor against FVIII. LA is one of the inhibitors against phospholipids, and its expression leads to antiphospholipid antibody syndrome. Furthermore, Patent Document 4 describes extracting multiple parameters from the first-order and second-order derivative waveforms of the coagulation waveform, determining the predicted concentration of each blood coagulation factor from these multiple parameters using multivariate correlation, and using a neural network trained for this prediction. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Patent No. 6994528 [Patent Document 2] International Publication No. 2020 / 158948 [Patent Document 3] Patent No. 6811975 [Patent Document 4] U.S. Patent No. 6524861 [Non-patent literature]

[0007] [Non-Patent Document 1] D.Shimomura et al., The First-Derivative Curve of the Coagulation Waveform Reveals the Cause of aPTT Prolongation, Clinical and Applied Thrombosis / Hemostasis, Volume 26(2020)P1-8 [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] (1) The challenge is to use the characteristics extracted from the coagulation waveform obtained by measuring the coagulation reaction in the reaction solution produced by mixing the test sample and reagents to differentiate and estimate whether the prolonged blood coagulation time in the test sample is due to the FVIII deficiency group, the FIX deficiency group, or the LA-positive group, which are important for diagnosing hemophilia A, hemophilia B, or antiphospholipid antibody syndrome.

[0009] (2) The blood clotting time of hemophilia patients increases with the severity of the disease. To improve throughput, it is common practice for technicians in this field to determine when the blood clotting reaction has finished for each sample and then stop measuring, before moving on to the next sample. As a result, the data length of the clotting waveform differs for each sample.

[0010] Patent Document 4 describes determining the predicted concentration of each blood coagulation factor from multiple parameters using multivariate backward correlation, and using a neural network for case estimation. Clustering methods including neural networks use fixed-length feature quantities as input data. Currently known clustering methods, such as neural networks, Gaussian mixture distributions, and K-means, are techniques that assume a constant number of input data points. When using existing clustering methods, including these, for coagulation waveform analysis, the clustering method only functions after appropriate preprocessing is applied to the variable-length coagulation waveform data to make it a fixed-length data point. Examples of such preprocessing for variable-length data include data compression techniques and zero-padding techniques, which are widely known as excellent preprocessing for image and speech recognition. If it becomes possible to estimate the cause of prolonged blood coagulation time using a neural network directly with a series of coagulation waveforms obtained from cross-mixing tests using only the coagulation waveform of the test sample, or multiple mixed samples of the test sample and normal sample mixed in predetermined ratios, then it will be possible to estimate the cause of prolonged blood coagulation time by including hidden information contained in the coagulation waveform, in addition to experimentally defined feature quantities. To achieve this, it is necessary to provide a suitable preprocessing method for discriminating coagulation waveforms with variable data length using a neural network, to demonstrate that a practical discrimination accuracy can be obtained using a neural network with a coagulation waveform or a group of coagulation waveforms, and to disclose the specific numerical values ​​of the neural network configuration and hyperparameters required to achieve this.

[0011] Therefore, this disclosure aims to estimate the cause of prolonged blood coagulation time in a test sample by using characteristic quantities extracted from a coagulation waveform that shows the change in light intensity over time due to the coagulation reaction of a reaction solution produced by mixing a test sample and a reagent.

[0012] Furthermore, this disclosure aims to estimate the causes of prolonged blood coagulation time with practical accuracy using a neural network by performing suitable preprocessing on coagulation waveforms with variable data length. [Means for solving the problem]

[0013] The method for estimating the cause of prolonged blood coagulation time according to this disclosure involves obtaining a coagulation waveform showing the change in light intensity over time due to the coagulation reaction of a reaction solution produced by mixing a test sample consisting of plasma obtained by separating blood acquired from a subject with a reagent; obtaining a first waveform by front-to-back difference processing of the coagulation waveform; obtaining a first fitted waveform by fitting processing of the coagulation waveform; obtaining a second fitted waveform by fitting processing of the first waveform; and obtaining a third waveform by front-to-back difference processing of the second fitted waveform. The method involves obtaining a first normalized waveform, a second normalized waveform, and a third normalized waveform by normalizing the light intensity axis and time axis of the first, second, and third fitting waveforms, respectively; extracting feature quantities from each of the first, second, and third normalized waveforms; and estimating the cause of prolonged blood coagulation time in a test sample based on known feature quantities extracted from a group of samples in which the cause of prolonged blood coagulation time is known, and the feature quantities extracted from each of the first, second, and third normalized waveforms.

[0014] The method for estimating the cause of prolonged blood coagulation time according to the present disclosure is an estimation method for estimating the cause of prolonged blood coagulation time of a test specimen from a coagulation waveform showing a change over time in the amount of light due to a coagulation reaction of a reaction solution generated by mixing a test specimen composed of plasma obtained by separating blood collected from a subject and a reagent, and includes aligning the data lengths of the coagulation waveform data, using a neural network to estimate a plurality of causes of prolonged blood coagulation time from the coagulation waveform data with aligned data lengths, and presenting the cause of prolonged blood coagulation time of the test specimen based on the probability of belonging of the estimated cause of prolonged blood coagulation time.

Effect of the Invention

[0015] According to the present disclosure, it is possible to estimate the cause of prolonged blood coagulation time of a test specimen using feature amounts extracted from a coagulation waveform showing a change over time in the amount of light due to a coagulation reaction of a reaction solution generated by mixing the test specimen and a reagent.

[0016] Also, according to the present disclosure, by performing suitable preprocessing on a coagulation waveform having a variable data length, it is possible to estimate the cause of prolonged blood coagulation time with practical accuracy by a neural network.

[0017] Problems, configurations, and effects other than those described above will be clarified by the description of the following embodiments.

Brief Description of the Drawings

[0018] [Figure 1] It is a diagram showing the coagulation waveforms of two FVIII-deficient specimens in Example 1 and waveforms corresponding to their first derivatives. [Figure 2] It is a diagram showing waveforms obtained by normalizing the coagulation waveforms, waveforms corresponding to the first derivatives of the coagulation waveforms, and waveforms corresponding to the second derivatives of the coagulation waveforms of the FVIII-deficient specimen group, FIX-deficient specimen group, and LA-positive specimen group in Example 1. [Figure 3A] It is a flowchart showing a method for estimating the cause of prolonged blood coagulation time in a comparative example. [Figure 3B]This flowchart shows the method for estimating the cause of the prolonged blood coagulation time in Example 1. [Figure 4] This figure shows an example of the overall configuration of the automated analyzer in Example 1. [Figure 5] This figure shows WaveNor0, 1, and 2, which are the normalized waveforms of Wave0, 1, and 2 from Example 1, along with their feature vectors. [Figure 6] This figure shows a two-dimensional map consisting of the WaveNor0, 1, and 2 feature quantities of Example 1. [Figure 7] This figure shows an example of plotting the characteristics of the test sample on a two-dimensional map created from the characteristics of samples for which the cause of prolonged blood coagulation time in Example 1 is known. [Figure 8] This figure shows the normalized Wave 0, 1, and 2 of the test specimens from Example 1, an indicator showing the probability of belonging to the FVIII-deficient specimen group, FIX-deficient specimen group, LA-positive specimen group, normal specimen group, and unknown, and an example of the assay estimation results. [Figure 9] This figure shows the relationship between the analysis unit, PC, and communication interface in Example 1, and an example of how the results of the test sample are displayed on the display unit. [Figure 10] This figure shows examples of WaveSrc1 and Wave1 in Example 1. [Figure 11A] This figure shows the preprocessing method for neural networks in Example 2. [Figure 11B] This figure shows the preprocessing method for neural networks in Example 2. [Figure 11C] This figure shows the preprocessing method for neural networks in Example 2. [Figure 12] This figure shows the structure of the neural network-based sorter in Example 2. [Figure 13A] This is an experimental result showing the relationship between the preprocessing method of Example 2 and the dropout and detection accuracy of the neural network-based detection system. [Figure 13B] This is an experimental result showing the relationship between the preprocessing method of Example 2 and the dropout and detection accuracy of the neural network-based detection system. [Figure 13C] This is an experimental result showing the relationship between the preprocessing method of Example 2 and the dropout and detection accuracy of the neural network-based detection system. [Figure 14A] This is an experimental result showing the relationship between the preprocessing method of Example 2, the hyperparameters of the neural network-based detector, and the detection accuracy. [Figure 14B] This is an experimental result showing the relationship between the preprocessing method of Example 2, the hyperparameters of the neural network-based detector, and the detection accuracy. [Figure 15] This experimental result shows that the preprocessing method and neural network-based classifier in Example 2 suppress overfitting. [Figure 16] This experimental result shows the difference between the FVIII-deficient sample and the FIX-deficient sample observed in the coagulation waveform of Example 4. [Figure 17A] This experimental result shows the relationship between the maximum value of the first derivative, a well-known indicator, and the log solidification time. [Figure 17B] This is an experimental result showing the relationship between vec1_y and log solidification time in Example 4. [Figure 18A] This is an experimental result showing representative waveforms for each sample group, with WaveNor0, which is the coagulation waveform of the commercially available sample from Example 4 normalized according to the NRC method. [Figure 18B] This is an experimental result showing representative waveforms for each sample group, based on WaveNor1, which is the standardized coagulation waveform of the commercially available sample from Example 4 according to the NRC method. [Figure 18C] This is an experimental result showing representative waveforms for each sample group, based on WaveNor2, which is the standardized coagulation waveform of the commercially available sample from Example 4 according to the NRC method. [Figure 19] This is an experimental result showing the frequency spectrum of WaveNor2 in Example 4. [Figure 20] This is a summary of an example feature set from Example 4. [Figure 21] This is the experimental result of displaying the commercially available sample group from Example 4 on feature space #6. [Figure 22A]This is an example of a display showing a summary plotting the commercially available sample group from Example 4 on the feature space. [Figure 22B] This is an example of a display showing the differential diagnosis result for one FVIII-deficient sample from Example 4. [Figure 23] This is a schematic diagram illustrating a method for dimensionality reduction of a series of coagulation waveform data obtained for a single test specimen in the cross-mixing test of Example 5 into two area indices F1 and F2. [Figure 24] This figure shows the results of identifying samples in a cross-mixing test using the multidimensional Gaussian mixture distribution of Example 5. [Figure 25A] This figure shows the cluster regions of the indices (F1, F2) in Example 5. [Figure 25B] This figure shows the experimental results of adding data points to the cluster regions of the indices (F1, F2) in Example 5. [Figure 26] This is a diagram illustrating the indicators (G1, G2) for Example 5. [Figure 27A] This figure shows experimental results plotting immediate (F1, G1) and delayed (F2, G2) indices calculated from a series of coagulation waveforms obtained from one sample in Example 5, and indices calculated from feature space #1 using the same one sample, with index F. [Figure 27B] This figure shows the experimental results plotting the indices (F1, G1) and (F2, G2) obtained from the 34 test samples in Example 5. [Figure 28A] This experimental result shows the average shape of the trajectory data for each cause of prolonged blood coagulation time, obtained by the cross-mixing test in Example 5. [Figure 28B] This figure shows the experimental results of separating and differentiating the trajectory data of each sample in the cross-mixing test of Example 5 into a 2D x14 space. [Figure 29] This diagram summarizes the conditions for determining the total number of samples required for the identification method in Example 5. [Figure 30]This is a schematic diagram showing the relationship between the number of coagulation waveforms used for differentiation in Example 5 and the worst-case probability (the minimum value of Pti across all samples, where Pti is the probability of the i-th sample belonging to the correct sample cluster). [Figure 31] This is an example of a display showing the identification results of the cross-mixing test in Example 5. [Modes for carrying out the invention]

[0019] This embodiment will be described below with reference to the drawings. In the drawings, functionally identical elements may be indicated by the same number. The drawings show embodiments and implementation examples in accordance with the principles of this disclosure, but they are for the purpose of understanding this disclosure and are not to be used in any way to restrict the interpretation of this disclosure. The descriptions herein are merely typical examples and do not limit the claims or applications of this disclosure in any way.

[0020] While this embodiment is described in sufficient detail for those skilled in the art to implement the disclosure, it is important to understand that other implementations and forms are possible, and that the configuration and structure can be modified and various elements replaced without departing from the scope and spirit of the technical idea of ​​this disclosure. Therefore, the following description should not be construed as limiting to this.

[0021] [Example 1] In Example 1, the cause of prolonged blood coagulation time is estimated using features extracted from waveforms (WaveNor0, WaveNor1, WaveNor2) that have been normalized based on the NRC (Normalization based on the Reaction Constants) method described later, using Wave0, Wave1, and Wave2 obtained by waveform fitting.

[0022] The samples used to acquire coagulation waveforms contain multiple blood coagulation factors, and differences in the concentration of each factor lead to differences in the rate of the coagulation reaction, resulting in diversity (variation) in blood coagulation time. In this field, the first derivative of the coagulation waveform is treated as the coagulation rate, and the second derivative as the coagulation acceleration. Even with the same group of samples, the coagulation waveforms and waveforms related to coagulation rate do not match, making it difficult to visually identify the characteristics of the sample group. As an example, Figure 1 shows the coagulation waveforms of two FVIII-deficient samples (hereinafter, the coagulation waveforms will be appropriately referred to as "WaveSrc0") and the first waveform corresponding to the first derivative of the two coagulation waveforms (WaveSrc0) (hereinafter, the waveform corresponding to the first derivative of the coagulation waveform will be appropriately referred to as "WaveSrc1"). In WaveSrc0, the rise time of the waveform (position of the arrowhead a1, a2), the total change in light intensity (arrow width b1, b2), and the blood coagulation time (for example, the time when the change in light intensity (scattered light intensity (count) in Figure 1) becomes 50%, c1, c2) differ from sample to sample. The total change in light intensity depends on the concentration of fibrinogen, one of the blood coagulation factors. Blood coagulation time is affected by the rate of the coagulation reaction, and in the case of hemophilia samples, it is obvious that blood coagulation time will be longer as the symptoms become more severe. In WaveSrc1, there are differences in the height and width of the waveform for each sample. It is also an experimental fact that waveform features such as coagulation rate (change in light intensity per unit time (scattered light intensity (counts))) decrease inversely proportional to the blood coagulation time as it increases, and thus differ from sample to sample due to differences in blood coagulation time. However, these two samples are both FVIII-deficient samples, and it is necessary to find common characteristics to classify both samples into the FVIII-deficient sample group.

[0023] The inventors believed that suppressing the influence of variations in blood coagulation time within each sample group would be effective in making common characteristics of the same sample group apparent. Based on this idea, this disclosure shows that by performing normalization based on the NRC method, which normalizes not only the light intensity axis but also the time axis simultaneously, it is possible to converge WaveSrc0 of the same sample group, WaveSrc1 obtained by processing the difference between WaveSrc0 and WaveSrc2 obtained by processing the difference between WaveSrc1 and WaveSrc2, and to make the waveform characteristics within the same sample group apparent when comparing waveforms for each sample group. The specific method of normalization will be described in detail later in (Sample Detection Program). Note that WaveSrc2 is the waveform equivalent to the first derivative of WaveSrc1 (equivalent to the second derivative of the coagulation waveform).

[0024] Figure 2 shows WaveSrcNor0, WaveSrcNor1, and WaveSrcNor2, which are normalized waveforms of WaveSrc0, WaveSrc1, and WaveSrc2 for the FVIII-deficient sample group, FIX-deficient sample group, and lupus anticoagulant (LA)-positive sample group, respectively. In all graphs, it can be seen that the waveforms generally converge within the same sample group. Furthermore, when comparing waveforms of the same type (e.g., WaveSrcNor0) within each sample group, differences in shape were observed. For example, in WaveSrcNor0, differences were observed in the rising edge position of the waveform for each sample group. The auxiliary line L1 was positioned to the center of the rising edge position of the waveform in the LA-positive group. The rising edge position of the waveform in the FVIII-deficient sample group was to the right of this auxiliary line L1, and the rising edge position of the waveform in the FIX-deficient sample group was to the left of this auxiliary line L1, showing a difference. In WaveSrcNor1 and WaveSrcNor2, differences were observed in the width between the peaks of the two peaks and in the height of the peaks. In both WaveSrcNor1 and WaveSrcNor2 waveforms, auxiliary lines L2 and L4 were positioned at the center of the peak on the left side of the LA-positive group, and auxiliary lines L3 and L5 were positioned at the center of the peak on the right side of the LA-positive group. When the peak position on the right side was confirmed using auxiliary lines L3 and L5 as a guide, no significant differences were observed between the sample groups. On the other hand, when the peak position on the left side was confirmed using auxiliary lines L2 and L4 as a guide, the peak of the FVIII-deficient sample group was located to the right of auxiliary lines L2 and L4, and the peak of the FIX-deficient sample group was located to the left of auxiliary lines L2 and L4, showing a difference. Since these differences between sample groups were larger than the variations within the same sample group, the inventors came up with the idea of ​​quantifying these differences as features and using them for sample detection.

[0025] The samples used to obtain WaveSrc0, the source data for WaveSrc0 shown in Figure 1 and WaveSrcNor0 shown in Figure 2, were all commercially available. Specifically, they were as follows: FVIII-deficient samples were Factor VIII Deficient Plasma (Precision BioLogic, Inc.), FVIII Deficient Plasma (George King Bio-Medical, Inc.), and Factor VIII Deficient Plasma (Affinity Biologicals, Inc.). FIX-deficient samples were Factor IX Deficient Plasma (Precision BioLogic, Inc.), FIX Deficient Plasma (George King Bio-Medical, Inc.), and Factor IX Deficient Plasma (Affinity Biologicals, Inc.). The LA-positive samples are Lupus Positive Control and Weak Lupus Positive Control (Precision BioLogic, Inc.) and Positive LA Plasma (George King Bio-Medical, Inc.).

[0026] The specimens covered by this disclosure are not particularly limited as long as they are specimens that produce a coagulation reaction originating from the subject, but specimens consisting of plasma obtained by separating blood acquired from a subject are preferred. The specimens consisting of plasma may contain trace amounts of impurities, as long as this does not pose a problem for analysis using the automated analyzer described in this disclosure, such as for measuring coagulation waveforms. Furthermore, the reagents are not particularly limited as long as they are reagents for measuring thromboplastin time (PT), activated partial thromboplastin time (APTT), and fibrinogen (Fbg). The instrument should be configured to measure the change in light intensity (light intensity is an index including turbidity) of the reaction solution over time due to the coagulation reaction. The reagent used for measuring the coagulation waveform shown in this embodiment is Coagupia APTT-N (manufactured by Sekisui Medical Co., Ltd.), and the instrument is a Hitachi Automated Analyzer 3500 (manufactured by Hitachi High-Tech Corporation).

[0027] This disclosure is characterized not only by the implementation of standardization based on the NRC method to make differences in waveform shape apparent, but also by the implementation of waveform fitting to quantify the apparent differences in waveform shape while suppressing the effects of noise.

[0028] Next, we will provide a detailed explanation of the automated analyzer used to acquire coagulation waveforms, the method for acquiring coagulation waveforms, and the sample screening program.

[0029] (Automatic analyzer 100) The automated analyzer 100 used to acquire coagulation waveforms will be described. Figure 4 shows an example of the overall configuration of the automated analyzer 100 in Example 1. Here, the basic operation of the device will be explained using Figure 4, but it is not limited to the following example.

[0030] As shown in Figure 4, the automated analyzer 100 is generally composed of an analysis unit 130, an operating computer 118, a storage unit 119, and a control computer 120. The analysis unit 130 is generally composed of a sample dispensing mechanism 101, a sample disk 102, a reagent dispensing mechanism 106, a reagent disk 107, a reaction vessel stock unit 111, a reaction vessel transport mechanism 112, a detection unit 113, and a reaction vessel disposal unit 117. The control computer 120 is connected to the analysis unit 130 in a communication manner.

[0031] The sample dispensing mechanism 101 aspirates the sample 103a contained in the sample container 103, which is placed on a sample disk 102 that rotates clockwise and counterclockwise, and dispenses it into the reaction vessel 104. The sample dispensing mechanism 101 performs the sample aspiration and dispensing operations through the operation of a sample syringe pump 105 controlled by a control computer 120.

[0032] The reagent dispensing mechanism 106 aspirates reagent 108a contained in reagent container 108 placed on reagent disk 107 and dispenses it into reaction vessel 104. The reagent dispensing mechanism 106 performs the reagent aspiration and dispensing operations through the operation of a reagent syringe pump 110 controlled by a control computer 120.

[0033] Furthermore, the reagent dispensing mechanism 106 incorporates a reagent heating mechanism 109, and the temperature of the reagent 108a drawn up by the reagent dispensing mechanism 106 is raised to an appropriate temperature (a predetermined temperature) by the reagent heating mechanism 109, which is controlled by the control computer 120.

[0034] The reaction vessel transport mechanism 112 transports and installs the reaction vessel 104. The reaction vessel transport mechanism 112 transports and installs the reaction vessel 104 from the reaction vessel stock section 111 to the reaction vessel installation section 114 of the detection unit 113 by holding the reaction vessel 104 and rotating it horizontally.

[0035] The detection unit 113 has one or more reaction vessel mounting sections 114 (in this embodiment, one is shown as an example) for placing the reaction vessel 104, and measures the light intensity of the reaction solution (a mixture of sample 103a and reagent 108a) inside the reaction vessel 104 inserted into the reaction vessel mounting section 114. The detection unit 113 controls the temperature so that the reaction solution inside the reaction vessel 104 inserted into the reaction vessel mounting section 114 is, for example, 37°C. In this embodiment, the case where one detection unit 113 is arranged is shown, but it may be configured to have multiple detection units 113. An example of the detection principle in the detection unit 113 is described below. Light irradiated from the light source 115 is scattered by the reaction solution inside the reaction vessel 104. The detection unit (light sensor) 116 receives the scattered light scattered by the reaction solution inside the reaction vessel 104. For example, a halogen lamp or an LED can be used as the light source 115. The detection unit (optical sensor) 116 is composed of a photodiode and the like. The signal received by the detection unit (optical sensor) 116 is converted into a digital signal of light intensity by the A / D converter 121, and this is input to the control computer 120 as reaction process data (coagulation waveform data representing the time change of the detected light intensity) and stored in the memory unit 119. The operation of the detection unit 113 is controlled by the control computer 120. Here, a detector that utilizes light scattering is used, but there are other types as well, such as those that utilize transmitted light.

[0036] The reaction vessel transport mechanism 112 holds the reaction vessel 104 after measurement is complete and disposes of it in the reaction vessel disposal section 117.

[0037] To improve processing capacity, the system may be equipped with a detector-less incubator 122 for warming the sample before adding the measurement starter reagent.

[0038] The control computer 120 is an information processing device having a computer system including a processor and memory. It not only controls the operation of the automated analyzer, such as dispensing samples 103a and reagents 108a, relocating the reaction vessel 104, and discarding the reaction vessel 104, but also calculates the blood coagulation time from measured values ​​of light intensity (coagulation waveform data) that change over time according to the coagulation reaction of the reaction solution, and performs diagnostic estimation of the cause of prolonged blood coagulation time. The calculated blood coagulation time and the results of the diagnostic estimation of the cause of prolonged blood coagulation time are output to the display unit 118c and stored in the storage unit 119. These results may also be printed out by the printer 123 via the operation computer 118. The storage unit 119 connected to the control computer 120 stores programs such as control programs, measurement programs, calibration curve generation programs, quantitative programs, and sample diagnostic programs, as well as measured waveform data, quantitative results, and sample diagnostic results. Various programs are read and executed according to requests entered into the operation computer 118 or sent from the communication interface 124. Input to the operating computer 118 can be done by touching the display unit 118c, or by using the connected keyboard 118b. Alternatively, input can be made by selecting what is displayed on the display unit 118c with the mouse 118a. Waveform data, quantitative results, specimen analysis results, etc., stored in the memory unit 119 are output to the display unit 118c, or sent to the communication interface 124, or printed by the printer 123 as needed. The communication interface 124 is connected to, for example, the hospital network and communicates with the HIS (Hospital Information System) and LIS (Laboratory Information System).

[0039] As described above, the estimation of the cause of prolonged blood coagulation time in the test specimen is performed by calling the control computer 120 from the sample detection program and the waveform data of the test specimen stored in the memory unit 119. Alternatively, the waveform data of the test specimen may be transferred to the analysis computer 125 which executes the sample detection program, and the analysis computer 125 may estimate the cause of prolonged blood coagulation time in the test specimen and display the result on the screen of the analysis computer 125. Alternatively, the waveform data stored in the memory unit 119 may be written to another external storage medium via the control computer 120 and the operation computer 118, and an independent analysis computer with a sample detection program may read the written waveform data and estimate the cause of prolonged blood coagulation time in the test specimen.

[0040] (Method for obtaining coagulation waveforms) Next, we will explain the analysis operation of blood coagulation time parameters and the acquisition of coagulation waveforms (WaveSrc0). There are mainly three blood coagulation time parameters: thromboplastin time (PT), activated partial thromboplastin time (APTT), and fibrinogen (Fbg). For each parameter, first, set the parameters necessary for the analysis. When setting the parameters, input the parameter to be analyzed, sample volume, reagent volume, output unit, etc. For the PT and Fbg parameters, input the calibration method and perform calibration as necessary. After setting the parameters, place the sample container 103a and reagent container 108, which contain the sample 103a and reagent 108a respectively, on the sample disk 102 and reagent disk 107, respectively, and perform the analysis. For reagent 108a, commercially available PT measurement reagents, APTT measurement reagents, and Fbg measurement reagents can be used, for example.

[0041] (Analysis of PT items) The analysis operation for the PT item will now be explained. Sample 103a is dispensed into an empty reaction vessel 104 located in the reaction vessel stock section 111 by the sample dispensing mechanism 101. The reaction vessel 104 containing sample 103a is moved to the reaction vessel installation section 114 of the detection unit 113 by the reaction vessel transport mechanism 112. Reagent 108a is drawn from the reagent container 108 by the reagent dispensing mechanism 106 and heated to an appropriate temperature by the reagent heating mechanism 109. A temperature of 37°C is desirable at this time. The heated reagent 108a is discharged into the reaction vessel 104 containing sample 103a, which has already been installed in the reaction vessel installation section 114. Simultaneously with the start of discharge, the change in turbidity over time of the reaction solution, which is a mixture of sample 103a and reagent 108a, is measured optically and becomes WaveSrc0.

[0042] (Analysis behavior of APTT items) The analysis procedure for the APTT item is described below. Sample 103a is dispensed into an empty reaction vessel 104 located in the reaction vessel stock section 111 by the sample dispensing mechanism 101. The reaction vessel 104 containing sample 103a is moved to the reaction vessel installation section 114 of the detection unit 113 by the reaction vessel transport mechanism 112. The first reagent contains activators such as ellagic acid and kaolin, and is drawn from the reagent container 108 by the reagent dispensing mechanism 106 and heated to an appropriate temperature by the reagent heating mechanism 109. A temperature of 37°C is desirable at this time. The heated first reagent is discharged into the reaction vessel 104 containing sample 103a, which has already been installed in the reaction vessel installation section 114. The mixture of sample 103a and the first reagent is temperature-controlled within the reaction vessel 104. A temperature of 37°C is desirable at this time. At this time, the mixture of sample 103a and the first reagent may be stirred by a stirring mechanism (not shown). The second reagent is a calcium chloride solution, which is drawn from the reagent container 108 by the reagent dispensing mechanism 106 and heated to an appropriate temperature by the reagent heating mechanism 109. A temperature of 37°C is desirable at this time. The heated second reagent is discharged into the reaction vessel 104, which contains a mixture of sample 103a and the first reagent, already installed in the reaction vessel installation section 114. Before the discharge of the second reagent, the mixture of sample 103a and the first reagent is temperature-controlled for a certain period of time, preferably 180 seconds. Simultaneously with the start of the discharge of the second reagent, the change in turbidity over time of the reaction solution, which is a mixture of sample 103a and the first and second reagents, is measured optically and becomes WaveSrc0.

[0043] (Analysis behavior of Fbg items) The analysis procedure for the Fbg item is described below. Sample 103a is dispensed into an empty reaction vessel 104 located in the reaction vessel stock section 111 by the sample dispensing mechanism 101. The reaction vessel 104 containing sample 103a is moved to the reaction vessel installation section 114 of the detection unit 113 by the reaction vessel transport mechanism 112. Sample diluent is drawn from the reagent container 108 by the reagent dispensing mechanism 106 and heated to an appropriate temperature by the reagent heating mechanism 109. A temperature of 37°C is desirable at this time. The heated sample diluent is discharged into the reaction vessel 104 containing sample 103a, which has already been installed in the reaction vessel installation section 114. The mixture of sample 103a and sample diluent is temperature-controlled within the reaction vessel 104. A temperature of 37°C is desirable at this time. At this time, the mixture of sample 103a and sample diluent may be stirred by a stirring mechanism (not shown). The reagent is drawn from the reagent container 108 by the reagent dispensing mechanism 106 and heated to an appropriate temperature by the reagent heating mechanism 109. A temperature of 37°C is desirable at this time. The heated reagent is discharged into the reaction vessel 104, which contains a mixture of sample 103a and sample diluent, already installed in the reaction vessel installation section 114. Simultaneously with the start of reagent discharge, the change in turbidity over time of the reaction solution, which is a mixture of sample 103a, sample diluent, and reagent, is measured optically and becomes WaveSrc0.

[0044] In all cases, the measurement ends when the coagulation reaction is complete (no change in turbidity is observed). Alternatively, the measurement ends after a certain period of time has elapsed. This period of time is, for example, 5 minutes. The time-dependent turbidity change measured by light is acquired as a coagulation waveform in the storage unit 119 via the control computer 120.

[0045] An example of a coagulation waveform is shown in Figure 1, where the horizontal axis represents measurement time and the vertical axis represents light intensity. Figure 1 is an example of a coagulation waveform for an FVIII-deficient sample (Precision BioLogic Inc.) in the APTT parameter, with the vertical axis representing scattered light intensity.

[0046] (Specimen screening program) Figure 3B is a flowchart showing a method for estimating the cause of prolonged blood coagulation time, which is performed by executing a sample detection program. The computer system of the control computer 120 performs the estimation of the cause of prolonged blood coagulation time in Example 1 by executing the sample detection program. This will be explained in detail with reference to Figure 3B. Each step of the flowchart in Figure 3B may be performed by the control computer 120, the operation computer 118, the analysis computer 125, or other computers on the network. Here, we will describe an example in which the computer system of the control computer 120 performs each step.

[0047] The control computer 120 reads the coagulation waveform (WaveSrc0) data acquired by the coagulation reaction measurement (Figure 3B-S301b). The specific analysis operations related to the acquisition of WaveSrc0 are as described above in (Method for acquiring coagulation waveform).

[0048] Next, the control computer 120 generates WaveSrc1 by processing the difference between WaveSrc0 and WaveSrc0 (Figure 3B-S302b). WaveSrc1 can be calculated, for example, using Equation 1. Here, Δt in Equation 1 is the measurement interval for the coagulation reaction, for example, 0.1 seconds.

[0049]

number

[0050] Next, the control computer 120 generates a first fitted waveform Wave0 (hereinafter, the waveform that has been fitted to the coagulation waveform will be referred to as the first fitted waveform, and will be appropriately called "Wave0") by waveform fitting of WaveSrc0 (Figure 3B-S303b), and generates Wave1 (hereinafter, the waveform that has been fitted to the first waveform will be referred to as the second fitted waveform, and will be appropriately called "Wave1") by waveform fitting of WaveSrc1 (Figure 3B-S304b). The fitting equation for WaveSrc0 is preferably an equation based on a sigmoid curve, considering that the coagulation waveform is a sigmoid curve (S-shaped curve). For example, equation 2. In equation 2, the amount of light after reagent dispensing is Y dc This formula is designed to follow the coagulation waveform.

[0051]

number

[0052] Y in formula 2 above dc Y0 is the base light intensity [counts], Y0 is the base scattered light intensity of the reaction solution itself [counts], Y1 is the saturated scattered light intensity [counts], t is the measurement time [seconds], T0 is the time [seconds] of (Y0 + Y1) ÷ 2, and T1 is the reaction time constant [seconds].

[0053] Equation 2 assumes fitting of a coagulation waveform measured by scattered light. However, in the case of a coagulation waveform measured by absorbance, the waveform is inverted vertically from the scattered light measurement waveform. Therefore, Equation 2 should be modified to conform to the coagulation waveform measured by absorbance.

[0054] The fitting equation for WaveSrc1 should preferably be one that better reproduces the bimodal shape. For example, it is good to represent the left peak of the bimodal shape with a Gaussian distribution and the right peak with a sigmoid derivative, and connect the two with a cubic spline function. The shape of WaveSrc1 is often bimodal, and this tendency is particularly pronounced in samples with prolonged blood coagulation time.

[0055] Next, the control computer 120 generates Wave2 (hereinafter, the waveform equivalent to the first derivative of the first fitting waveform will be referred to as "Wave2" as appropriate) by processing the difference between Wave1 and Wave2 (Figure 3B-S305b). Wave2 can be obtained, for example, by equation 3.

[0056]

number

[0057] Next, the control computer 120 normalizes both the time axis and the light intensity axis of Wave0, 1, and 2 based on the NRC method (Figure 3B-S306b). The time axis is normalized by Equation 4. For the light intensity axis, Wave0, 1, and 2 are normalized by Equations 5, 6, and 7, respectively. The waveform obtained by normalizing the time axis and light intensity axis of Wave0 is the first normalized waveform and will be referred to as "WaveNor0" below as appropriate. The waveform obtained by normalizing the time axis and light intensity axis of Wave1 is the second normalized waveform and will be referred to as "WaveNor1" below as appropriate. The waveform obtained by normalizing the time axis and light intensity axis of Wave2 is the third normalized waveform and will be referred to as "WaveNor2" below as appropriate. Normalization is based on reaction constants and is called the NRC (Normalization based on the Reaction Constants) method. The reaction constants used in the NRC method can be the values ​​obtained by fitting WaveSrc0. Waveform fitting is performed to quantify the aforementioned differences while suppressing the influence of noise and extract them as unique features.

[0058]

number

[0059]

number

[0060]

number

[0061]

number

[0062] Here, n is the number of data points (n≧1), and the relationship between t(n) and n is t(n)=n / 10 [seconds]. Also, T0, T1, Wave0(n), Y dc Y1 is obtained from equation 2.

[0063] Finally, the control computer 120 extracts features from the normalized waveform (Figure 3B-S307b). The features must be numerical representations of the parts where waveform shape differences are observed for each sample group. Now, let's refer to Figure 2 again. Figure 2 shows the normalized waveforms (WaveSrcNor0, 1, 2) of the original waveforms (WaveSrc0, 1, 2) before waveform fitting. Normalization follows equations 4 to 7, but in Figure 2, the original waveform is normalized, and WaveSrc0, WaveSrc1, and WaveSrc2 are substituted into Wave0(n) in equation 5, Wave1(n) in equation 6, and Wave2(n) in equation 7, respectively.

[0064] As mentioned above, for example, in WaveSrcNor0, differences were observed in the rise time of the waveform. The order of rise times, from the shortest normalization time, was FIX-deficient sample group, LA-positive sample group, and FVIII-deficient sample group. According to equation 4 of the time normalization formula, the smaller the reaction time constant T1 (i.e., the faster the reaction), the further away the normalization time was from 0. In other words, the difference in reaction rate among each sample group was reflected in the difference in rise time. FVIII, in the form of activated FVIII, enhances a part of the reaction within the coagulation reaction mechanism, so it is thought that the reaction time constant became larger and the rise time was delayed in the FVIII-deficient sample group. LA, due to its property of binding to phospholipids, which are the scaffold for the coagulation reaction, and inhibiting the binding of coagulation factors, uniformly inhibits the coagulation reaction regardless of the activity of blood coagulation factors. The degree of inhibition varies depending on the quantity and quality (titer) of LA, resulting in diverse coagulation reaction rates and a diversity in the reaction time constant T1, which is thought to have placed the group between the FIX-deficient and FVIII-deficient sample groups. Thus, the onset timing is thought to reflect differences originating from the cause of prolonged blood coagulation time, and it was determined that a feature that quantifies the difference in onset timing is useful. Here, a vector was used as the feature, with the normalized time of each waveform at 0 as the starting point and the rising point of the waveform as the ending point.

[0065] Furthermore, in WaveSrcNor1, differences were observed in the width between the peaks of the two peaks and in the height of the peaks. In the FVIII-deficient sample group, it is thought that the small amount of FVIII present was used in the initial stages of the reaction and then became depleted, resulting in the first peak being larger (higher). On the other hand, in the FIX-deficient sample group and the LA-positive sample group, FVIII was present in sufficient quantities, so the reaction rate was faster in the amplification reaction called positive feedback that occurs later in the reaction than in the initial stages, resulting in the second peak being larger (higher). Also, focusing on the normalization time, there was a difference in the timing of the rise of the waveform for the same reasons as in WaveSrcNor0, which led to a difference in the normalization time at which the first peak occurred. The difference in the occurrence time of the first peak and the difference in the height of the two peaks are thought to reflect differences originating from the cause of the prolongation of blood coagulation time, and it was judged that features that quantify these differences would be useful. Here, a vector with a normalization time of 0 as the starting point and the peak point of the first peak as the ending point was used as the feature.

[0066] In WaveSrcNor2, a difference was observed between the height of the first peak and its occurrence time, influenced by waveform differences equivalent to the first derivative. Here, information on the peak height of the first peak and the normalization time was combined, and a vector information was used as a feature, with the point where the normalization time is 0 as the starting point and the peak point of the first peak as the ending point.

[0067] Vector features are extracted from the normalized waveforms (WaveNor0, 1, 2) of the fitted waveforms (Wave0, 1, 2) in order to quantify them while suppressing the effects of noise. Figure 5 shows the normalized waveforms WaveNor0, 1, 2 and their feature vectors. The horizontal axis represents the normalization time, and the vertical axis represents the normalization level. The feature vectors of WaveNor0, 1, and 2 are denoted as vectors 0, 1, and 2, respectively. The x and y components of each vector are taken as the features (X, Y) of each waveform. Here, the x component of WaveNor0 is modified by subtracting T0 / T1 to emphasize the differences in waveform shape between the sample groups. The features (X, Y) of WaveNor0, 1, and 2 are expressed as (vec0_x, vec0_y), (vec1_x, vec1_y), and (vec2_x, vec2_y), respectively.

[0068] Finally, the extracted features of the test samples are compared with the features of samples with known sample background (causes of prolonged blood clotting time) to detect and estimate the cause of prolonged blood clotting time in the test samples. Using the features (X, Y) extracted from samples with known causes of prolonged blood clotting time independently, a probability density distribution is assigned to the feature clusters for each sample group in a two-dimensional map plotting these features. For example, a probability density distribution based on a Gaussian mixture model is used. Figure 6 shows an example of a two-dimensional map consisting of features extracted from normalized waveforms WaveNor0, 1, and 2. In the map, these are represented as Wave0, 1, and 2. The features were extracted by analyzing waveforms obtained from samples with known causes of prolonged blood clotting time. The ellipses shown for each sample group (FVIII deficiency, FIX deficiency, LA positive, normal) represent the spread of the feature data. To prevent overlap between the ellipses of normal and abnormal samples, the size of each ellipse was set to 4σ from the center (σ: standard deviation in each direction, called Mahalanobis distance in two or more dimensions). These ellipses are represented by a Gaussian probability density distribution. In the figure, regions that do not belong to any ellipse indicate that the sample does not belong to any sample group within the given data range, and represent novel coagulation waveforms depending on factors such as the artificial mixing of two or more test samples or the medication status of patients undergoing treatment. This region is defined as the "unknown" group. The ellipses of abnormal samples overlap in all detectors, and the boundary can be set to any value of the membership probability calculated according to the probability density distribution. Figure 6 shows the case where the boundary is a 50% membership probability. The abnormal samples used here are the commercially available products mentioned above. Commercially available products were also used as substitutes for normal samples. Specifically, these are Normal Reference Plasma and Pooled Normal Plasma (manufactured by Precision BioLogic, Inc.), and Factor Assay Control Plasma and Borderline Factor Assay Control Plasma (manufactured by George King Bio-Medical, Inc.).

[0069] Features extracted from WaveNor0, 1, and 2 of the test sample are applied to these three 2D maps to calculate the probability of the test sample belonging to each map (the probability of which sample group it belongs to). The test sample is then estimated for screening according to the calculated probability of belonging. For example, it may be estimated for screening as the sample group with the highest probability of belonging, or it may be estimated for screening as the sample group with the highest probability after averaging the probability of belonging in each map, or it may be estimated for screening as the sample group with the highest probability after weighting the probability of belonging in each map.

[0070] Figures 7 and 8 show examples of the display of the assay estimation results for a test sample. Figure 7 is an example of where the features of the test sample are plotted on a 2D map created from the features of known samples. Figure 8 shows the WaveNor0, 1, and 2 of the test sample (labeled Wave0, Wave1×2, and Wave2 in the diagram), indicators showing the probability of belonging to the FVIII-deficient sample group, FIX-deficient sample group, LA-positive sample group, normal sample group, and unknown, and the assay estimation results. Figure 9 shows the relationship between the analysis unit 130, the control computer 120, and the communication interface 124, and an example of the display of the test sample results on the display unit 118c. The display unit 118c displays fields for selecting the sample number (S_No.) of the test sample whose results are to be displayed and the 2D map. It is desirable that the display unit 118c displays the blood coagulation time, assay estimation results, and estimated blood coagulation factor activity results for the selected sample number. While it is not strictly necessary to display the estimated results of blood coagulation factor activity, it is beneficial to do so. For example, blood coagulation factor VIII deficiency refers to a condition where the activity level of blood coagulation factor VIII is less than 1%, and blood coagulation factor IX deficiency refers to a condition where the activity level of blood coagulation factor IX is less than 1%. In Figure 9, the selected map and the feature plot of the test sample within that map are shown in the lower left of the screen. This type of display allows for a highly visual and intuitive representation of the cause of the prolonged blood coagulation time in the test sample. On the other hand, it is also important to quantitatively indicate the cause of the prolongation, and in Figure 9, "Hemophilia A (Probability: 95%)" is displayed as the "Calibration Estimate Result." In addition, as shown in the lower right of the screen in Figure 9, it is beneficial to display the waveform and detailed belonging probability (for example, the content of Figure 8) for the sample with the selected sample number, and to show an indicator that visualizes the waveform and displays the detailed belonging probability.

[0071] In Example 1, preprocessing such as smoothing may be performed as appropriate, for example, after reading WaveSrc0 (Figure 3B-S301b) and / or after generating WaveSrc1 (Figure 3B-S302b). Preprocessing (a first filtering process to remove noise from WaveSrc0 and a second filtering process to remove noise from WaveSrc1) refers to filtering for noise reduction for each waveform. This processing step is not necessarily required. However, it is desirable to perform it before waveform fitting in order to achieve more accurate waveform fitting. Specific examples of filtering include N-point moving average processing (N is a positive integer and odd, 3 ≤ N ≤ number of measurement points) and smoothing processing by selecting the minimum value in a certain interval. In addition, a T1-point moving average may be used, using the reaction time constant of the fitting parameter according to Equation 2, while taking care not to lose the original waveform shape for variations in blood coagulation time values. In this case, for example, it is best to round the value of T1 to one decimal place to make it an integer, and if it is an even number, add 1 or subtract 1 to make it an odd number before using it.

[0072] (Example of results from a sample screening program) In the sample detection program, the results of detecting and estimating the test samples showed that the FVIII-deficient sample group, FIX-deficient sample group, and LA-positive sample group were detected with 85.7% (=12 / 14*100%), 100% (=9 / 9*100%), and 66.7% (=4 / 6*100%), respectively. Overall, the probability was 86% (=25 / 29*100%). Normal samples were detected with 100% (=10 / 10*100%) accuracy, confirming that normal samples were not misclassified as FVIII-deficient, FIX-deficient, or LA-positive samples. The test samples used were different lot numbers of the commercially available product mentioned above.

[0073] (Effects of Example 1) Figure 10 shows an example of WaveSrc1 and Wave1. It can be seen that the waveform of WaveSrc1 contains noise. Preprocessing methods such as smoothing have been published to remove noise, but excessive processing can lead to loss of waveform shape, and insufficient processing can result in multiple values ​​existing when extracting a certain feature numerically, making it impossible to determine uniquely. When extracting features inherent in a solidified waveform numerically, multiple values ​​existing and it is not possible to determine uniquely, for example, if you want to extract the maximum value of the left peak of the two peaks in Figure 10, in the raw data containing noise, there are multiple maximum values ​​at different times, and it is not possible to determine a single point. The method of extracting features by performing waveform fitting solves this problem by first determining waveform model parameters so that, for example, the RMS error is minimized from the observed waveform containing noise, and then quantifying the features inherent in the solidified waveform from the fitted waveform that does not contain noise.

[0074] Furthermore, the cause of prolonged blood clotting time is estimated using extracted features. Specifically, features are extracted using samples for which the cause of prolonged blood clotting time is known, and a probability density distribution is assigned to the feature clusters (clusters) for each sample group in a 2D map plotting these features. By applying the features of the test sample to this 2D map, the probability of the test sample belonging to a particular sample group (the probability of which sample group it belongs to) is calculated, and the cause of prolonged blood clotting time in the test sample can be estimated intuitively and with high visual clarity according to the probability of belonging.

[0075] This disclosure presents a novel method that solves the problems of the aforementioned conventional technology by extracting the characteristics inherent in the coagulation waveform as numerical values ​​through a fitting model, and by making the characteristics of each cause of prolonged blood coagulation time apparent through normalization processing using the NRC method based on reaction constants.

[0076] Here, Figure 3A shows an example of the sample detection flow for the comparative example. In the example of the sample detection flow for the comparative example (Figure 3A), first, coagulation waveform data obtained by coagulation reaction measurement is read (Figure 3A-S301a). Next, preprocessing such as smoothing is performed (Figure 3A-S302a). Next, normalization processing (sometimes expressed as correction processing) is performed (Figure 3A-S303a). Normalization processing is performed only on the light intensity axis. Next, the time derivative (sometimes simply expressed as derivative) or further time derivative of the waveform data is calculated to obtain a waveform related to coagulation rate or coagulation acceleration (Figure 3A-S304a). Finally, the waveform parameters are extracted as features (Figure 3A-S305a), and for example, the detection estimation of the test sample is performed from the correlation between the features of a sample for which the cause of prolonged blood coagulation time is known and the features of the test sample. At this time, preprocessing and normalization processing of the light intensity axis are not necessarily required, and the flow may be skipped. Furthermore, although not shown in the flowchart, it is disclosed that the axis representing the rate of change of light intensity of the waveform related to solidification rate or solidification acceleration may be normalized, and that normalization is performed by setting the maximum value of each waveform to 100%.

[0077] In contrast, this disclosure differs significantly from the comparative example in that it performs waveform fitting, normalizes both the time axis and the light intensity axis, and enables highly visible and intuitive detection estimation from the extracted features. In the example of the sample detection flow for Example 1 (Figure 3B), first, coagulation waveform (WaveSrc0) data obtained by coagulation reaction measurement is read (Figure 3B-S301b). Next, WaveSrc1 is generated by front-to-back difference processing of WaveSrc0 (Figure 3B-S302b). Next, Wave0 is generated by waveform fitting of WaveSrc0 (Figure 3B-S303b), and Wave1 is generated by waveform fitting of WaveSrc1 (Figure 3B-S304b). Next, Wave2 is generated by front-to-back difference processing of Wave1 (Figure 3B-S305b). In this disclosure, WaveSrc1 or Wave1 corresponds to the waveform related to coagulation rate in the comparative example, and Wave2 corresponds to the waveform related to coagulation acceleration in the comparative example. Next, Wave0, 1, and 2 are normalized based on the NRC method (Figure 3B-S306b). In this disclosure, normalization based on the NRC method means normalization based on reaction constants, and refers to normalization of both the time axis and the light intensity axis. As disclosed in the comparative example, when normalizing the light intensity axis, the waveforms related to coagulation waveform, coagulation velocity, or coagulation acceleration are not normalized by the maximum value of each waveform, but in this disclosure, normalization is performed that gives a relationship to Wave0, 1, and 2. Finally, features are extracted from the normalized waveforms (Figure 3B-S307b), and the features of the test sample are applied to a 2D map having a probability density distribution created using features extracted from samples for which the cause of prolonged blood coagulation time is known, and the detection estimation of the test sample is performed. In this disclosure as well, preprocessing such as smoothing may be performed as appropriate, for example, after loading WaveSrc0 (Figure 3B-S301b) and / or after generating WaveSrc1 (Figure 3B-S302b).

[0078] The primary inventive step of this disclosure is that, by performing normalization in the time axis direction using reaction constants measured with a fitting model of WaveSrc0, and by introducing a difference using the time constant included in the reaction constant (hereinafter referred to as the reaction time constant difference) instead of the time derivative of the coagulation waveform disclosed in Patent Documents 1-4 and Non-Patent Document 1 (which is simply described as the derivative), it becomes possible to estimate the cause of prolonged blood coagulation time while reducing the influence of the severity of symptoms expressed by the length of blood coagulation time.

[0079] [Example 2] Example 2 describes the detection of coagulation waveforms using a neural network (limited to examples using WaveSrc0 only, from the viewpoint of identity of the invention).

[0080] Here, we present an example of a method for estimating the cause of prolonged blood coagulation time using a neural network that takes coagulation waveform data as input, based on the time axis normalization method of the present disclosure. As mentioned above, estimation of the cause of prolonged blood coagulation time using a neural network that directly uses coagulation waveform data has not been disclosed as a set of techniques that can be easily reproduced by engineers in the relevant field. Therefore, the following three points constitute the core of this example. (1) Since the measured coagulation waveform is variable-length data, disclose a preprocessing technique that is suitable for a neural network. (2) To demonstrate that clustering of coagulation waveforms using neural networks is practical, demonstrate a detection accuracy of >90%. (3) To facilitate the reproduction of the above neural network and its application to measurement data from various devices and to samples from patients undergoing treatment, the configuration parameters such as the number of nodes and the number of hidden layers, as well as hyperparameters such as the number of epochs and batch size related to the learning conditions, should be presented.

[0081] (1) Disclosure of preprocessing steps First, we will discuss preprocessing techniques suitable for neural networks.

[0082] Figures 11A-C show experimental results illustrating the coagulation waveform (WaveSrc0) of commercially available samples and the coagulation waveform pre-processed to fit a neural network. Figure 11A shows WaveSrc0 of commercially available samples. Here, 103 waveforms were used, consisting of FVIII-deficient samples (46 waveforms), FIX-deficient samples (22 waveforms), LA-positive samples (6 waveforms), and normal samples (29 waveforms). The measurement device used was a Hitachi 3500 automated analyzer (manufactured by Hitachi High-Tech Corporation). The vertical axis represents the count value of the quantized data from the A / D converter corresponding to the scattered light intensity.

[0083] Figure 11B shows the waveform obtained by processing WaveSrc0 using the zero-padding method, which is well known in the fields of image and speech recognition, as a preprocessing step for the neural network. Here, the time axis was normalized according to Equation 4, which normalizes the time axis using the reaction time constant and response time, and the solidified waveform was resampled by linear interpolation as 501 equally spaced data points in the range of -5τ to +10τ. The vertical axis is normalized so that the value at the start of measurement is zero and a change in scattered light intensity of 10,000 counts is set to 1. This normalization is intended to be adapted to sigmoid and hyperblictangent, which are common activation functions for neural networks. The normalization used here is intended to use the change in scattered light intensity as a feature, and the normalization is Equation 8, which is different from Equation 5 described above.

[0084]

number

[0085] Furthermore, for time ranges -5τ to +10τ, data for periods where measurements were terminated and no data is available is interpolated with data "0" to equalize the data length.

[0086] Figure 11C shows waveform data obtained when a fitting model is used to imputate missing data. By using the parameters τ, T0, Y0, and Y1 obtained from the fitting model in Equation 2, it is easy to obtain continuous numerical data even for time after the measurement is completed. In this sense, the introduction of a fitting model is highly compatible with the detection of coagulation waveforms using neural networks. As a similarly simple method, it is also possible to imputate missing data by copying the values ​​of the last measured data point. Figure 11C differs from Figure 11B in that the data showing the scattered light intensity after the measurement end point is smoother. As preprocessing for use in neural networks, Figures 11B and 11C, in which the data length is fixed, are suitable.

[0087] (2) Proof of the neural network's structure and practical performance Next, we will discuss the structure of neural networks.

[0088] Figure 12 shows the structure of the neural network of this disclosure. Figure 12 is a summary by Keras, which is included in TensorFlow, a common framework for studying neural networks. As can be seen in the figure, the neural network of this disclosure can provide the practical performance shown above with a configuration of 4 hidden layers, 512 nodes, and tightly coupled layers throughout. The training epochs are 10,000, the batch size is 32, and the hyperblic tangent is used as the activation function. Regarding the activation function, it should be easy for engineers in this field to understand that equivalent detection performance can be obtained by using the sigmoid instead of the hyperblic tangent.

[0089] The following explanation assumes that the hyperparameters are constant, with 10,000 epochs and a batch size of 32. Furthermore, the coagulation waveform data from commercially available samples is divided into 60% for training, 20% for validation, and 20% for testing, and cross-entropy is used as the loss function. The reason for dividing the data into three parts is explained below. Since parameters such as weights that constitute the neural network are automatically updated for the training data, hyperparameters such as the number of hidden layers, number of nodes, dropout, number of epochs, and batch size are appropriately selected to minimize the cross-entropy of the validation data. To objectively evaluate the performance of the neural network determined in this way, it is necessary to evaluate the cross-entropy and detection accuracy using the test data not used in the above procedure.

[0090] Figures 13A-C show experimental results illustrating the relationship between training results and dropout. Here, we show results with 256 nodes and 3 hidden layers. Figure 13A shows the results when using a solidified waveform with data length equalized by the zero-padding method shown in Figure 11B. It can be seen that even when the dropout amount, which is effective in suppressing overfitting, is varied, the detection accuracy of the test data never exceeds 90%.

[0091] Figure 13B shows the results when using coagulation waveforms preprocessed by data extrapolation interpolation using the fitting model shown in Figure 11C. As can be seen in the figure, an accuracy of 90% or more, which is a practical benchmark for detection accuracy, can be obtained under conditions where the dropout amount is 0.4 or less. This demonstrates that the preprocessing method and the sample detection method based on coagulation waveform analysis using a neural network configuration are detection methods that can be easily reproduced and extended by technicians in the field.

[0092] Figure 13C shows the results when using coagulation waveforms with standardized data lengths achieved by interpolation using copies of the final measurement data points. Similar to the above, it can be seen that the detection accuracy is 90% or higher under conditions where the dropout amount is 0.4 or less. This method has the advantage of requiring less computation compared to the interpolation method shown in Figure 13B. When the sample data is further increased, or when noise is introduced during measurement for any reason, data extrapolation interpolation using a fitting model is superior in terms of the stability of detection performance. Hereafter, unless otherwise specified, preprocessing will be performed by data extrapolation interpolation using a fitting model.

[0093] (3) Presentation of configuration parameters such as the number of nodes and the number of hidden layers, and hyperparameters related to the learning conditions such as the number of epochs and batch size. Next, we will discuss the range of parameters suitable for implementing this disclosure.

[0094] The optimal number of epochs and dropout rate for training a neural network are as described above. Here, we will discuss the number of nodes and provide examples as numerical values ​​that indicate the configuration of a neural network.

[0095] Figures 14A and 14B show experimental results illustrating the hyperparameter range at which a detector combining the preprocessing method and neural network of this disclosure exhibits practical performance. Figure 14A shows the relationship between the number of hidden layers and detection accuracy. Here, the number of nodes is 1024 and the dropout is 0.25. As can be seen in the figure, the detection method of this disclosure can achieve a practical detection accuracy of 90% or more with 3 or more hidden layers.

[0096] Figure 14B shows experimental results illustrating the relationship between the number of nodes and detection accuracy. Here, the case with 3 hidden layers is shown. As can be seen in the figure, the detection method of this disclosure can achieve a practical detection accuracy of 90% or more with 16 or more nodes. These results demonstrate that the detection method of this disclosure can achieve practical detection accuracy by constructing a neural network with 3 or more hidden layers and 16 or more nodes.

[0097] Figure 15 shows experimental results demonstrating that the preprocessing method and neural network-based screening method of this disclosure achieve sufficient suppression of overfitting. Here, the relationship between the number of nodes and the cross-entropy for training, validation, and test samples is shown. In the fields of speech and facial recognition, where neural networks are widely used, the cross-entropy is generally around 0.1 to 0.3. In contrast, in the screening method of this disclosure, when the number of nodes is 16 or more, the cross-entropy is approximately 0.1 or less, and it can be seen that the cross-entropy value is almost the same for each sample group of training, validation, and test. This result indicates that overfitting can be sufficiently suppressed by the neural network configuration shown in this disclosure.

[0098] As described above, according to this disclosure, after preprocessing the coagulation waveform obtained from a sample to a predetermined data length using the reaction time constant and response time quantified by a fitting model, it is possible to estimate the cause of prolonged blood coagulation time with practical detection accuracy using a neural network with 3 or more hidden layers and 16 or more nodes. Furthermore, it goes without saying that since the neural network detector outputs the probability of belonging to each cluster of causes of prolonged blood coagulation time as output, it is obvious that the neural network detector shown here can also support the display of belonging probabilities illustrated in Figure 8.

[0099] Furthermore, by applying the NRC method to the waveform data used for training, it is possible to further improve the detection accuracy.

[0100] (Effects of Example 2) By performing appropriate preprocessing on coagulation waveforms with variable data lengths, it is possible to estimate the causes of prolonged blood coagulation time with practical accuracy using a neural network.

[0101] [Example 3] Example 3 describes the application to a cross-mixing test.

[0102] The above describes the detection of causes of prolonged blood coagulation time using coagulation waveforms obtained by measuring samples taken from subjects. On the other hand, when the blood coagulation time of a sample taken from a subject is prolonged, the cross-mixing test, which measures the blood coagulation time of multiple mixed plasma samples obtained by mixing the sample in question with a normal sample in a predetermined ratio, and screens for the cause of prolonged blood coagulation time from the graph shape plotted against the mixing ratio of the samples, is recognized as a representative method for detecting causes of prolonged blood coagulation time. According to this disclosure, from a series of coagulation waveforms obtained from multiple mixed plasma samples, it is possible to obtain not only the dependence of blood coagulation time on the mixing ratio, but also multiple feature quantities attributable to the cause of prolonged blood coagulation time included in the series of coagulation waveforms and the probability of belonging to each cause of prolongation. Therefore, the sample detection method of this disclosure can be easily applied to coagulation waveform data obtained by the cross-mixing test, and is expected to improve detection accuracy compared to conventional methods. When prioritizing compatibility with conventional methods, for example, it is preferable to quantify the dependence of blood coagulation time on the mixing ratio, and simultaneously quantify blood coagulation factor VIII deficiency and blood coagulation factor IX deficiency from the coagulation waveform of the sample collected from the subject, and map them in a two-dimensional space.

[0103] [Example 4] In Example 4, the characteristic features of the coagulation waveform obtained by normalization processing using the NRC method based on reaction constants are expanded.

[0104] In other words, in Example 4, features are extracted not only from the first normalized waveform (WaveNor0), the second normalized waveform (WaveNor1), and the third normalized waveform (WaveNor2), but also from the waveform data obtained by applying nonlinear calculations to them. Based on the distribution information of features for each cause of prolonged blood coagulation time, which has been prepared in advance, and the extracted features, the cause of the prolonged blood coagulation time in the test sample is estimated.

[0105] The method for estimating the cause of the prolonged blood coagulation time in the test specimen of Example 4 is as follows: (A) Obtain a coagulation waveform that shows the change in light intensity over time due to the coagulation reaction of the reaction solution produced by mixing the test sample, which consists of plasma obtained by separating blood acquired from the subject, with the reagent. (B) Obtain the first waveform by differential processing on the coagulation waveform. (C) Obtain a first fitted waveform by fitting the coagulation waveform. (D) Obtain a second fitted waveform by fitting the first waveform. (E) Obtain a third waveform by performing a front-to-back difference process on the second fitting waveform. (F) Obtain the first normalized waveform, the second normalized waveform, and the third normalized waveform by normalizing the light intensity axis and time axis of the first fitted waveform, the second fitted waveform, and the third waveform, respectively. (G) Extracting feature quantities from at least one of the first normalized waveform, the second normalized waveform, the third normalized waveform, and waveform data obtained by applying nonlinear calculations to them. (H) Estimate the cause of prolonged blood coagulation time in a test sample based on pre-prepared distribution information of characteristic quantities for each cause of prolonged blood coagulation time and characteristic quantities extracted from the coagulation waveform of the test sample. It has.

[0106] Hereinafter, in order to simplify the diagrams and aid in understanding the technology, the FVIII-deficient sample group will be referred to as "FVIII-deficient," the FIX-deficient sample group as "FIX-deficient," the LA-positive sample group as "LA," and the normal sample group as "Normal." Here, coagulation waveforms measured using a Hitachi 3500 automated analyzer (manufactured by Hitachi High-Tech Corporation) with commercially available samples are used. The measurement reagent is Coagupia APTT-N (manufactured by Sekisui Medical Co., Ltd.). In addition to the commercially available products described in Example 1, Coagupia Control PN I (manufactured by Sekisui Medical Co., Ltd.) and Coagtrol N (manufactured by Sysmex Corporation) were used as normal samples.

[0107] FIG. 16 compares WaveSrc0 of one representative sample each of the FVIII-deficient sample and the FIX-deficient sample. As can be seen in FIG. 16, the blood coagulation times are almost the same at about 100 seconds for both. On the other hand, it can be seen that the FVIII-deficient sample has a slower convergence of the reaction (the time from the initial value to the saturation value of the scattered light intensity is longer) compared to the FIX-deficient sample. This represents the difference in the influence of the deficient factors of both on the reaction process. This feature is, for example, in Equation 2, where T0 of both is approximately equal, and T1 is such that FIX deficiency < FVIII deficiency. In the present disclosure, (T0 / T1) was subtracted from vec0_x.

[0108] Note that the coagulation time is calculated as the time until the level reaches 50% when the initial value of the scattered light intensity is 0% and the saturation value is 100%, but the method for calculating the blood coagulation time is not limited to this method. For example, the time until reaching an arbitrary level other than 50% may be calculated as the blood coagulation time.

[0109] FIG. 17A shows the experimental results of the relationship between the maximum value of the first derivative of the coagulation waveform and the log coagulation time. This maximum value of the first derivative corresponds to the index |min1| described in Patent Document 1. As shown in FIG. 17A, the R2 value obtained by least-squares fitting with a quadratic function was 0.9323. That is, it can be seen that the maximum value of the first derivative has a very high correlation with the blood coagulation time. It can be seen that the maximum values of the first derivatives of the FVIII-deficient sample group, FIX-deficient sample group, and LA-positive sample group (LA) in the figure are distributed in a very small region compared to the overall change amount of the index including the normal sample group.

[0110] Figure 17B shows the experimental results illustrating the relationship between vec1_y and log coagulation time in Example 4. The R² value obtained by least-squares fitting with a quadratic function was 0.12. This indicates that vec1_y in Example 4 has almost no correlation with blood coagulation time. The distribution of vec1_y in the FVIII-deficient sample group, FIX-deficient sample group, and LA-positive sample group (LA) in the figure shows a large change compared to the overall change in the index including the normal sample group. Moreover, it can be seen that each sample group is separated into independent regions where the vec1_y values ​​are in the relationship FVIII deficiency > LA > FIX deficiency.

[0111] The results for the maximum value of the second derivative of the coagulation waveform obtained by scattered light measurement, which corresponds to another index |min2| described in Patent Document 1, were the same as in Figure 17A. As is clear from the above experimental facts, by normalization based on the NRC method of this disclosure, a waveform independent of blood coagulation time is generated and feature quantities are extracted, thereby improving the ability to detect the cause of prolonged blood coagulation time compared to conventional indices.

[0112] Figure 18A shows the experimental results for WaveNor0, which is the coagulation waveform of the commercially available sample from Example 4 normalized according to the NRC method, illustrating the representative waveforms for each sample group. It can be seen that there are significant differences among the sample groups in the rising edge of the waveform.

[0113] Figure 18B shows the experimental results for WaveNor1, which is the coagulation waveform of the commercially available sample from Example 4 normalized according to the NRC method, illustrating the representative waveforms for each sample group. Similarly, it can be seen that the differences in waveforms for each sample group are clearly evident.

[0114] Figure 18C shows the experimental results for WaveNor2, which is the coagulation waveform of the commercially available sample from Example 4 normalized according to the NRC method, illustrating the representative waveforms for each sample group. Similarly, it can be seen that the differences in waveforms for each sample group are clearly evident.

[0115] Figure 19 shows the experimental results of the WaveNor2 frequency spectrum for Example 4. Here, for representative samples from each sample group, a WaveNor2 waveform array was generated with i=0 to 2047 in the normalized time interval -50T1 to +50T1, and a Fourier transform was performed. As feature quantities for each sample group, for example, the amplitude spectral values ​​of frequency FREQ1 (minimum value of HB, i=40) and FREQ2 (maximum value of HA, i=115) can be added.

[0116] Figure 20 summarizes an example of the feature set in Example 4. Generally, a larger number of features is preferable because it allows for differentiation of the causes of prolonged blood coagulation time from different perspectives. However, there are limitations. For example, when there are features α and β, it is not preferable to add a new feature γ = aα + bβ (where a and b are constants). This is because γ is a linear combination of α and β, and adding γ to the features does not increase the amount of information for differentiation, and differentiation of the causes of prolonged blood coagulation time from different perspectives cannot be performed. Furthermore, the addition of γ increases the influence of measurement noise and numerical calculation errors, impairing the reliability of differentiation of the causes of prolonged blood coagulation time. On the other hand, waveform data obtained by mapping (converting into new information) normalized waveforms (WaveNor0, WaveNor1, WaveNor2) to another space using nonlinear operations such as power operations, differential operations, Fourier transform operations, Galois transform operations, and convolution operations is easy to add effective features to because mathematical orthogonality with respect to the original data is guaranteed. Of course, it is also possible to list multiple candidate features from the same waveform data, calculate the cross-correlation coefficient for all combinations of these features, and select features by eliminating those with large cross-correlation coefficients (which reduce the amount of additional information for discrimination). Below, we show an example of adding values ​​obtained from waveform data mapped to separate, independent spaces. As shown in the figure, here we can use two quantities from each of the seven feature spaces (#0 to #6) as features. (1) Feature Space #0 These are vec0_x-(T0 / T1) and vec0_y from Example 1. (2) Feature Space #1 These are vec1_x and vec1_y from Example 1. (3) Feature Space #2 These are vec2_x and vec2_y from Example 1. (4) Feature Space #3 These are the RMS errors between WaveNor2 and the representative waveform of the FIX-deficient sample group, and between WaveNor2 and the representative waveform of the normal sample group. (5) Feature Space #4 These are the x and y coordinates of the intersection point of WaveNor1 and WaveNor2. (6) Feature Space #5 Figure 19 shows the amplitude spectral values ​​of FREQ1 and FREQ2 in the frequency spectrum of WaveNor2. (7) Feature Space #6 When the coagulation time is expressed in terms of t50, the following values ​​are defined by the empirical formula representing the severity (Severity_Index) in equation 9, and the RMS error between WaveNor1 and the representative waveform of the normal sample group.

[0117] Note that the representative waveform described above may also be an average waveform. An average waveform is obtained by normalizing waveforms obtained from a group of samples with similar blood coagulation factor activity levels based on the NRC method, and then averaging the normalization levels for each normalization time.

[0118]

number

[0119] Furthermore, Figure 20 includes recommended x and y ranges for illustrating each feature space, which is convenient when implementing numerical analysis. In Example 4, the figures will be disclosed according to these ranges.

[0120] In Example 4, as a feature differentiation process, differentiation is performed in each feature space according to a probability density distribution based on a two-dimensional Gaussian mixture distribution. The basic Gaussian mixture distribution is well known to engineers in this field, so it will not be explained here. In Example 4, in order to improve the reliability of the differentiation results, a Gaussian mixture distribution that allows the influence of measurement error to be physically inherent in accordance with the central limit theorem was used as the basis, and an upper limit of the Mahalanobis distance was added to the probability density distribution of each cluster. The Mahalanobis distance is a quantity equivalent to the standard deviation of a one-dimensional Gaussian distribution, and if we refer to it using σ as the unit according to convention, an upper limit of about 3 to 6σ is preferable. By adding an upper limit to the Mahalanobis distance, the basic shape of each cluster region in the feature space becomes an ellipse. When two clusters are adjacent, the boundary is the condition that the probability of belonging to each is 50%. The data of one sample is displayed as one point in each feature space, and at the same time, the probability of belonging to each sample group in the cluster is calculated. The identification process in Example 4 is performed by averaging the probability of belonging to a cluster in each feature space, and the cluster and probability of belonging to the sample group to which the sample belongs with the highest probability are obtained as the identification result. In this example, an example was shown in which the probability of belonging to each sample group cluster of the sample was obtained by averaging the probability of belonging to each sample group cluster in each feature space. However, for example, the probability of belonging to a cluster in each feature space may be weighted and the sample group with the highest probability of belonging may be identified, or identification may be performed based on the probability of belonging to a sample group cluster in any of the feature spaces.

[0121] The introduction of an upper limit on the Mahalanobis distance in Example 4 means that, in terms of the standard deviation of the measured data, for example, data points that are more than 3σ away from the center of a sample group cluster are not assigned to that cluster. In general statistical distributions and classification methods such as SVM, K-means, and deep neural networks, the space is divided into a specified number of clusters, so new or abnormal data are classified into one of the known clusters.

[0122] Data generation for each cluster can be performed by using coagulation waveform data from FVIII-deficient samples (FVIII deficiency), FIX-deficient samples (FIX deficiency), LA-positive samples (LA), and normal samples (Normal). From these, two feature data points are extracted in each of the seven feature spaces, and distribution data consisting of the central position, height, and variance-covariance matrix can be calculated to form clusters for each sample group.

[0123] A note is added regarding feature space #6. Figure 21 shows the experimental results of the commercially available sample group from Example 4 displayed on feature space #6. As shown in Figure 21, this uses the Severity_Index shown in Equation 9 as the x-axis feature, and while other feature spaces differentiate clusters of four sample groups, this is specialized for differentiating between normal samples and others. Among the prepared commercially available samples, it was confirmed that Borderline Factor Assay Control Plasma (George King Bio-Medical, Inc.) with a coagulation time of 42 seconds and LA-positive samples with a coagulation time of 46 seconds could be separated by a Mahalanobis distance of 3σ or more and differentiated almost perfectly. According to the attached assay values, the FVIII and FIX activity values ​​of Borderline Factor Assay Control Plasma were 41% and 55%, respectively. Since this sample could not be classified into any of the following groups: LA-positive sample group, coagulation factor inhibitor-positive sample group, FVIII-deficient sample group, or FIX-deficient sample group, for convenience, it was treated as a normal sample in this example. The normal range can be adjusted by changing the upper limit of the Mahalanobis distance.

[0124] Figure 22A is an example of a display showing a summary plot of the commercially available sample group from Example 4 on the feature space. In Figure 22A, the commercially available sample group used in the experiment of Example 4 is plotted on seven feature spaces, and the WaveNor1 obtained by normalization and a summary of the accuracy of the discrimination results are displayed. Here, the upper limit of the Mahalanobis distance was set to 4σ. The accuracy was 100% (10⁶ / 10⁶*100%).

[0125] Figure 22B shows an example of the identification results for one FVIII-deficient sample from Example 4. In the feature space of Example 4, the sample is represented as a single point, and the probability of belonging to each sample group cluster is presented. In this case, the probabilities are FVIII deficiency (83%), LA (17%), FIX deficiency (0%), and Normal (0%).

[0126] [Example 5] Example 5 describes a specific method for applying the technology of the present invention to a cross-mixing test, following the overview shown in Example 3. Features are extracted from the first normalized waveform (WaveNor0), second normalized waveform (WaveNor1), and third normalized waveform (WaveNor2) of each of a series of sample specimens obtained by mixing plasma (obtained by separating blood from a subject) with normal plasma at multiple mixing ratios, and from the waveform data obtained by applying nonlinear calculations to these features. At least one of these features is then used to estimate the cause of the prolonged blood coagulation time in the subject specimen.

[0127] Regarding the cross-mixing test, paragraph 0121 of Patent Document 1 states that "the graph patterns of LA-positive samples and coagulation factor inhibitor-positive samples were similar. Therefore, differentiation between the two based on changes in the graph pattern is inevitably a qualitative evaluation, and it is clear that it is difficult for non-experts to make a judgment." It is a well known fact that it is difficult to differentiate between LA-positive samples and coagulation factor inhibitor-positive samples based on the relationship between the mixing ratio of normal plasma and test plasma and coagulation time. In this embodiment, the objective is to apply the technology for solving problem (1) to the cross-mixing test and provide a method that enables quantitative differentiation of the cause of prolonged blood coagulation time, not only in the case of LA-positive samples and coagulation factor inhibitor-positive samples, but also in the case of FVIII-deficient samples and FIX-deficient samples, even for non-experts.

[0128] The method for estimating the cause of the prolonged blood coagulation time in the test specimen of Example 5 is: (A) Prepare a series of sample specimens by mixing plasma obtained by separating blood from the subject with normal plasma at multiple mixing ratios. (B) Obtain a series of coagulation waveforms showing the change in light intensity over time due to the coagulation reaction of the reaction solution produced by mixing a series of sample specimens and reagents. (C) Obtaining a series of first waveforms by performing a before-and-after difference processing on each of the series of coagulation waveforms. (D) Obtaining a series of first fitted waveforms by fitting each of the series of coagulation waveforms, (E) Obtaining a series of second fitted waveforms by fitting each of the series of first waveforms, (F) Obtaining a series of third waveforms by performing front-to-back difference processing on each of the series of second fitting waveforms. (G) Obtaining a series of first normalized waveforms, a series of second normalized waveforms, and a series of third normalized waveforms by normalizing the light intensity axis and time axis of a series of first fitted waveforms, a series of second fitted waveforms, and a series of third waveforms. (H) Extracting features from at least one of a series of first normalized waveforms, a series of second normalized waveforms, a series of third normalized waveforms, and a series of waveform data obtained by applying nonlinear operations to them. (I) Using at least one of a series of features extracted from a series of coagulation waveforms, estimate the cause of the prolonged blood coagulation time in the sample specimen. It has. Furthermore, the computer system, such as the control computer 120 described in Example 1, executes each of the processes (A) to (I) described above.

[0129] In this embodiment, for the sake of simplifying the diagrams and aiding in understanding the technology, the FVIII-deficient sample group will be referred to as "FVIII-deficient," the FIX-deficient sample group as "FIX-deficient," the LA-positive sample group as "LA," and the coagulation factor inhibitor-positive sample group (here, a commercially available product to which a coagulation factor inhibitor has been added to the FVIII-deficient sample group) as "FVIII-deficient inh." Commercially available samples were used in the experiment. Specifically, they were as follows: The FVIII-deficient sample was Factor VIII Deficient Plasma (Precision BioLogic, Inc.). The FIX-deficient sample was Factor IX Deficient Plasma (Precision BioLogic, Inc.). The LA-positive samples were Lupus Positive Control and Weak Lupus Positive Control (Precision BioLogic, Inc.). Coagulation factor inhibitor-positive samples are Mild Factor VIII Inhibitor Plasma and Strong Factor VIII Inhibitor Plasma (Affinity Biologicals, Inc.). Normal samples are Pooled Normal Plasma (Precision BioLogic, Inc.). Here, FVIII-deficient samples, FIX-deficient samples, LA-positive samples, and coagulation factor inhibitor-positive samples were used as test samples. In this example, data from repeated measurements were treated as one sample, resulting in a total of 34 test samples (FVIII deficiency: 7, FIX deficiency: 7, LA: 8, FVIII-deficient inhibitor: 12). The measuring instrument was a Hitachi 3500 automated analyzer (Hitachi High-Tech Corporation), and the reagent was Coagupia APTT-N (Sekisui Medical Co., Ltd.).

[0130] In this example, a series of plasma samples were prepared by mixing the test subject and the normal subject at different ratios (10:0, 9:1, 8:2, 5:5, 2:8, 1:9, 0:10). Immediately after sample preparation (immediate type), and after heating the sample at 37°C for 2 hours (delayed type), measurement of the coagulation waveform was performed. For one test subject, measurement of 14 (7 series of immediate type and delayed type) series of coagulation waveforms and calculation of the feature quantities extracted from the waveforms standardized based on the NRC method were performed. Hereinafter, in order to represent the mixed plasma sample, seven mixing ratios (0.0, 0.1, 0.2, 0.5, 0.8, 0.9, 1.0) of the test subject and the measurement conditions (immediate or delayed) are used.

[0131] Figure 23 is a schematic diagram showing a method of dimensionally compressing a series of coagulation time data obtained for one test subject in the cross-mixing test of Example 5 into two area indices F1 and F2. In the cross-mixing test, there are conditions where the change in blood coagulation time is small with respect to the change in the mixing ratio of the test subject. At this time, the obtained blood coagulation time data is greatly affected by the mixing conditions and measurement errors and varies. Therefore, the average value of the two coagulation times (immediate and delayed) in the case of mixing ratio = 0.0 is t s , and the average value of the two coagulation times (immediate and delayed) in the case of mixing ratio = 1.0 is t e . Based on the following formula, by quantifying the difference from the linearly approximated coagulation time as the area indices F1 and F2, standardization is performed by reducing the variation of the measurement results by dimensional compression and averaging. In the following formula, n is an integer (n = 1, 2), N is the number of mixing ratios (here N = 7), i is an index integer representing the mixing ratio, t i is the coagulation time, t^ i is t s and t e linearly approximated coagulation time from, Δx i is the integration interval corresponding to the i-th mixing ratio x i , and is the width connecting each midpoint with x i-1 , x i+1 (in the case of the left end, x i-1 = 0.0, in the case of the right end, x i+1 = 1.0). Under the experimental conditions implemented in this example, the mixing ratio x iFor =(0.0, 0.1, 0.2, 0.5, 0.8, 0.9, 1.0), Δx i =(0.05, 0.1, 0.2, 0.3, 0.2, 0.1, 0.05).

number

[0132] According to the above formula, if we calculate the index F1 for immediate measurement and the index F2 for delayed measurement, we can obtain indices of the same dimension, for example, even if the mixing ratios used for immediate measurement and delayed measurement are different, by setting -1 ≤ F1 ≤ +1 and -1 ≤ F2 ≤ +1.

[0133] Figure 24 shows the results of cross-mixing test sample identification using the multidimensional Gaussian mixture distribution of Example 5. Here, we describe the case where a series of 14 test samples are prepared and tested by varying the mixing ratio of one original sample and a normal sample, and the immediate and delayed time conditions. The method disclosed in Example 1 was the case where one coagulation waveform was associated with one original test sample. Extending this to the case where 14 coagulation waveforms are associated with one original test sample results in 28 features being quantified per feature space. In each feature space, the identification accuracy was quantified by extending from the 2D Gaussian mixture distribution shown in Example 1 to a 28-dimensional Gaussian mixture distribution. Furthermore, as a conventional example, raw blood coagulation time data was identified using a 14-dimensional Gaussian mixture distribution, and the dimensionality-reduced indices (F1, F2) shown above were identified using a 2D Gaussian mixture distribution, and the results were summarized. As shown in Figure 24, the differentiation accuracy ranged from 32.4% to 88.2%, with the lowest differentiation accuracy when using raw blood coagulation time data and the highest when using the indices (F1, F2). Although the two are essentially the same, as mentioned above, higher differentiation accuracy was obtained when the influence of variability was reduced (F1, F2). Therefore, in the cross-mixing test, the preparation conditions of the mixed series of plasma samples are difficult to handle by simply extending the method disclosed in Example 1.

[0134] When using the indices for each feature space (28 dimensions) obtained by normalization based on the NRC method in this embodiment, a maximum discrimination accuracy of 76.5% was obtained in feature space #3. We believe that measurement variability plays a strong role in this as well.

[0135] Figure 25A shows the cluster regions of the indicators (F1, F2) in Example 5. As can be seen in the figure, the FVIII deficiency cluster and the FIX deficiency cluster are formed in close proximity, almost overlapping. Although they are independent clusters as blood coagulation factor deficiency groups, it is difficult to make a highly accurate distinction between FVIII and FIX factors. While it is difficult to identify which blood coagulation factor is deficient, it is possible to determine that it is a factor deficiency type, which is characteristic of the conventional cross-mixing test. Furthermore, the LA-positive cluster (LA) is formed overlapping the central part of the coagulation factor inhibitor-positive cluster (FVIII deficiency inh), and the experimental results reproduce the description in paragraph 0121 of Patent Document 1, which states, "The changes in the graph patterns of LA-positive samples and coagulation factor inhibitor-positive samples were similar. Therefore, differentiation between the two based on changes in the graph patterns is necessarily a qualitative evaluation, and it is difficult to make a judgment unless you are an expert."

[0136] Figure 25B shows the experimental results of adding data points to the cluster region of the indicators (F1, F2) in Example 5. As can be seen in the figure, regarding the coagulation factor inhibitor-positive cluster (FVIII deficiency inh), the commercially available samples prepared had two types of inhibitor titers, "Mild" and "Strong," and were distributed in the lower left and upper right, flanking the LA-positive cluster (LA). Therefore, assuming a patient sample with a titer somewhere between "Mild" and "Strong," FVIII deficiency inh ≈ LA in this method, making differentiation impossible.

[0137] Figure 26 illustrates the indices (G1, G2) of Example 5. As described above, it is difficult to separate the FVIII-deficient cluster from the FIX-deficient cluster using only blood coagulation time. Therefore, we decided to use the length of the perpendicular line drawn from the position vector to the data point P representing 100% of the test sample, with the midpoint of the line segment connecting the center of the FVIII-deficient cluster and the center of the FIX-deficient cluster in each feature space as the origin. The indices (G1, G2) corresponding to immediate measurement and delayed measurement are defined by the following equations.

[0138]

number

[0139] In the above equation, the center of the FVIII-deficient cluster is +0.5, and the center of the FIX-deficient cluster is -0.5. A higher value indicates a greater likelihood of an FVIII-deficient sample.

[0140] Figure 27A is an example of an experimental result plotting immediate (F1, G1) and delayed (F2, G2) indices, which are a combination of index F calculated from a series of coagulation waveforms obtained from one sample in Example 5 and index G calculated from the same one sample using feature space #1 according to equation 11. Here, one sample is represented as two points (F1, G1) and (F2, G2) and the straight line connecting them. The appearance is similar to a schematic diagram of a dipole. Hereafter, this will be referred to as a dipole. As can be seen in the figure, the clusters of the four sample groups (FVIII deficiency, FIX deficiency, LA, FVIII deficiency inh) are formed in almost independent regions. The upper limit of the Mahalanobis distance was set to 6σ. The reason why the region of FVIII deficiency inh is relatively large is, as mentioned above, because there are two types of inhibitor titers, "Mild" and "Strong".

[0141] Figure 27B shows the experimental results (example display) plotting the indices (F1, G1) and (F2, G2) obtained from 34 samples in Example 5. Here, index G was calculated from the data for feature space #1 according to formula 11. It can be seen that the dipoles representing each sample are plotted separately according to the cause of prolonged blood coagulation time. The differential accuracy was 97%. It can be seen that the "Mild" and "Strong" groups of the coagulation factor inhibitor-positive group (FVIII deficiency inh), the LA-positive group (LA), the FVIII deficiency group, and the FIX deficiency group are plotted spatially separated. It was confirmed that this method significantly improves the differential accuracy compared to conventional differential diagnosis using only blood coagulation time.

[0142] Next, as a specific example of a suitable method for applying the technology of the present invention to a cross-mixing test, we will show a method for discriminating the results of a cross-mixing test using the feature space #0 to #5.

[0143] In the feature space, the sequence of seven points obtained from immediate and delayed measurements of a single sample, linked in ascending or descending order of the mixing ratio of the test samples, is referred to here as trajectory data. Figure 28A shows the experimental results (example of display) showing the average shape of the trajectory data for each cause of prolonged blood coagulation time obtained by the cross-mixing test in Example 5. As can be seen in the figure, the shape of the trajectory data is characteristic for each cause of prolonged blood coagulation time. For example, in terms of morphology (shape similarity), the FVIII deficiency group and the coagulation factor inhibitor-positive group (FVIII deficiency inh) are similar, but the large discrepancy between immediate (Wait=0h) and delayed (Wait=2h) in the coagulation factor inhibitor-positive group (FVIII deficiency inh) can be judged as a correct result in light of the essence of the cross-mixing test.

[0144] One effective method for identifying the cause of prolonged blood clotting time from trajectory data is to apply handwriting recognition technology such as CNN (Convolutional Neural Network). This method involves connecting each point constituting the trajectory data using straight lines or spline interpolation to generate image data, which is then discriminated using the same techniques as handwriting recognition. This method has the ability to differentiate many causes of prolonged blood clotting time, corresponding not only to the clusters of the four sample groups discussed here, but also to the patient's condition during treatment (including medication, etc.).

[0145] Another method for identifying the cause of prolonged blood coagulation time from trajectory data is to use a multidimensional Gaussian mixture distribution. However, as shown in Figure 24, this method has been found to be inaccurate. The main reason is that, according to the aforementioned conditions, the clusters of each sample group are defined as being inside a 28-dimensional solid. However, the mathematical constraint that this solid must follow a Gaussian distribution regardless of which plane it is sliced ​​from results in insufficient degrees of freedom for differentiating trajectory data from the cross-mixing test. As is well known, a multidimensional Gaussian mixture distribution must be non-degenerate, but this constraint is relaxed in a 2-dimensional Gaussian mixture distribution. Therefore, by separating the 28-dimensional trajectory data into a 2-dimensional x14 space and using the product of the probability of belonging to the clusters of the sample groups obtained in each space to identify the cause of prolonged blood coagulation time, it is possible to identify the cause of prolonged blood coagulation time without being constrained by the limitations of the continuity of trajectory data with respect to changes in the mixing ratio of the samples.

[0146] Figure 28B shows the experimental results (example) of separating and differentiating the trajectory data of each sample in a cross-mixing test into a 2D x14 space. Here, differentiation was performed based on the probability of belonging to the normalized sample group into clusters, using the product of the belonging probabilities obtained in each individual space, and the trajectory data was plotted in four regions. In the figure, the results for feature space #4 are shown as an example. A significant improvement was demonstrated compared to Figure 24 (differentiation accuracy = 32.4%~88.2%), which is a simple extension of the method disclosed in Example 1, and the differentiation accuracy was 100%.

[0147] The complexity of sample preparation and incubation associated with cross-mixing tests places a burden on the tester. Furthermore, additional blood draws for testing increase the burden on the patient. Therefore, it is preferable to use as few sample conditions as possible for cross-mixing tests (total number of samples prepared due to differences in mixing ratios and incubation).

[0148] Figure 29 is a diagram summarizing the conditions for determining the total number of samples required for the identification method in Example 5. As mentioned above, in this case, 14 coagulation waveforms (7 series of immediate and delayed types) were obtained for one sample. Each row shows the conditions for the coagulation waveform, and hatched conditions indicate that they are not used for coagulation waveform analysis. In the figure, "Condition A" is the case where all 14 coagulation waveforms are used, and "Condition B" is the condition in which identification is performed using the 7 coagulation waveforms excluding the hatched waveforms.

[0149] Figure 30 is a schematic diagram showing the relationship between the number of coagulation waveforms used for discrimination in Example 5 and the worst-case probability (the minimum value of Pti across all samples, where Pti is the probability of the i-th sample belonging to the correct sample cluster). If the worst-case probability on the vertical axis is 50% or higher, it can be assumed that no discrimination error occurs, and if it is 50% or lower, a discrimination error occurs. Here, the performance of three methods was compared: (1) discrimination using indices F1 and F2 (F Factor Method), (2) discrimination using a combination of indices F1, G1, F2, and G2 (FG Factor Method), and (3) discrimination by dividing trajectory data into two-dimensional sets (Orbit Discrimination Method). (1) For the F Factor Method, the worst-case probability was 50% or lower under all conditions, and a discrimination error always occurred. (2) For the FG Factor Method, the discrimination performance improved, and the worst-case probability became 50% or higher when five or more coagulation waveforms were used. (3) The Orbit Discrimination Method exhibited the best performance, and when using two or more coagulation waveforms, the worst-case probability was 77% or higher.

[0150] The F-Factor Method uses an index that quantifies the relationship between the mixing conditions of the test sample and the normal sample and the blood coagulation time, according to the definition described above. Therefore, its differential performance can be considered to be equivalent to the differential results of the conventional cross-mixing test.

[0151] The FG Factor Method improves the diagnostic performance as shown in the figure by incorporating the shape characteristics of the coagulation waveform quantified by the NRC method of the present invention as indicators G1 and G2, in addition to indicators F1 and F2. Furthermore, the amount of data processed is small, at four per sample, and information processing calculations can be performed in a shorter time compared to the Orbit Discrimination Method. Therefore, the FG Factor Method is suitable for use in conjunction with conventional methods that differentiate the cause of prolonged blood coagulation time based on the relationship between the mixing ratio of the sample and the blood coagulation time, and is suitable for use as auxiliary information when judging the diagnostic results.

[0152] The Orbit Discrimination Method performs discrimination of cross-mixing test results using only the index of the feature space normalized and quantified by the NRC method of the present invention, thereby achieving high discrimination performance. At the same time, since there is no calculation of index F using formula 10, Δx i The relationship with adjacent data points via this method enables independent information processing, allowing for flexible adaptation to any selection of mixing ratios. Due to these advantages, the Orbit Discrimination Method is a suitable means of achieving fully automated discrimination of cross-mixing tests using machine learning techniques.

[0153] Figure 31 shows an example of the display of the identification results of the cross-mixing test in Example 5. The relationship between the analysis unit 130, the control computer 120, and the communication interface 124, and the display of the results of the test specimen on the display unit 118c, follow the results in Figure 9.

[0154] According to Example 5, a series of coagulation waveforms obtained from the cross-mixing test enable quantitative differentiation based on the probability of LA-positive samples, coagulation factor inhibitor-positive samples, FVIII-deficient samples, and FIX-deficient samples.

[0155] [Differentiation] This disclosure is not limited to the embodiments described above, but includes various modifications. The embodiments described above are explained in detail for the purpose of making this disclosure easy to understand, and are not necessarily limited to those having all the configurations described. It is possible to replace parts of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add configurations from other embodiments to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace parts of the configuration of each embodiment with the same or other configurations.

[0156] For example, in Example 1, the cause of prolonged blood coagulation time was estimated from the features of WaveNor0, 1, and 2. However, in this disclosure, the cause of prolonged blood coagulation time may be estimated from at least two features of WaveNor0, 1, and 2. For example, the cause of prolongation may be estimated from the features of the x and y components of WaveNor0, or from the features of the x component of WaveNor0 and the y component of WaveNor2. [Explanation of symbols]

[0157] 100: Automatic analyzer, 101: Sample dispensing mechanism, 102: Sample disk, 103: Sample container, 103a: Sample, 104: Reaction vessel, 105: Syringe pump for samples, 106: Reagent dispensing mechanism, 107: Reagent disk, 108: Reagent container, 108a: Reagent, 109: Reagent heating mechanism, 110: Syringe pump for reagents, 111: Reaction vessel stock section, 112: Reaction vessel transport mechanism, 113: Detection unit, 114: Reaction vessel installation section, 115: Light source, 116: Detection unit (optical sensor), 117: Reaction vessel disposal section, 118: Operating computer, 118a: Mouse, 118b: Keyboard, 118c: Display unit, 119: Memory unit, 120: Control computer, 121: A / D converter, 122: Incubator, 123: Printer, 124: Communication interface, 125: Analysis computer, 130: Analysis unit

Claims

1. To obtain a coagulation waveform showing the change in light intensity over time due to the coagulation reaction of a reaction solution produced by mixing a test sample consisting of plasma obtained by separating blood acquired from a subject with a reagent, and to obtain a first waveform by performing a before-and-after difference processing on the coagulation waveform. A first fitting waveform is obtained by fitting the aforementioned coagulation waveform. A second fitted waveform is obtained by fitting the first waveform. A third waveform is obtained by performing a front-to-back difference processing on the second fitting waveform. By normalizing the light intensity axis and time axis of the first fitting waveform, the second fitting waveform, and the third waveform, respectively, a first normalized waveform, a second normalized waveform, and a third normalized waveform are obtained. Extracting feature quantities from each of the first normalized waveform, the second normalized waveform, and the third normalized waveform, and The method involves estimating the cause of prolonged blood coagulation time in the test sample based on known characteristics extracted from a group of samples in which the cause of prolonged blood coagulation time is known, and characteristics extracted from each of the first normalized waveform, the second normalized waveform, and the third normalized waveform. A method for estimating the cause of prolonged blood coagulation time, characterized by the features described above.

2. The aforementioned normalization is a normalization based on a numerical value obtained by fitting the coagulation waveform. The method for estimating the cause of prolonged blood coagulation time according to feature 1.

3. Estimating the cause of the prolonged blood coagulation time includes estimating that the cause of the prolonged blood coagulation time in the test sample is a deficiency of coagulation factor VIII, a deficiency of coagulation factor IX, or a positive lupus anticoagulant (LA) test. The method for estimating the cause of prolonged blood coagulation time according to feature 1.

4. The aforementioned blood coagulation factor VIII deficiency refers to a condition in which the activity level of blood coagulation factor VIII is less than 1%, and the aforementioned blood coagulation factor IX deficiency refers to a condition in which the activity level of blood coagulation factor IX is less than 1%. The method for estimating the cause of prolonged blood coagulation time according to feature 3.

5. Before the fitting process to the solidification waveform, a first filtering process is performed to remove noise from the solidification waveform, and The method further comprises performing a second filtering process to remove noise from the first waveform before the fitting process to the first waveform. The method for estimating the cause of prolonged blood coagulation time according to feature 1.

6. In the second filtering process described above, a moving average processing is performed using the reaction time constant, which is one of the parameters obtained by the fitting process to the solidification waveform. The method for estimating the cause of prolonged blood coagulation time according to feature 5.

7. The system further includes displaying the estimated cause of the extension on the display unit. The method for estimating the cause of prolonged blood coagulation time according to feature 1.

8. To create a map showing the relationship between known features extracted from the aforementioned sample group, whose cause of prolonged blood coagulation time is known, and the cause of prolonged blood coagulation time, and The system further includes displaying the map and information plotting the characteristic quantities of the test subject on the map on a display unit. The method for estimating the cause of prolonged blood coagulation time according to feature 1.

9. The system further includes displaying the estimated results of the examination based on the estimated cause of the prolonged blood coagulation time of the test specimen on the display unit. The method for estimating the cause of prolonged blood coagulation time according to feature 1.

10. An automated analyzer and an information processing device capable of communicating with each other acquire a coagulation waveform that shows the change in light intensity over time due to the coagulation reaction of a reaction solution produced by mixing a test sample consisting of plasma obtained by separating blood acquired from a subject with a reagent, The aforementioned information processing device comprises a computer system having a processor and memory, The aforementioned computer system, A process for acquiring the coagulation waveform from the automatic analyzer, A process to obtain a first waveform by performing a front-to-back difference processing on the aforementioned coagulation waveform, A process to obtain a first fitted waveform by fitting the aforementioned coagulation waveform, and a process to obtain a second fitted waveform by fitting the aforementioned first waveform, A process to obtain a third waveform by performing a front-to-back difference processing on the second fitting waveform, A process to obtain a first normalized waveform, a second normalized waveform, and a third normalized waveform by normalizing the light intensity axis and time axis of the first fitting waveform, the second fitting waveform, and the third waveform, respectively. A process for extracting feature quantities from each of the first normalized waveform, the second normalized waveform, and the third normalized waveform, and The process of estimating the cause of prolonged blood coagulation time in the test sample is performed based on known features extracted from a group of samples in which the cause of prolonged blood coagulation time is known, and the features extracted from each of the first normalized waveform, the second normalized waveform, and the third normalized waveform. An information processing device characterized by the following:

11. The aforementioned computer system, Normalization is performed based on the numerical values ​​obtained by the fitting process to the aforementioned solidification waveform. The information processing apparatus according to feature 10.

12. The process for estimating the cause of the prolonged blood coagulation time includes a process for estimating that the cause of the prolonged blood coagulation time in the test sample is a deficiency of coagulation factor VIII, a deficiency of coagulation factor IX, or a positive lupus anticoagulant (LA) test. The information processing apparatus according to feature 10.

13. The aforementioned blood coagulation factor VIII deficiency refers to a condition in which the activity level of blood coagulation factor VIII is less than 1%, and the aforementioned blood coagulation factor IX deficiency refers to a condition in which the activity level of blood coagulation factor IX is less than 1%. The information processing apparatus according to feature 12.

14. The aforementioned computer system, Prior to the fitting process on the solidification waveform, a first filtering process is performed to remove noise from the solidification waveform, and Prior to the fitting process on the first waveform, a second filtering process is further performed to remove noise from the first waveform. The information processing apparatus according to feature 10.

15. The aforementioned computer system, In the second filtering process described above, a moving average process is performed using the reaction time constant, which is one of the parameters obtained by the fitting process to the solidification waveform. The information processing apparatus according to feature 14.

16. The system further includes a display unit that displays the estimated cause of the extension. The information processing apparatus according to feature 10.

17. The aforementioned computer system, A process to create a map showing the relationship between known features extracted from the group of samples in which the cause of prolonged blood coagulation time is known and the cause of prolonged blood coagulation time, and Further, the process of displaying the map and the information obtained by plotting the characteristic quantities of the test subject on the map on the display unit is performed. The information processing apparatus according to feature 10.

18. The system further includes a display unit that displays the estimated results of the examination based on the estimated cause of the prolonged blood coagulation time of the test specimen. The information processing apparatus according to feature 10.

19. A method for estimating the cause of prolonged blood coagulation time in a test specimen, which is obtained by mixing a test specimen consisting of plasma obtained by separating blood from a test subject with a reagent, and a coagulation waveform showing the change in light intensity over time due to the coagulation reaction of the reaction solution, wherein the method involves estimating the cause of prolonged blood coagulation time in a test specimen from a coagulation waveform showing the change in light intensity over time due to the coagulation reaction of the reaction solution produced by mixing the test specimen, which consists of plasma obtained by separating blood obtained from a test subject, with the reagent, To align the data length of the coagulation waveform data with the time axis of the coagulation waveform, By applying the coagulation waveform data, of which the data lengths are uniform, to a neural network, multiple causes of prolonged blood coagulation time can be estimated, and The system includes presenting the cause of the prolonged blood coagulation time in the test sample based on the estimated probability of belonging to the cause of the prolonged blood coagulation time. A method for estimating the cause of prolonged blood coagulation time, characterized by the features described above.

20. The neural network in the estimation method described above is A neural network with 16 or more nodes and 3 or more hidden layers. The method for estimating the cause of prolonged blood coagulation time according to feature 19.

21. To obtain a coagulation waveform showing the change in light intensity over time due to the coagulation reaction of a reaction solution produced by mixing a test sample consisting of plasma obtained by separating blood from a subject with a reagent. A first waveform is obtained by performing a front-to-back difference processing on the aforementioned coagulation waveform. A first fitting waveform is obtained by fitting the aforementioned coagulation waveform. A second fitted waveform is obtained by fitting the first waveform. A third waveform is obtained by performing a front-to-back difference processing on the second fitting waveform. By normalizing the light intensity axis and time axis of the first fitting waveform, the second fitting waveform, and the third waveform, respectively, a first normalized waveform, a second normalized waveform, and a third normalized waveform are obtained. Extracting feature quantities from at least one of the first normalized waveform, the second normalized waveform, the third normalized waveform, and waveform data obtained by applying nonlinear calculations to them, and The method involves estimating the cause of prolonged blood coagulation time in a test sample based on pre-prepared distribution information of the characteristic quantities for each cause of prolonged blood coagulation time and the characteristic quantities extracted from the coagulation waveform of the test sample. A method for estimating the cause of prolonged blood coagulation time, characterized by the features described above.

22. To prepare a series of sample specimens by mixing plasma obtained by separating blood from a subject with normal plasma at multiple mixing ratios. To obtain a series of coagulation waveforms showing the change in light intensity over time due to the coagulation reaction of the reaction solution produced by mixing the series of sample specimens and reagents, A series of first waveforms are obtained by performing a front-to-back difference processing on each of the series of coagulation waveforms. A series of first fitting waveforms are obtained by fitting each of the series of coagulation waveforms. A series of second fitted waveforms is obtained by fitting each of the series of first waveforms. A series of third waveforms is obtained by performing a front-to-back difference process on each of the series of second fitting waveforms. By normalizing the light intensity axis and time axis of the series of first fitting waveforms, the series of second fitting waveforms, and the series of third waveforms, a series of first normalized waveforms, a series of second normalized waveforms, and a series of third normalized waveforms are obtained. Extracting feature quantities from at least one of the series of first normalized waveforms, the series of second normalized waveforms, the series of third normalized waveforms, and a series of waveform data obtained by applying nonlinear calculations to them, and The method involves using at least one of the series of features extracted from the series of coagulation waveforms to estimate the cause of the prolonged blood coagulation time of the sample. A method for estimating the cause of prolonged blood coagulation time, characterized by the features described above.

23. The feature quantities extracted from the aforementioned series of waveform data are: An index G is calculated from the position where a feature quantity containing multiple components, obtained from the immediate measurement of the coagulation waveform of a sample sample in which the plasma mixture ratio obtained by separating blood from the subject is 100% from the series of sample samples mentioned above, is plotted in the feature quantity space representing the multiple components. １ And the index F calculated from the series of coagulation times of the series of immediate measurements of the aforementioned series of sample specimens. １ combination with (F １ G １ ), and An index G is calculated from the position where the feature quantities containing the multiple components, obtained from the delayed measurement coagulation waveform of the sample specimen with a plasma mixing ratio of 100%, are plotted in the feature space. ２ And the index F calculated from the series of coagulation times of the delayed measurements of the series of sample specimens. ２ combination with (F ２ G ２ ) and The combination (F １ , G １ ) and the combination (F ２ , G ２ ) are used to estimate the cause of the prolongation of the blood coagulation time of the sample specimen The method for estimating the cause of prolonged blood coagulation time according to feature 22.

24. The feature quantities extracted from the aforementioned series of waveform data are: A series of feature quantities including multiple components obtained from a series of coagulation waveforms of immediate measurements of the series of sample specimens is plotted in a feature quantity space representing the multiple components, and a first trajectory data is obtained by connecting the series of feature quantities in the feature quantity space in ascending or descending order of the plasma mixing ratio, and A second trajectory data obtained by plotting a series of feature quantities including the plurality of components obtained from a series of coagulation waveforms of delayed measurements of the series of sample specimens in the feature quantity space, and connecting the series of feature quantities in the feature quantity space in ascending or descending order of plasma mixing ratio, Using the first and second orbital data, the cause of the prolonged blood coagulation time of the sample is estimated. The method for estimating the cause of prolonged blood coagulation time according to feature 22.

25. An automated analyzer and an information processing device capable of communicating with each other acquire a coagulation waveform that shows the change in light intensity over time due to the coagulation reaction of a reaction solution produced by mixing a test sample consisting of plasma obtained by separating blood acquired from a subject with a reagent, The aforementioned information processing device comprises a computer system having a processor and memory, The aforementioned computer system, A process for acquiring the coagulation waveform from the automatic analyzer, A process to obtain a first waveform by performing a front-to-back difference processing on the aforementioned coagulation waveform, A process to obtain a first fitting waveform by fitting the aforementioned coagulation waveform, A process to obtain a second fitted waveform by fitting the first waveform described above, A process to obtain a third waveform by performing a front-to-back difference processing on the second fitting waveform, A process to obtain a first normalized waveform, a second normalized waveform, and a third normalized waveform by normalizing the light intensity axis and time axis of the first fitting waveform, the second fitting waveform, and the third waveform, respectively. A process for extracting feature quantities from at least one of the first normalized waveform, the second normalized waveform, the third normalized waveform, and waveform data obtained by applying nonlinear calculations to them, and The process of estimating the cause of prolonged blood coagulation time in a test sample is performed based on pre-prepared distribution information of the characteristic quantities for each cause of prolonged blood coagulation time and the characteristic quantities extracted from the coagulation waveform of the test sample. An information processing device characterized by the following:

26. A process to create a graph showing the relationship between the distribution information of the feature quantities for each cause of prolonged blood coagulation time, which was prepared in advance, and the feature quantities extracted from the coagulation waveform of the test sample, and The process of displaying the aforementioned graph and the estimated cause of the prolonged blood coagulation time of the test sample on the display unit is further executed. The information processing apparatus according to claim 25.

27. An information processing device that can communicate with an automated analyzer, which prepares a series of sample specimens by mixing plasma obtained by separating blood from a subject with normal plasma at multiple mixing ratios, and acquires a series of coagulation waveforms showing the change in light intensity over time due to the coagulation reaction of a reaction solution produced by mixing the series of sample specimens with reagents, The aforementioned information processing device comprises a computer system having a processor and memory, The aforementioned computer system, A process for acquiring the series of coagulation waveforms from the automatic analyzer, A process to obtain a series of first waveforms by performing a front-to-back difference processing on each of the series of coagulation waveforms, A process to obtain a series of first fitting waveforms by fitting each of the series of coagulation waveforms, A process to obtain a series of second fitted waveforms by fitting each of the series of first waveforms, A process to obtain a series of third waveforms by performing a front-to-back difference process on each of the series of second fitting waveforms, A process to obtain a series of first normalized waveforms, a series of second normalized waveforms, and a series of third normalized waveforms by normalizing the light intensity axis and time axis of each of the series of first fitting waveforms, a series of second fitting waveforms, and a series of third waveforms. A process for extracting feature quantities from at least one of the series of first normalized waveforms, the series of second normalized waveforms, the series of third normalized waveforms, and a series of waveform data obtained by applying nonlinear calculations to them, and A process is performed to estimate the cause of the prolonged blood coagulation time of the sample specimen using at least one of the series of features extracted from the series of coagulation waveforms. An information processing device characterized by the following:

28. The feature quantities extracted from the aforementioned series of waveform data are: An index G is calculated from the position where a feature quantity containing multiple components, obtained from the immediate measurement of the coagulation waveform of a sample sample in which the plasma mixture ratio obtained by separating blood from the subject is 100% from the series of sample samples mentioned above, is plotted in the feature quantity space representing the multiple components. １ And the index F calculated from the series of coagulation times of the series of immediate measurements of the aforementioned series of sample specimens. １ combination with (F １ G １ ), and An index G is calculated from the position where the feature quantities containing the multiple components, obtained from the delayed measurement coagulation waveform of the sample specimen with a plasma mixing ratio of 100%, are plotted in the feature space. ２ And the index F calculated from the series of coagulation times of the delayed measurements of the series of sample specimens. ２ combination with (F ２ G ２ ) and The process for estimating the cause of the extension is the combination (F １ G １ ) and the above combination (F ２ G ２ This process uses the method described above to estimate the cause of the prolonged blood coagulation time of the sample specimen. The aforementioned index G １ and the aforementioned index F １ The relationship, and the aforementioned index G ２ and the aforementioned index F ２ The process of creating a graph showing the relationship, and Further, the process of displaying the graph and the estimated cause of the prolonged blood coagulation time of the sample on the display unit is performed. The information processing apparatus according to feature 27.

29. The feature quantities extracted from the aforementioned series of waveform data are: A series of feature quantities including multiple components obtained from a series of coagulation waveforms of immediate measurements of the series of sample specimens is plotted in a feature quantity space representing the multiple components, and a first trajectory data is obtained by connecting the series of feature quantities in the feature quantity space in ascending or descending order of the plasma mixing ratio, and A second trajectory data obtained by plotting a series of feature quantities including the plurality of components obtained from a series of coagulation waveforms of delayed measurements of the series of sample specimens in the feature quantity space, and connecting the series of feature quantities in the feature quantity space in ascending or descending order of plasma mixing ratio, The process for estimating the cause of the prolongation is a process for estimating the cause of the prolongation of the blood coagulation time of the sample using the first orbital data and the second orbital data. A process for creating a graph showing the first orbital data and the second orbital data, and Further, the process of displaying the graph and the estimated cause of the prolonged blood coagulation time of the sample on the display unit is performed. The information processing apparatus according to feature 27.

Images