Pain identification method, computer program product, recording medium, and pain identification system

Abstract

The present invention provides a pain estimation method and device that can objectively and accurately discriminate the pain of an estimated object. Specifically, a pain monitoring method for monitoring the pain of an estimated object based on the brain waves of the estimated object, a device and a system for implementing the method, and a method for generating a model for discriminating the pain of the object, as well as a device and a system for implementing the method are provided. In addition, an improved method for machine learning, as well as a device and a system for implementing the method are provided. The present invention aims to efficiently provide high-precision pain discrimination, and therefore also provides a method for judging pain using at least one of the brain wave feature quantities including the correlation of brain wave feature quantities and the complexity index, and a system and a device for implementing the method.

Description

Technical Field

[0001] The present invention relates to discriminating the quality and quantity of pain by using trend analysis of biological signals such as brain waves obtained from a subject. More specifically, the present invention relates to summarizing biological signal values ​​(especially brain wave signals) for each time period (block), taking the measured average, and analyzing the transition of the average value, so as to perform an analysis that is more consistent with subjective pain. The present invention also relates to an economical discrimination model for the technology of discriminating pain using brain waves. The present invention also relates to a technology for implementing machine learning in a personalized (tailor-made) manner. The present invention also relates to new brain wave feature quantities for pain discrimination. Background Art

[0002] Pain is essentially subjective, but it is desirable to evaluate it objectively for treatment. It is common for patients to suffer disadvantages due to underestimation of pain. Therefore, a method for objectively estimating pain using brain waves has been proposed (for example, see Patent Document 1).

[0003] However, the intensity of pain is subjective and difficult to evaluate objectively. In addition, the fluctuation of brain wave signals is large and may not necessarily correspond to subjective feelings. In addition, there is no established method to effectively monitor the changes of pain over time. Regarding the discrimination of pain, there is only a beginning, and it is not possible to provide efficient model generation and discrimination.

[0004] Machine learning is widely used as a method for realizing artificial intelligence. As a method for constructing an algorithm capable of extracting information representing the characteristics of output data from input data with high accuracy, a method called ensemble learning is known (for example, Patent Documents 2 to 3).

[0005] Ensemble learning is a method of generating multiple relatively low-precision information extraction devices (hereinafter referred to as weak information extraction units (weak learners)) using multiple teaching data, and obtaining a high-precision information extraction device by combining the outputs of the generated multiple weak information extraction units.

[0006] Prior art literature

[0007] Patent Literature

[0008] Patent Document 1: Japanese Patent Application No. 2010-523226

[0009] Patent Document 2: Japanese Patent Application Publication No. 2005-44330

[0010] Patent Document 3: Japanese Patent Application Publication No. 2013-164863 Summary of the invention

[0011] Means for solving technical problems

[0012] In one embodiment, the present invention relates to performing trend analysis (time series analysis) on biological signals such as brain waves obtained from a subject and using the trend analysis (time series analysis) to determine the quality and quantity of pain. More specifically, the present invention is based on the discovery that if the biological signal values ​​(especially brain wave signals) are aggregated for each time period (block) and the average is taken, and the transition of the average is analyzed, it can better correspond to the transition of subjective pain. The present invention also provides a technology for generating such a trend analysis.

[0013] In this aspect, the present invention provides, for example, the following items.

[0014] (Item 1) A pain monitoring method is a method for monitoring pain experienced by an estimated subject based on brain waves of the estimated subject, comprising:

[0015] a) obtaining electroencephalogram data or analysis data thereof from the estimation subject by performing electroencephalogram measurement in response to stimulation;

[0016] b) a step of obtaining a time-lapse average value of the brain wave data or its analysis data in a specific time width; and

[0017] c) A step of evaluating or monitoring the level of pain that the estimated subject has based on the electroencephalogram data or analysis data thereof, based on the temporal transition of the average value.

[0018] (Item 2) The pain monitoring method according to Item 1,

[0019] The specific time width is at least 20 seconds.

[0020] (Item 3) The pain monitoring method according to Item 1 or 2,

[0021] The specific time width is at least 30 seconds.

[0022] (Item 4) The pain monitoring method according to any one of Items 1 to 3,

[0023] The specific time width is at least 40 seconds.

[0024] (Item 5) The pain monitoring method according to any one of Items 1 to 4, characterized in that:

[0025] If the time transition of the average value is monotonically increasing, the pain is judged to be severe; if the time transition of the average value is monotonically decreasing, the pain is judged to be mild.

[0026] (Item 6) The pain monitoring method according to any one of Items 1 to 5, wherein:

[0027] The average value in the specific time width is calculated by a non-overlapping block averaging method.

[0028] (Item 7) The pain monitoring method according to any one of Items 1 to 6, characterized in that:

[0029] As the average value, an average value in two or more time widths is used.

[0030] (Item 8) The pain monitoring method according to Item 7, wherein:

[0031] The two or more time widths include at least time widths of 10 to 120 seconds and 30 to 300 seconds, and further include a time width of more than 120 seconds as necessary.

[0032] (Item 9) The pain monitoring method according to any one of Items 1 to 8, characterized in that:

[0033] A pain index was used in monitoring the pain level.

[0034] (Item 10) The pain monitoring method according to claim 9, characterized in that:

[0035] The pain index includes numerical values ​​that facilitate interpretation of temporal changes in pain.

[0036] (Item 11) The pain monitoring method according to Item 10, wherein:

[0037] The values ​​are expressed continuously, ordinal or nominally with the brain characteristic quantity of the intense pain level as a baseline.

[0038] (Item 12) The pain monitoring method according to any one of Items 1 to 11, wherein:

[0039] The method also includes the step of generating a pain index based on the brain wave data or analysis data thereof.

[0040] (Item 13) The pain monitoring method according to any one of Items 1 to 12, wherein:

[0041] The method further includes the step of calculating a baseline according to the pain index, and the step of determining or estimating pain based on the baseline during the monitoring.

[0042] (Item 14) The pain monitoring method according to Item 13, wherein:

[0043] In addition to calculating the baseline, a baseline correction is further performed.

[0044] (Item 14A) The pain monitoring method according to Item 13, wherein:

[0045] The step (c) further includes referring to a pain ratio, wherein the pain ratio indicates the certainty of determining that the pain estimation value is painful.

[0046] (Item 14B) The pain monitoring method according to Item 13, wherein:

[0047] Regarding the pain rate,

[0048] (a) Calculate the discriminant estimates of pain and no pain using the discriminant model;

[0049] (b) dividing the pain estimate into different ranges and calculating the ratio of the pain estimate to the estimated values ​​in each range; and

[0050] (c) The pain ratio is calculated by converting the ratio into a function to generate a pain ratio function, and the pain ratio is calculated by the pain ratio function.

[0051] (Item 15) An apparatus for monitoring pain experienced by an estimated subject based on brain waves of the estimated subject, comprising:

[0052] A) an electroencephalogram data measuring unit that obtains electroencephalogram data of the estimation object or analysis data thereof;

[0053] B) a value feature quantity extraction unit for plotting a time transition of an average value in a specific time width of the electroencephalogram data or its analysis data; and

[0054] C) A pain level monitoring unit that monitors the level of pain that the estimated subject has based on the mapping and the electroencephalogram data or analysis data thereof.

[0055] (Item 16) The device according to Item 15, wherein:

[0056] Includes one or more of the features of items 2 to 14.

[0057] (Item 17) A program for causing a computer to execute a method of monitoring pain experienced by an estimation subject based on brain waves of the estimation subject, the method comprising:

[0058] a) causing the computer to obtain brain wave data from the estimation subject by performing brain wave measurement in response to stimulation;

[0059] b) causing the computer to extract a feature quantity for plotting a time transition of an average value at a specific time width of the electroencephalogram data or its analysis data; and

[0060] c) causing the computer to monitor the level of pain that the subject is estimated to have based on the mapping according to the brain wave data or analysis data thereof.

[0061] (Item 18) The procedure according to Item 17, wherein:

[0062] Includes one or more of the features of items 2 to 14.

[0063] (Item 19) A recording medium storing a program for causing a computer to execute a method of monitoring pain experienced by an estimation subject based on brain waves of the estimation subject, the method comprising:

[0064] a) causing the computer to measure electroencephalogram data or analysis data thereof from the estimation subject by performing electroencephalogram measurement in response to stimulation;

[0065] b) causing the computer to extract a feature quantity for plotting a time transition of an average value at a specific time width of the electroencephalogram data or its analysis data; and

[0066] c) causing the computer to monitor the level of pain that the subject is estimated to have based on the mapping according to the brain wave data or analysis data thereof.

[0067] (Item 20) The recording medium according to Item 19, wherein

[0068] Includes one or more of the features of items 2 to 14.

[0069] In another aspect, the present inventors have found, after intensive research, a technology that can efficiently provide high-precision discrimination models for various senses.

[0070] In this aspect, the present invention provides, for example, the following items.

[0071] (Item X1) A method for generating a model for determining pain of a subject, comprising:

[0072] a) obtaining brain wave data or analysis data thereof from the subject;

[0073] b) a step of reducing a feature value related to the pain in the electroencephalogram data or the analysis data thereof;

[0074] c) a step of performing discriminant analysis by machine learning and cross validation from the top ranking in the ranking of the reduced weighted feature quantities or their combinations; and

[0075] d) The steps to determine the model to achieve the specified accuracy.

[0076] (Item X2) The method according to Item X1, wherein:

[0077] This includes selecting a model with fewer types of feature quantities among the models that achieve the predetermined accuracy.

[0078] (Item X3) The method according to Item X1, wherein:

[0079] After step c), it also includes: calculating the difference value (Diff) of the adjacent model of the discrimination accuracy obtained by discriminant analysis, wherein the adjacent model is a model including n-1 feature quantities and n feature quantities, n is greater than 2, and the difference value is also considered in the judgment of the discriminant model in step d).

[0080] (Item X4) A method according to Item X3, wherein:

[0081] The consideration of the difference value includes a process of re-ranking the feature quantities from the value having the largest difference value and re-calculating the discrimination accuracy to thereby generate a model with high discrimination accuracy.

[0082] (Item X5) A method according to Item X3 or X4, wherein:

[0083] The determination based on the difference value includes classifying the feature amount into a main feature amount and an auxiliary feature amount, and re-ranking the auxiliary feature amounts.

[0084] (Item X6) A method according to Item X5, wherein:

[0085] The method comprises the step of changing the feature amount after the re-ranking and repeating the step c).

[0086] (Item X7) A method according to Item X5 or X6, wherein:

[0087] The method includes the steps of changing the feature amount after the re-ranking and performing machine learning and cross-validation to calculate the discrimination accuracy of each model.

[0088] (Item X8) The method according to any one of Items X3 to X7, wherein

[0089] At least one step among items X3 to X7 is repeated at least once.

[0090] (Item X9) The method according to any one of Items X1 to X8, wherein

[0091] The specified accuracy includes the case of highest accuracy.

[0092] (Item X10) The method according to any one of Items X1 to X9, characterized in that

[0093] Regarding the reduction, fitting is performed using a discriminant function model corresponding to the discriminant stage, a model approximation index is calculated, and an arbitrary number is selected from the top rank of the sorting.

[0094] (Item X11) The method according to any one of Items X1 to X10, characterized in that

[0095] The reduction extracts effective feature quantities according to the discrimination stage.

[0096] (Item X12) The method according to any one of Items X1 to X11, wherein

[0097] The weighting coefficient is selected from R 2 Values, correlation coefficients, regression coefficients, and residual sums of squares.

[0098] (Item X13) A method according to Item X12, wherein:

[0099] The discriminant function model is selected from a Sigmoid function and a step function.

[0100] (Item X14) A method according to Item X10 or X12, wherein:

[0101] The model approximation index is a partial set of weighting coefficients.

[0102] (Item X15) A method according to Item X11, wherein:

[0103] In the case of binary classification, the effective feature quantity is an all-or-none feature quantity corresponding to the presence or absence of pain, that is, a binary feature quantity, or a feature quantity that is closer to a discriminant function.

[0104] (Item X16) A method for determining pain in a subject, comprising:

[0105] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0106] b) a step of reducing a feature value related to the pain in the electroencephalogram data or the analysis data thereof;

[0107] c) a step of performing discriminant analysis by machine learning and cross validation starting from the top ranking of the weighted coefficients of each feature quantity;

[0108] d) the steps to determine the model to achieve the specified accuracy;

[0109] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0110] f) applying the model to the test subject to discern pain.

[0111] (Item X17) A method for determining pain in a subject, comprising:

[0112] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0113] f) applying the pain identification model to the test subject to identify pain,

[0114] The model is generated through the following steps:

[0115] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0116] b) a step of reducing a feature quantity of the electroencephalogram data or its analysis data with respect to the pain;

[0117] c) a step of performing discriminant analysis by machine learning and cross validation from the top ranking of the sorted weighted coefficients of each feature quantity; and

[0118] d) The steps to determine the model to achieve the specified accuracy.

[0119] (Item X18) A method for determining pain in a subject, comprising the steps of:

[0120] The pain of the subject is determined using a model for determining the pain of the subject including feature quantities at at least two electrodes around Fz or Fpz or both.

[0121] (Item X19) A method according to Item X18, wherein:

[0122] The surroundings include at least one of F3, F4, Fp1 and Fp2.

[0123] (Item X20) A program causing a computer to execute a method for generating a model for determining pain of a subject, the method comprising:

[0124] a) obtaining brain wave data or analysis data thereof from the subject;

[0125] b) a step of reducing a feature value related to the pain in the electroencephalogram data or the analysis data thereof;

[0126] c) a step of performing discriminant analysis by machine learning and cross validation from the top ranking of the order of the weighted feature quantities or their combinations after the reduction; and

[0127] d) The steps to determine the model to achieve the specified accuracy.

[0128] (Item X20A) According to the procedures set out in Item X20,

[0129] Also includes one or more features of items X1 to X19.

[0130] (Item X21) A recording medium storing a program for causing a computer to execute a method for generating a model for distinguishing pain of a subject, the method comprising:

[0131] a) obtaining brain wave data or analysis data thereof from the subject;

[0132] b) a step of reducing a feature value related to the pain in the electroencephalogram data or the analysis data thereof;

[0133] c) a step of performing discriminant analysis by machine learning and cross validation from the top ranking of the order of the weighted feature quantities or their combinations after the reduction; and

[0134] d) The steps to determine the model to achieve the specified accuracy.

[0135] (Item X21A) The recording medium of Item X21, wherein:

[0136] Also includes one or more features of items X1 to X19.

[0137] (Item X22) A system or apparatus for generating a model for determining pain in a subject, the system or apparatus comprising:

[0138] A) an electroencephalogram data obtaining unit that obtains electroencephalogram data or analysis data thereof from the subject;

[0139] B) a feature quantity reducing unit that reduces a feature quantity with respect to the pain in the electroencephalogram data or its analysis data; and

[0140] C) A pain discrimination estimation model generation unit performs discriminant analysis from the top ranking of the weighted feature quantities or their combinations after the reduction through machine learning and cross validation to determine a model that achieves a predetermined accuracy.

[0141] (Item X22A) A system or apparatus according to Item X22, wherein:

[0142] Also includes one or more features of items X1 to X19.

[0143] (Item X23) A program causing a computer to execute a method of determining pain in a subject, the method comprising:

[0144] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0145] b) a step of reducing a feature value related to the pain in the brain wave data or its analysis data;

[0146] c) a step of performing discriminant analysis by machine learning and cross validation starting from the top ranking of the weighted coefficients of each feature quantity;

[0147] d) the steps to determine the model to achieve the specified accuracy;

[0148] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0149] f) applying the model to the test subject to discern pain.

[0150] (Item X23A) The procedure of Item X23, wherein:

[0151] Also includes one or more features of items X1 to X19.

[0152] (Item X24) A recording medium storing a program causing a computer to execute a method of determining pain of a subject, the method comprising:

[0153] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0154] b) a step of reducing a feature value related to the pain in the brain wave data or its analysis data;

[0155] c) a step of performing discriminant analysis by machine learning and cross validation starting from the top ranking of the weighted coefficients of each feature quantity;

[0156] d) the steps to determine the model to achieve the specified accuracy;

[0157] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0158] f) applying the model to the test subject to discern pain.

[0159] (Item X24A) The recording medium of Item X24, wherein:

[0160] Also includes one or more features of items X1 to X19.

[0161] (Item X25) A system or apparatus for determining pain in a subject, the system or apparatus comprising:

[0162] A) an electroencephalogram data acquisition unit that obtains electroencephalogram data or analysis data thereof from a reference subject;

[0163] B) a feature quantity reduction unit that reduces a feature quantity related to the pain in the electroencephalogram data or its analysis data;

[0164] C) a pain discrimination estimation model generation unit that ranks the weighted coefficients of each feature quantity from the top of the ranking, performs discriminant analysis through machine learning and cross-validation, and determines a model that achieves a predetermined accuracy;

[0165] D) an electroencephalogram data measuring unit that obtains electroencephalogram data or analysis data thereof from a test subject; and

[0166] E) A pain discrimination estimation unit applies the model to the test subject to discriminate pain.

[0167] (Item X25A) A system or apparatus as described in Item X25, wherein:

[0168] Also includes one or more features of items X1 to X19.

[0169] (Item X26) A program causing a computer to execute a method of determining pain of a subject, the method comprising:

[0170] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0171] f) applying the pain identification model to the test subject to identify pain,

[0172] The model is generated through the following steps:

[0173] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0174] b) a step of reducing a feature quantity of the brain wave data or its analysis data with respect to the pain;

[0175] c) a step of performing discriminant analysis by machine learning and cross validation from the top ranking of the sorted weighted coefficients of each feature quantity; and

[0176] d) The steps to determine the model to achieve the specified accuracy.

[0177] (Item X26A) The procedure of Item X26, wherein:

[0178] Also includes one or more features of items X1 to X19.

[0179] (Item X27) A recording medium storing a program causing a computer to execute a method of determining pain of a subject, the method comprising:

[0180] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0181] f) applying the pain identification model to the test subject to identify pain,

[0182] The model is generated through the following steps:

[0183] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0184] b) a step of reducing a feature quantity of the brain wave data or its analysis data with respect to the pain;

[0185] c) a step of performing discriminant analysis by machine learning and cross validation from the top ranking of the sorted weighted coefficients of each feature quantity; and

[0186] d) The steps to determine the model to achieve the specified accuracy.

[0187] (Item X27A) The recording medium of Item X27, wherein:

[0188] Also includes one or more features of items X1 to X19.

[0189] (Item X28) A system or apparatus for determining pain in a subject, the system or apparatus comprising:

[0190] D) an electroencephalogram data measuring unit that obtains electroencephalogram data or analysis data thereof from a test subject; and

[0191] E) a pain discrimination estimation unit, which applies a pain discrimination model to the test subject to discriminate pain,

[0192] The model is generated through the following steps:

[0193] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0194] b) a step of reducing a feature quantity of the brain wave data or its analysis data with respect to the pain;

[0195] c) a step of performing discriminant analysis by machine learning and cross validation from the top ranking of the sorted weighted coefficients of each feature quantity; and

[0196] d) The steps to determine the model to achieve the specified accuracy.

[0197] (Item X28A) A system or apparatus according to Item X28, wherein:

[0198] Also includes one or more features of items X1 to X19.

[0199] In another embodiment, the present invention also provides the following items.

[0200] (Item A1) A method for generating a model for determining pain of a subject, comprising:

[0201] a) obtaining brain wave data or analysis data thereof from the subject;

[0202] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0203] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof; and

[0204] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0205] (Item A2) The method according to Item A1, characterized in that

[0206] In step b), after the feature quantity coefficients are determined, the reduction is performed by repeatedly performing discrimination estimation, calculating the average of the feature quantity coefficients for the discrimination estimation and ranking them, and selecting the feature quantities based on a predetermined threshold.

[0207] (Item A3) The method according to Item A1 or A2, wherein:

[0208] The characteristic quantity coefficient is determined by machine learning.

[0209] (Item A4) The method according to any one of Items A1 to A3, wherein

[0210] Step b) includes: when determining the feature quantity coefficient, determining the hyperparameter with the highest discrimination accuracy, determining the feature quantity coefficient based on the hyperparameter, and excluding feature quantities that are invalid for discrimination or have a low contribution rate.

[0211] (Item A5) The method according to Item A4, wherein:

[0212] The characteristic quantity coefficients and hyperparameters are determined by machine learning.

[0213] (Item A6) The method according to any one of Items A1 to A5, wherein

[0214] The steps b) to d) include:

[0215] (C1) a step of dividing the data corresponding to the feature quantity and the pain corresponding to the feature quantity into learning data and test data;

[0216] (C2) performing machine learning using the learning data to create a discriminant model;

[0217] (C3) a step of calculating the discrimination accuracy of the discrimination model using the test data; and

[0218] (C4) When there exists an object sample in the object whose discrimination accuracy is below the chance level, the sample is excluded and C1 to C3 are repeated, and the process ends when there is no sample below the chance level, thereby determining the steps of the discrimination model, wherein the chance level is a value obtained by dividing 100% by the number to be classified.

[0219] (Item A7) The method according to any one of Items A1 to A6, wherein

[0220] This includes selecting a model with fewer types of feature quantities among the models that achieve the predetermined accuracy.

[0221] (Item A8) The method according to any one of Items A1 to A7, wherein

[0222] After step c), it also includes: calculating the difference value (Diff) of the adjacent model of the discrimination accuracy obtained by discriminant analysis, wherein the adjacent model is a model including n-1 feature quantities and n feature quantities, n is greater than 2, and the difference value is also considered in the judgment of the discriminant model in step d).

[0223] (Item A9) The method according to Item A8, wherein:

[0224] The consideration of the difference value includes a process of re-ranking the feature quantities from the value having the largest difference value and re-calculating the discrimination accuracy to thereby generate a model with high discrimination accuracy.

[0225] (Item A10) The method according to Item A8 or A9, wherein:

[0226] The determination based on the difference value includes classifying the feature amount into a main feature amount and an auxiliary feature amount, and re-ranking the auxiliary feature amounts.

[0227] (Item A11) The method according to Item A10, wherein:

[0228] The method comprises the step of changing the feature amount after the re-ranking and repeating the step c).

[0229] (Item A12) The method according to Item A10 or A11, wherein:

[0230] The method includes the steps of changing the feature amount after the re-ranking and performing machine learning and cross-validation to calculate the discrimination accuracy of each model.

[0231] (Item A13) The method according to any one of Items A8 to A12, wherein

[0232] At least one step in items A8 to A12 is repeated at least once.

[0233] (Item A14) The method according to any one of Items A1 to A13, wherein

[0234] The specified accuracy includes the case of highest accuracy.

[0235] (Item A15) The method according to any one of Items A1 to A14, characterized in that:

[0236] Regarding the reduction, fitting is performed using a discriminant function model corresponding to the discriminant stage, a model approximation index is calculated, and an arbitrary number is selected from the top rank of the sorting.

[0237] (Item A16) The method according to any one of Items A1 to A15, characterized in that:

[0238] The reduction extracts effective feature quantities according to the discrimination stage.

[0239] (Item A17) The method according to any one of Items A1 to A16, wherein

[0240] The weighting coefficient is selected from R 2 values, correlation coefficients, regression coefficients, and residual sums of squares.

[0241] (Item A18) The method according to Item A17, wherein:

[0242] The discriminant function model is selected from a Sigmoid function and a step function.

[0243] (Item A19) The method according to any one of Items A15 to A18, wherein

[0244] The model approximation index is a partial set of weighting coefficients.

[0245] (Item A20) The method according to any one of Items A16 to A19, wherein

[0246] In the case of binary classification, the effective feature quantity is an all-or-none feature quantity corresponding to the presence or absence of pain, that is, a binary feature quantity, or a feature quantity that is closer to a discriminant function.

[0247] (Item A21) A method for determining pain in a subject, comprising:

[0248] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0249] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0250] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof;

[0251] d) steps for determining a discriminant analysis model that achieves a specified accuracy;

[0252] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0253] f) A step of substituting the brain wave data or analysis data of the test subject into the model to discriminate pain.

[0254] (Item A22) A method for determining pain in a subject, comprising:

[0255] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0256] f) the step of substituting the brain wave data or the analysis data of the test subject into a pain discrimination model to discriminate pain,

[0257] The model is generated through the following steps:

[0258] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0259] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0260] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof; and

[0261] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0262] (Item A23) A method for determining pain in a subject, wherein:

[0263] The method includes the step of discriminating the pain of the subject using a model for discriminating the pain of the subject, wherein the model includes feature quantities at at least two electrodes around Fz or Fpz or both.

[0264] (Item A24) The method according to Item A23, wherein:

[0265] The surroundings include at least one of F3, F4, Fp1 and Fp2.

[0266] (Item A25) A program causing a computer to execute a method for generating a model for determining pain of a subject, the method comprising:

[0267] a) obtaining brain wave data or analysis data thereof from the subject;

[0268] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0269] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof; and

[0270] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0271] (Item A25A) The procedure of Item A25, wherein:

[0272] Also includes one or more features of items A1-A24 and A34-A36.

[0273] (Item A26) A recording medium storing a program for causing a computer to execute a method of generating a model for determining pain of a subject, the method comprising:

[0274] a) obtaining brain wave data or analysis data thereof from the subject;

[0275] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0276] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof; and

[0277] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0278] (Item A26A) The recording medium according to Item A26, wherein

[0279] Also includes one or more features of items A1-A24 and A34-A36.

[0280] (Item A27) A system for generating a model for determining pain of a subject, the system comprising:

[0281] A) an electroencephalogram data obtaining unit that obtains electroencephalogram data or analysis data thereof from the subject;

[0282] B) a feature quantity reduction unit that determines a feature quantity coefficient related to the pain for a feature quantity based on the electroencephalogram data or its analysis data, and performs reduction on the pain; and

[0283] C) A pain discrimination estimation model generation unit that creates a discrimination analysis model through machine learning and verification based on the reduced weighted feature quantities or a combination thereof.

[0284] (Item A27A) The system of Item A27, wherein:

[0285] Also includes one or more features of items A1-A24 and A34-A36.

[0286] (Item A28) A program causing a computer to execute a method for determining pain in a subject, the method comprising:

[0287] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0288] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0289] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof;

[0290] d) steps for determining a discriminant analysis model that achieves a specified accuracy;

[0291] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0292] f) A step of substituting the brain wave data or analysis data of the test subject into the model to discriminate pain.

[0293] (Item A28A) The procedure of Item A28, wherein:

[0294] Also includes one or more features of items A1-A24 and A34-A36.

[0295] (Item A29) A recording medium storing a program causing a computer to execute a method of determining pain of a subject, the method comprising:

[0296] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0297] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0298] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof;

[0299] d) steps for determining a discriminant analysis model that achieves a specified accuracy;

[0300] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0301] f) A step of substituting the brain wave data or analysis data of the test subject into the model to discriminate pain.

[0302] (Item A29A) The recording medium according to Item A29, wherein

[0303] Also includes one or more features of items A1-A24 and A34-A36.

[0304] (Item A30) A system for determining pain in a subject, the system comprising:

[0305] A) an electroencephalogram data acquisition unit that obtains electroencephalogram data or analysis data thereof from a reference subject;

[0306] B) a feature quantity reduction unit that determines a feature quantity coefficient related to the pain for a feature quantity based on the electroencephalogram data or its analysis data, and performs reduction on the pain;

[0307] C) a pain discrimination estimation model generation unit, which creates a discrimination analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof;

[0308] D) an electroencephalogram data measuring unit that obtains electroencephalogram data or analysis data thereof from a test subject; and

[0309] E) A pain discrimination estimation unit that substitutes the brain wave data of the test subject or its analysis data into the discrimination analysis model to discriminate pain.

[0310] (Item A30A) The system of Item A29, wherein:

[0311] Also includes one or more features of items A1-A24 and A34-A36.

[0312] (Item A31) A program causing a computer to execute a method for determining pain in a subject, the method comprising:

[0313] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0314] f) the step of substituting the brain wave data or the analysis data of the test subject into a pain discrimination model to discriminate pain,

[0315] The model is generated through the following steps:

[0316] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0317] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0318] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof; and

[0319] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0320] (Item A31A) The procedure of Item A31, wherein:

[0321] Also includes one or more features of items A1-A24 and A34-A36.

[0322] (Item A32) A recording medium storing a program causing a computer to execute a method of determining pain of a subject, the method comprising:

[0323] e) a step of obtaining brain wave data or analysis data thereof from a test subject; and

[0324] f) the step of substituting the brain wave data or the analysis data of the test subject into a pain discrimination model to discriminate pain,

[0325] The model is generated through the following steps:

[0326] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0327] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0328] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof; and

[0329] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0330] (Item A32A) The recording medium according to Item A32, wherein:

[0331] Also includes one or more features of items A1-A24 and A34-A36.

[0332] (Item A33) A system for determining pain in a subject, the system comprising:

[0333] D) an electroencephalogram data measuring unit that obtains electroencephalogram data or analysis data thereof from a test subject; and

[0334] E) a pain identification and estimation unit, which substitutes the brain wave data or analysis data of the test subject into a pain identification model to identify pain,

[0335] The model is generated through the following steps:

[0336] a) a step of obtaining brain wave data or analysis data thereof from a reference subject;

[0337] b) determining a characteristic quantity coefficient related to the pain for a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0338] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof; and

[0339] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0340] (Item A33A) The system of Item A33, wherein:

[0341] Also includes one or more features of items A1-A24 and A34-A36.

[0342] (Item A34) The method according to Items A1 to A24, wherein:

[0343] The method also includes a step of filtering the brain wave data or its analysis data.

[0344] (Item A35) The method, program, recording medium or system according to Item A34, wherein:

[0345] The filtering includes filtering for at least one selected from removing eye movements and myoelectric potential attenuation.

[0346] (Item A36) The method according to item A1 to A24, A34 or A35, wherein:

[0347] The brain wave data or analysis data thereof includes at least one selected from amplitude, frontal-parietal potential correlation, frequency power and complexity index, namely multi-scale entropy.

[0348] (Item A37) A program causing a computer to execute a method of determining pain of a subject, the method comprising:

[0349] a) obtaining brain wave data or analysis data thereof from a subject;

[0350] b) a step of generating a discriminant model based on the brain wave data or its analysis data; and

[0351] c) A step of substituting the brain wave data or analysis data from the subject into the discrimination model to discriminate pain.

[0352] (Item A37A) The procedure of Item A37, wherein:

[0353] Includes one or more of the features listed above.

[0354] (Item A37B) A recording medium storing the program according to Item A37 or Item A37A.

[0355] (Item A38) A system for determining pain in a subject, the system comprising:

[0356] X) an electroencephalogram data obtaining unit that obtains electroencephalogram data or analysis data thereof from a subject;

[0357] Y) a pain discrimination estimation model generating unit that generates a discrimination model based on the electroencephalogram data or its analysis data; and

[0358] Z) A pain discrimination estimation unit that substitutes the brain wave data from the subject or the analysis data thereof into the model to discriminate pain.

[0359] (Item A38A) The system of Item A38, wherein:

[0360] Includes one or more of the features listed above.

[0361] (Item A39) A system for determining pain of an object, the system comprising X) a pain determination terminal and Y) a pain determination estimation server,

[0362] The pain discrimination terminal comprises:

[0363] X-1) an electroencephalogram data obtaining terminal which obtains electroencephalogram data or analysis data thereof from a subject; and

[0364] X-2) module, which sends the brain wave data or its analysis data to the pain discrimination estimation server and receives the discrimination result from the pain discrimination estimation server,

[0365] The pain discrimination estimation server includes:

[0366] Y-1) A pain discrimination estimation model generation module, which is generated based on the brain wave data or its analysis data

[0367] Discriminant model;

[0368] Y-2) a pain discrimination estimation module, which substitutes the brain wave data or its analysis data from the subject into

[0369] The model generates a discriminant result for discriminating pain; and

[0370] Y-3) A judgment result sending and receiving module receives the brain wave data or its analysis data and sends the judgment result.

[0371] (Item A39A) The system of Item A39, wherein:

[0372] Includes one or more of the features listed above.

[0373] (Item A40) A system for determining pain of an object, the system comprising X) a pain determination terminal and Y) a pain determination estimation server,

[0374] The pain discrimination terminal comprises:

[0375] X-1) an electroencephalogram data obtaining terminal that obtains electroencephalogram data from a subject; and

[0376] X-2) module, which sends the brain wave data to the pain discrimination estimation server and receives the discrimination result from the pain discrimination estimation server,

[0377] The pain discrimination estimation server includes:

[0378] Y-1) A pain discrimination estimation model generation module, which is generated based on the brain wave data or its analysis data

[0379] Discriminant model;

[0380] Y-1') a brain wave feature extraction module, which extracts analysis data from the brain wave data;

[0381] Y-2) a pain discrimination estimation module, which substitutes the brain wave data or its analysis data from the subject into

[0382] The model generates a discriminant result for discriminating pain; and

[0383] Y-3) A judgment result sending and receiving module receives the brain wave data or its analysis data and sends the judgment result.

[0384] (Item A40A) The system of Item A40, wherein:

[0385] Includes one or more of the features listed above.

[0386] (Item A41) A system for determining pain of a subject, the system comprising X) a pain determination terminal and Y) a pain determination estimation server,

[0387] The pain discrimination terminal comprises:

[0388] X-1) an electroencephalogram data acquisition terminal, which obtains electroencephalogram data or analysis data thereof from a subject;

[0389] X-2) module, which sends the brain wave data or its analysis data to the pain discrimination estimation server,

[0390] The pain discrimination estimation server receives the discrimination model; and

[0391] X-3) module is a discriminant model module for storing a discriminant model, and the brain wave data or its analysis data from the subject is substituted into the discriminant model to discriminate pain.

[0392] The pain discrimination estimation server includes:

[0393] Y-1) A pain discrimination estimation model generation module, which is generated based on the brain wave data or its analysis data

[0394] Discriminant model; and

[0395] Y-2) a model sending module, which sends the discrimination model to the pain discrimination terminal; and

[0396] The electroencephalogram feature extraction module extracts analysis data from the electroencephalogram data as needed.

[0397] (Item A41A) The system of Item A41, wherein:

[0398] Includes one or more of the features listed above.

[0399] In another scenario, machine learning is a technology that is not explicitly programmed but is intended to give computers the ability to learn, repeatedly performing predictions, searches, verifications, etc. Although there are cases where labels to be distinguished are given (supervised learning) and cases where they are not (unsupervised learning), in the present disclosure, it is important to implement this in a customized manner.

[0400] Therefore, in this aspect, the present invention provides, for example, the following items.

[0401] (Project B1) An improved method for machine learning, comprising:

[0402] A) Steps for performing machine learning to generate a discriminant model;

[0403] B) Steps to determine hyperparameters in all samples through cross-validation;

[0404] C) using the hyperparameters to generate a discriminant model and calculating the discriminant accuracy for each sample;

[0405] D) a step of grouping each sample based on the discrimination accuracy;

[0406] E) determining independent hyperparameters for each group; and

[0407] F) Using the independent hyperparameters to generate independent discriminant models for each group, calculating the discrimination accuracy of each sample, and generating a maximum discriminant (MAX) model therefrom.

[0408] (Item B2) The method according to Item B1, wherein:

[0409] The step of generating the maximum discriminant (MAX) model includes an integration method and its pruning.

[0410] (Item B3) The method according to item B1 or B2, characterized in that

[0411] The maximum discriminant (MAX) model is used as the main model, and the remaining ones are used as auxiliary models, and the auxiliary models are added one by one to improve the discrimination accuracy.

[0412] (Item B4) The method according to any one of Items B1 to B3, wherein

[0413] The maximum discriminant (MAX) model is used as the main model, and the remaining ones are used as auxiliary models. The auxiliary models are added to the main model in a one-by-one manner, so that the discrimination accuracy is improved in stages.

[0414] (Item B5) The method according to Item B4, wherein:

[0415] The staged improvement of the discrimination accuracy includes:

[0416] A) when one of the N-1 auxiliary models is selected to be added to the main model, the two model sets with the highest improvement in discrimination accuracy through the ensemble method are used as a step to update the main model;

[0417] B) excluding N-2 models after updating the main model from all N models as a step of updating the auxiliary model; and

[0418] C) For the model set with the highest improvement in discrimination accuracy through the ensemble method, steps A) to B) are repeated in the same manner to update the update main model and the update auxiliary model, and the steps are repeated until the update auxiliary model disappears.

[0419] (Item B6) The method according to any one of Items B1 to B5, wherein

[0420] The following steps are also included:

[0421] All the samples are divided into reference samples and test samples. Using the reference samples, the maximum discriminant model (MAX) or the optimal discriminant model among the multiple discriminant model sets including the maximum model is determined as the reference discriminant model. The reference discriminant model is used to perform discriminant estimation using the test samples, thereby verifying the reference discriminant model, and the ideal final discriminant model is determined based on the result of the discriminant estimation.

[0422] (Item B7) The method according to any one of Items B1 to B6, wherein

[0423] The step A) includes providing a reference sample and generating the discriminant model based on the reference sample.

[0424] (Item B8) The method according to any one of Items B1 to B7, wherein

[0425] The machine learning is aimed at the identification of pain.

[0426] (Item B9) The method according to any one of Items B1 to B5, wherein

[0427] The machine learning is aimed at discriminating pain, the samples are brain wave data of the subject in response to experimental stimulation, and the discrimination model is calculated based on the brain wave data.

[0428] (Item B10) The method according to any one of Items B1 to B9, wherein

[0429] The machine learning aims at discriminating pain, the sample is the brain wave data of the subject in response to experimental stimulation, the discriminant model is calculated based on the brain wave data, and the method also includes the following steps: dividing all the samples into reference samples and test samples, using the reference sample, determining the maximum discriminant (MAX) model or the optimal discriminant model in a set of multiple discriminant models including the maximum model as the reference discriminant model, using the reference discriminant model to perform discriminant estimation using the test sample, thereby verifying the reference discriminant model, and determining the ideal final discriminant model based on the result of the discriminant estimation.

[0430] (Item B11) The method according to Item B10, characterized in that

[0431] The earliest interval in the electroencephalogram data of the sample is used as the reference sample, and the other intervals are used as the test samples.

[0432] (Item B12) The method according to any one of Items B1 to B11, wherein

[0433] The step A) includes providing a reference sample of a brain wave obtained by applying a reference stimulus to a body part of the subject, and generating the discrimination model based on the reference sample.

[0434] (Item B13) A program for causing a computer to perform an improved method for machine learning, the method comprising:

[0435] A) Steps for performing machine learning to generate a discriminant model;

[0436] B) Steps to determine hyperparameters in all samples through cross-validation;

[0437] C) using the hyperparameters to generate a discriminant model and calculating the discriminant accuracy for each sample;

[0438] D) a step of grouping each sample based on the discrimination accuracy;

[0439] E) determining independent hyperparameters for each group; and

[0440] F) Using the independent hyperparameters to generate independent discriminant models for each group, calculating the discrimination accuracy of each sample, and generating a maximum discriminant (MAX) model therefrom.

[0441] (Item B14) A recording medium storing a program for causing a computer to execute an improved method for machine learning, the method comprising:

[0442] A) Steps for performing machine learning to generate a discriminant model;

[0443] B) Steps to determine hyperparameters in all samples through cross-validation;

[0444] C) using the hyperparameters to generate a discriminant model and calculating the discriminant accuracy for each sample;

[0445] D) a step of grouping each sample based on the discrimination accuracy;

[0446] E) determining independent hyperparameters for each group; and

[0447] F) Using the independent hyperparameters to generate independent discriminant models for each group, calculating the discrimination accuracy of each sample, and generating a maximum discriminant (MAX) model therefrom.

[0448] (Item B15) A system for implementing an improved method for machine learning, the system comprising:

[0449] A) a discriminant model generation module, which performs machine learning to generate a discriminant model;

[0450] B) a hyperparameter determination module, which determines the hyperparameters in all samples through cross-validation;

[0451] C) a discrimination accuracy calculation module, which uses the hyperparameters to generate a discrimination model and calculates the discrimination accuracy for each sample;

[0452] D) a grouping module, which groups each sample based on the discrimination accuracy;

[0453] E) an independent hyperparameter module, which determines independent hyperparameters for each group; and

[0454] F) A maximum discriminant model generation module generates an independent discriminant model for each group using the independent hyperparameters, calculates the discriminant accuracy of each sample, and generates a maximum discriminant (MAX) model therefrom.

[0455] (Item B16) A system for performing an improved method of machine learning, the system comprising a machine learning execution module, the machine learning execution module being configured to perform the method comprising the following steps:

[0456] A) Steps for performing machine learning to generate a discriminant model;

[0457] B) Steps to determine hyperparameters in all samples through cross-validation;

[0458] C) using the hyperparameters to generate a discriminant model and calculating the discriminant accuracy for each sample;

[0459] D) a step of grouping each sample based on the discrimination accuracy;

[0460] E) determining independent hyperparameters for each group; and

[0461] F) Using the independent hyperparameters to generate independent discriminant models for each group, calculating the discrimination accuracy of each sample, and generating a maximum discriminant (MAX) model therefrom.

[0462] (Item B17) A method for identifying pain, comprising:

[0463] A) Steps for performing machine learning to generate a discriminant model;

[0464] B) Steps to determine hyperparameters in all samples through cross-validation;

[0465] C) using the hyperparameters to generate a discriminant model and calculating the discriminant accuracy for each sample;

[0466] D) a step of grouping each sample based on the discrimination accuracy;

[0467] E) determining independent hyperparameters for each group; and

[0468] F) A step of generating an independent pain discrimination model for each group using the independent hyperparameter, calculating the pain discrimination accuracy of each sample, and generating a maximum pain discrimination model therefrom.

[0469] (Item B18) The method according to Item 17,

[0470] The discriminant model is generated by a processing method selected from supervised machine learning models including LASSO (Least Absolute Shrinkage and Selection Operator), linear regression, logistic regression, support vector machine, neural network, random forest, Bayesian, clustering, decision tree, or unsupervised machine learning models including deep learning.

[0471] (Item B19) A program for causing a computer to execute a pain identification method, the method comprising:

[0472] A) Steps for machine learning to generate a pain discrimination model;

[0473] B) Steps to determine hyperparameters in all samples through cross-validation;

[0474] C) using the hyperparameters to generate a pain discrimination model and calculating the pain discrimination accuracy for each sample;

[0475] D) a step of grouping each sample based on the discrimination accuracy;

[0476] E) determining independent hyperparameters for each group; and

[0477] F) The step of using the independent hyperparameters to generate an independent pain discrimination model for each group, calculating the pain discrimination accuracy of each sample, and generating a maximum pain discrimination (MAX) model therefrom.

[0478] (Item B20) A recording medium storing a program for causing a computer to execute a pain identification method, the method comprising:

[0479] A) Steps for machine learning to generate a pain discrimination model;

[0480] B) Steps to determine hyperparameters in all samples through cross-validation;

[0481] C) using the hyperparameters to generate a pain discrimination model and calculating the pain discrimination accuracy for each sample;

[0482] D) a step of grouping each sample based on the discrimination accuracy;

[0483] E) determining independent hyperparameters for each group; and

[0484] F) The step of using the independent hyperparameters to generate an independent pain discrimination model for each group, calculating the pain discrimination accuracy of each sample, and generating a maximum pain discrimination (MAX) model therefrom.

[0485] (Project B21) A pain identification system comprising:

[0486] A) a discriminant model generation module, which performs machine learning to generate a pain discriminant model;

[0487] B) a hyperparameter determination module, which determines the hyperparameters in all samples through cross-validation;

[0488] C) a discrimination accuracy calculation module, which uses the hyperparameters to generate a pain discrimination model and calculates the pain discrimination accuracy for each sample;

[0489] D) a grouping module, which groups each sample based on the discrimination accuracy;

[0490] E) an independent hyperparameter module, which determines independent hyperparameters for each group; and

[0491] F) A maximum discrimination (MAX) model generation module, which uses the independent hyperparameters to generate an independent pain discrimination model for each group, calculates the pain discrimination accuracy of each sample, and generates a maximum pain discrimination (MAX) model therefrom.

[0492] (Item B22) A pain identification system, the system comprising a machine learning execution module, the machine learning execution module being configured to execute a method comprising the following steps:

[0493] A) Steps for machine learning to generate a pain discrimination model;

[0494] B) Steps to determine hyperparameters in all samples through cross-validation;

[0495] C) using the hyperparameters to generate a pain discrimination model and calculating the pain discrimination accuracy for each sample;

[0496] D) a step of grouping each sample based on the discrimination accuracy;

[0497] E) determining independent hyperparameters for each group; and

[0498] F) The step of using the independent hyperparameters to generate an independent pain discrimination model for each group, calculating the pain discrimination accuracy of each sample, and generating a maximum pain discrimination (MAX) model therefrom.

[0499] In another scheme, the inventors have conducted intensive research and discovered new parameters that can be used as brain wave feature quantities for pain discrimination, and found that among them, complexity indicators such as entropy and the mutual relationship between brain wave feature quantities (such as average amplitude (potential), frequency power, complexity indicators, etc.) (mutual relationship between brain wave feature quantities; such as synchronization, correlation, etc.) can efficiently provide high-precision pain discrimination technology.

[0500] In this aspect, the present invention provides, for example, the following items.

[0501] (Item C1A) A method for determining pain using one or more characteristic quantities of brain waves.

[0502] (Item C1B) The method according to Item C1, wherein the characteristic quantity includes at least one selected from amplitude, correlation between brain wave characteristic quantities, frequency power, and complexity index.

[0503] (Item C1C) A method according to any one of the above items, wherein the amplitude includes an average amplitude (such as an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as an amplitude median, an amplitude mode, an amplitude maximum, a peak amplitude and / or an quartile amplitude, the mutual relationship between the brain wave feature quantities includes potential correlation (such as frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality and their subspecies)) or phase synchronization between electrodes (such as coherence, phase-locking value and their subspecies), the frequency power includes spectral density, power spectrum and / or their subspecies, and the complexity index includes at least one selected from entropy (such as multi-scale entropy (MSM), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy, etc.), and biological potential feature quantities (such as eye movement potentials reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain.

[0504] (Item C1D) The method according to any one of the above items, further comprising the step of creating a discrimination model using the feature amount.

[0505] (Item C1) A method for determining pain using an index of brain wave complexity.

[0506] (Item C2) The method according to item C1A-C1D or item C1, wherein the complexity indicator includes entropy.

[0507] (Item C3) A method according to any one of the above items, wherein the complexity indicator includes at least one selected from multi-scale entropy (MSE), sample entropy, self entropy, average entropy, joint entropy, relative entropy and conditional entropy.

[0508] (Item C4) A method according to any one of the above items, wherein the complexity index is a complexity index at at least one electrode selected from electrode Fp1, electrode Fp2, electrode F3, electrode F4, electrode C3, electrode C4 and electrode Pz.

[0509] (Item C5) A method according to any one of the above items, comprising the step of generating a decision model through machine learning using the complexity indicator.

[0510] (Item C6A) A method for generating a model for determining pain in a subject, comprising:

[0511] a) obtaining a characteristic amount of brain waves from the subject;

[0512] b) determining a characteristic quantity coefficient related to the pain for the characteristic quantity, and performing reduction on the pain;

[0513] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced feature quantities or their combination; and

[0514] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0515] (Item C6B) A method according to Item C6A above, wherein the characteristic quantity includes at least one selected from amplitude, correlation between brain wave characteristic quantities, frequency power, and complexity index.

[0516] (Item C6C) A method according to any one of the above items, wherein the amplitude includes an average amplitude (for example, an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as the median amplitude, the mode amplitude, the maximum amplitude, the peak amplitude and / or the quartile amplitude, the mutual relationship between the brain wave feature quantities includes potential correlation (for example, frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality and their subspecies)) or phase synchronization between electrodes (for example, coherence, phase-locking value and their subspecies), the frequency power includes spectral density, power spectrum and / or their subspecies, and the complexity index includes at least one selected from entropy (for example, multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy, etc.), and biological potential feature quantities (eye movement potential reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain.

[0517] (Item C6) A method for generating a model for determining pain of a subject, comprising:

[0518] a) obtaining a complexity index of brain waves from the subject;

[0519] b) determining a characteristic quantity coefficient related to the pain according to the complexity index, and reducing the pain;

[0520] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced feature quantities or their combination; and

[0521] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0522] (Item C7A) A method for determining pain in a subject, comprising:

[0523] a) obtaining a characteristic amount of brain waves from a reference object;

[0524] b) determining a characteristic quantity coefficient related to the pain for the characteristic quantity, and performing reduction on the pain;

[0525] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced feature quantities or their combination;

[0526] d) steps for determining a discriminant analysis model that achieves a specified accuracy;

[0527] e) steps for obtaining a complexity indicator from the test subject; and

[0528] f) Substituting the complexity index of the test subject into the step of the model to judge the pain.

[0529] (Item C7B) A method according to Item C7A, wherein the characteristic quantity includes at least one selected from amplitude, correlation between brain wave characteristic quantities, frequency power and complexity index.

[0530] (Item C7C) A method according to any one of the above items, wherein the amplitude includes an average amplitude (for example, an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as the median amplitude, the mode amplitude, the maximum amplitude, the peak amplitude and / or the quartile amplitude, the mutual relationship between the brain wave feature quantities includes potential correlation (for example, frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality and their subspecies)) or phase synchronization between electrodes (for example, coherence, phase-locking value and their subspecies), the frequency power includes spectral density, power spectrum and / or their subspecies, and the complexity index includes at least one selected from entropy (for example, multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy, etc.), and biological potential feature quantities (eye movement potential reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain.

[0531] (Item C7) A method for determining pain in a subject, comprising:

[0532] a) Steps to obtain complexity indicators from reference objects;

[0533] b) determining a characteristic quantity coefficient related to the pain according to the complexity index, and reducing the pain;

[0534] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced feature quantities or their combination;

[0535] d) steps for determining a discriminant analysis model that achieves a specified accuracy;

[0536] e) steps for obtaining a complexity indicator from the test subject; and

[0537] f) Substituting the complexity index of the test subject into the step of the model to judge the pain.

[0538] (Item C8A) A method for determining pain in a subject, comprising:

[0539] e) a step of obtaining a characteristic amount of brain waves from a test subject; and

[0540] f) a step of substituting the characteristic value of the test subject into a model for distinguishing pain to distinguish pain,

[0541] The model is generated through the following steps:

[0542] a) obtaining the feature quantity from a reference object;

[0543] b) determining a characteristic quantity coefficient related to the pain for the characteristic quantity, and performing reduction on the pain;

[0544] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced feature quantities or their combination; and

[0545] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0546] (Item C8B) The method according to Item C8 above, wherein the characteristic quantity includes at least one selected from amplitude, correlation between brain wave characteristic quantities, frequency power and complexity index.

[0547] (Item C8C) A method according to any one of the above items, wherein the amplitude includes an average amplitude (for example, an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as the median amplitude, the mode amplitude, the maximum amplitude, the peak amplitude and / or the quartile amplitude, the mutual relationship between the brain wave feature quantities includes potential correlation (for example, frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality and their subspecies)) or phase synchronization between electrodes (for example, coherence, phase-locking value and their subspecies), the frequency power includes spectral density, power spectrum and / or their subspecies, and the complexity index includes at least one selected from entropy (for example, multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy, etc.), and biological potential feature quantities (such as eye movement potentials reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain.

[0548] (Item C8) A method for determining pain in a subject, comprising:

[0549] e) steps for obtaining a complexity indicator from the test subject; and

[0550] f) Substituting the complexity index of the test subject into a model for distinguishing pain to distinguish pain, wherein the model is generated by the following steps:

[0551] a) Steps to obtain complexity indicators from reference objects;

[0552] b) determining a characteristic quantity coefficient related to the pain with respect to the complexity index, and performing a reduction step on the pain;

[0553] c) a step of creating a discriminant analysis model based on the reduced feature quantities or their combination through machine learning and verification; and

[0554] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0555] (Item C9) A method according to any one of the above items, wherein the reduced feature quantities or their combination include alternating comprehensive brain wave feature quantities (this feature quantity set is extracted from limited electrodes (Fp1, Fp2, etc.) through a variety of signal processing methods, and is adaptively alternating according to the monitoring environment and / or individual differences through a reduction process, for example, it is so called because it "comprehensively" contributes to the pain discrimination model through a linear regression model).

[0556] (Item C10) A program for causing a computer to execute the method described in any one of the above items.

[0557] (Item C11) A recording medium storing a program for causing a computer to execute the method described in any one of the above items.

[0558] (Item C12) A method for determining pain using a correlation between brain wave feature quantities.

[0559] (Item C13) The method according to any one of the above items, wherein the mutual relationship between the electroencephalogram feature quantities is a mutual relationship between the electroencephalogram feature quantities in the same or different electrodes.

[0560] (Item C14) A method according to any of the above items, wherein the relationship is selected from time correlation, spatial correlation, spatiotemporal synchronicity, spatial relation or combination, irrelevance or uncorrelation, delayed or time-related collapse, positive or negative or correlated nature, level of similarity or correlation coefficient, and consistent or complete correlation.

[0561] (Item C15) The method according to Item C14, wherein the synchronization includes phase synchronization (Phase Locking Value) or coherence.

[0562] (Item C16) The method according to any one of the above items, wherein the electroencephalogram feature quantity correlation is a correlation of electroencephalogram feature quantities between different electrodes.

[0563] (Item C17) The method according to Item C16, wherein the different electrodes are located in a relatively front-to-rear position relationship on the head.

[0564] (Item C18) A method according to Item C16 or C17, wherein at least one of the different electrodes is present in the front of the head.

[0565] (Item C19) The method according to any one of Items C16 to C18, wherein at least one of the different electrodes is present at the back of the head.

[0566] (Item C20) A method according to any one of Items C16 to C19, wherein at least one of the different electrodes is present at the front of the head and the other electrodes are present at the back of the head.

[0567] (Item C21) A method according to any of the above items, wherein the correlation includes a correlation between electrodes at the front of the head and electrodes at the back of the head.

[0568] (Item C22) A method according to any one of the above items, wherein the correlation includes a correlation between electrodes in the frontal lobe and electrodes in the parietal lobe.

[0569] (Item C23) A method according to any one of the above items, wherein the frontal lobe portion includes at least one selected from the frontal pole Fp1, the frontal pole Fp2, the frontal lobe portion F3, the frontal lobe portion F4, the frontotemporal lobe portion F7, the frontotemporal lobe portion F8, the mid-frontal lobe portion Fz and adjacent parts, and the parietal lobe portion includes the mid-parietal lobe portion Pz, the parietal lobe portion P3 and the parietal lobe portion P4.

[0570] (Item C24) The method according to any one of the above items, wherein the brain wave feature amount includes at least one selected from average amplitude, frequency power, and complexity index.

[0571] (Item C25) The method according to any one of the above items, wherein the average amplitude depends on the stimulus presentation time and is the average value of the absolute value of the amplitude during 15 seconds after the stimulus presentation.

[0572] (Item C26) A method according to any one of the above items, wherein the frequency power includes at least one of the five frequency bands of δ, θ, α, β and γ of the frontal pole Fp1, frontal pole Fp2, frontal lobe F3, frontal lobe F4, central part C3, central part C4 and mid-parietal part Pz.

[0573] (Item C27) The method according to any one of the above items, further comprising the step of generating a determination model by machine learning using the electroencephalogram feature amount.

[0574] (Item C28) A method for determining pain in a subject, comprising the steps of:

[0575] The pain of the subject is determined using a model for determining the pain of the subject using feature quantities at at least two electrodes including at least one electrode at the front of the head.

[0576] (Item C29) A method for determining pain in a subject, comprising the steps of:

[0577] The pain of the subject is determined using a model for determining the pain of the subject including feature quantities at at least two electrodes around Fz or Fpz or both.

[0578] (Item C30) The method according to any one of the above items, wherein the surroundings include at least one of F3, F4, Fp1 and Fp2.

[0579] (Item C31) A method for generating a model for determining pain of a subject, comprising:

[0580] a) obtaining a step related to a brain wave characteristic quantity from the subject;

[0581] b) determining a characteristic quantity coefficient related to the pain based on the brain wave characteristic quantity correlation, and reducing the pain;

[0582] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced feature quantities or their combination; and

[0583] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0584] (Item C32) A method for determining pain in a subject, comprising:

[0585] a) obtaining a step related to the brain wave characteristic quantity from a reference object;

[0586] b) determining a characteristic quantity coefficient related to the pain based on the brain wave characteristic quantity correlation, and reducing the pain;

[0587] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced feature quantities or their combination;

[0588] d) steps for determining a discriminant analysis model that achieves a specified accuracy;

[0589] e) a step of obtaining a characteristic value of an electroencephalogram from a test subject; and

[0590] f) The step of substituting the characteristic amount of the brain wave of the test subject into the model to judge the pain.

[0591] (Item C33) A method for determining pain in a subject, comprising:

[0592] e) a step of obtaining a characteristic value of an electroencephalogram from a test subject; and

[0593] f) a step of replacing the brain wave characteristic quantity of the test subject with a pain discrimination model to discriminate pain,

[0594] The model is generated through the following steps:

[0595] a) obtaining a step related to the brain wave characteristic quantity from a reference object;

[0596] b) determining a characteristic quantity coefficient related to the pain based on the brain wave characteristic quantity correlation, and reducing the pain;

[0597] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced feature quantities or their combination; and

[0598] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0599] (Item C34) A method according to any one of Items 31 to 33, wherein the reduced feature quantities or their combination include alternating comprehensive brain wave feature quantities (this feature quantity set is extracted from limited electrodes (Fp1, Fp2, etc.) by various signal processing methods, and is adaptively alternating according to the monitoring environment and / or individual differences through the reduction process, for example, it is so called because it "comprehensively" contributes to the pain discrimination model through a linear regression model).

[0600] (Item C35) A program for causing a computer to execute the method described in any one of the above items.

[0601] (Item C36) A recording medium storing a program for causing a computer to execute the method described in any one of the above items.

[0602] (C37A) A device for evaluating or determining pain experienced by a subject, the device comprising:

[0603] A) a head-mounted device comprising at least one electrode for obtaining a brain wave signal; and

[0604] B) Basic unit,

[0605] The basic unit calculates a parameter of a characteristic amount of at least one brain wave,

[0606] generating a discriminant model relating the parameter to the subject's pain level,

[0607] By applying the parameter of the subject to the discriminant model, the pain level of the subject is calculated and displayed.

[0608] (Item C37B) An apparatus according to Item 37A, wherein the characteristic quantity includes at least one selected from amplitude, correlation between brain wave characteristic quantities, frequency power, and complexity index.

[0609] (Item C37C) A method according to any one of the above items, wherein the amplitude includes an average amplitude (such as an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as an amplitude median, an amplitude mode, an amplitude maximum, a peak amplitude and / or an quartile amplitude, the interrelationships between the brain wave feature quantities include potential correlation (such as frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality and their subspecies) or phase synchronization between electrodes (such as coherence, phase-locking value and their subspecies), the frequency power includes spectral density, power spectrum and / or their subspecies, and the complexity index includes at least one selected from entropy (such as multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy, etc.), and biological potential feature quantities (such as eye movement potentials reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain.

[0610] (Item C37) A device for evaluating or determining pain experienced by a subject, the device comprising:

[0611] A) a head-mounted device comprising at least one electrode for obtaining a brain wave signal; and

[0612] B) Basic unit,

[0613] The basic unit calculates at least one parameter selected from the group consisting of a correlation between brain wave feature quantities and a complexity index of brain waves.

[0614] generating a discriminant model relating the parameter to the subject's pain level,

[0615] By applying the parameter of the subject to the discriminant model, the pain level of the subject is calculated and displayed.

[0616] (Item C38) The device according to Item C37, wherein the correlation between the brain wave feature quantities and the brain wave complexity index further include any one or more features of the above items.

[0617] (Item C39A) A computer program for causing a device to perform a process for evaluating or determining pain experienced by a subject, the process:

[0618] calculating a parameter including a characteristic amount of at least one brain wave,

[0619] generating a discriminant model relating the parameter to the subject's pain level,

[0620] The pain level of the subject is calculated and performed by applying the parameter of the subject to the discriminant model.

[0621] (Item C39B) A program according to Item C39A, wherein the characteristic quantity includes at least one selected from amplitude, correlation between brain wave characteristic quantities, frequency power and complexity index.

[0622] (Item C39C) A method according to any one of the above items, wherein the amplitude includes an average amplitude (such as an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as an amplitude median, an amplitude mode, an amplitude maximum, a peak amplitude and / or an quartile amplitude, the interrelationships between the brain wave feature quantities include potential correlation (such as frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality and their subspecies) or phase synchronization between electrodes (such as coherence, phase-locking value and their subspecies), the frequency power includes spectral density, power spectrum and / or their subspecies, and the complexity index includes at least one selected from entropy (such as multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy, etc.), and biological potential feature quantities (such as eye movement potentials reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain.

[0623] (Item C39) A computer program for causing a device to perform a process for evaluating or determining pain experienced by a subject, the process:

[0624] calculating a parameter including at least one selected from the group consisting of a correlation between electroencephalogram feature quantities and an index of electroencephalogram complexity,

[0625] generating a discriminant model relating the parameter to the subject's pain level,

[0626] By applying the parameter of the subject to the discriminant model, the pain level of the subject is calculated and displayed.

[0627] (Item C40) The program according to Item C39, wherein the correlation between the electroencephalogram feature quantities and the electroencephalogram complexity index further include any one or more features in the above items.

[0628] (Item C41A) A recording medium storing a computer program for causing a device to execute a process for evaluating or determining pain experienced by a subject, the process:

[0629] calculating a parameter including a characteristic amount of at least one brain wave,

[0630] generating a discriminant model relating the parameter to the subject's pain level,

[0631] The parameter of the subject is applied to the discrimination model to calculate the pain level of the subject and display it.

[0632] (Item C41B) A recording medium according to Item C37A, wherein the feature quantity includes at least one selected from amplitude, electroencephalogram feature quantity correlation, frequency power, and complexity index.

[0633] (Item C41C) A method according to any one of the above items, wherein the amplitude includes an average amplitude (such as an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as an amplitude median, an amplitude mode, an amplitude maximum, a peak amplitude and / or an quartile amplitude, the interrelationships between the brain wave feature quantities include potential correlation (such as frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality and their subspecies) or phase synchronization between electrodes (such as coherence, phase-locking value and their subspecies), the frequency power includes spectral density, power spectrum and / or their subspecies, and the complexity index includes at least one selected from entropy (such as multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy, etc.), and biological potential feature quantities (such as eye movement potentials reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain.

[0634] (Item C41) A recording medium storing a computer program for causing a device to execute a process for evaluating or determining pain experienced by a subject, the process:

[0635] calculating a parameter including at least one selected from the group consisting of a correlation between electroencephalogram feature quantities and an index of electroencephalogram complexity,

[0636] generating a discriminant model relating the parameter to the subject's pain level,

[0637] By applying the parameter of the subject to the discriminant model, the pain level of the subject is calculated and displayed.

[0638] (Item C42) The recording medium according to Item C41, wherein the electroencephalogram feature quantity mutual relationship and the electroencephalogram complexity index further include any one or more features of the above items.

[0639] (C43A) A method for assessing or determining pain experienced by a subject, the method comprising:

[0640] a step of calculating a parameter including a characteristic amount of at least one brain wave;

[0641] the step of generating a discriminant model relating the parameter to the subject's pain level; and

[0642] The step of applying the parameter of the object to the discrimination model to calculate the pain level of the object and displaying it.

[0643] (Item C43B) A method according to Item C43A, wherein the characteristic quantity includes at least one selected from amplitude, correlation between brain wave characteristic quantities, frequency power and complexity index.

[0644] (Item C43C) A method according to any one of the above items, wherein the amplitude includes an average amplitude (such as an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as an amplitude median, an amplitude mode, an amplitude maximum, a peak amplitude and / or an quartile amplitude, the interrelationships between the brain wave feature quantities include potential correlation (such as frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality and their subspecies) or phase synchronization between electrodes (such as coherence, phase-locking value and their subspecies), the frequency power includes spectral density, power spectrum and / or their subspecies, and the complexity index includes at least one selected from entropy (such as multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy, etc.), and biological potential feature quantities (such as eye movement potentials reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain.

[0645] (Item C43) A method for assessing or determining pain experienced by a subject, the method comprising:

[0646] A step of calculating a parameter including at least one selected from the group consisting of a correlation between electroencephalogram feature quantities and an index of electroencephalogram complexity;

[0647] the step of generating a discriminant model relating the parameter to the subject's pain level; and

[0648] The step of calculating and displaying the pain level of the subject by applying the parameter of the subject to the discriminant model.

[0649] (Item C44) The method according to Item C43, wherein the correlation between the brain wave feature quantities and the brain wave complexity index further include any one or more features in the above items.

[0650] In the present invention, one or more of the above features are intended to be provided in combination in addition to the combinations explicitly described. Other embodiments and advantages of the present invention will be recognized by those skilled in the art upon reading and understanding the following detailed description as needed.

[0651] Effects of the Invention

[0652] The present invention can accurately monitor the subjective pain transition. The present invention can also perform more precise treatment and / or surgery in accordance with the subjective pain transition based on the accurately recorded pain transition, and is useful in the medical-related industry.

[0653] The present invention can also efficiently identify pain. Pain can be identified at a very high level of accuracy, and treatment and / or surgery that conforms to the subjective feeling can be performed more precisely, which is useful in the medical-related industry.

[0654] The present invention realizes a method that can customize machine learning. This is very important in all fields that require customized machine learning, and applications in various fields such as communications, electrical, acoustic, mechanical, chemical, and biological fields can be imagined. For example, it is particularly useful in fields where customization is important, such as personalized medicine (Precision Medicine). In this example, the present invention can efficiently distinguish pain. Pain can be distinguished at a level with exceptionally high accuracy, and for pain that is difficult to grasp by subjective performance, accurate diagnosis, treatment, and / or surgery can be performed more finely, which is useful in medical-related industries.

[0655] When the present invention is used, pain can be efficiently determined. Pain can be determined at a very high level of accuracy, and treatment and / or surgery that is subjectively appropriate and objectively reasonable can be performed more finely, which is useful in medical-related industries.

[0656] In addition, the present invention effectively applies the interrelationships (phase synchronization, coherence, combination, causality) of brain wave feature quantities to pain discrimination, thereby producing an effect that includes not only the pattern of "intensified pain → increased EEG feature quantity" but also the pattern of "intensified pain → decreased EEG feature quantity". The discovery that it is possible to suppress the phenomenon that pain discrimination cannot be covered by a single feature quantity is considered valuable. The feature quantity of the present invention, which can also be called "comprehensive EEG feature quantity", can achieve such an effect. In addition, with reference to the description of this specification, those skilled in the art can appropriately and adaptively implement it according to the situation during the reduction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0657] Figure 1A is a graph showing the relationship between electrical stimulation and pain level (VAS).

[0658] Figure 1B is a graph showing the relationship between electrical stimulation and pain level (paired comparison).

[0659] Figure 1C is a graph showing the relationship between electrical stimulation and brain wave amplitude.

[0660] Figure 1D This is a graph showing an example of an electroencephalogram waveform.

[0661] Figure 1E This is a graph showing the relationship between the pain level (VAS) caused by electrical stimulation and the brain wave amplitude.

[0662] Figure 1F This is a graph showing the relationship between the pain level (paired comparison) caused by electrical stimulation and the brain wave amplitude.

[0663] Figure 1G This is a graph showing the relationship between the pain level (VAS) caused by thermal stimulation and the brain wave amplitude.

[0664] Figure 1H This is a graph showing the relationship between the pain level due to thermal stimulation (paired comparison) and the brain wave amplitude.

[0665] Fig. 1I The comparison between the EEG amplitude (absoluteized and standardized) and the change in cold pain (A) and the comparison between the EEG amplitude and the change in subjective pain intensity (B) at the average time interval of 10 seconds (top), 20 seconds (middle) and 40 seconds (bottom) are shown. In any case, the correlation coefficient is improved as the average time interval is longer. The horizontal axis represents the time window, and the vertical axis represents the absolute EEG amplitude (standardized) or the cold pain intensity (standardized).

[0666] Figure 2 The comparison of EEG changes (absoluteized and standardized) under high temperature pain stimulation conditions and low temperature pain stimulation conditions and the average correlation coefficient (n=32) between temperature or EEG changes and pain evaluation changes is shown. Under both high temperature stimulation and low temperature stimulation conditions, statistical differences in correlation coefficients can be seen between adjacent time intervals around 30 seconds to 40 seconds. A shows low temperature stimulation, B shows subjective low temperature pain intensity, C shows high temperature intensity, and D shows subjective high temperature pain intensity. The vertical axis represents the correlation coefficient and the horizontal axis represents the average interval (seconds). * indicates statistical significance (p<0.05).

[0667] Figure 3 Comparison of average correlation coefficient values ​​between the non-overlapping time direction averaging method and the overlapping time direction averaging method is shown. In the non-overlapping, overlapping (moving average) and mixed methods (non-overlapping average after moving average), the correlation coefficient with the pain index (subjective evaluation, low temperature intensity) is higher when the non-overlapping method is used than when the overlapping method is used. The vertical axis represents the absolute correlation coefficient, the left group detection (left panel) represents the subjective pain evaluation, and the right group detection (right panel) represents the low temperature intensity. Black columns represent non-overlapping, white columns represent overlapping, and black and white mixed represents mixing.

[0668] Figure 4The difference between durations of less than about 15 seconds and longer durations of more than 40 seconds is shown. In particular, the difference in correlation coefficient values ​​similar to the Sigmoid function can be seen between the average interval durations of less than 15 seconds and more than 40 seconds. The vertical axis represents the correlation coefficient and the horizontal axis represents the interval duration of the average EEG. The bold line represents the S-shaped fitting function.

[0669] Figure 5 The pain index based on EEG is shown. It can be expected that the change in pain level under analgesics and sedatives in the clinical field is mainly moving from strong to weak. Here, the characteristic value (average of the absolute value of the amplitude (standardized)) of the condition with the highest pain level is used as the baseline to make the pain index, and the trend is expressed by the 40-second non-overlapping time averaging method. The vertical axis represents the pain index and the horizontal axis represents the time window (40-second average interval). The black dots represent the changes in EEG, and the dotted line strung together by white squares represents the transition of subjective evaluation.

[0670] Figure 6 It shows how the correlation between the average potential change and the pain index changes according to the change of the average time interval. Figure 6 A shows the subjective evaluation of pain (correlation between the mean potential amplitude (absolute value) and the subjective evaluation of pain under low temperature stimulation), Figure 6 B shows the change of the correlation coefficient of cold stimulation intensity (correlation between the average potential amplitude (absolute value) and the low temperature stimulation intensity). The vertical axis represents the correlation coefficient, and the horizontal axis represents the average time interval (seconds).

[0671] Figure 7 This is a random high temperature stimulation test paradigm in which the temperature and stimulation duration are changed. The temperature varies from 38℃ to 48℃, and the duration varies from 15 seconds to 240 seconds.

[0672] Figure 8 The change in the correlation coefficient when the arithmetic mean time interval is changed from 10 seconds to 120 seconds using the high temperature stimulation test data (6 levels of pain) is shown. It can be seen that the correlation coefficient tends to increase when the average time interval becomes longer. A represents the correlation between the arithmetic mean potential amplitude (absolute value) and the subjective pain evaluation, and B represents the correlation between the arithmetic mean potential amplitude (absolute value) and the high temperature stimulation intensity. The vertical axis represents the correlation coefficient, and the horizontal axis represents the arithmetic mean time interval (seconds).

[0673] Fig. 9The figure shows the change in the correlation coefficient when the geometric mean time interval is changed from 10 seconds to 120 seconds using high temperature stimulation test data (6 levels of pain). It can be seen that the correlation coefficient tends to increase when the average time interval becomes longer. A represents the correlation between the geometric mean potential amplitude (absolute value) and the subjective pain evaluation, and B represents the correlation between the geometric mean potential amplitude (absolute value) and the intensity of high temperature stimulation. The vertical axis represents the correlation coefficient, and the horizontal axis represents the geometric mean time interval (seconds).

[0674] Fig.10 The trend reduction process (here 10 seconds → 60 seconds, 60 seconds → 120 seconds) in the "grading pain trend monitoring method" used in pain level trend monitoring is illustrated. This is a method that covers everything from local monitoring to global monitoring by reducing the trend unit (here 10 seconds) along with the time lapse.

[0675] Fig.11 It is a schematic diagram of the scheme of the present invention.

[0676] Fig.12 It is a block diagram of the device structure of the present invention.

[0677] Fig.13 is Fig.12 A block diagram of a "graded pain trend monitoring method" is incorporated into the device composition.

[0678] Fig.14 Results obtained by analyzing the EEG data in the presentation of pain stimulation using a discriminant model (multiple regression model) using average amplitude are shown. The discriminant model was created using an experimental protocol different from the EEG data to be tested. Different pain stimulation protocols for high temperature stimulation and low temperature stimulation were used for modeling. A 15-second time period was set for the visualization of pain trends. A trend similar to the subjective evaluation of pain was shown. Regarding the number of samples, 51 people were observed. The data for the top 10 are shown graphically.

[0679] Fig.15 An embodiment of ID328 is shown.

[0680] Fig.16 An embodiment of ID358 is shown.

[0681] Fig.17 An embodiment of ID224 is shown.

[0682] Fig.18 An embodiment of ID343 is shown.

[0683] Fig.19 An embodiment of ID371 is shown.

[0684] Fig. 20 An embodiment of ID345 is shown.

[0685] Fig.21 An embodiment of ID416 is shown.

[0686] Fig. 22 An embodiment of ID383 is shown.

[0687] Fig.23 An embodiment of ID405 is shown.

[0688] Fig.24 An embodiment of ID208 is shown.

[0689] Fig.25 The summary of the analysis method of Example 2 of Example 7 is shown. In addition, the time-varying characteristics of the frequency characteristic quantity (β band) representing the delayed pain characteristics of C fibers are illustrated (represented by time-frequency analysis). In particular, level 6 starts to be delayed from 5 seconds after the stimulus presentation, and the frequency power increases compared to other levels.

[0690] Fig.26 The feature extraction process (1) EOG removal processing is shown. Process 1: Perform principal component analysis on the 4 electrode data and extract the EOG component (first component: gray). Then, implement a 1-30 Hz bandpass filter to reduce slow changes and rapid changes to form a smooth EOG waveform (black). Process 2: Regress the original data of each electrode through the EOG data and obtain the β coefficient of each electrode. Remove the component obtained by multiplying EOG and the β coefficient from the original recorded data. Gray (data with jitter): original Fp1 data. Black (data that changes near the center): data after EOG removal processing.

[0691] Fig. 27 The feature extraction process (2) EMG removal processing is shown. Process 1: In the waveform after EOG removal, the length of the stimulation interval × 3 is set to the low-end cutoff frequency and the high-end cutoff frequency is set to 30Hz to remove the noise component from the waveform. * The stimulation interval is set to × 3 times because the brain wave activity continues after the stimulus is presented, so a weak setting is made. At this stage, components above 30Hz can be excluded from the original data, so the "γ band" is not used as a feature this time.

[0692] Fig.28 The feature extraction process (3) EOG feature addition is shown. Process: "Noise inversion" is performed on the main component data 1ch for EOG removal and added to the feature. * As a process, bandpass filtering is implemented. After performing processes (1) to (3), the interval until 5 to 15 seconds after the stimulus presentation is divided into 1 second intervals, and the feature values ​​of amplitude and frequency (δ, θ, α, β) are produced by each electrode. That is, a total of 25 feature values ​​× 6 levels × 3 experiments × 10 intervals.

[0693] Fig.29 The discriminant model (normalized in the resting interval, with EOG feature values) is shown. Subjects below the chance level were excluded, and the model was re-created four times (81 subjects, 4860 samples). The basic model was created by discriminating L1 (40°C) and L6 (50°C). The refined feature value coefficients (the coefficients of the default pain discrimination model) are shown on the left. The graph on the right shows the overall distribution of pain estimates after four refinements.

[0694] Fig.30 An overview of the discrimination model is shown. Model creation was performed using L1 and L6 (30 samples each), and a total of 10,200 samples were used. The refinement benchmark was 81 subjects after four runs in a manner that left only subjects with a discrimination accuracy of 70%. Using the refined model, the discrimination accuracy of all 170 subjects for L1 and L6 (feature quantities 5 to 10 seconds after stimulus presentation, 6 samples) was "79.4±26.4%". There were "93" subjects with a discrimination accuracy of more than 70%, which was "54.7%", and there were "147" subjects with a discrimination accuracy of more than 50%, which was "86.5%".

[0695] Fig.31 The pain percentage is shown. This is a concept that represents the certainty of judging the pain estimate as "painful". The calculation method is as follows. i. Using the refined discrimination model of 81 people (discrimination accuracy > 70%), the discrimination estimates of no pain (30 samples × 81 people) and pain (30 samples × 81 people) are calculated. ii. The minimum value (-1) and the maximum value (2) of all pain estimates are divided by a width of, for example, 0.1, and the percentage of estimated values ​​with pain in each estimate is calculated. "Pain percentage (%) = the number of pain estimated values ​​ / (the number of pain estimated values ​​+ the number of no pain estimated values)" iii. The "pain percentage function" is obtained by fitting the Sigmoid function, and a percentage model is created. An example of the pain percentage function is shown below.

[0696] Fig.32An example of pain trend analysis is shown. The feature values ​​of other experimental data are substituted into the pain model, and a "pain trend display" is performed on the pain score according to the time change. The EEG data is processed as follows. In the resting period of 30 seconds from the start time of recording, the EEG feature values ​​are standardized and then substituted into the refined discrimination model. Corresponding to the increase in subjective evaluation and stimulation intensity, the estimated value increases, and the incidence rate (certainty) also increases. The correlation between the estimated value and the subjective evaluation of pain (COVAS) is as follows. :r=0.45,p≒0. Through the dual evaluation method of trend, the estimated value of the model also increases corresponding to the increase in the evaluation of inappropriateness, and auxiliary information on the degree of pain (percentage) caused at this time can also be obtained. By referring to the percentage, the authenticity of the discrimination judgment can be objectively judged.

[0697] Fig.33 The high temperature stimulation paradigm used in Example 8 and the like is shown. The high temperature stimulation is 6 levels, and each level is stimulated three times. The stimulation presentation time is 15 seconds, and the time interval between level blocks is 100 seconds. The characteristic quantity is the average absolute amplitude and average frequency power (δ, θ, α, β, γ) of three stimulations (15 seconds) of each level. The two levels to be judged are mild pain (levels 1 to 3) and severe pain (levels 4 to 6), and the number of samples is strength × 3 levels × 40 people = 240.

[0698] Fig.34 The flow of discriminant analysis with the reduction process of the number of feature quantities is shown.

[0699] Fig.35 The results of the discriminant analysis (Example 8) with the feature reduction process (feature ranking and discrimination accuracy) are shown. This is an economical pain discrimination estimation model with low computational cost. It can be seen that previously all 24 feature quantities were used for discrimination, but this time only the sigmoid function was used to approximate the higher two upper feature quantities, and the discrimination accuracy of "71.3%" for the two levels of pain was shown.

[0700] Fig.36The results of the comparative example are shown. The results (feature ranking and discrimination accuracy) of the discriminant analysis with feature reduction process (existing SVM-RFE, Support Vector Machine Recursinve Feature Elimination: Guyon et al., (2002). Gene Selection for Cancer Classification using Support Vector Machine. Machine Learning 46: 389-422)). In the economic pain discrimination estimation model, the following is known. 1) The types of the first four feature quantities of the existing support vector machine (SVM) with feature quantity reduction include the same types as in the case of S-type reduction; 2) Regarding the discrimination accuracy, "71.3%" was achieved using only one feature quantity, but the difference was only one feature quantity; 3) Regarding the computational cost, 240 leave-one-out cross-validations were performed for SVM model construction, and this process was performed 24 times for ranking the 24 feature quantities, so the computational cost was high.

[0701] Fig.37 The difference in computational cost between S-type reduction and reduction in SVM-RFE is shown. A shows the conventional SVM-RFE method. B shows an example of SVM with S-type reduction of the present invention.

[0702] Fig.38 The verification process of the economic discrimination model of the present invention is shown. 240 samples were divided into 10 parts (4 people × 10 groups), 90% (36 people) were used as data for machine learning (SVM) model making, and the pain level of the remaining 10% (4 people) was discriminated and estimated. Only the first two feature quantities were used.

[0703] Fig.39 The block paradigm of high temperature stimulation used in the experiment of model making for discriminating 32160 samples is shown. It includes three blocks: no pain, weak pain, and strong pain. Each stimulus is presented for 15 seconds. The time interval between blocks is 20 seconds. There are 5 groups in total, and each group includes 64 blocks.

[0704] Fig.40 The following is a diagram showing the process of extracting the feature quantities used in the discriminant model. This process includes 1) EEG data collection, 2) data extraction, 3) selection of two feature quantities, and 4) combination of feature quantities. Through these processes, a total of 49 feature quantities can be obtained.

[0705] Fig.41 The discriminant model determination process is shown in FIG. 1 through a double reduction process when selecting feature quantities and discriminant models. Fig.34 Compared with the discriminant model determination process of , the same process is included until 1) quantification of the discriminant characteristics of the feature quantity, 2) sorting of the feature quantity, and 3) hierarchical discriminant analysis that is included in the model in order from the top ranked feature quantity. On the other hand, the point that the economic discriminant model is selected by approximating the change in discrimination accuracy using a sigmoid function or the like at the final discriminant model selection time point is newly included. In addition, the numerical values ​​in the figure are all examples.

[0706] Fig.42 The variation patterns of the discrimination accuracy of all 49 discrimination models are shown. Fig.42 A shows the model making process. Fig.42 B shows the discriminant accuracy variation pattern of the final test data. As can be understood from the two graphs, there are two step-shaped inflection areas in the variation pattern, and the variation range of the first inflection area is larger. Therefore, after performing Sigmoid function approximation on the inflection area and investigating the minimum feature quantity exceeding the maximum value "63.7%" of the approximate function, it is determined that the discriminant model ( Fig.42 C). In addition, it can be seen that the discrimination accuracy from the inflection point on accounts for 54% of the overall discrimination accuracy improvement ( Fig.42 D).

[0707] Fig.43 The following table shows a list of 23 feature quantities included in the model that showed the highest gain in improving discrimination accuracy. The electroencephalogram feature quantities of Fp1 and Fp2 account for 9 (40%) of the total 23.

[0708] Fig.44 FIG2 shows a discriminant model determination process 2 implemented by a double reduction process when selecting feature quantities and discriminant models. Fig.41 The difference between the discriminant model determination process of is that after calculating the discrimination accuracy for the first time, the difference (Diff) of the discrimination accuracy between the n-1 feature quantity model (n≥2) and the n feature quantity model, and between adjacent models, is obtained, and the feature quantities with the largest difference value when the model is substituted are re-sorted. In this case, the feature quantity group that ranks first or is significantly related in the first discrimination accuracy is regarded as the "main feature quantity", and the others are regarded as "auxiliary feature quantities", and only the latter feature quantities are re-sorted. For example, in Fig.43In the feature quantity list, the maximum gain of 1.4% in discrimination accuracy was shown in the modeling from feature quantity No. 14 to No. 15. Therefore, in the re-ranking after excluding the high γ band feature quantity at the Pz electrode ranked first at the front, feature quantity No. 15 was listed at the top of the ranking and was substituted into the model at the initial stage of modeling. By adding this process, the main feature quantity that strongly approximates the discriminant function (such as sigmoid function) and the auxiliary feature quantity that has a low approximation but fine-tunes the model are combined at an early stage, and high discrimination accuracy can be obtained with a small number of feature quantities.

[0709] Fig.45 Shown in Fig.44 The figure describes how the feature re-ranking and discrimination accuracy change using the difference value (Diff) of the discrimination accuracy of adjacent models (n-1 feature quantity models and n feature quantity models). A shows the change in discrimination accuracy during model verification using cross-validation of learning data, and B shows the change in accuracy during test data discrimination accuracy verification. It is shown that: regarding the change after two re-rankings, the discrimination accuracy basically reaches the limit at the time point when the number of feature quantities is 10 to 15, which is the most economical discrimination model. This result provides the following inspiration: By re-ranking the number of feature quantities according to the difference value, the main feature quantity that strongly approximates the discriminant function (such as Sigmoid function, etc.) and the auxiliary feature quantity that fine-tunes the model can be effectively combined at an early stage to improve the discrimination accuracy.

[0710] Fig.46 A flowchart (A) showing the discrimination model refinement and improvement method and the actual situation of the improvement in discrimination accuracy from before refinement (B) to after refinement (C) in an embodiment (Example 13) using electrical stimulation.

[0711] Fig.47 A schematic diagram showing a discrimination device with a feature reduction process.

[0712] Fig.48 This is a schematic diagram showing another example of the reduction process based on feature quantity coefficients. The numerical values ​​in the figure are all examples.

[0713] Fig.49 A schematic diagram showing the reduction process based on hyperparameters. The values ​​in the figure are all examples.

[0714] Fig.50 This is a schematic diagram of the apparatus of the present invention, assuming that all parts are equipped as an example.

[0715] Fig.51This is a schematic diagram of the device of the present invention, showing an implementation mode in which the device part (left side) performs brain wave acquisition and data transmission and reception, and only has a visualization function. This is an implementation mode in which the generation of a hypothesis analysis and / or judgment discrimination model is performed through a cloud or a server. In this model, the extraction of brain wave feature quantities (analysis data) is performed on the terminal side.

[0716] Fig.52 This is a schematic diagram of the device of the present invention, showing an embodiment in which the device part (left side) performs brain wave acquisition and data transmission and reception and has only a visualization function. This is an embodiment in which the generation of the analysis and / or determination discrimination model is performed through the cloud or server. In this model, the extraction of brain wave feature quantities (analysis data) is performed on the server side.

[0717] Fig.53 This is a schematic diagram of the device of the present invention, showing an embodiment in which the device part (left side) acquires brain waves and sends and receives data, can save the discrimination model for on-site discrimination, and has the function of visualizing these. Such an embodiment is an embodiment that is implemented in facilities and / or places such as hospitals where it is difficult to send and receive radio waves. The generation of the discrimination model is performed through the cloud or server, while the actual substitution of measured data into the model is performed through the device. The extraction of brain wave feature quantities (analysis data) can be performed on both the terminal side and the server side.

[0718] Fig.54 The high temperature stimulation presentation paradigm commonly used in Example 16 and the like is shown. There are 3 types of high temperature blocks, including painless blocks (40°C), mild pain blocks (44°C), and intense pain blocks (48°C). Each block includes 2 stimuli, one stimulus for 15 seconds, and a total of 30 seconds. The time interval between blocks is 20 seconds. A total of 5 groups of subjects were stimulated in all 64 blocks.

[0719] Fig.55 The following is a diagram showing the process of extracting the feature quantities used in the discriminant model. This process includes 1) collecting EEG data, 2) extracting data, 3) selecting two feature quantities, and 4) combining feature quantities. A total of 49 feature quantities were obtained through these processes.

[0720] Fig.56The following is an overview of the "Customized Pain Discrimination Estimation Method 1". Here, as step 1, hyperparameters (C and γ in the case of a support vector machine (SVM)) in all samples are determined by cross validation (also called cross verification. Cross Validation; CV), as a step, a model is created based on all samples using the hyperparameters of step 1 and the discrimination accuracy of the individual level is calculated, as a step 3, a group of subjects with similar discrimination accuracy is created based on the ranking of the discrimination accuracy of the individual level, as a step 4, the hyperparameters of each group are determined by CV, as a step 5, a model is created based on the samples of each group using the hyperparameters obtained in step 4, the discrimination accuracy of the individual level is calculated, and the maximum discrimination model (expressed as "discrimination MAX model") is determined.

[0721] Fig.57 : is a schematic diagram briefly explaining step 1. In step 1, the hyperparameters (for example, C and γ in the case of SVM) in all samples are determined by cross-validation. Thus, the optimal hyperparameters are determined.

[0722] Fig.58 This is a schematic diagram for briefly explaining step 2. In step 2, the hyperparameters of step 1 are used to create a model based on all samples, and the discrimination accuracy of the individual level is calculated. In the schematic diagram, C=2 and γ=0.125 used in this embodiment 16 are shown. A chart is shown in which the subjects are sorted in descending order from the subjects with high discrimination accuracy.

[0723] Fig.59 This is a schematic diagram for briefly explaining step 3. In step 3, a group of subjects having similar discrimination accuracy is created based on the ranking of discrimination accuracy at the individual level.

[0724] Fig.60 This is a schematic diagram briefly explaining step 4. In step 4, the grouped samples are used to determine the optimal hyperparameters for each group (for example, C and γ in the case of SVM).

[0725] Fig.61 This is a diagram briefly explaining step 5. In step 5, the discrimination accuracy of individuals is calculated according to the number of models using the set of discrimination models obtained in step 4, and the maximum discrimination model ("discrimination MAX model") is determined for each person. Through this method, the average discrimination accuracy of all subjects increased by about 14% from 64% to 77%.

[0726] Fig.62The discriminant accuracy of the maximum discriminant model ("discriminant MAX model") is summarized by socio-demographic characteristics, namely gender and age group. It can be predicted that pain response characteristics vary by gender and age group. For example, the discrimination accuracy of a single discriminant model using both genders and all age groups does not reach 60% for men in their 60s, and the highest accuracy is 70% for women in their 20s. On the other hand, if the maximum discriminant (MAX) model is specified for each person of each gender and age group and the discrimination estimation is performed, it shows that the discrimination accuracy is improved by 10% even for men in their 60s, and reaches 77% and 80% for women in their 20s. This is an empirical example of the accuracy improvement effect achieved through customized machine learning based on grouping.

[0727] Fig.63 The following is an overview of the "customized pain discrimination estimation method 2" performed in Example 17. Steps 6 and 7 are performed on the basis of steps 1 to 5 of Example 16. The purpose is to improve the discrimination accuracy achieved by the pain discrimination model set technology (ensemble method). As step 6, the discrimination models are arranged in descending order for each person according to the discrimination accuracy obtained by using 62 models one by one. As step 7, the discrimination models are added one by one, and the discrimination accuracy is calculated by the "ensemble method (majority voting system)". The number of discrimination models with the highest discrimination accuracy is adopted ("ensemble pruning" is performed).

[0728] Fig.64 The conceptual diagram of the ensemble method is shown in Figure 1. The relationship between the training data set and the new test data is shown, and the process of making the final prediction is shown. The training data set is divided into multiple parts for sampling (grouping), and many independent discriminant models are created. The final discriminant result of the test data is determined by majority voting based on the aggregate of the discriminant results of these independent models.

[0729] Fig.65 Another concept showing the integration method Figure 2 . The process of calculating the discrimination accuracy is shown. In the example, by setting pain as the label "1" and painlessness as the label "0", the test data (individual pain samples) are predicted using each discrimination model obtained by grouping the learning data (samples). A majority vote is taken on the result, and the final discrimination prediction is determined as "pain" when there are more votes for the label "1" and as "no pain" when there are more votes for the label "0". By comparing with the actual pain label, the discrimination accuracy of the integrated discrimination model is determined. In addition, this example is an example of guided clustering algorithm (bagging).

[0730] Fig.66Example 1 shows the improvement of discrimination accuracy achieved by ensemble pruning. In this example, the result of subject 1 is shown. In the ensemble method, the discrimination estimation model achieved by majority voting of 15 discrimination models achieved an improvement in accuracy. Compared with the maximum discrimination (MAX) method achieved by a single model in Example 16, the discrimination accuracy of the ensemble method performed in Example 17 was improved by about 7%.

[0731] Fig.67 Example 2 of improvement in discrimination accuracy achieved by integrated pruning is shown. In this example, the result of subject 2 is shown. In the integrated method, the discrimination accuracy achieved by majority voting of four discrimination models achieved an improvement in accuracy. Compared with the maximum discrimination (Max) method achieved by a single model in Example 16, the discrimination accuracy of the integrated method in another example performed in Example 17 was also improved by about 7%.

[0732] Fig.68 A summary of the improvement in discrimination accuracy achieved by the integrated method 1 performed in Example 17 is shown. A comparison of the three discrimination methods for all subjects is presented. The discrimination accuracy of the integrated method is improved by about 16% over the discrimination accuracy of a single discrimination model using all samples, and is also improved by about 3% over the discrimination accuracy of a single discrimination (maximum) Max model.

[0733] Fig.69 The following is an overview of the "customized pain discrimination estimation method 3" performed in Example 18. In this method, the purpose is to improve the discrimination accuracy achieved by a new discrimination model set technology (ensemble method). In this example, as step 6A, the models are sorted in descending order according to the discrimination accuracy obtained by using 62 models one by one, and as step 7A, the maximum discrimination (MAX) model with the highest discrimination accuracy is set as the "main model", and the remaining "auxiliary models" are added one by one in a comprehensive manner, and the discrimination accuracy is repeatedly improved step by step.

[0734] Fig.70 The following is a detailed description of the improved integration method implemented in Example 18. Specifically, 1. The model with the maximum discrimination (MAX) is selected from all models (N) and is set as the "main model"; 2. The N-1 models after excluding the main model from all models are set as "auxiliary models"; 3. When one model is selected from the N-1 auxiliary models and added to the main model, the two model sets with the highest improvement in discrimination accuracy through the integration method are set as the "updated main model"; 4. The N-2 models after excluding the updated main model from all models are set as the "updated auxiliary model"; 5. The above steps 3-4 are repeated in the same way for the model set with the highest improvement in discrimination accuracy through the integration method to update the updated main model and the updated auxiliary model, and these steps are repeated until the updated auxiliary model disappears.

[0735] Fig.71 An example of improvement in discrimination accuracy achieved by the improved ensemble pruning method is shown. Subjects 1 to 4 performed in Example 18 are shown. For all subjects, discrimination accuracy was improved by 5 to 10% in the early stages of the iteration.

[0736] Fig.72 A summary of the improvement in discrimination accuracy achieved by Ensemble Method 2 (improved method) is shown. A comparison of the five discrimination methods for all subjects is shown. Regarding Ensemble Method 2 (improved method), it goes without saying that the discrimination accuracy is further improved by about 4% compared to the improved Ensemble Method 1 of Example 17, achieving a discrimination accuracy of about 83%.

[0737] Fig.73 This is an overview of the customized pain discrimination estimation method 4 with calibration. When actual pain monitoring is assumed, individual samples (a total of n) are divided into reference (m) and test (nm) samples to create a reference discrimination model, and discrimination estimation of the test samples assuming online monitoring samples is performed. The reference samples are samples in the model determination process before assuming actual online monitoring, and the other nm samples are samples assuming online pain monitoring.

[0738] Fig.74 This is a comparison of the discrimination accuracy results with calibration for one subject. In the three models (ensemble 1, ensemble 2, and maximum discrimination (MAX)) other than the single model for all samples, the accuracy of the reference discrimination model stabilized until the number of reference samples reached about 50. It can be seen that the discrimination accuracy of the test using the reference discrimination model gradually increased, but there was no significant change. Regarding the presentation of reference stimuli before the implementation of pain monitoring, it is believed that the upper limit is about 50 times.

[0739] Fig.75 This is a comparison of the discriminant accuracy results with calibration for all 132 people. Fig.74 The individual patterns of the three models (ensemble 1, ensemble 2, and maximum discrimination (MAX)) other than the single model for all samples, the accuracy gradient of the reference discrimination model stabilized until the number of reference samples was about 50. The test discrimination accuracy using the reference discrimination model showed a gradual improvement until the number of samples was about 50, except for ensemble 1 and the single model. It can be said that the overall results also show that the upper limit of the reference stimulus presentation before the implementation of pain monitoring is about 50 times.

[0740] Fig.76It shows how the discrimination accuracy of the ensemble method with calibration changes when the number of references changes. The discrimination accuracy of the test samples is similar for the single model and ensemble method 1 even when the number of reference samples increases to 10 to 50, but the discrimination accuracy of ensemble method 2 and the maximum discrimination (MAX) method increases by 4 to 5%.

[0741] Fig.77 It is a schematic diagram of a customized pain discrimination device equipped with the customized discrimination model preparation method realized by machine learning of the present invention.

[0742] Fig.78 This is a schematic diagram of the head viewed from above, showing a typical electrode configuration. In addition, the boundaries between the front and back of the auricle are also shown.

[0743] Fig.79 The example of the flow of the determination process of the discriminant analysis with the reduction process of the number of feature quantities is shown. It is an example of 1) quantification of the discriminant characteristics of the feature quantities, 2) sorting of the feature quantities, 3) hierarchical discriminant analysis included in order from the highest feature quantity, and 4) determination of the discriminant model. In addition, the numerical values ​​in the figure are all examples.

[0744] Fig.80 This is a diagram showing the discriminant model determination process achieved through a double reduction process when selecting feature quantities and discriminant models. Fig.79 Compared with the discriminant model determination process of , the same process is included until 1) quantification of the discriminant characteristics of the feature quantity, 2) sorting of the feature quantity, and 3) hierarchical discriminant analysis in which the feature quantity is included in the model in order from the upper feature quantity. On the other hand, the economic discriminant model is newly included in the selection of the economic discriminant model by approximating the change in discrimination accuracy using a sigmoid function or the like at the final discriminant model selection time point. In addition, the numerical values ​​in the figure are all examples.

[0745] Fig.81 The high temperature stimulation paradigm used in Example 20 is shown. There are 6 levels of high temperature stimulation, and 3 stimulations are performed at each level. The stimulation presentation time is 15 seconds, and the time interval between level blocks is 100 seconds. The characteristic quantities are the average absolute amplitude and average frequency power (δ, θ, α, β, γ) of 3 stimulations (15 seconds) at each level. The two levels to be judged are mild pain (levels 1 to 3) and severe pain (levels 4 to 6), and the number of samples is strength × 3 levels × 40 people = 240.

[0746] Fig.82 The results in Example 20 are shown. Fig.79 Results of discriminant analysis with feature reduction process (feature ranking and discriminant accuracy).

[0747] Fig.83The high temperature stimulation block paradigm used in the experiment of model making for discriminating 32160 samples in Example 21 is shown. It includes 3 blocks of no pain, weak pain, and strong pain, each stimulation is 15 seconds, and presented twice. The time interval between blocks is 20 seconds. There are 5 groups in total, each group includes 64 blocks. In 1 group, 1 to 3 of the 3 heat blocks are randomly presented at a ratio of 12:10:10.

[0748] Fig.84 The following is a process for extracting the feature quantities used in the discriminant model in Example 21. The process includes 1) EEG data collection, 2) data extraction, 3) selection of two feature quantities, and 4) combination of feature quantities. Through these processes, a total of 49 feature quantities are obtained.

[0749] Fig.85 The difference value (Diff) of the discrimination accuracy due to the use of adjacent models (n-1 feature quantity model and n feature quantity model) is shown (refer to Fig.80 ) and how the discrimination accuracy changes. The lower left graph shows the accuracy change when the test data discrimination accuracy is verified. It shows that regarding the change when the re-ranking is performed twice, the discrimination accuracy basically reaches the limit at the time point when the number of feature quantities is 10 to 15, which is the most economical discrimination model. The following inspiration is given by this result: by re-ranking the number of feature quantities according to the differential value, the main feature quantity that strongly approximates the discriminant function (such as Sigmoid function, etc.) and the auxiliary feature quantity that fine-tunes the model are effectively combined at an early stage to improve the discrimination accuracy. That is, when the feature quantities are repeatedly re-ranked by the differential value (Diff) of adjacent discrimination accuracies, the feature quantities with high contribution tend to be concentrated in the front ranking, and at the earliest stage of the second re-ranking, the discrimination accuracy is roughly limited by at least 12 feature quantities. The standard deviation of the discrimination accuracy for rankings 1 to 12 is "1.4", and for rankings 13 and above it is "0.11", which actually shows an improvement of "about 10 times" (the DCV accuracy figure is obtained by calculating the SD of the first 12 rankings and the SD of rankings 13 and above and comparing them).

[0750] Fig.86 The distribution of the subject sample used in Example 22 by age group and gender is shown.

[0751] Fig.87 The example of the hyperthermia stimulation used in Example 21 is shown. The hyperthermia stimulation has 6 levels, and each level is stimulated three times. The stimulation presentation time is 15 seconds, and the time interval between level blocks is 100 seconds. The brain wave feature quantity and feature quantity correlation used are Fig.88The two levels to be judged are mild pain (levels 1 to 3) and severe pain (levels 4 to 6), and the number of samples is strength × 3 levels × 158 people = 948.

[0752] Fig.88 These are the EEG feature quantities measured in Example 21 and the types of feature quantity correlation.

[0753] Fig.89 is a schematic diagram of the calculation of the multiscale entropy (MSE) measured in Example 21. MSE is the sample entropy (S E ) is a modified version of . Initially, the time series data is segmented and coarsened by the time scale (τ: number of data segments). The negative of the natural logarithm of the ratio (ratio in the m+1 time period / ratio in the m time period) is calculated by calculating the ratio of the distance data from m points in each segmented data to the standard deviation of the incoming data within r% in the m and m+1 time periods. The activities in the two time periods are often similar, so the value of "0" is often taken, but the value becomes higher when the complexity of the activities is high. S E The time scale is divided into several parts, and the sum is calculated and used as the complexity index (C). Generally, m = 2 and r = 0.15 are used. In addition, τ includes 200 data points in order to calculate a reliable MSE, and sometimes includes 600 points at most (Busa & Emmerik, Journal of Sport and Health Science 5 (2016) 44-51).

[0754] Fig.90 The figure shows the changes of the monotonically increasing Sigmoid and monotonically decreasing Sigmoid feature quantities. A represents the absolute average amplitude, and B represents the frequency power. C represents the amplitude correlation (also called potential correlation) of the frontal and parietal lobes. D represents the multi-scale entropy (MSE). Monotonically increasing Sigmoid: At A, the monotonically increasing Sigmoid feature quantity: there is an inflection point at levels 5 to 6, and the feature quantity increases. Monotonically decreasing Sigmoid: At C and D, the monotonically decreasing Sigmoid feature quantity: there is an inflection point at levels 5 to 6, and the feature quantity decreases.

[0755] Fig.91 is included in Fig.90 Another form of "discrete feature quantity" of Sigmoid feature quantity shown in . Before pain monitoring, a plurality of pain test stimuli are applied to the subject, and the distribution characteristics of the feature quantities of painlessness and pain are examined. Fig.91The following is a schematic diagram showing the case of curve approximation using a histogram of feature quantities. The value at which the distribution of painless samples and painful samples intersect is set as a discrimination threshold, and the classification scale is transformed in such a way that samples smaller than the threshold are set to "-1" and samples larger than the threshold are set to "1". As a sample distribution for determining the threshold, other than this, Poincare distribution or the like can also be used.

[0756] Fig.92 The extraction of dominant feature quantities (binary classification) is shown. The feature quantities are sorted in descending order according to the size of the standardized coefficient, and the feature quantities that include zero coefficients in more than 30% of the 100 discrimination estimates are excluded, thereby extracting the priority feature quantities. In this way, 6 feature quantities can be extracted. This feature quantity set is called "alternative integrated EEG features", and feature quantities are extracted from limited (Fp1, Fp2, etc.) electrodes worn during monitoring through various signal processing methods. The feature quantities are adaptively alternated according to the monitoring environment and / or individual through a reduction process, and are comprehensively helpful to the pain discrimination model through, for example, linear models, etc., so it is called this. It was found that the frequency power at the parietal lobe, the frontal lobe-parietal lobe potential correlation, the MSE at the frontal lobe, the frequency power, and the average amplitude are particularly useful as EEG indicators for binary classification of pain.

[0757] Fig.93 The results of discriminant estimation using the dominant feature quantity (linear model) are shown. Fig.91 The six feature quantities (parameters) calculated in the above method can distinguish the two levels of “painful” and “no pain” with an accuracy of “83%”.

[0758] Fig.94 This is the discrimination accuracy obtained by a generalized discrimination model that includes both high-temperature and low-temperature pain stimulation samples. The process of using n-1 samples to create a discrimination model and estimating the pain level of the remaining sample was repeated according to the number of people. The discrimination accuracy of the measured samples reached 74%, which is about 20% higher than the 56% of the random samples. In this embodiment, the characteristic quantity related parameters of the phase synchronization between the electrodes of the frequency are also used for model creation.

[0759] Fig.95An example of verifying the discrimination accuracy of each electrode using a generalized discrimination model that includes both high-temperature and low-temperature pain stimulation samples is shown. The process of using n-1 samples to create a discrimination model and estimating the pain level of the remaining sample (2 levels × 3 trials) was repeated for each person. The discrimination accuracy of each of the 6 electrodes in the frontal lobe and the central part reached "about 70-75%", which shows that even with a limited number of electrodes, at least 1, a general pain discrimination model can be created as long as a combination of at least frontal lobe and central electrodes and parameters is used.

[0760] Fig.96 This is a schematic diagram of a discriminant device with a feature reduction process. At the time of model creation (white arrow), a discriminant function corresponding to the number of classification levels is created and approximated with each feature according to R. 2 The feature quantity is selected by using weighting coefficients such as the value of to create a discrimination model. This information is stored in the feature quantity extraction unit and the discrimination estimation unit, and the pain level is estimated online during actual pain monitoring (black arrow). DETAILED DESCRIPTION

[0761] The present invention is described below. In the full text of this specification, the expression of singular form should be understood as also including the concept of its plural form unless otherwise mentioned. Therefore, the article in the singular form (for example, "a", "an", "the", etc. in English) should be understood as also including the concept of its plural form unless otherwise mentioned. In addition, the terms used in the specification should be understood as being used in the sense commonly used in the field unless otherwise mentioned. Therefore, all professional terms and scientific terms used in this specification have the same meaning as those generally understood by those skilled in the art, unless otherwise defined. In the event of a conflict, this specification (including definitions) takes precedence.

[0762] (definition)

[0763] First, the terms and general techniques used in the present invention are described.

[0764] (About information processing)

[0765] In this specification, "machine learning" refers to a technology that gives computers the ability to learn without being explicitly programmed. It is a process of improving its own performance by acquiring new knowledge and skills in functional units or reconstructing existing knowledge and technology. Programming a computer allows it to learn from experience, thereby reducing the effort required to program a lot of details. In the field of machine learning, methods are being discussed to build computer programs that can automatically achieve improvements based on experience. The role of data analysis and machine learning is a basic technology that is parallel to the field of algorithms and is the basis of intelligent processing. It is usually used in conjunction with other technologies and requires knowledge of the combined field (domain-specific, field-specific knowledge; for example, the medical field). As its scope of application, it has functions such as prediction (collecting data, predicting what will happen), search (finding any obvious features from the collected data), and verification description (investigating the relationship between various elements in the data). Machine learning is based on indicators that represent the degree of goal achievement in the real world. Users of machine learning must master the goals in the real world. Then, when the goal is achieved, the improved indicators need to be formalized. Machine learning is an inverse problem, an ill-posed problem that is not clear whether there is an answer. The behavior of the learned rules is not deterministic but probabilistic. It is necessary to work on the application under the premise that there are some uncontrollable parts. The customization method of the present invention can also be said to be a solution. While observing the performance indicators during training and application, it is also useful for the user of machine learning to make choices on data and / or information in accordance with the goals of the real world.

[0766] As machine learning, linear regression, logistic regression, support vector machine, etc. can be used, and cross validation (also called cross validation, cross confirmation. Cross Validation; CV) can be performed to calculate the discrimination accuracy of each model. After sorting, the feature quantities can be added one by one and machine learning (linear regression, logistic regression, support vector machine, etc.) and cross validation can be performed to calculate the discrimination accuracy of each model. In this way, the model with the highest accuracy can be selected. In the present invention, any type of machine learning can be used. As supervised machine learning, linear, logistic, support vector machine (SVM), etc. can be used.

[0767] Machine learning performs logical reasoning. Logical reasoning can be roughly divided into three categories: deduction, induction, abduction, and analogy. Regarding deduction, when there is the assumption that Socrates is a human being and all people are mortal, it can be concluded that Socrates is also mortal, which can be said to be a special conclusion. Regarding induction, when there is the assumption that Socrates is mortal and Socrates is a human being, it can be concluded that all people are mortal, which derives a general rule. Regarding abduction, when there is the assumption that Socrates is mortal and all people are mortal, it can be concluded that Socrates is a human being, which is equivalent to a hypothesis / explanation. However, it should be noted that even if induction is performed, how to generalize depends on the premise, so it may not be objective. Regarding analogy, it is a probabilistic thinking method as follows: there are objects A and B, and when object A has four characteristics and object B has three of the above characteristics in common, it is inferred that object B also has the remaining one characteristic, and object A and object B are of the same species or have similar close relatives.

[0768] There are three basic principles of impossibility: impossible, very difficult, and unsolved. In addition, there are generalization error, no free lunch theorem, and ugly duckling theorem. Since observation of the true model is impossible, it is necessary to pay attention to the fact that this ill-posed problem cannot be verified.

[0769] In machine learning, a feature or attribute indicates the state of the object to be predicted when observed in a certain aspect. A feature vector or attribute vector is obtained by summarizing the features (attributes) describing the object to be predicted into a vector form.

[0770] In this specification, "model" or "hypothesis" is used with the same meaning, which is to describe the mapping from the input prediction object to the object corresponding to the prediction result, or a candidate set of these mappings, expressed using mathematical functions or logical expressions. In learning using machine learning, the model that is considered to be the closest to the true model is selected from the model set with reference to the training data.

[0771] As models, there are generative models, recognition models, functional models, etc. The model represents the difference in the strategy of the classification model that expresses the mapping relationship between input (prediction object) x and output (prediction result) y. The generative model expresses the conditional distribution of output y when input x is given. The recognition model expresses the simultaneous distribution of input x and output y. The mapping relationship between the recognition model and the generative model is probabilistic. The mapping relationship of the functional model is deterministic, and it expresses a definite functional relationship between input x and output y. Between the recognition model and the generative model, it is sometimes believed that the recognition model has a slightly higher accuracy, but based on the no free lunch theorem, there is basically no difference between the two.

[0772] Model complexity: The degree to which the relationship between the predicted object and the predicted result can be described in more detail and complexity. The more complex the model set, the more training data is generally required.

[0773] When a mapping relationship is represented by a polynomial, a high-order polynomial can express a more complex mapping relationship. It can be said that a high-order polynomial is a model that is more complex than a linear equation.

[0774] When a decision tree is used to represent a mapping relationship, a deep decision tree with a large number of layers can express a more complex mapping relationship. Therefore, it can be said that a decision tree with a large number of layers is a more complex model than a decision tree with a small number of layers.

[0775] They can also be classified according to the strategy used to express the correspondence between input and output. In parametric models, the shape of the distribution and / or function is completely determined by parameters; in non-parametric models, its shape is basically determined by the data, and the parameters are limited to smoothness.

[0776] Parameter: An input used to specify one of the model distribution and function sets. To distinguish it from other inputs, it is also recorded as Pr[y|x;θ], y=f(x;θ), etc.

[0777] In the parametric model, the shape of the Gaussian distribution is determined by the mean / variance parameters regardless of the amount of training data. In the non-parametric model, only the smoothness in the histogram is determined by the number of groupings (bins) parameter, which is more complicated than the parametric model.

[0778] In learning using machine learning, the model that is considered to be the closest to the true model is selected from the model set with reference to the training data. There are various learning methods according to what kind of "approximation" is performed. A representative method is the maximum likelihood estimation method, which is a learning benchmark for selecting the model with the highest probability of generating training data from the probability model set. The model that is closest to the true model can be selected by maximum likelihood estimation. Regarding KL divergence, the larger the likelihood, the smaller the KL divergence relative to the true distribution. There are many types of estimation, which vary depending on the type of form in which the estimated predicted value and / or parameter is obtained. Point estimation only finds the value with the highest certainty, and uses the mode of distribution and / or function such as maximum likelihood estimation and MAP estimation (maximum a posteriori probability estimation), which is most commonly used. On the other hand, for interval estimation, it is commonly used in the field of statistics to find the range of estimated values ​​and the probability that the estimated value exists in this range is 95%. For distribution estimation, it is used in Bayesian estimation, etc. with a generative model set that has introduced a prior distribution, and the prior distribution finds the distribution of the estimated value.

[0779] In machine learning, over-learning (overfitting) may occur. In over-learning, because a model that over-matches the training data is selected, although the empirical error (prediction error relative to the training data) is small, the generalization error (prediction error relative to the data from the true model) becomes large, and the original learning purpose cannot be achieved. Generalization error can be divided into three types: bias (error caused by not including the true model in the candidate model set; the simpler the model set, the larger it is), variance (error caused by selecting different prediction models when the training data is different; the more complex the model set, the larger it is), and noise (the deviation of the true model that is independent of the choice of the model set and cannot be reduced in essence). Bias and variance cannot be reduced at the same time, so the overall error is reduced by balancing bias and variance.

[0780] In this specification, "ensemble (also called ensemble learning, ensemble method, etc.)" is also called group learning. It uses a relatively simple learning model and a learning rule with reasonable computational effort to select various hypotheses through the weights of the assigned examples and / or the differences in initial values, and combine them to form the final hypothesis, in order to perform learning equivalent to learning a complex learning model. In ensemble learning, multiple learners that learn independently are fused to improve generalization ability (prediction ability for unlearned data) to form a learner ( Fig.64). Ensemble learning can be used not only as a machine learning algorithm for regression and / or classification, but also as an auxiliary method for obtaining learning coefficients of other machine learning algorithms. Its application occasions are very wide. In ensemble learning, since the performance is improved by combining many hypotheses in this way, the learning algorithm used in ensemble learning is called a weak learning algorithm or a weak learner, and the hypothesis is called a weak hypothesis, a weak discriminator, a weak determinator, etc.

[0781] In this specification, "pruning" is used in ensemble methods, etc., and refers to pruning the branches of the tree that overfits the model. It can be performed by cross-validation. The procedure is usually only used for decision trees that work independently. On the other hand, in the case of decision tree ensembles, by forming a sufficiently small tree, the protection mechanism against overfitting itself can play a role. Attempts to combine simple learning machines to make complex learning machines have been ongoing in the field of neural networks. To refer to the learning machines or their algorithms produced in this way, words such as combining predictor, combining learner, committee machine, modular network, voting network, and ensemble learning can be used, but in this specification, the term ensemble learning is used.

[0782] When different parameters are given through such learning, the final output is determined by majority vote. Here, when the hypothesis hi is not given a weight wi (usually the sum of the weights is normalized to 1), the weight indicates which hypothesis output is prioritized, and when a majority vote is constructed based on this, it is called a weighted majority vote. In addition, when the weights are the same, it becomes an equal majority vote. When the output is quantitative, the result of weighted addition of the hypothesis output is the final output.

[0783] The learning machine obtained by ensemble learning is also called the final hypothesis or strong discriminator, which includes many weak hypotheses and a combiner that combines the weak hypotheses. Here, it is classified into several learning algorithms according to whether the combiner works dynamically or statically with respect to the input, and whether the weak hypotheses are generated in parallel or sequentially.

[0784] Examples of ensemble methods include boosting and bagging.

[0785] In boosting, each teaching data is weighted, and a weak information extraction unit is generated using all the teaching data. Next, the weight of the teaching data is lowered for the teaching data that correctly extracted information by the generated weak information extraction unit, and the weight of the teaching data that incorrectly extracted information is increased, thereby updating the weight of the teaching data. In addition, a weak information extraction unit is generated again using all the teaching data after the weight update. The following is a method of repeatedly performing the same process to generate multiple weak information extraction units, and a high-precision information extraction device is obtained by combining the outputs of the generated multiple weak information extraction units according to their respective weights. Boosting is a method in which, for example, thousands of samples (example questions) called learning samples, including inspection objects and non-inspection objects with answers (labels) in advance, such as facial images and non-facial images, are used, and different learning machines (weak hypotheses) are generated while changing the weights of the examples in sequence, and they are combined to form a high-precision learning machine (final hypothesis). The word boosting is used according to the meaning of the accuracy of the enhanced learning algorithm.

[0786] In guided bagging, a specified number of teaching data is randomly extracted from all teaching data to generate a teaching data group, and a weak information extraction unit is generated using the teaching data group. By repeatedly performing this process, multiple weak information extraction units are generated, and the outputs of the generated multiple weak information extraction units are combined to obtain a high-precision information extraction device. Hypotheses are generated in parallel. It can also be said that it is a method that does not use the information of all learning data, but uses the information of part of the learning data for learning and finally combines them. Regarding boosting, the use of part of the learning data and finally merging it is common with guided bagging. The difference is that boosting reuses the previously used data and performs boosting as literally recorded. For this reason, it cannot be processed in parallel like guided bagging.

[0787] In order to improve the recognition performance, a bootstrap method can be used. The bootstrap method is a method of selecting weak learners and merging them into the final learner.

[0788] In this specification, "reduction" means reducing or summarizing a variable called a feature quantity. For example, factor analysis means that when there are multiple variables, it is assumed that there is an abstract concept (construct) behind them that affects them and the relationship between multiple variables is explained through a small number of latent variables. It is a form of transformation, i.e., reduction, of a small number of variables. The latent variable that explains the abstract concept is called a factor. Factor analysis is about reducing variables that can be assumed to be common factors behind them and creating new quantitative variables.

[0789] In this specification, the "discriminant function" is a sequence, i.e., a function, which assigns continuous numerical values ​​to the number of levels to be discriminated and is produced corresponding to the arrangement of the samples to be discriminated. For example, when the discrimination level is 2 levels and the samples to be discriminated are arranged accordingly to the levels, the sequence, i.e., the discriminant function, is generated by, for example, obtaining the Sigmoid function type. In addition, when there are more than 3 levels, a step function can be used. The model approximation index is a numerical representation of the correspondence between the discriminant function and the discrimination level of the sample to be discriminated. When the difference between the two is used, the fluctuation range is limited, and the smaller the absolute value of the difference, the higher the approximation. In addition, when performing correlation analysis, the higher the correlation coefficient (r), the higher the approximation. In addition, when performing regression analysis, R 2 The higher the value, the higher the similarity of the judgment.

[0790] In this specification, "weighting coefficient" is used to set in the calculation of the present invention so that important elements are calculated as more important elements, including approximation coefficients. For example, a function can be approximated to data to obtain a coefficient, but it itself is only a descriptive quantity indicating the degree of approximation. When it is sorted by size or selected, the contribution difference within the model is set for a specific feature quantity, so it can be called a weighting coefficient. The weighting coefficient is used with the same meaning as the approximation index of the discriminant function, and can be exemplified by R 2 value, correlation coefficient, regression coefficient, and residual sum of squares (the difference between the discriminant function and the characteristic quantity).

[0791] In this specification, the "discriminant function model" refers to a function model used when discriminating pain, etc. For example, a sigmoid function and a step function can be mentioned, but it is not limited to these functions.

[0792] In this specification, "hyperparameter" refers to a (hyper) parameter that is above machine learning and is not learned by machine learning. In the model used for machine learning, there are more or less parameters that the analyst must set, and the parameters are usually accompanied by machine learning. Each machine learning algorithm has one or more hyperparameters for controlling its operation. Hyperparameters are different from the coefficients (parameters) contained in the model, and their values ​​are not determined by machine learning. Before the machine learning algorithm is executed, they are assigned values ​​by solving optimization problems. As examples of hyperparameters, the number of decision trees generated in random forests, the fitting accuracy of regression analysis, the order of polynomials contained in the model, etc. can be cited. As the value of a hyperparameter, there are both cases where a fixed value is used and cases where a value specified by the user is used.

[0793] (About brain waves)

[0794] In this specification, the so-called "object" is used in the same meaning as a patient or a subject, and refers to any organism or animal that is taken as an object by the techniques disclosed in this disclosure, such as pain measurement and brain wave measurement. As an object, it is preferably a human, but not limited to this. In this specification, when pain estimation is performed, it is sometimes referred to as an "estimation object", but it has the same meaning as an object, etc. There can be multiple "objects". In such a case, each example is sometimes referred to as a "sample" (of the object).

[0795] In this specification, "brain wave" has the same meaning as that commonly used in the technical field, and refers to the electric current generated by placing a pair of electrodes on the scalp due to the potential difference associated with the neural activity of the brain. The brain wave includes an electroencephalogram (EEG) obtained by deriving and recording the time change of the electric current. When quiet, the wave with an amplitude of about 50μV and a frequency of about 10Hz is the main component. It is called an alpha wave. During mental activity, the alpha wave is suppressed and a fast wave with a small amplitude of 17 to 30Hz appears, which is called a beta wave. During light sleep, the alpha wave gradually decreases and the theta wave of 4 to 8Hz appears. During deep sleep, a delta wave of 1 to 4Hz appears. These brain waves can be expressed by specific amplitudes, frequencies, complexity indicators, correlations, etc. In the present invention, they can be expressed by specific amplitudes and frequencies, or analysis of amplitudes.

[0796] In this specification, "brain wave data" refers to any data related to brain waves (also called "brain activity amount", "brain characteristic amount", etc.), including amplitude data (EEG amplitude), frequency characteristics, etc. The "analysis data" obtained by analyzing these brain wave data can be used in the same way as the brain wave data, so in this specification, it is sometimes collectively referred to as "brain wave data or its analysis data". As analysis data, for example, the average amplitude and / or peak amplitude of brain wave data (for example, Fz, Cz, C3, C4), frequency power (for example, Fz(δ), Fz(θ), Fz(α), Fz(β), Fz(γ), Cz(δ), Cz(θ), Cz(α), Cz(β), Cz(γ), C3(δ), C3(θ), C3(α), C3(β), C3(γ), C4(δ), C4(θ), C4(α), C4(β), C4(γ), etc.) can be cited. Of course, other data that are usually used as brain wave data or its analysis data are not excluded. For example, if data obtained by cutting out raw data at a certain time can be used for discrimination, it is also a feature amount and can therefore also be used in the present invention.

[0797] In this specification, "brain wave characteristic quantity" or "characteristic quantity of brain waves" refers to any characteristic quantity of brain waves, and may include "brain wave data or its analysis data", for example, it may include amplitude, correlation between brain wave characteristic quantities, frequency power, and complexity index. As examples of these, the amplitude may include an average amplitude (e.g., an absolute average amplitude, a relative average amplitude, etc.) and / or amplitude distribution characteristic values ​​such as the median amplitude, the mode amplitude, the maximum amplitude, the peak amplitude, and the quartile amplitude. The correlations between the brain wave characteristic quantities may include potential correlation (e.g., frontal-parietal potential correlation (correlation coefficient, partial correlation coefficient, connectivity, causality, and their subspecies)) or phase synchronization between electrodes (e.g., coherence, phase-locking value, and their subspecies). The frequency power may include spectral density, power spectrum, and / or their subspecies. The complexity index may include at least one selected from entropy (e.g., multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy, and conditional entropy, etc.), and biological potential characteristic quantities (e.g., eye movement potentials reflecting eye movements such as blink reflex, etc.) generated in an event-related manner in conjunction with pain generation.

[0798] In this specification, "amplitude data" is a type of "brain wave data", which refers to data on the amplitude of brain waves. Sometimes it is simply referred to as "amplitude", or "EEG amplitude". Such amplitude data is an indicator of brain activity, and therefore is sometimes referred to as "brain activity data", "brain activity amount", etc. Amplitude data can be obtained by measuring the electrical signals of brain waves, and can be represented by a potential (which can be represented by μV, etc.). As amplitude data, the average amplitude can be used, but is not limited to this.

[0799] In this specification, the so-called "frequency power" is to express the frequency component of a waveform as energy, also known as the power spectrum. Regarding frequency power, the frequency component of a signal buried in a noise-containing signal in the time domain can be extracted and calculated using a high-speed Fourier transform (FFT) (an algorithm for high-speed calculation of discrete Fourier transform (DFT) on a computer). Regarding the FFT of a signal, its output can be standardized using, for example, a periodogram of a function in MATLAB to calculate the power spectral density PSD, or the power spectrum that is the source of power measurement. PSD represents how the power of a time signal is distributed with respect to frequency, in units of watts per hertz (w / Hz). The power spectrum is calculated by integrating each point of the PSD over the frequency range that defines the point (i.e., over the resolution bandwidth of the PSD). The unit of the power spectrum is watts. Regarding the value of power, it can be read directly from the power spectrum instead of integrating over the frequency range. Both PSD and the power spectrum are real numbers, and therefore do not contain any phase information. In this way, the calculation of frequency power can be calculated using standard functions of MATLAB. As an analytical method, such as Fig.25 Time-frequency analysis can be used as illustrated in . For example, a time unit such as 1 second can be determined, and while moving the point, the time variation of the frequency power is obtained. Thus, as Fig.25 As shown in FIG. 1 , the delayed characteristic of heat pain is obvious, for example, a specific time interval that is sensitive to pain (for example, 5 seconds after the stimulus is presented), and a feature quantity that can be distinguished from the "delayed pain feature quantity" in terms of category can also be created in the frequency feature quantity, and the actual situation is shown in the embodiment (refer to Figure 25 to Figure 32 ).

[0800] The "complexity" used in the present invention refers to a situation where there are an excess of logically possible combinations of elements, and selection has to be made from among them. When used in brain waves, it also refers to a state where there are an excess of possible combinations of each brain wave, and its index is called a "complexity index."

[0801] In this specification, "complexity index" refers to a scale of seemingly irregular behavior that is complex due to the large number of constituent elements and / or nonlinearity, and can be represented by entropy, etc. Entropy refers to a scale of disorder in a system, and refers to the average amount of information transmitted by informing that an event has occurred from a finite complete event system. In addition, in informatics, entropy also refers to the amount of information and is an indicator of the complexity of an activity. Therefore, it can be said that complexity has a broader concept than entropy in terms of disorder. As entropy, for example, multi-scale entropy (MSE), sample entropy, self-entropy, average entropy, joint entropy, relative entropy and conditional entropy can be cited, but are not limited to these.

[0802] Multiscale entropy (MSE) overcomes the problems of existing nonlinear analytical methods (data stability, i.e., the state in which the overall characteristics of the data (dispersion, etc.) cannot be reproduced locally), and is a new nonlinear analytical method that has attracted attention. It is easily affected by artifacts and requires high-resolution and long-term data. However, in order to overcome the difficulty of performing measurements that meet these conditions using clinically measured data, approximate entropy (ApEn; Pincus SM. Approximate entropy as a measure of system complexity. Proc Natl Acad Sci USA 1991; 88: 2297-2301.) and its improved version, sample entropy (SampEn; Richman JS. Moorman JR. Physiological time-series analysis using approximate entropy and sample entropy. Am J Physiol Heart Circ Physiol. Med. Am J., 2003; 105: 217-219.) were developed as practical nonlinear analytical methods. 2000; 278: 2039-2049), was developed by Costa et al. (Costa M. Goldberger AL. Peng CK. Multiscale entropy analysis of complex physiologic timeseries. Phys Rev Lett 2002 89: 068102.). Regarding MSE, it is obtained by reconstructing the original data by adding and averaging in a non-repetitive manner, and calculating the approximate entropy / sample entropy of the data reconstructed by multiple addition numbers (time axis). The complexity of high frequency bands is represented by a low number of additions, and the complexity of low frequency bands is represented by a high number of additions. Therefore, the MSE analysis can perform nonlinear extraction that is difficult to achieve through existing nonlinear analysis methods. Regarding brain waves, there is a recent report from Busa & Emmerik (Journal of Sport and Health Science Volume 5, Issue 1, March 2016, Pages 44-51).

[0803] In this specification, "correlation between brain wave feature quantities" refers to the correlation between two or more brain wave feature quantities. The brain wave feature quantity used can be any feature quantity as long as it is brain wave data or its analysis data, and can include amplitude (including average amplitude, etc.), frequency power, potential, complexity index (including MSE, etc.), etc., but is not limited to these.

[0804] In this specification, "correlation" refers to any relationship between two characteristic quantities. It is a broad concept including the relationship between different characteristic quantities of the same electrode, the relationship between the same characteristic quantity at different times, etc., and includes correlation. There are phase synchronization indicators such as taking out the phases of multiple activities and investigating the synchronization between cycles, which are also included in the broad concept of correlation. In this way, although the correlation coefficient is not used, the same kind of relational indicators are also included, that is, correlation.

[0805] In this specification, "correlation" is a concept or value generally representing the correlation between two or more variables in mathematical statistics and / or biostatistics. As an example of correlation, the scale in statistics can be represented by a correlation coefficient (correlation coefficient, r, ρ, etc.). The value is between -1 and +1, and the closer to +1 means the more positive correlation (positive correlation), the closer to -1 means the more negative correlation (negative correlation), and the closer to 0 means the more irrelevant (non-correlation). For example, there is a certain r value between the height and weight of a person, and the positive correlation is established. When the positive and negative correlations are established, the functional relationship (experimental formula) between the variables can be obtained by the form of a regression line (regression line), and further, the method can also be expanded to the case of nonlinear regression. In addition, for each variable, in the case of only asking the qualitative correlation of size (+1 or -1), it is sometimes called association in English.

[0806] In one embodiment, in addition to correlation, mutual relationships also include indices that cannot be called correlations, such as phase synchronization indices for investigating synchronization between cycles, and as correlations in a narrow sense, various forms of correlations (synchronicity, irrelevance, delay, positive and negative, similarity, and consistency), in other words, time correlation, space correlation, spatiotemporal synchronization, spatial relationship or combination, irrelevance or non-correlation, delay or collapse of time correlation, positive and negative or related properties, similarity or correlation coefficient level, and consistency or complete correlation, etc. In this way, synchronization can be understood as time correlation, combination can be understood as spatial (e.g., brain part) relationship, irrelevance can be understood as non-correlation, delay can be understood as collapse of time correlation, positive and negative can be understood as related properties, similarity can be understood as high correlation coefficient, and consistency can be understood as complete correlation.

[0807] In this specification, "synchronicity" refers to the degree of synchronization. Here, synchronization refers to two periodically changing quantities having the same or different vibration numbers (frequencies), and the phase difference between them is constant or varies around a constant average value. Regarding synchronization, phase synchronization (phase locking value) or coherence can be cited. The coherence feature quantity involves multiple aspects, and the frequency feature quantities corresponding to the power spectrum, bispectrum, high-order spectrum, self-power spectrum, self-bispectrum, and self-high-order spectrum are the mutual power spectrum, mutual bispectrum, mutual high-order spectrum, self / mutual bicoherence, self / mutual bispectral density, self / mutual real-time triple product power spectrum (Japanese original: オート / クロスリアルトリプルプロダクト), self / mutual biphase code power spectrum (Japanese original: オート / クロスバイフェーズ). In the past, regarding the correlation of these characteristic quantities, if the consciousness level decreases, the numerical value decreases, which is understood to be an effective indicator of the physiological and psychological state of the organism. However, contrary to expectations, in the present invention, the numerical value related to the characteristic quantity of the potential shows a decrease in numerical value even in a situation such as when the pain level becomes very strong and the consciousness level rises instead of falling. Therefore, not only the increase pattern of the characteristic quantity correlation mentioned above, but also the decrease pattern is applied to the discrimination model in an opposing manner as "pain level inversely correlated characteristic quantity parameter" and "pain level positively correlated characteristic quantity parameter". Thus, with respect to both, for example, in the linear multiple regression model used in the present embodiment, if the positive correlation parameter acts as an "acceleration", the negative correlation parameter acts as a "brake" to adjust it, thereby making it possible to create a more sophisticated discrimination model.

[0808] In this specification, similarity refers to the sum of similarity (similarity, resemblance, affinity), dissimilarity (dissimilarity) or distance (distance) between objects obtained from feature data. Usually, the range of similarity and dissimilarity is a real number "above 0 and below 1" when the distance is scaled, but the range of distance can be a non-negative real number (from above 0 to infinity). In addition, regarding distance, for example, when considering the distance between two extreme stars in the universe, it is close to infinity, but if the distance is scaled and set as "1" as a similarity index, it is not infinite. Distance is ultimately the difference between data, but after it is scaled and transformed into 0 to 1, it can be used as a similarity index. Distances that can be used in this specification include Euclidean distance, city block distance, Minkowski distance, Chebyshev distance, Mahalanobis distance, and the value obtained by subtracting the correlation coefficient from 1, etc., all of which can be used.

[0809] In this specification, "potential correlation" refers to the correlation between two or more potentials that are electroencephalogram feature quantities. Since the potential is expressed in terms of amplitude, it is also referred to as "amplitude correlation."

[0810] For example, in the case of an event-related potential as a kind of brain wave, a negative potential called N100 reflecting automatic attention arousal is observed around 100 milliseconds after the sensory stimulus is presented. This component is easily affected by the internal state of the organism such as habit and drowsiness, so, for example, the amplitude will decrease when consciousness decreases. Moreover, this consciousness decrease will also have an impact on the higher-order cognition after N100, and it can be predicted that the reduction of positive potentials such as P300 reflecting the evaluation and / or judgment of consciousness can be predicted. Therefore, assuming that a subject is subjected to 100 sensory stimuli and consciousness decreases as the examination proceeds, it can be predicted that the negative potential of N100 will decay and the positive potential of P300 will decrease. Therefore, if the relationship between the potential of 100 N100 (the average potential or peak potential of a certain time period) and the potential of P300 is analyzed by Pearson's correlation analysis method, it can be predicted that the more positive the N100 potential is, the more negative the P300 potential is, and this negative potential correlation will be observed. Here, the relationship between such brain potential activities is defined as potential correlation.

[0811] In this specification, "electroencephalogram feature quantities" and "correlations between electroencephalogram feature quantities" are collectively referred to as "parameters" related to electroencephalograms.

[0812] In this specification, "alternating comprehensive brain wave feature quantity" refers to the feature quantity extracted from limited electrodes (the above-mentioned front Fp1, Fp2, etc.) through a variety of signal processing methods, which is adaptively alternating according to the monitoring environment and / or individual differences through a reduction process, and comprehensively acts on the pain discrimination model, for example through the linear regression model in this embodiment.

[0813] In this specification, "pain" and "pain" have the same meaning, and both refer to the sensation produced when there is a generally strong invasion such as injury or inflammation on a part of the body as a stimulus. Pain is not a disease but a symptom, and its form is determined by the combination of three main characteristics: central pain, nociceptive pain, and neuropathic pain. In addition, acute pain is different from chronic pain, and the two have differences in the associated brain part network (combination). In the case of chronic pain, there are also cases where it is not actually painful but is subjectively reported as painful, and there are also psychogenic causes that cannot be explained by the intensity of the sensation of pain stimulation.

[0814] For humans, pain is a feeling accompanied by a strong sense of discomfort and is also included in general sensations. Moreover, skin pain and the like also have the characteristics of external receptivity to a certain extent, and cooperate with other skin sensations and tastes to make qualitative judgments such as hardness, sharpness, heat (hot pain), coldness (cold pain), and spiciness of external objects. In addition to the skin and mucous membranes, human pain may also occur in almost all parts of the body (for example, pleura, peritoneum, viscera (visceral pain, excluding the brain), teeth, eyes, and ears, etc.), all of which can be perceived in the brain as brain waves or their changes. In addition, internal pain represented by visceral pain is also included in pain. The above-mentioned pain is called somatic pain relative to visceral pain. In addition to somatic pain and visceral pain, there are also reports of pain such as "referred pain", which is a phenomenon of surface pain in a part different from the actual injured part. The present invention provides a method for expressing the temporal changes of these various types of pain in a trend and monitoring the subjective pain level, as well as other methods recorded in this specification.

[0815] Regarding pain, there are individual differences in sensitivity (pain threshold), and there are qualitative differences depending on the way the stimulus occurs and / or the difference in the receptor site. There are classifications such as dull pain and sharp pain, but in the present disclosure, any type of pain can be measured, estimated and classified. In addition, it is also possible to deal with acute pain (A pain), chronic pain (B pain), (acute) local pain and (chronic) diffuse pain. The present invention can deal with allodynia such as general abnormal hypersensitivity. Among the peripheral nerves that transmit pain, two nerve fibers, "Aδ fibers" and "C fibers", are known. For example, when clapping your hands, the initial pain is transmitted through the conduction of Aδ fibers, transmitting sharp pain with clear localization (primary pain; sharp pain). Afterwards, through the conduction of C fibers, a tingling pain with unclear localization is felt (secondary pain; dull pain). Pain is classified into "acute pain" that lasts within 4-6 weeks and "chronic pain" that lasts for more than 4-6 weeks. Although pain is an important vital sign along with heartbeat and / or body temperature, blood pressure, and respiration, it is difficult to characterize it as objective data. Representative pain rating methods such as the visual analogue scale (VAS) and the faces pain rating scale are subjective evaluation methods and cannot compare pain between patients. On the other hand, the inventors of the present invention have deduced that as an indicator for objectively evaluating pain, if the changes in the amplitude / latency of brain waves that are not easily affected by the peripheral circulatory system are observed and trend analysis is performed, it is possible to distinguish and classify pain. Both instantaneous stimulation and continuous stimulation can be detected. In particular, instantaneous pain and continuous pain of stabbing pain can also be distinguished by the trend analysis of the present invention. Instantaneous pain is pain in a short time interval, so if the time averaging method of at least tens of seconds in the trend analysis is used, the associated brain activity may be attenuated (for example, no significant correlation with pain evaluation is seen). On the other hand, in the case of continuous pain, because it is continuous, the significant correlation with pain evaluation may become stronger by the time averaging method. The inventors of the present invention have also deduced that as an indicator for objectively evaluating pain, if we focus on brain waves that are less susceptible to the peripheral circulatory system and observe the changes in their amplitude / latency in response to pain stimulation, we can also classify the type of pain (comfortable or not). This classification can be performed for both instantaneous stimulation and continuous stimulation.

[0816] In the present invention, it is more important to distinguish whether the pain is "needing treatment" than the intensity itself. Therefore, it is also important to be able to clearly classify "pain" based on the concept of "treatment". For example, it can also be said to be a "qualitative" classification of pain related to "comfortable or not" and "unbearable". For example, the positioning of the "pain index", the baseline and the relationship between them can also be defined. In addition to being able to assume the case of n=2, it is also possible to assume the case of n=3 or more. In addition, in the case of more than three, it can be divided into "no pain", "painful but comfortable" and "painful". For example, it is possible to distinguish between "unbearable, needing treatment" pain, "in between", and "painful but not minding". When using the trend analysis of the present invention, by specifying the threshold of the duration of the signal associated with strong pain, it is possible to identify whether it is "unbearable" pain or "painful but tolerable" pain.

[0817] In this specification, "subjective pain level" refers to the level of pain felt by the subject, which can be expressed using conventional techniques such as the computerized visual analog scale (COVAS) or other well-known techniques, such as the Support Team Assessment Schedule (STAS-J), Numerical Rating Scale (NRS), Faces Pain Scale (FPS), Abbey pain scale (Abbey), Check in list of Nonverbal Pain Indicatiors (CNPI), Non-communicative Patient's Pain Assessment Instrument (NOPPAIN), Doloplus2, etc.

[0818] In this specification, "trend analysis" refers to a method that places more emphasis on "time series changes" than on the distribution of data and analyzes the changing tendency of data. An attempt is made to select by comparing the characteristic changes of parameters such as the slope of the change, the proportion of the change in the slope, etc. with the assumed changes, and to find out a certain cause and / or tendency. In view of the present disclosure, the changing tendency of subjective pain is analyzed or estimated by trend analysis of electrical signals. In the present invention, the "time changes" of subjective pain can be analyzed in detail and accurately by trend analysis of electrical signals.

[0819] In this specification, "average potential under time width" refers to the value obtained by averaging the potential of brain waves under a certain time width (for example, 30 seconds, etc., but not limited to this). Regarding the average, the arithmetic mean is usually used, but an average other than the arithmetic mean, such as the geometric mean, etc., can also be used. In addition, other central tendencies (median, mode) can also be used. In addition, a single or multiple peak amplitudes in a specific time period can also be used. In the present invention, it can be understood that any averaging method can be used (for example, refer to Fig. 9 ).

[0820] The time width that can be used in this specification includes, for example, 10 seconds or more, 15 seconds or more, 20 seconds or more, 25 seconds or more, 30 seconds or more, 35 seconds or more, 40 seconds or more, 45 seconds or more, 50 seconds or more, 55 seconds or more, 60 seconds or more, etc., but is not limited to these. Even if the actual individual pattern of pain is unclear when measured individually, it is preferable to use the average value obtained by using various time periods when estimating it.

[0821] In a preferred embodiment, the time width that can be used can be more than 30 seconds, or more than 40 seconds. As the upper limit, 20 seconds, 30 seconds, 40 seconds, 50 seconds, 60 seconds, 70 seconds, 80 seconds, 90 seconds, 100 seconds, 110 seconds, 120 seconds, etc. can be cited, but it is not limited to these. Although I do not want to be bound by theory, significant effects have been seen using these values. However, especially in the case where the duration of pain exceeds tens of minutes, a longer time width for calculating the average value of more than 60 seconds (for example, more than 70 seconds, more than 80 seconds, more than 90 seconds, more than 100 seconds, more than 110 seconds, more than 120 seconds, more than 150 seconds, more than 180 seconds, etc.) is effective (refer to Figure 8 , 9 ).

[0822] In one example, it can be distinguished into 60 seconds and 120 seconds, but it is not limited to this. In the short case, the time width is 10 to 120 seconds, and in the long case, the time width is 30 to 300 seconds. As a longer time width, it can include a time width of 120 seconds or more as needed.

[0823] In addition, in the clinical field, regarding the pain monitoring time, a variety of conditions from short-term to long-term can be envisioned, and therefore, it is preferred to use a "graded pain trend monitoring method." In the initial stage of pain monitoring, a "short-time reference unit" of, for example, 10 seconds is used to monitor pain at part of the time. As time passes, the time units are reduced in the direction of the time axis to visualize the trend (change) of the brain characteristic quantity in the middle period where the correlation with pain perception increases. Furthermore, in the case of long-term monitoring, the pain trend is visualized by reducing it to the level of several minutes. Using this method, local changes and global changes in pain are monitored in parallel and graded, which enables complex monitoring of pain.

[0824] Here, as the time width of the "short time reference unit", for example, in addition to 10 seconds, 5 seconds, 15 seconds, 20 seconds, 25 seconds, 30 seconds, 35 seconds, 40 seconds, 45 seconds, 50 seconds, 55 seconds, 60 seconds, etc. can be cited. They can also be reduced to 2 times, 3 times, 4 times, 5 times, 6 times, 7 times, 8 times, 9 times, etc., and the time width after the aggregation is set (for example, in the case of 10 seconds, it can be reduced to 20 seconds, 30 seconds, 40 seconds, 50 seconds, 60 seconds, 70 seconds, 80 seconds, 90 seconds, etc.). The reduced time width can be further reduced to a level of several minutes (for example, 2 minutes, 2.5 minutes, 3 minutes, 3.5 minutes, 4 minutes, 4.5 minutes, 5 minutes, 6 minutes, 7 minutes, 8 minutes, 9 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, 30 minutes, etc.) for observing global changes. For example, the time width may be reduced to 4 or more levels in addition to 2 or 3 levels. In this case, each reduction may be reduced to 2, 3, 4, 5, 6, 7, 8, 9, 10, etc. or multiples thereof.

[0825] In this specification, the term "non-overlapping block average method" means calculating the average value for each time block, which means using only blocks having values ​​for calculating the average value (eg, arithmetic / geometric mean potential). In contrast, there is an "overlapping moving average method".

[0826] In this specification, "stimulus" refers to causing a certain reaction in an object. When the object is a living organism, it refers to the main cause that causes a temporary change in the physiological activity of the organism and / or a part thereof.

[0827] The situations of events related to pain as exemplified as specific examples of "stimulation" include any stimulation that can produce pain. For example, electrical stimulation, cold stimulation, heat stimulation, physical stimulation, chemical stimulation, etc. In the present invention, stimulation can be any stimulation. Regarding the evaluation of stimulation, for example, conventional techniques such as computerized visual analog scale (COVAS) or other well-known techniques, such as Support Team Assessment Schedule (STAS-J), Numerical Rating Scale (NRS), Faces Pain Scale (Faces Pain Scale (FPS)), Abbey pain scale (Abbey), Check inlist of Nonverbal Pain Indicatiors (CNPI), Non-communicative Patient's Pain Assessment Instrument (NOPPAIN), Alzheimer's disease pain assessment scale (Doloplus2) J, etc. can be used to make it correspond to the subjective pain sensation level. Examples of values ​​that can be used as stimulus intensity include the nociceptive threshold (the threshold at which nociceptive fibers generate nerve impulses), the pain detection threshold (the intensity of an injurious stimulus that a person can perceive as pain), and the pain tolerance threshold (the strongest stimulus intensity among injurious stimuli that a person can tolerate experimentally).

[0828] In this specification, the "classification" of pain can be performed from various viewpoints. As a representative example, it includes classification according to the estimated object "pain" or "no pain", but in addition to this, it is also possible to conceive of a method of classifying the pain felt into strong pain, weak pain, "tolerable" pain, and "unbearable" pain. In addition, a method of classification into "painful but uncomfortable" and "painful but comfortable" can also be cited. Using the present disclosure, by observing monotonic increase and / or monotonic decrease, it is possible to determine / estimate in time series whether the object is experiencing unbearable strong pain or weak pain.

[0829] In this specification, "pain index" refers to an index obtained by appropriately processing brain wave data or its analysis data. As long as the description of the assigned index is recorded, any processing method can be used, and a method that can visualize and track the characteristics (trend) of pain persistence or change over time is very important, and this method is shown in this disclosure. The pain index can also be called a "pain level index" or the like. In addition, the term "pain index" refers to subjective evaluation, stimulation intensity, and related brain characteristic quantities, etc.

[0830] In this specification, "baseline" is a benchmark or reference for facilitating the interpretation of pain levels, and means, for example, a characteristic value related to the intensity of pain levels at the start of monitoring, or a value obtained by averaging and / or standardizing it, or a method of using a pain index as zero, and a calculated value.

[0831] In this specification, "head-mounted device" refers to a device used to obtain brain waves from the head. The shape is arbitrary, and any acquisition method can be used as long as the brain waves can be obtained directly or indirectly. As a shape, it is preferably a shape that can be worn on the head, but it is not limited to this. As an example, in addition to wireless shapes such as boxing headgear, existing shapes such as hat shapes, net shapes, and belt shapes can also be cited. If further improvements are made, if brain waves can be directly obtained from the head through electrodes in the form of hairpins, any shape can be used, and it can also be obtained from the outside in a non-contact manner. The above can be collectively referred to as head-mounted devices.

[0832] In this specification, the "base unit" refers to a part that obtains information such as brain wave signals from a head-mounted device and performs operations such as analysis, discrimination, communication, and display. It may mainly include a processor, which is configured to perform the following steps: extracting quantitative features such as brain wave feature quantities from the brain electrical activity data (brain wave data or its analysis data, etc.) performed by the subject, and then generating and applying a discrimination model, discriminating pain, etc. It may also include an input device for inputting a memory device that is operatively coupled to the processor.

[0833] (Preferred embodiment)

[0834] The preferred embodiments of the present invention are described below. The embodiments provided below are provided for a better understanding of the present invention, and it should be understood that the scope of the present invention should not be limited to the following description. Therefore, those skilled in the art understand that they can be appropriately changed within the scope of the present invention with reference to the description in this specification. In addition, it should be understood that the following embodiments of the present invention can be used independently or in combination.

[0835] In addition, the embodiments described below are all inclusive or specific examples. The numerical values, shapes, materials, components, configuration positions and connection methods of components, steps, and the order of steps shown in the following embodiments are only examples and are not intended to limit the scope of the claims. In addition, among the components in the following embodiments, the components that are not recorded in the independent claims showing the highest concept are described as arbitrary components.

[0836] Therefore, the inventors of the present invention have evaluated various pains using various methods to clarify the relationship between pain and brain waves. The relationship between pain and brain waves clarified by the inventors of the present invention will be described below with reference to the accompanying drawings.

[0837] First, the relationship between pain caused by electrical stimulation and brain waves will be described. The data shown below is data of a representative subject among a plurality of subjects.

[0838] Figure 1A is a graph showing the relationship between electrical stimulation and pain level (VAS). Figure 1B is a graph showing the relationship between electrical stimulation and pain level (paired comparison). Figure 1C is a graph showing the relationship between electrical stimulation and brain wave amplitude. Figure 1D This is a graph showing an example of an electroencephalogram waveform.

[0839] Figure 1A , Figure 1B and Figure 1C The horizontal axis represents the current value of electrical stimulation. Figure 1A The vertical axis represents the pain level reported by the subjects through VAS. Figure 1B The vertical axis represents the pain level reported by the subjects through pairwise comparison. Figure 1C The vertical axis represents the amplitude of the brain wave. Figure 4 In the figure, the horizontal axis represents time and the vertical axis represents signal level.

[0840] Paired comparison is a method of using two electrical stimulations of different sizes as a group, and for each of the multiple electrical stimulation groups, the subject reports which electrical stimulation caused what degree of pain through a numerical value. In this case, the pain level is reported by comparing the two pains, thus reducing the influence of the subject's past experience on the pain level.

[0841] like Figure 1A and Figure 1B As shown, regardless of the VAS or paired comparison method, the relationship between the current value of the electrical stimulation (i.e., the intensity of the stimulation) and the pain level is generally represented by a Sigmoid curve. In addition, the shape of the Sigmoid curve (e.g., upper and lower limits, etc.) varies from subject to subject.

[0842] In addition, if Figure 1C As shown in FIG. 1 , the relationship between the current value of the electrical stimulation and the amplitude value of the brain wave can also be roughly represented by a Sigmoid curve. Here, the amplitude value of the brain wave uses the difference between the maximum peak value and the minimum peak value (i.e., peak-to-peak value). For example, Figure 1D In the example, the largest difference value (N1-P1) among the three difference values ​​(N1-P1, N2-P2, N1-P2) is used as the amplitude value.

[0843] In this way, the relationship between the intensity of electrical stimulation and the level of pain, and the relationship between the intensity of electrical stimulation and the amplitude of brain waves can be expressed by a Sigmoid curve. That is, the pain level and the amplitude of brain waves both have upper and lower limits relative to electrical stimulation, and show the same changes relative to the intensity of electrical stimulation. Therefore, after analyzing the relationship between the amplitude of brain waves and the level of pain, it is found that the relationship between the amplitude of brain waves and the level of pain is expressed as follows: Figure 1E and Figure 1F As shown.

[0844] Figure 1E This is a graph showing the relationship between the pain level (VAS) caused by electrical stimulation and the brain wave amplitude. Figure 1F This is a graph showing the relationship between the pain level (paired comparison) caused by electrical stimulation and the amplitude of brain waves. Figure 1E and Figure 1F In the figure, the horizontal axis represents the amplitude of brain waves and the vertical axis represents the pain level.

[0845] like Figure 1E and Figure 1F As shown in FIG. 1 , the pain level caused by electrical stimulation and the amplitude of the brain wave are linear in both VAS and paired comparison. That is, the amplitude of the brain wave is proportional to the pain level.

[0846] In addition, in the present disclosure, the term "linearity" includes not only strict linearity but also substantial linearity. That is, linearity includes a relationship that can be approximated to linearity within a specified error range. The specified error range is determined, for example, by the coefficient of determination R in regression analysis. 2 The coefficient of determination R 2 It is the value obtained by subtracting the sum of squares of the residuals divided by the sum of squares of the differences between the observed values ​​and the mean from 1. The specified error range is, for example, R 2 The range is 0.5 or more.

[0847] Regarding the relationship between pain caused by thermal stimulation and brain waves, similar to the case of electrical stimulation, there is a linear relationship between the pain level and the brain wave amplitude.

[0848] Figure 1G This is a graph showing the relationship between the pain level (VAS) caused by thermal stimulation and the brain wave amplitude. Figure 1H This is a graph showing the relationship between the pain level (paired comparison) caused by thermal stimulation and the brain wave amplitude. Figure 1G and Figure 1H In the figure, the horizontal axis represents the amplitude of brain waves and the vertical axis represents the pain level.

[0849] like Figure 1G and Figure 1H As shown, whether in the case of VAS or paired comparison, the pain level caused by thermal stimulation has a linear relationship with the amplitude of the brain wave. In addition, the upper and lower limits of the amplitude of the brain wave vary from subject to subject, but the inventor's experiments show that the upper limit of the amplitude does not exceed about 60μV.

[0850] As described above, the inventors analyzed the relationship between the pain levels obtained by evaluating various pains using multiple methods and the amplitude values ​​of brain waves, and found that the amplitude of brain waves has a specific relationship with pain. In addition, the inventors found that based on the specific relationship between the amplitude of brain waves and pain, a pain classification value that can estimate the magnitude of pain can be calculated.

[0851] (Trend Analysis Method)

[0852] In one embodiment, the present invention provides a method for monitoring pain of an estimated subject based on the brain waves of the estimated subject. The method includes: a) measuring or obtaining brain wave data or analysis data thereof from the estimated subject by performing brain wave measurement in response to stimulation; b) obtaining a time transition of an average value of the brain wave data or analysis data thereof in a specific time width; and c) monitoring the level of pain of the estimated subject according to the brain wave data or analysis data thereof based on the time transition of the average value.

[0853] In more detail, the method includes step a) of measuring or obtaining brain wave data or analysis data thereof from the estimation object by performing brain wave measurement for stimulation, step b) of obtaining the time transition of the average value (e.g., arithmetic / geometric mean potential) of the brain wave data or analysis data thereof in a specific time width, and the step includes: b-1) a step of extracting a feature quantity for the average value and b-2) a step of plotting the time transition of the average value (e.g., arithmetic / geometric mean potential) of the feature quantity in the specific time width, and step c) of evaluating or monitoring the pain level of the estimation object according to the brain wave data or analysis data thereof based on the time transition of the average value. The above steps a), b) and c) may use a pain index (e.g., a feature quantity degree indicating the intensity of the pain level, etc.) for evaluating or monitoring the pain level of the estimation object in a time series according to the brain wave data, or set a baseline for pain monitoring, as required. The "pain index" may be calculated by separately analyzing the feature quantity of the brain wave data or analysis data thereof in a reasonable manner.

[0854] The pain index can be expressed in a variety of ways. For example, the obtained characteristic quantity itself and / or the average value can be used directly to express the pain level. When it is determined that "high average value = strong pain level", the average value itself can become the pain index. In addition, the average value can be converted into a numerical range of "0 to 100" and presented in a digital form. In addition, the characteristic quantity itself and / or its average value can be standardized and normalized and used as a pain index.

[0855] Regarding the "baseline" used in this specification, the above-mentioned pain index is used to provide a judgment standard for the transition of pain levels. For example, according to the pain level determined by the pain judgment / estimation device, the pain index of the pain intensity level is used to determine the zero line, that is, the baseline. As a result, it is easier to understand the transition of pain levels. In addition, when the pain level is unknown, a certain interval in which no external stimulation is arbitrarily applied, a quiet state, or a "resting (rapid) state" close to this state can also be used as the baseline. The acquisition of the baseline (sometimes also referred to as baseline correction) can be performed, for example, as follows (this calculation method is also referred to as a baseline determination process).

[0856] The baseline correction of the pain index (in the case of the electroencephalogram characteristic quantity (EEG)) is expressed by the following [Formula 1] and [Formula 2].

[0857] [Mathematical formula 1]

[0858] EEG-BS

[0859] BS = average of the strongest pain EEG (normalized), etc.

[0860] [Mathematical formula 2]

[0861] (EEG-BS) / SD(BS)

[0862] SD(BS) = Standard deviation of BS

[0863] By performing trend analysis, subjective pain transitions can be evaluated more accurately. In addition, as a further feature of the present invention, by obtaining the average amplitude of each block of tens of seconds instead of obtaining the change of small data points such as every 1 second, it is possible to eliminate the external noise randomly mixed in and the internal noise from the biological body, and the time change pattern of the pain level, that is, the trend, can be evaluated more accurately (improving the correlation between the subjective pain evaluation and the brain feature quantity).

[0864] In the present invention, the method of the present invention further includes referring to a pain ratio, where the pain ratio indicates the certainty of discriminating a pain estimate value as pain. With respect to such a pain ratio, (a) a discrimination model is used to calculate a discrimination estimate value between pain and no pain, (b) the pain estimate value is divided into a classification range, and the ratio of the pain estimate value in each estimation value of the classification range is calculated, and (c) the ratio is functionalized to generate a pain ratio function, and the pain ratio is calculated by the pain ratio function. The pain ratio can be calculated, for example, by a specific process as shown in the embodiment.

[0865] The pain ratio (pain ratio) (%) was calculated by the following formula.

[0866] PO i = the estimated number of samples with pain (with pain) i / (The estimated number of samples with pain (with pain) i + Estimated number of pain-free (no pain) samples i )×100

[0867] i: the number of divisions of the pain estimate range (i is usually a positive integer)

[0868] In one embodiment, if the time lapse of the average value (e.g., arithmetic / geometric mean potential) is monotonically increasing, it can be judged that the pain is strong, and if it is monotonically decreasing, it can be judged to be weak. For example, in one embodiment, under the condition of 60 seconds stimulation and 60 seconds rest, the correlation coefficient of EEG, subjective evaluation, and stimulation intensity is calculated, and it is applied to the corresponding t-verification, and compared between 15 seconds and the adjacent 20 seconds. It is found that the correlation of 20 seconds is significantly higher than that of 15 seconds. This tendency can also be observed when it increases from 20 seconds to 30 seconds and from 30 seconds to 40 seconds. It can be concluded that it is preferable to use a time width of more than 20 seconds, more preferably more than 30 seconds, and further preferably more than 40 seconds. In particular, in the monitoring of long-term continuous pain, the effectiveness of this method is high. In addition, even for general pain, there is a situation where the uncomfortable pain is continuously or intermittently generated for more than tens of seconds, and the effectiveness of monitoring using a time width of more than 20 seconds is high. It was found that the method of the present invention can be effectively used for the analysis of uncomfortable pain when used in practice. Therefore, it can be understood that the time width that can be used in the present invention is, for example, more than 10 seconds, more preferably more than 15 seconds, and further preferably, more than 20 seconds, more than 25 seconds, more than 30 seconds, more than 35 seconds, more than 40 seconds, more than 45 seconds, more than 50 seconds, more than 55 seconds, and more than 60 seconds are successively preferred implementation methods. In addition, according to the change of pain monitoring time from short-term to long-term, a "graded pain trend monitoring method" is used. Using a short time unit of about 10 seconds, the time unit is reduced as the monitoring time passes, so that the trend (time change) of the brain characteristic quantity in the middle period where the correlation with the perception of pain becomes higher is visualized. Furthermore, in the case of long-term monitoring, it is reduced to the level of several minutes, showing an ultra-long-term pain trend.

[0869] In one embodiment, the result of the trend analysis is understood as follows: when the transition of the average value (e.g., arithmetic / geometric mean potential) is a monotonically increasing pattern, it can be judged as "severe pain", and on the other hand, when it is a monotonically decreasing pattern, it can be judged as "mild pain". Sometimes, in order to simplify monitoring, the "pain index" can be transformed ("baseline" correction process) as needed and set to an indicator that the pain is severe when the "pain index" is greater than 0, and the pain is weak when it is greater than 0 hours. When the pain level rises and the EEG potential increases monotonically, the waveform is not inverted but used directly, and the activity of intense pain can be set to 0. On the other hand, when it decreases monotonically, the waveform is inverted and set in a manner that the larger the number, the more intense the pain, so that it is easy to intuitively understand the change in pain level. The present invention is applicable regardless of how the index is calculated.

[0870] In one embodiment, the baseline correction of the pain index can be calculated by the following formula:

[0871] [Mathematical formula 3]

[0872] EEG －1 -BS

[0873] ( －1 BS = indicates the base value calculated using the strongest pain EEG (normalized). In this case,

[0874] For example, in another embodiment, the pain index can be calculated by the following formula:

[0875] [Formula 4]

[0876] EEG-BS

[0877] (BS=base value representing the strongest pain EEG (normalized)). In this case, the smaller the number, the stronger the pain, and the activity of strong pain is 0. At this time, the data has been normalized before the baseline correction, so normalization is not required at this stage.

[0878] Here, in this specification, the "pain index" or the "baseline" of the pain index is an index obtained by introducing a characteristic quantity reflecting a specific pain level (e.g., pain intensity level) as a reference for the characteristic quantity change related to continuous pain and taking into account qualitative judgments such as "pain" or "no pain" or pain "decreasing" or "increasing". Such an index can be understood similarly to the distinction between aqueous solutions such as "alkaline", "neutral" and "acidic" based on pH 7.

[0879] In the present disclosure, the time width for taking the average value (e.g., arithmetic / geometric mean potential) or the absolute value of the potential and the average value of other feature quantities can be arbitrary, for example, a moving average method, a non-overlapping block average method, a sparse average method (average discrete data), a method of obtaining multiple peak amplitudes and averaging, etc. It is preferred that the average value (e.g., arithmetic / geometric mean potential) at a specific time width is calculated by the non-overlapping block average method when it is statistically excellent. This is because: although it is not desired to be bound by theory, in the moving average method, there is a possibility that it is more difficult to discretely capture the change in pain level due to the high temporal continuity of the average data, so it is considered to use the non-overlapping block average method as a basis. The advantage of the time average method is that it includes as much data as possible in the time direction, so it can be assumed that the noise that varies with time is offset and / or reduced to leave the pain-related signal, and even if noise such as a single large burst pulse is mixed, the more data in the time direction, the less the impact can be reduced.

[0880] The following diagram illustrates the pain level monitoring method: Fig.11 a) step S100 of measuring or obtaining brain wave data or analysis data thereof from the estimated object by performing brain wave measurement for stimulation; b) step S200 of extracting a characteristic quantity of the time lapse of an average value (e.g., arithmetic / geometric mean potential) for plotting the brain wave data or analysis data thereof under a specific time width; step S300 of calculating a pain index (e.g., a characteristic quantity degree indicating the intensity of the pain level, etc.) for evaluating or monitoring the level of pain of the estimated object in a time series according to the brain wave data as needed, and calculating a monitoring baseline based on the pain index; and c) step S400 of monitoring the level of pain of the estimated object from the brain wave data based on the baseline or pain index as needed.

[0881] In step a), i.e., the step (S100) of measuring or providing brain wave data or analysis data thereof from the estimated object by performing brain wave measurement for stimulation, the brain wave data or analysis data thereof (e.g., amplitude data) of the estimated object is obtained by any method, and such brain wave data or analysis data thereof (e.g., amplitude data) can be obtained by any method known in the technical field. The exemplified amplitude data can be obtained by measuring the electrical signal of the brain wave, and can be represented by a potential (which can be represented by μV, etc.).

[0882] In step b), i.e., b) obtaining the time-lapse of the average value (e.g., arithmetic / geometric mean potential) of the brain wave data or its analysis data (e.g., amplitude data) at a specific time width, the feature quantity used to obtain the average value is extracted, and the average value of the extracted feature quantity at a constant time width (e.g., 40 seconds, etc.) is calculated (S200). Here, as preprocessing, attenuation of eye movement, correction of baseline fluctuation, removal of false signals (amplitude absolute value>100μV), normalization or standardization of the amplitude absolute value, etc. may also be performed.

[0883] Step b) is a step (S200) of extracting feature quantities in order to plot the time shift of the average value (e.g., arithmetic / geometric mean potential) at a specific time width of the brain wave data. In this step, the amplitude data obtained in step a is used to produce the shift data of the average value. The average value (average value (e.g., arithmetic / geometric mean potential), absolute value of potential, and other feature value average values) can be obtained using any method known in the technical field. As such specific methods, there can also be cited the moving average method, the non-overlapping block average method, the sparse average method, the method of obtaining multiple peak amplitudes and averaging them, and other types of central tendency values ​​including the average value (median, mode), but are not limited to these. Among them, methods that are statistically shown to have a high correlation and / or approximation with subjective pain changes (e.g., the non-overlapping block average method in the present disclosure) are also given high priority from the point of view of expressing the pain level trend more discretely.

[0884] As required, in order to evaluate or monitor the pain level of the estimated object in a time series based on the brain wave data, a pain index (e.g., a characteristic quantity degree indicating the intensity of the pain level, etc.) can be used to determine the monitoring baseline (S300). The pain index can be calculated in advance by a pain discrimination / estimation model, or it can be determined for each person before the start time of monitoring. Therefore, S300 is not a particularly necessary step in the trend analysis method of the present invention, and a pre-assigned pain index and / or baseline can also be used.

[0885] Step c) monitors or evaluates pain online based on the time lapse of the average value (e.g., arithmetic / geometric mean potential) and with reference to the monitored baseline as needed (S400). Here, the pain level can be represented and evaluated by a suitable method. Here, various evaluations can be performed based on the representation method of the data shown (trend representation, or single digital method). For example, based on the drawing, the pain level of the estimated object can be judged according to the brain wave data. In particular, the main purpose of the pain discrimination / estimation device is to monitor the pain level that changes moment by moment due to sedatives at the clinical site. Therefore, the pain trend analysis and visualization method in the present application is an important component technology.

[0886] The pain level discrimination / estimation device 1110 of the present invention for monitoring the pain of an estimated subject based on the brain waves of the estimated subject includes:

[0887] A) an electroencephalogram data measuring unit 1111 that obtains electroencephalogram data of the estimation object or analysis data thereof;

[0888] B) a feature quantity extraction unit 1112 for extracting an average value (e.g., arithmetic / geometric mean potential) in order to obtain a time transition of an average value (e.g., an average value (e.g., arithmetic / geometric mean potential)) in a specific time width of the electroencephalogram data or its analysis data, a pain index generation unit 1113 as needed, and a pain monitoring benchmark determination unit 1114 for determining a monitoring baseline using a pain index (e.g., a feature quantity degree indicating a pain intensity level, etc.) in order to evaluate or monitor the pain level of the estimated object based on the electroencephalogram data in a time series as needed; and

[0889] C) A pain level monitoring unit 1115 monitors the pain level of the estimated object based on the electroencephalogram data or analysis data thereof, based on the temporal transition of the average value.

[0890] The electroencephalogram data acquisition unit A) performs step a), the feature extraction unit B) including the average value (e.g., arithmetic / geometric mean potential) data performs step b), the pain index generation unit generates a pain index as needed, the benchmark determination unit determines the benchmark as needed, and the pain level monitoring unit C) performs step c). B) and C) and any pain index generation unit and benchmark determination unit can be configured so that the same part has three functions, or can be configured as separate parts. The pain index generation unit and benchmark determination unit can be any part, and the pain index and / or benchmark value (baseline) can be imported or received from the outside.

[0891] Fig.12 1 is a block diagram showing the functional configuration of a system 1100 including a pain level discriminating / estimating device according to one embodiment. The system 1100 includes a pain level discriminating / estimating device 1110 and an electroencephalogram 1120. In this way, the electroencephalogram and the pain level discriminating / estimating device may be configured separately.

[0892] The device 1110 includes a measuring unit 1111, a feature extraction unit 1112, a reference determination unit 1114, and a pain level monitoring unit 1115. The device 1110 is implemented, for example, by a computer having a processor and a memory. In this case, when the processor executes a program stored in the memory, the device 1110 causes the processor to function as the measuring unit 1111, the feature extraction unit 1112, and, if necessary, as a pain index generation unit 1113 and a reference determination unit 1114. In addition, the device 1110 may also be implemented, for example, by a dedicated electronic circuit. The dedicated electronic circuit may be either an integrated circuit or a plurality of electronic circuits. The amplitude data measuring unit 1111 and the pain index generation unit 1113 that provides the pain index to the reference determination unit may be configured either internally or externally.

[0893] The measuring unit 1111 obtains a plurality of brain wave data from the estimated object 1000 by performing brain wave measurement via the electroencephalogram 1120. The estimated object 1000 refers to a living body whose brain waves change due to pain, and it can be any living body having pain nerves (for example, vertebrates such as mammals and birds (including livestock, pet animals, etc.)), and is not limited to humans.

[0894] The feature extraction unit 1112 generates an average value of an appropriate time width of the electroencephalogram data. In addition, in order to perform pain level discrimination / estimation (machine learning, etc.) and / or online monitoring of the pain level based on a more advanced model, the feature value is provided to the pain index value generation unit 1113 and the reference determination unit 1114 for pain monitoring as needed. The average value can be calculated by a method known in the technical field. The algorithm of such a calculation method can be stored in the feature extraction unit 1112 in advance, or can be obtained when used through a medium or communication function. That is, the feature extraction unit 1112 can generate a feature value for monitoring pain trends including average data. The pain level monitoring unit 1115 helps the monitor monitor or evaluate the subjective pain of the subject by tracking the feature value including the calculated average data. In addition, the monitoring unit 1115 can also express the output result of the reference determination unit 1114 as a single number (0 to 100 pain level, etc.) and / or a change in the number.

[0895] The electroencephalogram 1120 measures the electrical activity occurring in the brain of the estimated subject using electrodes on the scalp, and outputs electroencephalogram data as the measurement result in conjunction with the measurement unit 1111 .

[0896] The processing or method of the device constructed as above can be performed Fig.11 The flowchart shown in the example shows a series of processing. That is, S100 to S400 can be performed by Fig.12 The invention is performed by the illustrated apparatus.

[0897] Fig.13 is Fig.12 Block diagram of the device configuration incorporating details of the "Graded Pain Trend Monitoring Method" and the selection process for baseline correction.

[0898] The present invention can be used as a method for obtaining Fig.12 and Fig.13 For example, data from subjects evaluating pain is obtained and averaged. The average value in this case can be calculated while monitoring the data, after accumulating a certain amount of data, or by using both methods. Fig.13In the apparatus using the graded pain trend monitoring method, an apparatus is installed. Here, an average value and / or other characteristic quantities can be calculated, and based on the pain index, a time point at which the pain level is strong can be identified and a baseline of the pain trend can be determined. That is, brain wave data (or its analysis data) is obtained from the electroencephalogram 1120 and measured by the measuring unit 1111. The characteristic quantity extraction unit 1112 extracts characteristic quantities and the like from the measured values. A pain index can be generated from the characteristic quantities and the like in the pain index generation unit 1113 as needed, and a baseline that becomes a benchmark for pain level monitoring can be determined in the benchmark determination unit 1114 as needed using the pain index. In addition, in the graded pain trend monitoring method, the trend of the pain level can be monitored online immediately after the baseline is obtained or determined.

[0899] (Other embodiments)

[0900] The pain estimation device according to one or more embodiments of the present invention has been described above based on the embodiments, but the present invention is not limited to the embodiments. As long as it does not deviate from the main purpose of the present invention, various modifications that can be thought of by a person skilled in the art to the present embodiment and the combinations of components in different embodiments may also be included in the scope of one or more embodiments of the present invention.

[0901] In addition, part or all of the components of the pain estimation device in each of the above-mentioned embodiments may be constituted by a system LSI (Large Scale Integration). For example, the device 1110 may be constituted by a system LSI having a measuring unit 1111, an average data obtaining unit 1112, a pain monitoring reference determining unit 1114, and a pain level monitoring unit 1115.

[0902] System LSI is a super multifunctional LSI manufactured by integrating multiple components into one chip. Specifically, it is a computer system composed of a microprocessor, ROM (Read Only Memory), RAM (Random Access Memory), etc. The ROM stores a computer program. The microprocessor works according to the computer program, so that the system LSI achieves its function.

[0903] In addition, although it is called system LSI here, it may also be called IC, LSI, super LSI, or large-scale LSI depending on the difference in integration. In addition, the method of integrated circuitization is not limited to LSI, and it can also be realized by a dedicated circuit or a general-purpose processor. After LSI is manufactured, a programmable FPGA (Field Programmable Gate Array) or a reconfigurable processor that can reconfigure the connection and / or setting of the circuit unit inside the LSI can also be used.

[0904] Furthermore, if integrated circuit technology that replaces LSI emerges due to advances in semiconductor technology or other derived technologies, it is of course possible to integrate functional blocks using this technology. Application of biotechnology is also possible.

[0905] In addition, one aspect of the present invention may be not only such a pain index generation and pain identification / classification device, but also a pain index value generation and pain level monitoring method that uses the characteristic components included in the pain identification / estimation device as steps. In addition, one aspect of the present invention may be a computer program that causes a computer to execute the characteristic steps included in the pain index value generation and pain level monitoring method. In addition, one aspect of the present invention may be a computer-readable non-volatile recording medium that records such a computer program.

[0906] In addition, in each of the above embodiments, each component can be formed by dedicated hardware or implemented by executing a software program suitable for each component. Each component can also be implemented by a program execution unit such as a CPU or a processor reading and executing a software program recorded in a recording medium such as a hard disk or a semiconductor memory. Here, the software that implements the pain discrimination / estimation device and the like of each of the above embodiments is the following program.

[0907] That is, the program causes a computer to execute a method for monitoring the pain of an estimated object based on the brain waves of the estimated object, the method comprising: a) a step of measuring or obtaining brain wave data or its analysis data from the estimated object by performing brain wave measurement in response to stimulation; b) a step of obtaining the time shift of an average value (e.g., arithmetic / geometric mean potential) of the brain wave data or its analysis data under a specific time width (here, a feature quantity for the time shift of the average value is extracted as needed), a step of generating a pain index (e.g., a feature quantity degree representing the pain intensity level, etc.) as needed in order to evaluate or monitor the pain level of the estimated object in a time series based on the brain wave data, a step of determining a monitoring baseline as needed; and c) a step of monitoring the pain level of the estimated object based on the brain wave data based on the baseline determination process as needed.

[0908] The present invention provides a recording medium storing a program, wherein the program causes a computer to execute a method for monitoring pain of an estimated object based on the brain waves of the estimated object. The program causes a computer to execute a method for monitoring pain of the estimated object based on the brain waves of the estimated object, the method comprising: a) a step of measuring or obtaining brain wave data or its analysis data from the estimated object by performing brain wave measurement for stimulation; b) a step of obtaining and plotting the time transition of an average value (e.g., arithmetic / geometric mean potential) of the brain wave data or its analysis data under a specific time width (here, extracting a feature quantity of the time transition for obtaining a suitable average value), generating a pain index (e.g., a feature quantity degree indicating the pain intensity level, etc.) as needed in order to evaluate or monitor the pain level of the estimated object in a time series based on the brain wave data, and determining a monitoring baseline as needed; and c) a step of monitoring the pain level of the estimated object based on the brain wave data or its analysis data based on the time transition of the average value.

[0909] It can be understood that the method implemented here can be used in combination with one or more of any of the embodiments described in the (trend analysis method) part of the present invention.

[0910] In another embodiment, the present invention provides a method for generating a model for determining pain of a subject, comprising:

[0911] a) obtaining brain wave data or analysis data thereof from the subject;

[0912] b) determining a characteristic quantity coefficient related to the pain with respect to a characteristic quantity based on the electroencephalogram data or its analysis data, and reducing the pain;

[0913] c) a step of creating a discriminant analysis model through machine learning and verification based on the reduced weighted feature quantity or a combination thereof; and

[0914] d) The step of determining a discriminant analysis model that achieves a specified accuracy.

[0915] In one embodiment, the present invention is characterized in that, in b), after the feature quantity coefficient is determined, the reduction is performed by repeatedly performing discrimination estimation, calculating the average of the feature quantity coefficient for the discrimination estimation and sorting them, and selecting the feature quantity based on a predetermined threshold. The determination of the feature quantity coefficient is preferably performed by machine learning, but is not limited thereto.

[0916] Machine learning includes support vector machine (SVM), neural network, deep learning, logistic regression, reinforcement learning, etc.

[0917] In another embodiment, the present invention includes in b): when determining the feature quantity coefficient, determining the hyperparameter with the highest discrimination accuracy, determining the feature quantity coefficient based on the hyperparameter, and excluding feature quantities that are invalid for discrimination or have a low contribution rate. The determination of the feature quantity coefficient and the hyperparameter is preferably performed by machine learning, but is not limited thereto.

[0918] In a further embodiment, regarding the present invention, the steps b) to d) include the following steps:

[0919] (C1) step of dividing the data corresponding to the feature quantity and the pain corresponding to the feature quantity into learning data and test data;

[0920] (C2) step, using the learning data to perform machine learning, create a discriminant model and calculate the optimal λ value (and partial regression coefficient, regression equation and model intercept);

[0921] (C3) step, using the test data to calculate the discrimination accuracy of the discrimination model; and

[0922] In step (C4), when there is an object sample in the object whose discrimination accuracy is below the chance level, the sample is excluded and steps C1 to C3 are repeated, and the process ends when there are no samples below the chance level, thereby determining the discrimination model, wherein the chance level is a value obtained by dividing 100% by the number to be classified.

[0923] Here, the chance level is a value obtained by dividing 100% by the number of classifications. It is a verification to avoid accidental coincidences. In the present invention, such a backward elimination method is also useful for generating a discriminant model.

[0924] In the model produced by the method of the present invention, although it is only one example, it is very meaningful that an accuracy of 70% can be achieved even with a small number of feature quantities. In addition to the method of the present invention, there are several other methods for sorting and selecting feature quantities. For example, the machine learning (SVM-RFE) presented as a comparative example is a more complex method that actually uses feature quantities to produce a discriminant model, calculates the weighting coefficients of each feature quantity and sorts them. In the present invention, in contrast, the sorting of feature quantities is not deduced by performing discrimination. One of its characteristics is: by reducing based on the most basic characteristics of classification (for example, the "all or nothing" law as the principle of neuronal discharge is also used as a selection criterion), the robustness of the change pattern of the brain wave feature quantity is investigated from the beginning before performing discriminant analysis.

[0925] With regard to whether reduction comes first or later, it is advantageous for the present invention to perform reduction first. The reason is that, in the present invention, after the fitting of the feature quantity is realized by reduction (e.g., Sigmoid function), all feature quantities are extracted or not, and a few elite discriminant models are determined by machine learning, so the characteristic is simply to end the calculation. On the other hand, in the case of fitting such as Sigmoid fitting, it is better to say that Sigmoid is used for reduction than to say that how many feature quantities used in a specific discriminant model have all patterns or no patterns (or approximations). In this case, machine learning is performed, and each feature quantity or multiple feature quantities are used to determine the feature quantity and / or discriminant model with the highest discrimination accuracy, and Sigmoid fitting is performed on the feature quantity with high discrimination accuracy, and the robustness of the feature quantity is investigated. Then, if an economical discriminant model is to be made, only the feature quantity with high fitting is selected to perform machine learning again, so the calculation cost becomes high, and at the same time waste is generated in the learning process. If the previous "machine learning process with reduction of feature quantity" is additionally implemented by Sigmoid fitting of feature quantity, it is necessary to further perform machine learning steps in vain, and the calculation cost becomes high. For this reason, it is advantageous to perform reduction first in the present invention.

[0926] Here, for reduction, for example, the correlation between feature quantities can be investigated and the highly correlated ones can be summarized into one or deleted, or factor analysis can be performed to find the main factors (for example, 10 factors can be found from 100 data), etc., or the feature quantity can be found according to a specific reflection pattern model using a Sigmoid function as in the above example and substituted into the model to achieve it. Regarding the specific pattern, various settings can be made, and in terms of pain, it can also be set to a discrimination such as "0 or 1" such as the presence or absence of pain. Here, if it is 0 or 1, a Sigmoid function (an example of logistic regression) and / or a step function can be used. A Sigmoid function can be made with "0 or 1" and approximated, and if a statistically significant approximation is found, it is substituted into the model. Regarding "all or nothing", in other words, it can also be expressed as a response to a specific stimulus with two values ​​of "0 or 1", which can also be said to be a feature quantity representing a discrete response of a number. When reducing, by determining the purpose of the object, the reduction can be converged to a certain range. For example, it can be expressed as "reduction related to pain" or "reduction related to pain stimulation" or the like.

[0927] The above example is a function approximation for the purpose of reducing feature quantities, but when specifying the optimal discriminant model, it is also possible to perform function approximation on the changes in the discrimination accuracy of multiple discriminant models (from the model with a small number of feature quantities to the model with a maximum number of feature quantities) as shown in Example 10 of the present invention, and select an economical model. For example, in function approximation, the above-mentioned Sigmoid function for realizing binary classification can be used. The Sigmoid function is an element function used at various levels. Of course, it is also the principle of neuronal discharge, a pain response function, a pain discriminator, and a pain incidence function (refer to Fig.31 ), and also the above-mentioned feature reduction tool (see Fig.34 ), further, can also be associated with the selection process of the discriminant model as described herein. For example, Fig.42 As shown in A, it can be observed that the variation curve of the discrimination accuracy of the discrimination model is a pattern including two inflection areas. Therefore, in a preferred embodiment, first, the inflection area with a relatively large inflection width (amplitude) is limited to derive the asymptotes of the minimum and maximum values ​​through S-shaped approximation. Next, the unevenness of the discrimination accuracy can be expressed by representative values ​​(minimum value, maximum value), and the improvement value of the discrimination accuracy (maximum value - minimum value) can be calculated, and the maximum value is used as the threshold, and the first feature quantity that exceeds this value is presented as the economic discrimination model with the highest gain in achieving the improvement rate.

[0928] Discriminant analysis is performed through machine learning and cross-validation starting from the top ranking of the sorted weighted coefficients (including approximate coefficients: for example, regression coefficients) of each feature quantity or its combination after reduction. This step can produce a sorted list of feature quantities (or their combination) and weighted coefficients after reduction, and perform cross-validation through machine learning, etc., to perform discriminant analysis.

[0929] In one embodiment, the present invention further includes, after step c), calculating the difference value (Diff) of the discrimination accuracy obtained by discriminant analysis with the adjacent model, wherein the adjacent model is a model including n-1 feature quantities and n feature quantities, n is greater than 2, and the difference is also considered in the judgment of the discriminant model in the step d). Here, regarding the difference value, as is well known in the technical field, it is installed in the software through the MATLAB function Diff, etc., which is a function of "continuously performing subtraction calculations on adjacent numerical values". In other words, the Diff function is one of the functions installed in MATLAB, and is a function that finds the difference value of adjacent values ​​while shifting one point at a time. This function corresponds to "differential" in mathematics and to "speed" in mechanics. The method of use is diverse, but in the case of signal processing, it can be used to find sharp or discontinuous (or staged) inflection points of time series and spatial data distribution. As for the Sigmoid function, the Diff function is effective for finding inflection areas. In terms of brain waves, it can also be used to find points in time or space where the characteristic quantity changes significantly.

[0930] In one embodiment, the consideration of the difference value includes the following process: re-sorting the feature quantities from the value with the largest difference value, and recalculating the discrimination accuracy, thereby generating a model with high discrimination accuracy.

[0931] In one embodiment, the judgment based on the differential value includes classifying the feature quantity into a main feature quantity and an auxiliary feature quantity, and re-ranking the auxiliary feature quantity. Here, as the main feature quantity, the feature quantity ranked first in the front or significantly related to it can be selected, and the others can be used as auxiliary feature quantities, and re-ranked based on the difference value between the n feature quantity model and the n-1 feature quantity model with the auxiliary feature quantity substituted. The function Diff can use a function installed in MATLAB, etc. For example, it is also possible to fix the first place in the front and set the ones after it as auxiliary feature quantities. Therefore, a main single feature quantity or a combination of the main feature quantity and the auxiliary feature quantity other than it can be used.

[0932] In addition, in this manual, pressing R 2 The values ​​are reordered and the feature quantity ranked first is called the main individual feature quantity. In addition, since the feature quantity ranked at the top and showing a significant correlation with the main individual feature quantity is used as a similar discriminant pattern, the main individual feature quantity is not limited to one, and their feature quantity group can also be collectively referred to as the main feature quantity. In this case, the feature quantity ranked first and showing a significant correlation with it can be used as the main feature quantity.

[0933] In one embodiment, the method of the present invention includes the step of changing the feature amount after the re-ranking and repeatedly performing the step c).

[0934] In another embodiment of the present invention, the following steps are included, namely: after the re-sorting, the feature quantity is changed to perform machine learning (linear regression, logistic regression, support vector machine, etc.) and cross-validation to calculate the discrimination accuracy of each model. After the re-sorting, the feature quantities are increased one by one to perform machine learning (linear regression, logistic regression, support vector machine, etc.) and cross-validation to calculate the discrimination accuracy of each model. Thus, the model with the highest accuracy can be selected. In the present invention, any machine learning can be used for machine learning, and linear regression, logistic regression, support vector machine (SVM), etc. can be used as supervised machine learning.

[0935] In one embodiment, the above steps may be repeated at least once. Here, because the order of the feature quantities is changing, only the feature quantities substituted are different.

[0936] When the candidate models are calculated, the target model can be generated by determining a model that achieves a specified accuracy, as needed.

[0937] The specified accuracy can be appropriately set according to the purpose, and a model with the highest accuracy and the least number of feature quantities, a model with the highest accuracy, etc. can be set. For example, when there are multiple discrimination models that achieve the same or equivalent degree of discrimination accuracy, it is preferred to select a model with the least number of feature quantities (this is also called an economic benchmark).

[0938] Therefore, one embodiment includes selecting a model with fewer types of feature quantities among models that achieve a predetermined accuracy.

[0939] In another embodiment, the predetermined discrimination accuracy includes the case of the highest accuracy. The reduction can be characterized by extracting effective feature quantities. More specifically, the reduction is characterized by selecting feature quantities close to all-or-nothing, i.e., binary feature quantities corresponding to pain or no pain. In addition, when extracting feature quantities related to subordinate classifications within pain or no pain, the same process can be repeated to further perform reflexive reduction.

[0940] In one embodiment, the weighting coefficient is used in the same sense as the approximate index of the discriminant function and can be selected from R 2 value, correlation coefficient, regression coefficient, residual sum of squares (difference between discriminant function and feature quantity), etc.

[0941] In another embodiment, the discriminant function (model) can be selected from a sigmoid function, a step function, and the like.

[0942] In yet another embodiment, the model approximation index is a partial set of weighting coefficients, which may be selected from R 2 Values, correlation coefficients, regression coefficients, residual sums of squares, and their partial sets.

[0943] In still another embodiment, in the case of binary classification, the effective feature quantity is an all-or-nothing feature quantity corresponding to the presence or absence of pain, or a feature quantity that is more similar to the discriminant function.

[0944] In another embodiment, the present invention provides a method for determining pain in an object, the method comprising: a) obtaining brain wave data or analysis data thereof from a reference object; b) reducing feature quantities in the brain wave data or analysis data thereof with respect to the pain; c) reducing weighted coefficients (also called model approximation indexes, such as R 2 value) by performing discriminant analysis through machine learning and cross-validation; d) determining a model that achieves a specified accuracy; e) obtaining brain wave data or analysis data thereof from a test subject; and f) applying the model to the test subject to discriminate pain.

[0945] In another embodiment, the present invention provides a method for determining pain in an object, the method comprising: e) a step of obtaining brain wave data or its analysis data from a test object; and f) a step of applying a model for determining pain to the test object to determine pain, wherein the model is generated by the following steps: a) a step of obtaining brain wave data or its analysis data from a reference object; b) a step of reducing feature quantities of the brain wave data or its analysis data with respect to the pain; c) a step of reducing the feature quantities of the brain wave data or its analysis data from weighted coefficients of each feature quantity (also called model approximation index, for example, R 2 value) by performing discriminant analysis through machine learning and cross-validation; d) determining a model that achieves a specified discriminant accuracy.

[0946] In a preferred embodiment, the present invention further comprises a step of filtering the brain wave data or its analysis data.

[0947] In a preferred embodiment, the filtering used in the present invention includes filtering for at least one selected from eye movement removal and myopotential attenuation.

[0948] In a preferred embodiment, the brain wave data or its analysis data used in the present invention includes at least one selected from average amplitude, correlation of inter-electrode (frontal-parietal, etc.) signals (potential, frequency, amount of information, etc.), frequency power, complexity index (multi-scale entropy), and phase synchronization (phase locking value, coherence).

[0949] (Backward sampling)

[0950] In one embodiment, in the present invention, the steps b) to d) include:

[0951] (C1) step of dividing the data corresponding to the feature quantity and the pain corresponding to the feature quantity into learning data and test data;

[0952] (C2) step, using the learning data to perform machine learning, create a discriminant model, and calculate the optimal λ value (and partial regression coefficient, regression equation and model intercept);

[0953] (C3) step, using the test data to calculate the discrimination accuracy of the discrimination model; and

[0954] In step (C4), when there exists an object sample whose discrimination accuracy is below the chance level in the object, the sample is removed and steps C1 to C3 are repeated. The process ends when there are no samples below the chance level, thereby determining a discrimination model, wherein the chance level is a value obtained by dividing 100% by the number to be classified.

[0955] Here, the chance level is the value obtained by dividing 100% by the number of classifications. This is a calculation to avoid accidental coincidences.

[0956] In such backward sampling, when making a prototype of a discrimination model used in an initial pain discrimination device, a "demonstration model" can be made based on a subject whose brain wave feature quantity changes when feeling pain are observed. For example, as shown in the embodiment, when the data of healthy subjects related to two levels ("pain, no pain") of acute neuropathic pain (electrical tingling) are discriminated and estimated one by one by machine learning (LASSO method), and the process of including subjects above the chance level in the model is repeated, as illustrated in the embodiment, it is shown that: the discrimination accuracy that remains at a not very high level such as "62%" in the initial stage is improved to a relatively high level such as "81%" in the final fifth stage. It is also found that the subjects included in the model are more sensitive to neuropathic pain than those not included, indicating that a demonstration model can be made based on samples sensitive to neuropathic pain through refinement. Progressive sampling is a method of discriminating with a small number of samples and improving the discrimination accuracy by gradually increasing the number of samples. Such a method is also useful for refinement and can be said to be an important scheme of the present invention. For example, there is also a method in which the samples are ranked based on the degree of difference in the characteristic quantity of "pain, no pain" of the individual samples, and the samples are progressively substituted into the model in sequence. The difference between progressive and backward elimination is that the progressive approach is "few samples → more samples", so the burden of calculation time is reduced. On the other hand, the advantage of backward elimination, which is another embodiment of the present invention, is that the model is refined by the information of the discrimination accuracy of the individual samples, so although it takes calculation time, it can be said to be comprehensive.

[0957] (New feature quantity for pain discrimination)

[0958] In another embodiment, the present invention provides a method for discriminating pain of a subject, the method comprising: using a model for discriminating pain of the subject including feature quantities at at least two electrodes at or around Fz or FPz, the step of discriminating pain of the subject. In addition, in the reduction learning of the present invention, specific information of the electrode is not required. Since the process of making the discriminant model is important, it can be said that what kind of feature quantity is entered is secondary information.

[0959] In one embodiment, the surrounding includes at least one of F3, F4, Fp1 and Fp2.

[0960] In the present invention, in addition to providing a model formation with good computational cost, it is unexpectedly determined that feature quantities that were not previously considered to be effective for pain discrimination can also be used for pain discrimination. In terms of the nature of brain waves, potential activity in the brain is conducted in a wave-like manner and is further attenuated by the skull. Therefore, if the amplitude of Fz is important, it can be said that the amplitude of peripheral electrodes such as Fpz and FCz can serve as a substitute. In addition, since existing monitoring devices are not attached to the scalp with hair, two electrodes (Fp1, Fp2) for brain wave recording are usually attached near the forehead. In the present invention, it is shown that "two electrodes" are the most effective in the reduction process of feature quantities, and Fz is effective. In addition, from the perspective of the possibility of potential propagation and diffusion, Fpz or peripheral electrodes may also be effective. It can be said that the results of "two electrodes" and "around the forehead" were obtained based on the data analysis of the device performance that determines the discrimination accuracy. Since there are reports that the frequency power of the gamma band in the anterior part of the prefrontal cortex may be associated with pain (Schulz et al., (2015). Prefrontal Gamma Oscillations Encode Tonic Pain in Humans. Cerebral Cortex 25: 4407-4414), it can be said that the pain analysis method of the present invention using Fz, Fpz or their surroundings as feature quantities should be recognized as a new discrimination means.

[0961] Each step is described below.

[0962] The following uses a scheme diagram to illustrate the discriminant analysis method with reduction of feature quantities ( Fig.34 ).

[0963] a) First, the discriminant characteristics of the characteristic quantities of the brain wave data or its analysis data are quantified (S10010). That is, here, the brain wave characteristic quantities are fitted using a sigmoid function with a binary change pattern of "0, 1" for each person, and the model approximation index (e.g., R 2Value). This step can also be said to be a step of reducing the characteristic quantity about the pain in the brain wave data or its analysis data. In the case where the purpose is to discriminate and estimate by each independent characteristic quantity, it can be used as a step of setting a threshold value and / or a judgment index in the model curve obtained by fitting. That is, the threshold value can be set by a numerical value such as a threshold potential and used as a judgment index. In detail, a discrimination / estimation model is made using brain wave characteristic quantities so that 2, 3 or 4 or more classifications can be performed in conjunction with conditional parameters. One method is to make a drawing and apply a suitable fitting function (fitting) such as a Sigmoid function pattern and a step function. Any method known in the art can be used. As such a specific fitting function, in addition to the above, Boltzmann function, double Boltzmann function, Hill function, logical dose response, Sigmoid Richard function, Sigmoid Weibull function, etc. can also be cited, but it is not limited to these. Among them, the standard logistic function is called the Sigmoid function, and the standard function or its deformation is a general case, which is preferred.

[0964] When the regression coefficient applied to the appropriate function pattern such as the Sigmoid function pattern is greater than the prescribed value as needed, a threshold for pain determination can be set based on the Sigmoid curve, etc. Here, in the case of the Sigmoid curve, it can be generated based on the inflection point, but is not limited to this. As needed, the pain classification value can also be calibrated to maximize the classification of the pain level.

[0965] b) Next, the feature quantities are sorted (S10020). Here, a weighting coefficient may be used. As the weighting coefficient, the R of each feature quantity may be used. 2 When the calculation is completed, the feature quantities are sorted.

[0966] Next, c) performs a hierarchical discriminant analysis (S10030) including the feature quantities ranked in sequence from the top. For example, it can be illustrated that the machine learning model such as support vector machine is substituted in sequence from the top-ranked feature quantities and the discrimination accuracy of all samples is investigated by leave-one-out cross-validation. b) and c) are equivalent to the following steps, that is, the steps of substituting from the top-ranked sort of the reduced weighted feature quantity or its combination, and performing discriminant analysis through machine learning and cross-validation.

[0967] d) Next, the discriminant model is determined (S10040). This is equivalent to the step of determining a model that achieves a specified accuracy. For example, the "economic discriminant model" with the highest accuracy or the least feature value if the accuracy is the same can be determined. However, as long as it achieves the specified accuracy (for example, 70% discrimination accuracy), it can be set to select any model, etc. In the present invention, steps c) and d) can be performed in the model generation unit. In the case where it is assumed that the model is predetermined using an existing database, the pain data can also be entered into the model during actual monitoring and a discriminant estimate can be performed. It can be assumed that the black arrow is a process that assumes actual monitoring.

[0968] In addition, in different optimal model selection processes, there may also be a method for re-ranking feature quantities as shown in Example 11. In this case, after determining the discriminant model in S10040, return to the feature quantity ranking in S10020, re-rank the feature quantities according to the difference in discrimination accuracy between the n-1 feature quantity model (n≥2) and the n feature quantity model, and perform S10030 to substitute the top-ranked feature quantities into the model one by one, and recalculate the discrimination accuracy. Afterwards, the discriminant model is determined through the second model determination process in S10040. For this, Fig.44 The following is explained with reference to the reference numerals. That is, after the discriminant model is determined, the feature quantity ranking is returned to S10060, and the feature quantities are re-ranked according to the difference value of the discrimination accuracy between the n-1 feature quantity model (n≥2) and the n feature quantity model (S10070-1), and the model is substituted one by one from the top ranking feature quantity in S10070-2 to recalculate the discrimination accuracy. After that, the discriminant model can be determined through the model determination process of S10080.

[0969] The characteristic quantity can be generated according to a certain stimulus (such as hot and cold stimulation, electrical stimulation, etc.), or obtained in a natural environment. It can also use various brain wave data, brain activity data, brain activity amount, amplitude data (EEG amplitude), frequency characteristics, etc. Such brain wave data can be obtained using any method known in the technical field. Brain wave data can be obtained by measuring the electrical signal of the brain wave, and represented by potential (which can be represented by μV, etc.) as amplitude data. The frequency characteristic can be represented by power spectrum density, etc. Regarding brain wave data, after basic signal processing such as filtering, eye movement correction, and pseudo signal removal, the corresponding part of the signal can be extracted in association with the conditional parameters and made into brain wave characteristic quantities. These include average values ​​(arithmetic mean, geometric mean), other representative values ​​(median, mode), entropy, frequency power, wavelet, average, and components of potential related to a single event, etc. Fig.35 and 36Shown is a comparison of the example of the present invention and a comparative example, and the difference in computational cost is obvious.

[0970] In a preferred embodiment, in order to implement the present invention, the electroencephalogram data is preferably recorded by a simple method that 1) the number of electrodes required for analysis is used, 2) the scalp with hair is avoided as much as possible, and 3) the recording can be performed even when the person is asleep. For example, although 24 electrodes are used as an example, this is not limited to this, and it can also be 12 to 24 or 24 to 36, or 6 to 12, or a smaller number (such as 3, 4, 5, etc.) is also possible.

[0971] As an example of reduction, in addition to Sigmoid fitting, a step function that is bent in steps, a linear function that changes continuously in steps, or the like may be used.

[0972] As weighting coefficients, in addition to regression coefficients, R 2 Value, correlation coefficient, residual sum of squares (difference between discriminant function and characteristic quantity), etc. However, in the discrimination of pain, it is important to be able to distinguish whether an individual feels pain and / or stress as accurately as possible, and effectiveness different from the detection of statistically significant differences is required, and this can also be used as a purpose.

[0973] In one embodiment, the data recording position of the brain wave data or its analysis data may include the positions on the scalp from the frontal lobe to the parietal lobe and then to the occipital part according to the international 10-20 standard or its expansion standard, or the positions separated by a specific uniform distance (e.g., 2.5 cm, etc.) may be included, and the time period for recording and analysis may be 0-100, 100-200, 200-300, 300-400, 400-500, 500-600, 600-700, 700-800 milliseconds (ms) in the case of short-term event-related potential activity, or it may be a more refined time division (10 milliseconds, etc.) and / or a longer time period (sometimes up to several seconds). The brain wave data or its analysis data includes at least one brain wave feature selected from their combination.

[0974] In another embodiment, the characteristic amount of the brain wave includes at least one selected from Fp1, Fp2, Fpz, F3, F4, Fz, C3, C4, Cz, P3, P4 and Pz, for example, including average amplitude Fz, C3, C4, frequency Fz (δ), Fz (β), Cz (δ), C3 (θ) and C4 (β). Preferably, it includes Cz (amplitude), C3 (α), Cz (β), Fz (δ), Cz (γ), but is not limited to this.

[0975] Fig.47 The schematic diagram (101000-108000) of the device (discrimination device with feature reduction process) of the present invention is recorded (in addition, it should be noted that several components in the composition diagram are arbitrary and can be omitted). In this schematic diagram, each step (S10100-S10600) is also appropriately recorded.

[0976] The device includes a feature reduction unit 101000, a feature extraction unit 102000, a pain discrimination estimation model generation unit 103000, a pain discrimination estimation unit (which may include a model correction process) 104000, a reference stimulus presentation unit 105000, an electroencephalogram data measurement unit 105200 and a pain level visualization unit 107000, and the object is identified by 106000.

[0977] In the discrimination with reduction process, the number of pain discrimination estimation stages (2 stages, 3 stages, etc.) is determined (S10100), and the discriminant function is generated (S10200, for example, Sigmoid function in binary classification). The feature quantity is obtained after the reference stimulus presentation unit 105000 gives the object 106000 a reference stimulus (electrical stimulation, etc.) according to the discrimination stage determined in S10100, collects the feature quantity associated with the pain stage (S10300), and performs reduction (S10400). In addition, the collected feature quantity is approximated by the discriminant function generated in S10200, and is sorted according to the size of the obtained approximate coefficient (regression coefficient, etc.). The features ranked in the front are used in sequence, and the pain level of the reference stimulus is discriminated and estimated in the pain discrimination estimation unit 104000, and the discriminant model with the highest number of feature quantities for discrimination accuracy is used for pain monitoring. This is one method of the feature quantity reduction process S10400. In addition, the discrimination model (algorithm) installed in the pain discrimination estimation unit used in the reduction process (hollow arrow) and the actual pain monitoring process (black arrow) is created in the pain discrimination estimation model generation unit 103000 and installed in the pain discrimination estimation unit 104000. After the above-mentioned preprocessing in the feature reduction unit 101000 is completed, the actual pain-related electroencephalogram data is collected from the subject 106000 by the electroencephalogram data measurement unit 105200 including an electroencephalogram (S10500). It is sent to the feature extraction unit 102000 and converted into the feature (e.g., the specific amplitude of a specific electrode, the frequency band power, etc.) selected in the amplitude and / or frequency reduction process. The extracted feature is taken from the feature extraction unit 102000 into the pain discrimination estimation unit 104000 (which may include a model correction process) to determine the pain level (S10600). The determination result is represented in the pain level visualization part 107000 using a change trend and / or a numerical value (eg, from 1 to 100).

[0978] When determining the pain discrimination estimation stage at S10100, the number of levels (for example, 2 stages, 3 stages, etc.) to be subjected to discrimination estimation is determined.

[0979] When generating the discriminant function in S10200, a discriminant function (such as a sigmoid function or a step function in binary classification) to be used is created according to the number of discrimination levels in S10100.

[0980] When collecting feature quantities associated with pain stages at S10300, reference stimulation presenting unit 105000 presents reference stimulation (electric stimulation, etc.) to subject 106000 a plurality of times according to the number of levels determined at S10100, and collects associated brain wave feature quantities.

[0981] When reducing the feature quantity in S10400, the feature quantity obtained in S10300 is approximated by the discriminant function, and the approximation index (such as R 2 The features with the highest values ​​(such as the value of the feature with the highest discrimination accuracy) are sorted, and the features ranked from the top are sequentially substituted into the pain discrimination estimation unit 104000 to perform discrimination estimation on the level of the reference stimulus. Among them, the model with the feature with the highest discrimination accuracy is used for the actual pain discrimination estimation.

[0982] When collecting pain-related electroencephalogram data in S10500, actual pain-related electroencephalogram data that is the target of pain monitoring is collected after the reduction process in the feature value reduction unit 101000 is completed. This step is data collection in the real-time pain monitoring process.

[0983] When determining the pain level in S10600, after the feature extraction unit 102000 processes the actual pain-related data obtained in S10500 and obtains a feature set, the pain discrimination estimation unit 104000 performs discrimination estimation, quantifies the pain level according to the estimated value, determines the pain level, and visualizes it through the pain level visualization unit 107000.

[0984] In order to obtain electroencephalogram data synchronized with the stimulation given to the subject (106000) by the electroencephalogram data measuring unit 105200 from the reference stimulation presenting unit 105000, the apparatus 108000 is provided with an electroencephalogram connected to the subject (106000) or connectable to the subject, or connected to the electroencephalogram.

[0985] The device 108000 may include an electroencephalogram measuring unit including an electroencephalogram recording sensor and, if necessary, an electroencephalogram amplifying unit internally or externally connected thereto, and the device 108000 may perform signal processing of pain-related electroencephalograms and estimation of pain discrimination.

[0986] In the device 108000, the feature extraction unit 102000 processes the collected brain wave signals to extract the brain wave feature. When extracting, the feature that has been reduced in advance in the feature reduction unit 101000 is selected. In addition, (if necessary) the pain is visualized by the pain level visualization unit 107000. In addition, a reference stimulus presentation unit 105000 is provided inside or outside. The reference stimulus presentation unit 105000 presents a reference stimulus such as multiple electrical stimulations according to the pain level determined in S10100 in order to reduce the feature effective for pain monitoring of the object 106000. The brain wave data associated with it is recorded by the brain wave data measurement unit 105200, and the associated brain wave feature is obtained by the feature extraction unit 102000, and the pain discrimination estimation unit 104000 discriminates and estimates the pain level of the reference stimulus according to the feature, and the feature reduction S10400 is performed based on the result. In addition, the reference stimulus presentation unit 105000 also issues pain stimulus information (stimulus type, environmental information, etc.) for discriminator production and actual unknown pain level discrimination. Regarding the reference stimulus presentation unit 105000, in addition to the reference stimulus presentation unit 105000, a stimulus information visualization unit may also be provided as needed to display images, numbers, and other information related to the stimulus and / or environment. In addition, the device 108000 may also be provided with a pain discrimination estimation unit 104000 that generates a discriminator and / or a judgment value outside or inside it.

[0987] Thus, the device 108000 includes an electroencephalogram data measuring unit 105200 and a pain discrimination estimation unit 104000, and also includes a reference stimulus presenting unit 105000 as required. The device 108000 is implemented, for example, by a computer having a processor and a memory. In this case, in the device 108000, when the program stored in the memory is executed by the processor, the processor is made to function as a feature reduction unit 101000, a feature extraction unit 102000, a pain discrimination estimation model generation unit 103000, a pain discrimination estimation unit 104000, etc. as required. Visualization of stimulation and / or environmental information is also performed as required. In addition, the device 108000 of the present invention can also be implemented, for example, by a dedicated electronic circuit. The dedicated electronic circuit can be either an integrated circuit or a plurality of electronic circuits. With regard to the electroencephalogram data acquisition unit and the comfort determination value generation unit, the same structure as that of the comfort determination device can be adopted.

[0988] The feature extraction unit 102000 may obtain a plurality of electroencephalogram data from an estimation target by performing electroencephalogram measurement a plurality of times via an electroencephalogram device (included in the electroencephalogram data measurement unit 105200). The target is a living body whose electroencephalogram changes due to stimulation and / or environment, and is not necessarily limited to humans.

[0989] The pain discrimination estimation unit 104000 uses the judgment value to discriminate and estimate discomfort. Even when the discriminator and / or judgment value are not pre-generated externally or internally, they are generated. The component that generates the discriminator and / or judgment value can also be configured as the pain discrimination estimation unit 104000 outside or inside the device 108000. The judgment value used for pain discrimination estimation is used to estimate or classify discomfort based on the amplitude of multiple brain wave data. In other words, the pain discrimination estimation unit 104000 or the pain discrimination estimation model generation unit 103000 can generate a judgment value for estimating or classifying discomfort of the object based on brain wave data.

[0990] The electroencephalogram recording sensor included in the electroencephalogram data measuring unit 105200 measures the electrical activity occurring in the brain of the estimated subject through electrodes on the scalp. Then, the electroencephalogram recording sensor outputs electroencephalogram data as the measurement result. The electroencephalogram data can be amplified as needed.

[0991] (Other embodiments)

[0992] The above is an explanation of the discrimination device with feature reduction process involved in one or more schemes of the present invention based on the implementation, but the present invention is not limited to the implementation. As long as it does not deviate from the main purpose of the present invention, the scheme obtained by applying various deformations that can be thought of by those skilled in the art to the present embodiment, and / or the method of combining the constituent elements in different implementations can also be included in the scope of one or more schemes of the present invention.

[0993] For example, in each of the above-mentioned embodiments, the peak-to-peak value may be used as the amplitude value of the electroencephalogram data, but the present invention is not limited thereto. For example, a simple peak value may be used as the amplitude value.

[0994] In addition, in the above-mentioned embodiment, it is assumed that the value Pmax of the discomfort magnitude corresponding to the upper limit value Amax of the characteristic quantity or its combination such as the brain wave amplitude is 1, and the value Pmin of the pain magnitude corresponding to the lower limit value Amin of the characteristic quantity or its combination is 0, and the range of the value of the discomfort magnitude is set, but it is not limited to this. For example, the magnitude of pain can also be expressed by 0 to 100. In this case, when the value Px of the pain magnitude is displayed by the pain level visualization unit 107000, the pain discrimination estimation unit 104000 can estimate the value Px of the pain magnitude by the following formula.

[0995] Px=Pmax×(Ax-Amin) / (Amax-Amin)

[0996] In addition, in the above description, as an example of generating a comfortableness judgment value by analyzing multiple brain wave data, curve fitting including Sigmoid fitting is described, but it is not limited to this. In addition, a predetermined value can also be used for the upper limit of the brain wave amplitude. For example, the predetermined value (absolute value) is, for example, 50μV to 100μV, which can be set by experiment or experience. In this way, in the usual analysis, as a method of removing false signals, data from positive and negative 50μV to about 100μV are excluded, but such false signal removal can also be implemented in the present invention as needed.

[0997] In addition, regarding the reference stimulus presentation unit 105000 (refer to Fig.47 ) The stimulus applied to the object 106000 may be any type of stimulus as long as the magnitude of discomfort felt by the object 106000 changes depending on the type of stimulus and / or the presenting environment.

[0998] In addition, part or all of the components of the device of the present invention in each of the above-mentioned embodiments may be constituted by a system LSI (Large Scale Integration). Fig.47 As shown, the device 108000 includes, in addition to a feature reduction unit 101000, a pain discrimination estimation model generation unit 103000, a pain discrimination estimation unit 104000, and a pain level visualization unit 107000, a system LSI having a feature extraction unit 102000 and a reference stimulus presentation unit 105000.

[0999] System LSI is a super multifunctional LSI manufactured by integrating multiple components into one chip. Specifically, it is a computer system composed of a microprocessor, ROM (Read Only Memory), RAM (Random Access Memory), etc. The ROM stores a computer program. The microprocessor works according to the computer program, so that the system LSI achieves its function.

[1000] Although it is called system LSI here, due to the difference in integration, it is also called IC, LSI, super LSI, and large-scale LSI. In addition, the method of integrated circuit is not limited to LSI, and it can also be realized by a dedicated circuit or a general-purpose processor. It is also possible to use a FPGA (Field Programmable Gate Array) that can be programmed after LSI manufacturing, or a reconfigurable processor that can reconfigure the connection and / or setting of the circuit unit inside the LSI.

[1001] Furthermore, if integrated circuit technology that can replace LSI emerges due to the progress of semiconductor technology or other derived technologies, it is of course possible to integrate functional blocks using this technology. Application of biotechnology is also possible.

[1002] In addition, one aspect of the present invention may be not only such a pain discrimination estimation model generation and a continuous pain discrimination estimation unit, but also a pain classification value generation and a pain discrimination classification method in which the characteristic components included in the pain estimation device are set as steps. In addition, one aspect of the present invention may also be a computer program that causes a computer to execute characteristic steps included in feature reduction, feature extraction, pain discrimination estimation model generation, and pain discrimination estimation. In addition, one aspect of the present invention may also be a computer-readable non-volatile recording medium that records such a computer program.

[1003] In addition, in each of the above-mentioned embodiments, each component can be formed by dedicated hardware or implemented by executing a software program suitable for each component. Each component can also be implemented by a program execution unit such as a CPU or a processor reading and executing a software program recorded in a recording medium such as a hard disk or a semiconductor memory. Here, the software that implements the pain estimation device and the like of each of the above-mentioned embodiments is the above-mentioned program in this specification.

[1004] (Implementation methods using cloud, IoT, and AI)

[1005] The pain determination technology of the present invention can be provided as a system or device in a manner including all the components (see Fig.50 Alternatively, as a pain discrimination device, it is also conceivable that the measurement of brain waves and the display of the results, calculation and / or calculation of the discrimination model are mainly performed through a server and / or cloud (see Fig.51 , Fig.52 , Fig.53 ). Some or all of them can be implemented using IoT (Internet of Things) and / or artificial intelligence (AI).

[1006] Alternatively, the pain discrimination device also stores the discrimination model and performs discrimination on the spot, but it is also conceivable that the main calculations such as the calculation of the discrimination model are performed by a server and / or a cloud, that is, a semi-independent solution ( Fig.53 ). Since transmission and reception may not always be possible in some implementation locations such as hospitals, a model that can be used even in shielded conditions is envisioned.

[1007] Therefore, in one embodiment, the present invention provides a program, a recording medium, a system and a device that enable a computer to execute a method for discriminating pain of an object, the method comprising: a) a step of obtaining brain wave data or its analysis data from the object; b) a step of generating a discriminant model based on the brain wave data or its analysis data; and c) a step of substituting the brain wave data or its analysis data from the object into the discriminant model to discriminate pain.

[1008] Regarding the system for realizing such a program, it is realized by understanding the whole as an implementation method of the system. In this scheme, the present invention is a system for determining pain of an object, and the system includes: X) an electroencephalogram data acquisition unit that obtains electroencephalogram data or its analysis data from the object; Y) a pain determination estimation model generation unit that generates a determination model based on the electroencephalogram data or its analysis data; and Z) a pain determination estimation unit that substitutes the electroencephalogram data or its analysis data from the object into the model to determine pain. In this case, Fig.50 In the example schematically shown in FIG. 1 , the electroencephalogram data acquisition unit is described as the electroencephalogram data measurement unit 110000 and the electroencephalogram feature extraction unit 140000. The pain discrimination estimation model generation unit is described as the pain discrimination model generation unit 160000. Moreover, the pain discrimination estimation unit is described as the pain level discrimination estimation unit 150000. Fig.50 In addition, a pain level visualization unit 130000 and a data storage unit 170000 are described. Such a visualization unit and a storage unit are not necessarily required, but it is better to have them when providing as a medical device.

[1009] As for the visualization part, any visualization part can be used as long as it is a structure that allows the user to recognize the pain discrimination result, and an input / output device, a display device, a television, a monitor, etc. can be used. As an alternative to the visualization part, other recognition means such as sound can also be used, which can include a sound generating device (such as a microphone), a vibration device, an electrode, or other devices that can present a challenge to the subject.

[1010] The storage unit may be a recording medium such as CD-R, DVD, Blueray, USB, SSD, or a hard disk, or may be stored on a server, or may be appropriately recorded on the cloud.

[1011] As in Fig.50As schematically illustrated in FIG. 1 , the present device can be used when making or determining a discrimination model (white arrow) and when monitoring actual pain. When making and determining a model, the brain wave data measuring unit 110000 obtains the brain waves of the subject when it is subjected to multiple pain stimuli. The recorded data is sent to the brain wave feature extraction unit 140000 and made into multiple feature quantities. The feature quantity is sent to the pain discrimination model generation unit 160000 to generate a discrimination algorithm. The discrimination algorithm is sent to the pain level discrimination estimation unit 150000, and the discrimination result obtained by the produced model is sent to the pain level visualization unit 130000 to be displayed to confirm the appropriateness of the pain level discrimination result. After the discrimination model is determined, the real-time monitoring of the actual pain level is carried out in the direction of the black arrow. That is, when pain monitoring starts, the brain wave data measuring unit 110000 obtains brain wave data from the subject, and the brain wave data is sent to the brain wave feature extraction unit 140000 to extract multiple brain wave feature quantities. The generated feature quantity is sent to the pain level discrimination estimation unit 150000, and the discrimination result is displayed by the pain level visualization unit 130000. It should be noted that the two processes can also be combined. If the discrimination result is inappropriate during real-time monitoring, the result of the pain level discrimination estimation unit 150000 will be fed back to the pain discrimination model generation unit 160000, and after the model is corrected, the corrected model will be sent to the pain level discrimination estimation unit 150000 again.

[1012] In one scheme, according to Fig.51 The present invention is a system for determining the pain of an object by a terminal that only obtains brain waves, and the system includes X) a pain determination terminal and Y) a pain determination estimation server. The pain determination terminal includes: X-1) a brain wave data acquisition terminal that obtains brain wave data or its analysis data from the object, and X-2) a module that sends the brain wave data or its analysis data to the pain determination estimation server and receives the determination result from the pain determination estimation server; the pain determination estimation server includes: Y-1) a pain determination estimation model generation module that generates a determination model based on the brain wave data or its analysis data; Y-2) a pain determination estimation module that substitutes the brain wave data or its analysis data from the object into the model and generates a determination result for determining pain; and Y-3) a determination result sending and receiving module that sends and receives the brain wave data or its analysis data and the determination result.

[1013] As in Fig.53As schematically illustrated in FIG. 1 , the present device can be used when making or determining a discrimination model (white arrow) and when monitoring actual pain. When making and determining a model, the brain wave data measuring un...

Claims

1. A pain identification method, comprising: A) Steps for performing machine learning to generate a discriminant model; B) Steps to determine hyperparameters in all samples through cross-validation; C) using the hyperparameters to generate a discriminant model and calculating the discriminant accuracy for each sample; D) a step of grouping each sample based on the discrimination accuracy; E) determining independent hyperparameters for each group; and F) A step of generating an independent pain discrimination model for each group using the independent hyperparameter, calculating the pain discrimination accuracy of each sample, and generating a maximum pain discrimination model therefrom.

2. The pain identification method according to claim 1, wherein: The discriminant model is generated by a processing method selected from supervised machine learning models including LASSO, linear regression, logistic regression, support vector machine, neural network, random forest, Bayesian, clustering, decision tree, or unsupervised machine learning models including deep learning.

3. A computer program product for causing a computer to execute a pain identification method, the method comprising: A) Steps for machine learning to generate a pain discrimination model; B) Steps to determine hyperparameters in all samples through cross-validation; C) using the hyperparameters to generate a pain discrimination model and calculating the pain discrimination accuracy for each sample; D) a step of grouping each sample based on the discrimination accuracy; E) determining independent hyperparameters for each group; and F) The step of using the independent hyperparameters to generate an independent pain discrimination model for each group, calculating the pain discrimination accuracy of each sample, and generating a maximum pain discrimination (MAX) model therefrom.

4. A recording medium storing a program for causing a computer to execute a pain identification method, the method comprising: A) Steps for machine learning to generate a pain discrimination model; B) Steps to determine hyperparameters in all samples through cross-validation; C) using the hyperparameters to generate a pain discrimination model and calculating the pain discrimination accuracy for each sample; D) a step of grouping each sample based on the discrimination accuracy; E) determining independent hyperparameters for each group; and F) The step of using the independent hyperparameters to generate an independent pain discrimination model for each group, calculating the pain discrimination accuracy of each sample, and generating a maximum pain discrimination (MAX) model therefrom.

5. A pain identification system, comprising: A) a discriminant model generation module, which performs machine learning to generate a pain discriminant model; B) a hyperparameter determination module, which determines the hyperparameters in all samples through cross-validation; C) a discrimination accuracy calculation module, which uses the hyperparameters to generate a pain discrimination model and calculates the pain discrimination accuracy for each sample; D) a grouping module, which groups each sample based on the discrimination accuracy; E) an independent hyperparameter module, which determines independent hyperparameters for each group; and F) A maximum discrimination model generation module, which uses the independent hyperparameters to generate an independent pain discrimination model for each group, calculates the pain discrimination accuracy of each sample, and generates a pain discrimination maximum (MAX) model therefrom.

6. A pain discrimination system, the system comprising a machine learning execution module, the machine learning execution module being configured to execute a method comprising the following steps: A) Steps for machine learning to generate a pain discrimination model; B) Steps to determine hyperparameters in all samples through cross-validation; C) using the hyperparameters to generate a pain discrimination model and calculating the pain discrimination accuracy for each sample; D) a step of grouping each sample based on the discrimination accuracy; E) determining independent hyperparameters for each group; and F) The step of using the independent hyperparameters to generate an independent pain discrimination model for each group, calculating the pain discrimination accuracy of each sample, and generating a maximum pain discrimination (MAX) model therefrom.

Images