Determination device, determination method, and program

Abstract

The judgment device comprises a state change means for receiving input of data relating to an object to be judged at each time step and changing a state at each time step, a reference time step identification means for identifying a reference time step which is a reference time step for changes in the state of the object to be judged, and a judgment means for making a judgment regarding the object to be judged based on the state of the state change means at a time step derived from the reference time step.

Description

[Technical field]

[0001] The present invention relates to a determination device, a determination method, and program Regarding. [Background technology]

[0002] It has been proposed to use a neural network to perform a process of determining the presence or absence of a cardiac abnormality based on an electrocardiogram (ECG). For example, Patent Document 1 describes a method of determining the presence or absence of an abnormality by applying a convolutional neural network to an ECG expressed as a matrix of real numbers. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] US Patent Application Publication No. 2017 / 0112401 Summary of the Invention [Problem to be solved by the invention]

[0004] A convolutional neural network is expected to be able to detect cases where a part of time series data has features with a relatively high degree of accuracy, since it performs a convolutional operation partially on the input data. On the other hand, a convolutional neural network generally requires a large amount of calculation, and therefore a device with high calculation capabilities is required. If it is possible to deal with cases where a part of time series data has features without the need to use a convolutional neural network, it is expected that the amount of calculation will be relatively small.

[0005] One example of the object of the present invention is to provide a determination device, a determination method, and program The purpose of this project is to provide [Means for solving the problem]

[0006] According to a first aspect of the present invention, a determination device includes: The state changes periodically a state change means for receiving an input of data relating to an object to be judged at each time step and changing a state at each time step; In one period of The system includes a reference time step identifying means for identifying a reference time step which is a reference time step, and a judgment means for making a judgment regarding the judgment object based on the state of the state change means in a time step derived from the reference time step.

[0007] According to a second aspect of the present invention, a determination method includes: The state changes periodically An apparatus including a state changing means for receiving an input of data relating to an object to be judged at each time step and changing a state at each time step, In one period of The method includes identifying a reference time step that is a reference time step, and making a decision regarding the decision object based on a state of the state changing means at a time step derived from the reference time step.

[0008] According to a third aspect of the present invention, program teeth, The state changes periodically A state change means for receiving an input of data relating to a judgment target at each time step and changing a state at each time step, and a device for executing a program, In one period of The program executes the steps of: identifying a reference time step that is a reference time step; and making a judgment regarding the judgment object based on the state of the state change means at a time step derived from the reference time step. Effect of the Invention

[0009] According to the present invention, it is possible to handle cases where part of the time series data has characteristics without the need to use a convolutional neural network. [Brief description of the drawings]

[0010] [Figure 1]An example is shown in which the reference time step identifying section 120 identifies the reference time step based on the sampling data from the sampling data acquiring section 110. [Diagram 2] 1 is a diagram showing an example of electrocardiogram data acquired by a sampling data acquisition unit 110. FIG. [Diagram 3] 13 is a diagram illustrating an example of the configuration of the determination device 100 in the case where the reference time step identifying unit 120 directly detects a trigger signal from a trigger signal generating unit 910. FIG. [Figure 4] 2 is a diagram illustrating an example of a data flow in the determination device 100. FIG. [Diagram 5] 1 is a diagram showing an example of the configuration of a calculation system 1 when a determination device 100 performs learning for determining a determination target 920. FIG. [Figure 6] FIG. 13 is a diagram illustrating another example of the configuration of the determination device according to the embodiment. [Figure 7] 10 is a flowchart showing an example of a processing procedure in a determination method according to the embodiment. [Figure 8] FIG. 1 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] The following describes embodiments of the present invention, but the following embodiments do not limit the scope of the invention. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention. Fig. 1 is a diagram showing an example of the configuration of a determination device according to an embodiment. In the configuration shown in Fig. 1, the determination device 100 includes a sampling data acquisition unit 110, a reference time step identification unit 120, a state change unit 130, a state storage unit 140, and a determination unit 150.

[0012] The determination device 100 performs a determination on a determination target 920. In particular, the state of the determination target 920 changes, and the determination device 100 determines the execution timing of the determination based on the timing related to the state of the determination target 920. In the following, an example will be described in which the determination object 920 is a heart, and the determination device 100 determines the presence or absence of a cardiac abnormality based on electrocardiogram data. However, the determination object by the determination device 100 is not limited to a specific one, and may be various things whose states change and whose values ​​according to the states can be measured. For example, the object 920 to be determined may be various objects, such as a factory machine, a human being, etc., for which classification may be performed on the movement of the object 920 to be determined.

[0013] The trigger signal generating unit 910 outputs a signal that serves as a trigger for a state change of the determination target 920. The signal that serves as a trigger for a state change of the determination target 920 is also called a trigger signal. When the determination target 920 is the heart, the sinoatrial node is an example of the trigger signal generating unit 910 .

[0014] The sampling data acquiring unit 110 measures a value related to the determination target 920. For example, the sampling data acquiring unit 110 may include a sensor to measure the output of the determination target 920 or the state of the determination target 920, and output the measurement value as sensing data. The sampling data acquiring unit 110 may directly measure the determination target 920, or may indirectly measure the determination target 920, for example, by measuring the temperature of the surrounding environment of the determination target 920. Direct measurement of the determination target 920 and indirect measurement of the determination target 920 are collectively referred to as measurement related to the determination target 920.

[0015] The sampling data acquisition unit 110 may be included in the determination device 100, or may be included in the determination target 920. Alternatively, the sampling data acquisition unit 110 may be configured outside the determination device 100 and outside the determination target 920. FIG. 1 illustrates an example in which the sampling data acquisition unit 110 is included in the determination device 100.

[0016] When the determination device 100 determines the presence or absence of cardiac abnormality using an electrocardiogram, the sampling data acquisition unit 110 may be configured using an electrocardiograph, and may measure electrocardiogram data at predetermined intervals and output the data as sampling data. The electrocardiogram data referred to here is, for example, measurement data of the potential (potential difference) between the electrodes of the electrocardiograph.

[0017] Fig. 2 is a diagram showing an example of electrocardiogram data acquired by the sampling data acquisition unit 110. The horizontal axis of the graph in Fig. 2 represents time, and the vertical axis represents potential. Line L11 shows an example of an electrocardiogram. The "X" marks on line L11 show examples of data sampled as electrocardiogram data.

[0018] FIG. 2 shows an example in which the sampling data acquisition unit 110 samples electrocardiogram data at regular intervals. The sampling interval of the electrocardiogram data by the sampling data acquisition unit 110 is not limited to a specific time interval. For example, if it is desired that the sampling data acquisition unit 110 measure the maximum and minimum values ​​of the potential more accurately, it is possible to have the sampling data acquisition unit 110 perform sampling at shorter time intervals.

[0019] In the following, the sampling interval of electrocardiogram data by the sampling data acquisition unit 110 is denoted as Δt. In addition, in the following, time may be expressed as a time step of Δt. Each step in the time step is also simply referred to as a time step or a time in a time step.

[0020] For example, the time step before time step t is denoted as t-Δt. The time step after time step t is denoted as t+Δt. The time step before time step t is the time step that is one step in the past of time step t. The time step after time step t is the time step that is one step in the future of time step t. The sampling data acquisition unit 110 acquires sampling data of an electrocardiogram for each time step.

[0021] FIG. 2 shows an electrocardiogram for one beat of the heart. In a typical electrocardiogram, the waveforms are observed in the order of P wave, Q wave, R wave, S wave, and T wave, as shown in FIG. 2. The potential is maximum in the P wave, R wave, and T wave. On the other hand, the potential is minimum in the Q wave and S wave. The Q wave, R wave, and S wave are collectively called the QRS complex.

[0022] The time from the start of the P wave to before the start of the Q wave is also called the PR Interval. The time from the start of the Q wave to the end of the T wave is also called the QT Interval. The time from the end of the P wave to before the start of the Q wave is also called the PR Segment. The time from the end of the S wave to before the start of the T wave is also called the ST Segment. Parts of the electrocardiogram can be associated with parts of the heart, for example, P waves represent atrial contractions, QSR complexes represent ventricular contractions, and T waves represent recovery from ventricular contractions.

[0023] The sampling data acquisition unit 110 outputs data indicating measurement values ​​obtained by measuring the determination target 920. The sampling data output by the sampling data acquisition unit 110 is also referred to as data related to the determination target, or simply as sampling data. In particular, the sampling data acquisition unit 110 acquires and outputs time series data of the sampling data by performing measurements on the determination target 920 at each sampling period. The time series data of the sampling data output by the sampling data acquisition unit 110 is also referred to as determination target time series data.

[0024] The reference time step identifying unit 120 identifies a reference time step. The reference time step is a reference time step regarding a change in the state of the determination target 920. For example, when the state of the determination target 920 changes periodically, the reference time step identifying unit 120 may identify the reference time step as a time step including a predetermined timing in one period of the state change of the determination target 920.

[0025] The reference time step identified by the reference time step identifying unit 120 is used as a guideline time step for the determining unit 150 of the determining device 100 to make a determination regarding the object 920 to be determined at a specific time step in response to a change in state of the object 920 to be determined. The reference time step identifying unit 120 corresponds to an example of a reference time step identifying means.

[0026] The reference time step identifying unit 120 may be able to directly detect the trigger signal. For example, the trigger signal generating unit 910 may be located outside the determination target 920, and the reference time step identifying unit 120 may be able to detect the trigger signal output by the trigger signal generating unit 910.

[0027] FIG. 3 is a diagram showing an example of the configuration of the determination device 100 in the case where the reference time step identification unit 120 directly detects the trigger signal from the trigger signal generation unit 910. As shown in FIG. The example in Fig. 3 differs from the case in Fig. 1 in that the reference time step identification unit 120 acquires a signal from a trigger signal generation unit 910 instead of the sampling data from the sampling data acquisition unit 110. In other respects, the configuration of the determination device 100 in Fig. 3 is the same as the configuration in Fig. 1. An example of the trigger signal generating unit 910 located outside the determination target 920 is a device that uses an external clock, such as an audio system that uses an external clock, but is not limited to this.

[0028] When the reference time step identifying unit 120 detects a trigger signal from the trigger signal generating unit 910, the reference time step may be identified as the time step at which the trigger signal is input to the reference time step identifying unit 120. Alternatively, when it takes time to detect a trigger signal, the reference time step identifying unit 120 may identify the reference time step as the time step next to the time step at which the trigger signal is input to the reference time step identifying unit 120.

[0029] A certain time step and a related time step that is shifted from that time step by a certain number of time steps are collectively referred to as a time step from that time step. The reference time step identifying unit 120 may identify a reference time step corresponding to a time step at which a trigger signal is input to the reference time step identifying unit 120.

[0030] It may be impossible or difficult for the reference time step identifying unit 120 to directly detect the trigger signal. An example of a case where it is impossible or difficult for the reference time step identifying unit 120 to directly detect the trigger signal is a case where it is impossible or difficult to directly observe the trigger signal generating unit 910. Also, an example of a case where it is impossible or difficult to directly detect the trigger signal is a case where the trigger signal is weak and impossible or difficult to detect.

[0031] For example, when the determination object 920 is a heart and the trigger signal generating unit 910 is a sinoatrial node, since the sinoatrial node is inside the body and the electrical signal output by the sinoatrial node is weak, it is considered difficult to directly detect the electrical signal output by the sinoatrial node in a non-invasive manner. In addition, it is generally not realistic to detect the electrical signal output by the sinoatrial node in an invasive manner.

[0032] When the reference time step identifying unit 120 does not directly detect a trigger signal, the reference time step may be identified based on the sampling data from the sampling data acquiring unit 110. Fig. 1 shows an example in which the reference time step identifying unit 120 identifies the reference time step based on the sampling data from the sampling data acquiring unit 110.

[0033] In this case, the reference time step identifying unit 120 may detect sampling data corresponding to a characteristic portion of the judgment target time series data. Then, the reference time step identifying unit 120 may identify the reference time step as the time step at which the sampling data corresponding to the characteristic portion is detected. It is expected that the reference time step identifying unit 120 can detect sampling data corresponding to the characteristic portion with high accuracy, and is unlikely to fail to identify the reference time step.

[0034] In the example of FIG. 2, if the reference time step identifying section 120 detects sampling data of an R wave, the potential difference between the preceding and succeeding sampling data is large, and it is expected that the sampling data can be detected with high accuracy. The reference time step identifying section 120 may compare the value of the sampling data (potential measurement value) with a predetermined threshold, and determine that the sampling data is an R wave sampling data when the value of the sampling data is equal to or greater than the threshold. In this case, the condition that the value of the sampling data is equal to or greater than the threshold corresponds to an example of the predetermined condition. The time step at which the value of the sampling data is equal to or greater than the threshold corresponds to an example of the time step at which the data related to the judgment target 920 satisfies the predetermined condition.

[0035] Alternatively, the reference time step identifying unit 120 may identify the reference time step based on both the trigger signal from the trigger signal generating unit 910 and the time series data to be determined from the determination target 920. For example, when the trigger signal from the trigger signal generating unit 910 is weak, the reference time step identifying unit 120 may identify the reference time step based on both the trigger signal from the trigger signal generating unit 910 and the time series data to be determined from the determination target 920. This is expected to reduce the possibility that the reference time step identifying unit 120 will fail to identify the reference time step.

[0036] The reference time step identification unit 120 may learn a process for detecting sampling data corresponding to a characteristic portion of the target time series data. The term "learning" used here refers to adjusting parameter values ​​of a learning model. For example, the reference time step identification unit 120 may input sampling data into a judgment formula to judge whether the sampling data corresponds to a characteristic part. In this case, the judgment formula corresponds to an example of a learning model, and adjusting parameter values ​​such as coefficients of the judgment formula corresponds to an example of learning.

[0037] The state change unit 130 receives an input of sampling data from the sampling data acquisition unit 110. The state change unit 130 has a state, and the state of the state change unit 130 changes depending on the input of the sampling data. The state change unit 130 receives an input of sampling data for each time step, and the state of the state change unit 130 changes for each time step. The state change unit 130 corresponds to an example of a state change means.

[0038] The state change unit 130 may be configured in the form of a reservoir layer in reservoir computing. That is, the state change unit 130 may be configured using a recurrent neural network (RNN). The values ​​of the weighting coefficients between the nodes of the state change unit 130 may be set in advance as being outside the learning target. For example, the connection relationships between the nodes of the state change unit 130 and the values ​​of the weighting coefficients may be set in advance randomly.

[0039] By configuring state change section 130 recursively, the state of state change section 130 reflects not only the current state of object 920 to be determined, but also past states of object 920 to be determined. The state x(t) of the reservoir layer in reservoir computing at time step t is expressed as equation (1).

[0040]

number

[0041] x(t-Δt) represents the state of the reservoir layer at time step t-Δt. As mentioned above, time step t-Δt is the time step one step before time step t. Both x(t) and x(t-Δt) are expressed as vertical vectors whose elements are the number of nodes in the reservoir layer.

[0042] W res is a matrix that indicates the weighting coefficient for the value of the intermediate node. res x(t-Δt)" indicates the weighting of the values ​​of the intermediate nodes at time step t-Δt. res is a square matrix whose number of rows and columns is equal to the number of elements of the vector x(t). res The calculation result of "x(t-Δt)" is expressed as a column vector with the same number of elements as the vector x(t).

[0043] u(t) is a vector that indicates the input value to the reservoir computing system at time step t. The vector u(t) is represented as a vertical vector. W in is a matrix that indicates the weighting coefficient for the input value. in "u(t)" denotes the weighting of the input value at time step t. The matrix W in The number of rows is equal to the number of elements in vector x(t), and the number of columns is equal to the number of elements in vector u(t). in The calculation result of "u(t)" is expressed as a column vector with the same number of elements as the vector x(t).

[0044] f indicates the activation function. res x(t-Δt)+W in The calculation result of "f(W res x(t-Δt)+W in u(t)" represents the application of an activation function to each element of this column vector. The output y(t) of the reservoir computing system at time step t is given by equation (2).

[0045]

number

[0046] W out is a matrix that indicates the weighting coefficients for calculating the output value from the values ​​of the intermediate nodes. out The number of rows is equal to the number of elements in vector y(t), and the number of columns is equal to the number of elements in vector x(t). out "x(t)" indicates that for each output node, the value of the intermediate node at time t is weighted and added. In the determination device 100, equation (2) or an equation obtained by modifying equation (2) corresponds to a determination regarding the determination target 920. However, the determination method by which the determination device 100 makes a determination regarding the determination target 920 is not limited to a specific method.

[0047] The state storage unit 140 stores past states of the state change unit 130. For example, the state storage unit 140 stores the states of the state change unit 130 at a predetermined number of immediately previous time steps. The combination of the state change unit 130 and the state memory unit 140 may be configured in the form of a reservoir layer.

[0048] Specifically, a storage node for storing the value of each node constituting state change unit 130 may be provided for each node constituting state change unit 130. In this case, the value of the weighting coefficient between nodes is set to 1, and the value of the node constituting state change unit 130 is output as is to the storage node. As a result, the value of the node constituting state change unit 130 at a certain time step is stored in the storage node at the next time step.

[0049] When it is desired to store the values ​​of the nodes constituting the state change unit 130 for a plurality of time steps, the storage nodes may be connected in series (in a daisy chain) for the number of time steps to be stored. By storing the node value in the next storage node in the series connection every time a time step passes, it is possible to store the values ​​of the nodes constituting the state change unit 130 for the number of time steps equal to the number of storage nodes connected in series.

[0050] The determination unit 150 makes a determination regarding the determination object 920 based on the state of the state change unit 130. In particular, the determination unit 150 makes a determination regarding the determination object 920 based on the state of the state change unit 130 at a time step derived from the reference time step identified by the reference time step identification unit 120. The determination unit 150 corresponds to an example of a determination means.

[0051] The determination unit 150 handles cases where the effect of the event of the determination object 920 appears only in data of some of the time steps by making a determination regarding the determination object 920 at a time step derived from the reference time step. For example, consider a case where the determination device 100 determines the presence or absence of a cardiac abnormality based on electrocardiogram data. In this case, depending on the correspondence between the electrocardiogram parts and the heart parts, it is considered that the part of the electrocardiogram that requires attention is limited in determining the presence or absence of a cardiac abnormality. For example, in the case of atrial fibrillation, a clear P wave may not appear in the electrocardiogram. In this case, it is considered effective to focus on the sampling data of the time step corresponding to the P wave in order to detect the abnormality.

[0052] In this way, the sampling data acquired by the sampling data acquisition unit 110 may include sampling data that indicates an abnormality and sampling data that does not show the influence of the abnormality, depending on the time step. In this case, let us consider a case where the judgment unit 150 continues to perform judgment at all time steps and output judgment results indicating an anomaly. In this case, the judgment unit 150 will output the same judgment result at both time steps where the sampling data indicates an anomaly and time steps where the sampling data is not affected by the anomaly. Since the judgment unit 150 always outputs the same judgment result regardless of whether the input data has an anomaly or not, it is considered that the judgment accuracy by the judgment unit 150 will be reduced.

[0053] Therefore, the reference time step identification unit 120 identifies a reference time step. The judgment unit 150 makes a judgment regarding the judgment target 920 based on the state of the state change unit 130 at a time step derived from the reference time step. This allows the judgment unit 150 to make a judgment for each time step in which the state of the judgment target 920 becomes a similar state. This allows the state change unit 130 to also become a similar state at each time step in which the judgment unit 150 makes a judgment, and it is expected that the judgment unit 150 can make a judgment with a relatively high degree of accuracy.

[0054] When the reference time step identification unit 120 detects sampling data corresponding to a characteristic portion of the time series data to be judged and identifies the reference time step, the reference time step does not necessarily coincide with the time step at which the influence of an anomaly appears in the sampling data. For example, when the reference time step identifying unit 120 detects a reference time step as a time step at which sampling data of an R wave is obtained, this reference time step is different from a time step at which sampling data of a P wave, at which the effect of atrial fibrillation appears, is obtained.

[0055] In order to deal with such differences in time steps, the determination device 100 reflects input data at past time steps in the state of the determination device 100 itself. Specifically, state change section 130 is configured recursively and assumes a state that reflects the past state of judgment target 920. Also, state storage section 140 stores the past states of state change section 130.

[0056] The judgment section 150 may make a judgment regarding the judgment target 920 based on the state of the state change section 130 at a time step prior to the reference time step, in addition to or instead of the state of the state change section 130 at the reference time step. Furthermore, the determination section 150 may make a determination regarding the determination target 920 based on the state of the state change section 130 at a time step that is future than the reference time step.

[0057] FIG. 4 is a diagram showing an example of a data flow in the determination device 100. As shown in FIG. 4, trigger signal generation unit 910 outputs a trigger signal to object to be determined 920. At time steps corresponding to the trigger signal from trigger signal generation unit 910, the state of object to be determined 920 changes.

[0058] The sampling data acquisition unit 110 measures the judgment target 920 for each time step and acquires sampling data. The sampling data acquisition unit 110 outputs the acquired sampling data to each of the reference time step identification unit 120 and the state change unit 130. As described above, the reference time step identification unit 120 may acquire a trigger signal from the trigger signal generation unit 910 in addition to or instead of the sampling data.

[0059] The state change unit 130 receives sampling data at each time step, and the state of the state change unit 130 changes at each time step. The state storage unit 140 stores the state of the state change unit 130 at past time steps. Furthermore, the reference time step identifying unit 120 identifies a reference time step based on either one or both of the sampling data and the trigger signal. The reference time step identifying unit 120 outputs a timing designation signal to the determining unit 150 at a time step derived from the identified reference time step. The timing designation signal is a signal that designates a time step at which the determining unit 150 makes a decision on the decision target 920.

[0060] In the example of FIG. 4, time step t represents the reference time step. Also, time step t+P + Δt represents the current time step in the example of FIG. + is a positive integer constant. In the example of FIG. 4, the combination of the state change unit 130 and the state storage unit 140 is - Δt to time step t+P + The state of the state change unit 130 at each time step up to Δt is stored. - is a positive integer constant. - The value of P + may be the same as or different from the value of

[0061] The determination unit 150 determines the time step t+P in response to the timing command signal from the reference time step determination unit 120. + At Δt, time step tP - Δt to time step t+P + A decision is made regarding the decision object 920 based on the state of the state change unit 130 at each time step up to Δt.

[0062] Time step tP - Δt to time step t+P +If the combined data indicating the state of state change unit 130 at each time step up to Δt is expressed as x^(t) and the judgment result by judgment unit 150 is expressed as y(t), the judgment by judgment unit 150 is expressed, for example, as in equation (3).

[0063]

number

[0064] "Classifier" indicates a classifier, and formula (3) shows an example of the case where the determination unit 150 performs a determination by class classification. For example, the determination unit 150 may determine whether or not there is a cardiac abnormality by classifying the data x^(t) into either a class indicating that there is a cardiac abnormality or a class indicating that there is no cardiac abnormality.

[0065] The method of classifying by the determination unit 150 is not limited to a specific method, and various methods capable of learning can be used. For example, the determination unit 150 may perform class classification by logistic regression or learning of the output layer of reservoir computing, but is not limited to these.

[0066] Furthermore, the determination made by the determination unit 150 is not limited to class classification, and various determinations can be made regarding the determination target 920. For example, the determination unit 150 may perform a quantitative determination regarding the determination target 920, such as quantitatively evaluating the possibility of a cardiac abnormality.

[0067] The data x^(t) is expressed as in equation (4).

[0068]

number

[0069] x(tP - Δt) to x(t+P +When each of the vectors Δt is expressed as a vertical vector, x^(t) may be expressed as a single vertical vector combining those vertical vectors. The time step of the state of the state change unit 130 used by the determination unit 150 for the determination is also called attention. In the example of FIG. 4, attention(t) is expressed as in equation (5).

[0070]

number

[0071] x(tP - x(t) to x(t-Δt) correspond to examples of the state of state change unit 130 at time steps earlier than the reference time step t. x(t) corresponds to an example of the state of the state change unit 130 at a reference time step. x(t+Δt) to x(t+P + Δt) correspond to examples of the state of state change unit 130 in time steps future than reference time step t.

[0072] As shown in equation (4), the judgment unit 150 may make a judgment based on the state of the state change unit 130 at each of a time step earlier than the reference time step t, the reference time step t, and a time step later than the reference time step t.

[0073] For example, the determination unit 150 may determine the presence or absence of a cardiac abnormality based on the state of the state change unit 130 in each time step of one cycle of heart rate fluctuation. As a result, if there is a cardiac abnormality, it is expected that a characteristic corresponding to the abnormality will appear in the state of the state change unit 130 in at least one or more time steps.

[0074] Furthermore, since the judgment unit 150 makes a judgment at a time step according to the timing designation signal from the reference time step identification unit 120, it is expected that similar characteristics will appear in the state of the state change unit 130 at time steps that have the same relative time from the reference time step for each period of heart rate fluctuation. In this respect, it is expected that the determining unit 150 can perform the determination with high accuracy.

[0075] Alternatively, the determination section 150 may make the determination based on the state of the state change section 130 only at the reference time step t and at a time step earlier than the reference time step t. The determination section 150 may make the determination based on the state of the state change section 130 only at the reference time step t and time steps future to the reference time step t.

[0076] The determination section 150 may make the determination based on the state of the state change section 130 only in time steps prior to the reference time step t. The determination unit 150 may make the determination based on the state of the state change unit 130 only at the reference time step t. The determination section 150 may make a determination based on the state of the state change section 130 only in time steps future than the reference time step t.

[0077] In order for the judgment section 150 to make a judgment based on a time step that is future than the reference time step t, either the reference time step identification section 120 or the judgment section 150, or both, may wait for a period of time. For example, the reference time step identification unit 120 identifies the time step t as the reference time step and then + Time step t+P has passed + A timing designation signal may be output to the determining section 150 at Δt.

[0078] Alternatively, the reference time step identification unit 120 may output a timing designation signal to the determination unit 150 at the reference time step t. Then, the determination unit 150 determines P + Time step t+P has passed + The determination may be made at Δt.

[0079] The decision unit 150 may learn decisions regarding the decision object 920 at time steps derived from the reference time step. FIG. 5 is a diagram showing an example of the configuration of the calculation system 1 when the determination device 100 performs learning for determining a determination target 920. As shown in FIG.

[0080] 5 differs from the case of FIG. 1 in that a learning control unit 210 is shown instead of the trigger signal generating unit 910 and the determination target 920. The learning control unit 210 may be configured as a device separate from the determination device 100, or may be configured as a part of the determination device 100. The combination of the determination device 100 and the learning control unit 210 is also referred to as a calculation system 1.

[0081] 5 shows an example in which the learning control unit 210 inputs training data including sampling data of an electrocardiogram to the determination device 100, and does not show a sampling data acquisition unit. The determination device 100 may or may not include the sampling data acquisition unit 110 during learning. In other respects, the configuration shown in FIG. 5 is similar to the configuration shown in FIG.

[0082] The learning control unit 210 controls the learning of the judgment by the judgment unit 150. In particular, the learning control unit 210 provides the judgment device 100 with training data for each time step, including sampling data of an electrocardiogram and correct answer label data indicating the correct answer of the judgment result, to cause the judgment unit 150 to learn the judgment. The training data may be stored in the learning control unit 210. Alternatively, the learning control unit 210 may acquire the training data from another device such as a database device.

[0083] The judgment unit 150 performs learning based on the state of the state change unit 130 in one or more time steps and the correct label data acquired from the learning control unit 210. In particular, the judgment unit 150 performs learning at a time step whose relative time from the time step at which the timing designation signal from the reference time step identification unit 120 was received is the same as that at the time of judgment. This is expected to enable the determination unit 150 to perform learning with high accuracy.

[0084] Of the state change section 130 and the determination section 150, only the determination section 150 may be the target of learning. This is expected to shorten the learning time. Furthermore, the determination unit 150 may be configured with a single-layer neural network, similar to the output layer of reservoir computing. This is expected to shorten the learning time. Also, even if the determination unit 150 has a simple configuration such as a single-layer neural network, it is expected to be able to perform learning with a relatively high degree of accuracy, similar to the case of reservoir computing.

[0085] In addition, in an experiment on the determination device 100, relatively good results were obtained. In the experiment, the determination device 100 was configured as software on a computer, and cardiac abnormality detection was performed using an electrocardiogram data set. 500 data were used as training data, and another 500 data were used as test data. Each data was used as one dimension of 2000 steps, and the first 200 steps of the 2000 steps were used as free run. Sinus rhythm data was used as non-abnormal data. Atrial fibrillation data was used as abnormal data. The state change unit 130 was configured as a reservoir layer with 50 neurons, and attention was set to the time steps of [-80, 60, ..., 160, 180] of the electrocardiogram data. The accuracy of anomaly detection was 3.4% during training and 6.2% during testing.

[0086] As described above, state change unit 130 receives input of data related to determination target 920 at each time step, and the state changes at each time step. Reference time step identification unit 120 identifies a reference time step that is a reference time step related to changes in the state of determination target 920. Determination unit 150 makes a determination related to the determination target based on the state of state change unit 130 at a time step corresponding to the reference time step.

[0087] Since the state of the state change unit 130 changes every time data related to the determination target 920 is input, even if a feature useful for determination appears only in a portion of the time-series data of the sampling data, the feature in the sampling data is expected to appear in the state of the state change unit 130. Moreover, the determination unit 150 makes a determination based on the state of the state change unit 130 at a time step corresponding to the reference time step, and can repeat determination at the same timing regarding the state change of the determination target 920. In this respect, the determination device 100 is expected to be able to perform a highly accurate determination. In this way, the determination device 100 can handle cases where part of the time-series data of the sampling data related to the determination target 920 has characteristics, without the need to use a convolutional neural network.

[0088] Furthermore, the reference time step identifying section 120 identifies the reference time step as the time step at which the data relating to the judgment target 920 satisfies a predetermined condition. As a result, the determination unit 150 performs determination based on the state of the state change unit 130 at a time step corresponding to the reference time step, and can repeat determination at similar timing regarding state changes of the determination target 920. In this respect, the determination device 100 is expected to be able to perform determination with high accuracy.

[0089] Furthermore, the reference time step identifying section 120 identifies the reference time step as a time step corresponding to the time step at which a signal is input to the object to be determined 920 from outside the object to be determined 920 . As a result, the determination unit 150 performs determination based on the state of the state change unit 130 at a time step corresponding to the reference time step, and can repeat determination at similar timing regarding state changes of the determination target 920. In this respect, the determination device 100 is expected to be able to perform determination with high accuracy.

[0090] Furthermore, the determination section 150 makes a determination regarding the determination target 920 based on the state of the state change section 130 at a time step that is earlier than the reference time step. As a result, even if a feature useful for judgment appears only in sampling data at a time step earlier than the reference time step among the time series data of the sampling data acquired by the sampling data acquisition unit 110, it is expected that the feature will be likely to appear in the state of the state change unit 130. In this respect, it is expected that the determination device 100 will be able to perform the determination with high accuracy.

[0091] Furthermore, the determination section 150 makes a determination regarding the determination target 920 based on the state of the state change section 130 at a time step that is future than the reference time step. As a result, even if a feature useful for judgment appears only in the sampling data at a time step future to the reference time step among the time series data of the sampling data acquired by the sampling data acquisition unit 110, it is expected that a feature corresponding to that feature will be likely to appear in the state of the state change unit 130. In this respect, it is expected that the determination device 100 will be able to perform the determination with high accuracy.

[0092] Furthermore, the judgment unit 150 performs learning regarding judgments regarding the judgment target 920 at time steps corresponding to the reference time steps. This allows the determination unit 150 to perform learning in the same time steps as in the case of judgment, in relation to the state change of the judgment target 920. This is expected to enable the determination unit 150 to perform learning with high accuracy.

[0093] 6 is a diagram showing another example of the configuration of the determination device according to the embodiment. In the configuration shown in FIG. 6, a determination device 610 includes a state change unit 611, a reference time step identification unit 612, and a determination unit 613. In this configuration, the state change unit 611 receives input of data related to the object to be determined at each time step, and the state changes at each time step. The reference time step identification unit 612 identifies a reference time step that is a reference time step related to the change in the state of the object to be determined. The determination unit 613 performs a determination on the object to be determined based on the state of the state change unit 611 at a time step corresponding to the reference time step.

[0094] Since the state of the state change unit 611 changes every time data related to the object to be determined is input, even if a feature useful for determination appears only in a portion of the time-series data of the sampling data related to the object to be determined, the feature in the sampling data is expected to appear in the state of the state change unit 611. Moreover, the determination unit 613 makes a determination based on the state of the state change unit 611 at a time step corresponding to the reference time step, and can repeat determination at the same timing regarding the state change of the object to be determined. In this respect, the determination device 610 is expected to be able to perform a highly accurate determination. In this way, the determination device 610 can handle cases where part of the time-series data of the sampling data related to the determination target has characteristics, without the need to use a convolutional neural network. The state change unit 611 corresponds to an example of a state change means, the reference time step identification unit 612 corresponds to an example of a reference time step identification means, and the determination unit 613 corresponds to an example of a determination means.

[0095] Fig. 7 is a flowchart showing an example of a processing procedure in a determination method according to an embodiment. Fig. 7 shows an example of a processing procedure in which an apparatus receives input of data related to a determination target at each time step and includes a state change unit whose state changes at each time step to perform a determination. The determination method shown in Fig. 7 includes identifying a reference time step (step S611) and performing a determination (step S612).

[0096] In specifying a reference time step (step S611), a reference time step that is a reference time step regarding a change in the state of the determination target is specified. In making the judgment (step S612), a judgment is made regarding the judgment target based on the state of the state changing means at the time step corresponding to the reference time step. The state change unit corresponds to an example of a state change means.

[0097] Since the state of the state change unit changes every time data related to the object of judgment is input, even if the feature useful for judgment appears only in a part of the time series data of the sampling data related to the object of judgment, the feature in the sampling data is expected to appear in the state of the state change unit. Furthermore, by performing judgment based on the state of the state change unit at the time step corresponding to the reference time step in the process of step S612, judgment can be repeated at the same timing regarding the state change of the object of judgment. In this respect, it is expected that the judgment can be performed with high accuracy according to the judgment method shown in FIG. In this way, the determination method shown in FIG. 7 can handle cases where part of the time-series data of the sampling data related to the determination target has characteristics, without the need to use a convolutional neural network.

[0098] FIG. 8 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. In the configuration shown in FIG. 8, a computer 700 includes a CPU (Central Processing Unit) 710, a main memory device 720, an auxiliary memory device 730, and an interface 740.

[0099] Any one or more of the above-mentioned determination device 100 and the determination device 610 or a part thereof may be implemented in the computer 700. In this case, the operation of each of the above-mentioned processing units is stored in the auxiliary storage device 730 in the form of a program. The CPU 710 reads the program from the auxiliary storage device 730, develops it in the main storage device 720, and executes the above-mentioned processing according to the program. The CPU 710 also secures a storage area corresponding to each of the above-mentioned storage units in the main storage device 720 according to the program. Communication between each device and other devices is executed by the interface 740 having a communication function and performing communication according to the control of the CPU 710.

[0100] When the determination device 100 is implemented in a computer 700, the operations of the sampling data acquisition unit 110, the reference time step identification unit 120, the state change unit 130, the state storage unit 140, and the determination unit 150 are stored in the form of a program in the auxiliary storage unit 730. The CPU 710 reads the program from the auxiliary storage unit 730, loads it into the main storage unit 720, and executes the above-mentioned processing according to the program.

[0101] Furthermore, the CPU 710 allocates a storage area for processing of the determination device 100, such as the state storage unit 140, in the main storage device 720 according to the program. Communication between the determination device 100 and other devices is performed by an interface 740 having a communication function and performing communication under the control of the CPU 710. Interaction between the determination device 100 and a user is performed by the interface 740 having a display device and an input device, displaying various images under the control of the CPU 710, and accepting user operations.

[0102] When the determination device 610 is implemented in the computer 700, the operations of the state change unit 611, the reference time step identification unit 612, and the determination unit 613 are stored in the form of a program in the auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing in accordance with the program.

[0103] Furthermore, the CPU 710 allocates a storage area for processing of the determination device 610 in the main storage device 720 according to the program. Communication between the determination device 610 and other devices is performed by an interface 740 having a communication function and performing communication under the control of the CPU 710. Interaction between the determination device 610 and a user is performed by the interface 740 having a display device and an input device, displaying various images under the control of the CPU 710, and accepting user operations.

[0104] Note that a program for executing all or part of the processing performed by the determination device 100 and the determination device 610 may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed to perform processing of each part. Note that the term "computer system" as used herein includes software such as an OS (Operating System) and hardware such as peripheral devices. Furthermore, the term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs (Read Only Memory), and CD-ROMs (Compact Disc Read Only Memory), as well as storage devices such as hard disks built into computer systems. The above-mentioned program may be one for implementing part of the above-mentioned functions, or may be a differential program (differential file) that can implement the above-mentioned functions in combination with a pre-installed program in the computer system.

[0105] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and the present invention also includes designs and the like within the scope of the gist defined in the appended claims. [Industrial Applicability]

[0106] The present invention may be applied to a determination device, a determination method, and a recording medium. [Explanation of symbols]

[0107] 1. Computing System 100, 610 Judgment device 110 Sampling data acquisition unit 120, 612 Reference time step identification part 130, 611 State change section 140 Status memory unit 150, 613 Judgment section 210 Learning control unit 910 Trigger signal generator 920 Judgment subject

Claims

1. A state changing means for receiving an input of data relating to a target object whose state changes periodically at each time step, and changing the state at each said time step; A reference time step specifying means for specifying a reference time step which is a reference time step in one period of the change in the state of the object to be judged; a determination means for making a determination regarding the determination object based on a state of the state change means at a time step derived from the reference time step; A determination device comprising:

2. the reference time step identification means identifies the reference time step as a time step at which the data to be judged that satisfies a predetermined condition is input; The determination device according to claim 1 .

3. the reference time step specifying means specifies the reference time step as a time step at which a signal is input to the object to be determined from outside the object to be determined. The determination device according to claim 1 or 2.

4. the determination means makes a determination regarding the determination object based on a state of the state changing means at a time step that is earlier than the reference time step. The determination device according to claim 1 .

5. the determination means makes a determination regarding the determination object based on a state of the state change means at a time step future to the reference time step. The determination device according to claim 1 .

6. The judgment means performs learning on judgment regarding the judgment target at a time step derived from the reference time step. The determination device according to claim 1 .

7. An apparatus having a state change means for receiving an input of data relating to a target object whose state changes periodically at each time step, the state change means changing the state at each time step, Identifying a reference time step which is a reference time step in one period of the change in the state of the object to be determined; making a judgment regarding the judgment object based on a state of the state changing means at a time step derived from the reference time step; A determination method comprising:

8. An apparatus for executing a program, comprising: a state changing means for receiving an input of data relating to a determination target whose state changes periodically at each time step, and changing the state at each time step, Identifying a reference time step which is a reference time step in one period of the change in the state of the object to be determined; making a judgment regarding the judgment object based on a state of the state changing means at a time step derived from the reference time step; A program for executing the above.

Images