Calculation device, detection device, calculation method, and computer program

Abstract

The calculation device (1) is a calculation device that corrects RRI data, which is data on the interval between adjacent R waves of an electrocardiogram signal, and detects a range to be corrected that includes noisy RRI data from the RRI data obtained in a time series, and corrects the RRI data in the range to be corrected by inputting the RRI data in the range to be corrected to a DAE (denoising autoencoder).

Description

Technical Field

[0001] The present disclosure relates to an arithmetic device, a detection device, an arithmetic method, and a computer program. This application claims priority based on Japanese Application No. 2018-090592 filed on May 9, 2018, and incorporates by reference all the descriptions described in the Japanese application.

Background Art

[0002] It is known that the interval between adjacent R waves (RRI: R-R Interval) on an electrocardiogram is related to autonomic nerve activity. The variation in RRI is called heart rate variability (HRV). Generally, the greater the HRV, the higher the responsiveness to internal or external stimuli, and the smaller the HRV, the lower the responsiveness.

[0003] Utilizing this feature, various health monitoring technologies using HRV analysis have been developed. For example, health monitoring technologies such as detection of the onset of epilepsy (Japanese Patent Application Laid-Open No. 2015-112423) and detection of apnea (Japanese Patent Application Laid-Open No. 2016-214491) are known.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

[0005] In health monitoring technologies using HRV analysis, accurate detection of R waves is a prerequisite. Therefore, if R waves are not accurately detected, there is a problem that the accuracy of HRV analysis decreases. The decrease in the accuracy of HRV analysis leads to a decrease in the performance of health monitoring.

[0006] One of the factors that reduce the reliability of R-wave detection is the false detection or missed detection of R-waves. Another example of a factor that reduces the reliability of R-wave detection is premature ventricular contraction (PVC), which is a common arrhythmia that can occur even in healthy individuals. This is because when PVC occurs, artifacts are mixed into the RRI.

[0007] According to an embodiment, the arithmetic unit is an arithmetic unit that corrects RRI data, which is data on the intervals between adjacent R-waves of an electrocardiogram signal. The arithmetic unit detects a correction target range including RRI data having noise among the RRI data obtained in time series, and corrects the RRI data in the correction target range by inputting the RRI data in the correction target range into a DAE (denoising autoencoder).

[0008] According to another embodiment, the detection device is equipped with the above arithmetic unit, and detects specific biological information based on the RRI data obtained in time series including the RRI data in the correction target range.

[0009] According to another embodiment, the arithmetic method is an arithmetic method for correcting RRI data, which is data on the intervals between adjacent R-waves of an electrocardiogram signal. The arithmetic method includes a step of detecting a correction target range including RRI data having noise among the RRI data obtained in time series, and a step of correcting the RRI data in the correction target range by inputting the RRI data in the correction target range into a DAE.

[0010] According to another embodiment, the computer program is a program that causes a computer to function as an arithmetic unit for correcting RRI data, which is data on the intervals between adjacent R-waves of an electrocardiogram signal. The computer program causes the computer to execute a step of detecting a correction target range including RRI data having noise among the RRI data obtained in time series, and a step of correcting the RRI data in the correction target range by inputting the RRI data in the correction target range into a DAE. Further details will be described as embodiments below.

Brief Description of Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Figure 22

Figure 23

Figure 24

Figure 25

Figure 26

Figure 27

Figure 28

[0012] [1. Overview of the arithmetic device, detection device, arithmetic method, and computer program]

[0013] (1) The arithmetic unit included in this embodiment is an arithmetic unit that corrects RRI data, which is data on the intervals between adjacent R waves of an electrocardiogram signal. It detects a correction target range including RRI data having noise among the RRI data obtained in time series, and corrects the RRI data in the correction target range by inputting the RRI data in the correction target range into a DAE (denoising autoencoder). Noise (artifact) is, for example, premature ventricular contraction (PVC) and missing detection of R waves.

[0014] The variation in RRI data is heart rate variability (HRV), and HRV analysis is used in various health monitoring technologies. Therefore, by correcting the RRI data in the correction target range including the RRI data having noise, the accuracy of HRV analysis can be improved. As a result, the performance of the health monitoring technology using HRV analysis can be improved.

[0015] (2) Preferably, the detection includes detecting a correction target range including RRI data in which the noise is due to PVC. In this case, since the range of the RRI data including the noise due to PVC is set as the correction target range and corrected, RRI data obtained by removing the noise due to PVC from the measured RRI data can be obtained. As a result, the accuracy of HRV analysis of the corrected RRI data is improved.

[0016] (3) Preferably, the activation function of the DAE is ReLU (Rectified Linear Unit). Through the implementation by the inventors, when the RRI data contains noise due to PVC, applying a DAE with the ReLU activation function to the RRI data results in a higher degree of tracking of the corrected RRI data to the RRI data without noise due to PVC than when applying other activation functions. That is, when the RRI data contains noise due to PVC, using ReLU as the activation function provides higher correction accuracy. Therefore, the accuracy of HRV analysis of the corrected RRI data can be further improved.

[0017] (4) Preferably, the detection includes detecting a correction target range including RRI data where the noise is caused by the missing detection of the R wave. In this case, since the range of RRI data including noise due to the missing detection of the R wave is set as the correction target range and corrected, RRI data with this noise removed from the measured RRI data can be obtained. As a result, the accuracy of HRV analysis of the corrected RRI data is improved.

[0018] (5) Preferably, the activation function of the DAE is the sigmoid function. Through the implementation by the inventors, when the RRI data contains noise due to the missing detection of the R wave, applying a DAE with the sigmoid activation function to the RRI data results in a higher degree of tracking of the corrected RRI data to the RRI data without noise due to the missing detection of the R wave than when applying other activation functions. That is, when the RRI data contains noise due to the missing detection of the R wave, using the sigmoid function as the activation function provides higher correction accuracy. Therefore, the accuracy of HRV analysis of the corrected RRI data can be further improved.

[0019] (6) Preferably, when the arithmetic unit corrects the RRI data in the correction target range, depending on whether the detected correction target range is a correction target range including RRI data in which the noise is due to PVC or a correction target range including RRI data in which the noise is caused by the missing detection of the R wave, the activation function of the DAE is switched between ReLU and the sigmoid function. Thereby, optimal correction can be performed according to the noise included in the RRI data.

[0020] (7) Preferably, the number of RRI data input to the DAE when correcting the RRI data in the correction target range matches the number of RRI data included in the correction target range. The number of RRI data included in the correction target range is, for example, when the noise is due to PVC and the noise appears in two consecutive beats, it is a total of four beats, one beat before and after those two beats. Thereby, correction in a necessary range can be enabled without unnecessarily enlarging the correction target range.

[0021] (8) Preferably, the detection includes detecting a correction target range including RRI data in which the noise is due to PAC. In this case, since the range of the RRI data including the noise due to PAC is set as the correction target range and corrected, RRI data with the noise due to PAC excluded from the measured RRI data can be obtained. As a result, the accuracy of the HRV analysis of the corrected RRI data is improved.

[0022] (9) The detection device included in the present embodiment is equipped with the arithmetic unit described in any one of (1) to (7), and detects a specific biological state based on the RRI data obtained in a time series including the RRI data in the correction target range. The specific biological state is, for example, a sign of an epileptic seizure, an apnea state, a dozing state, etc. By being equipped with the arithmetic unit described in any one of (1) to (7), this detection device exhibits the same effects as the arithmetic unit described in (1) to (7). That is, the detection accuracy of the biological state as described above using HRV analysis can be improved.

[0023] (10) The arithmetic method included in this embodiment is an arithmetic method executed in the arithmetic unit described in any one of (1) to (7). Therefore, this arithmetic method has the same effects as the arithmetic unit described in (1) to (7).

[0024] (11) The computer program included in this embodiment is a program that causes a computer to function as the arithmetic unit described in (1) to (7). Therefore, this computer program has the same effects as the arithmetic unit described in (1) to (7). The computer program is stored in a computer-readable non-transitory storage medium.

[0025] [2. Examples of Arithmetic Unit, Detection Device, Arithmetic Method, and Computer Program]

[0026] [First Embodiment]

[0027] <Configuration of Detection Device> FIG. 1 is a diagram showing an outline of the configuration of the detection device according to this embodiment. Referring to the figure, the detection device 100 includes an arithmetic unit 1 and a heart rate monitor 2. The detection device according to this embodiment is a detection device that detects a specific biological state based on the heart rate of a subject. The specific biological state is, for example, a premonition of epilepsy, apnea, drowsiness, etc.

[0028] The arithmetic unit 1 and the heart rate monitor 2 are communicable and communicate wirelessly as an example. The wireless communication is, for example, short-range wireless communication such as Bluetooth (registered trademark), communication via the Internet, etc.

[0029] <Heart Rate Monitor> The heart rate monitor 2 is a small and lightweight wearable device attached to the body of the subject P to measure the heart rate of the subject P. A plurality (three in FIG. 1) of electrodes 21 attached to the body surface of the subject P are connected to the heart rate monitor 2. The three electrodes 21 are, for example, a plus electrode, a minus electrode, and a ground electrode.

[0030] Figure 2(a) is a diagram showing an example of an electrocardiogram signal. The vertical axis in Figure 2(a) represents potential, and the horizontal axis represents time. When measuring the heartbeat using electrode 21, a potential change consisting of P to T waves as shown in Figure 2(a) appears periodically. The peak with the highest potential in the potential change of a unit period is called the R wave, and the heart beats at the timing of the R wave. The heartbeat measuring device 2 transmits R wave data indicating the R wave to the arithmetic unit 1.

[0031] Figure 2(b) shows the R wave data corresponding to the electrocardiogram signal in Figure 2(a). As shown in Figure 3, the R wave data is data representing a rectangular pulse train in which a period corresponding to the R wave in the electrocardiogram signal (a period during which the signal intensity I exceeds a predetermined intensity threshold Ith) is set to "1" and the other periods are set to "0".

[0032] <Arithmetic unit> The arithmetic unit 1 functions as a detection device for detecting a specific biological state such as a sign of epileptic seizure based on the R wave data transmitted from the heartbeat measuring device 2, and also functions as a correction device for correcting the RRI described later. The arithmetic unit 1 is, for example, a communication terminal such as a smartphone or a personal computer.

[0033] Figure 3 is a block diagram showing an example of the configuration of the arithmetic unit 1. Referring to the figure, the arithmetic unit 1 includes a processing unit 10 composed of a dedicated microcomputer or the like. The arithmetic unit 1 also has a memory 12, and a program 121 for performing arithmetic operations by the processing unit 10 is stored in the memory 12.

[0034] The arithmetic unit 1 has a communication unit 13 for communicating with the heartbeat measuring device 2. The communication unit 13, for example, receives a radio signal indicating the R wave transmitted from the heartbeat measuring device 2, generates data indicating the R wave from the received radio signal, and inputs the data to the processing unit 10. The communication unit 13 may be a reading device that accesses a recording medium in which data indicating the R wave is recorded and reads out the data from the recording medium. Obtaining data includes receiving data by communication and reading out data from a recording medium.

[0035] <Impact on HRV Analysis of PVC From the R-wave data, RRI data, which is time-series data of the RRI variable indicating the intervals between R-waves, is obtained. Artifacts (noise) may be mixed into the RRI data. The causes of the artifacts mixed into the RRI data include missed detection of R-waves and the occurrence of premature ventricular contractions (PVCs), which are arrhythmias. In the first embodiment, detection and correction are to be performed for the mixing of artifacts caused by PVCs (hereinafter referred to as PVC artifacts).

[0036] The indices used for HRV analysis (hereinafter referred to as HRV indices) include time-domain indices and frequency-domain indices. The time-domain indices include the following five indices. The time-domain indices are directly calculated from the RRI data. 1) meanNN: Mean value of RRI 2) SDNN: Standard deviation of RRI 3) Total Power: Variance of RRI 4) RMSSD: Root mean square of the differences between adjacent RRIs 5) NN50: Number of times the difference between adjacent RRIs exceeds 50 ms

[0037] The frequency-domain indices include the following three indices. The frequency-domain indices are calculated from the power spectral density (PSD) of the RRI data. Since the RRI data is not sampled at equal intervals, it is necessary to sample in order to obtain the PSD. The PSD is calculated using an autoregressive (AR) model or Fourier transform from the resampled RRI data. 1) LF: Power of the low frequency (0.04 - 0.15 Hz) of the PSD 2) HF: Power of the high frequency (0.15 - 0.40 Hz) of the PSD 3) LF / HF: Ratio of LF to HF

[0038] FIG. 4 is a diagram showing an example of an electrocardiogram signal when PVC occurs. The vertical axis of the figure indicates potential, and the horizontal axis indicates time. Comparing FIG. 4 with FIG. 2(a) (the electrocardiogram signal when PVC does not occur), it can be seen that significant fluctuations have occurred in the RRI, which was generally equidistant when PVC did not occur, due to the occurrence of PVC.

[0039] When significant fluctuations occur in the RRI, each HRV index calculated from the RRI is also affected. FIGS. 5 and 6 are diagrams showing SDNN and LF / HF among the above HRV indices calculated from RRI data mixed with artificial PVC artifacts. The horizontal axis of the figure indicates the passage of time, and the vertical axis indicates the value of the HRV index. The timing t in the figure indicates the timing when the PVC artifact was mixed. The dotted line and the solid line in the figure indicate the HRV indices calculated from the original RRI data without the PVC artifact and the HRV indices calculated from the PVC-RRI, respectively.

[0040] From the calculation results shown in FIGS. 5 and 6, it can be seen that changes have occurred in each HRV index immediately after the PVC artifact was mixed. Therefore, if the HRV index changed by the mixing of the PVC artifact is used in HRV analysis, the analysis accuracy may be reduced.

[0041] <Functional Configuration of the Processing Unit of the Arithmetic Unit> In the present embodiment, the processing unit 10 of the arithmetic unit 1 detects PVC artifacts mixed in the RRI data obtained from the measurement results. Then, the RRI data obtained from the measurement results is corrected so as to be RRI data from which the influence of the detected PVC artifacts has been excluded. The processing unit 10 uses the corrected RRI data for detecting a predetermined biological state.

[0042] To execute the above processing, the processing unit 10 includes an RRI calculation unit 101, a detection unit 102, a correction unit 103, and a detection unit 104 as functions realized by reading and executing the program 121 from the memory 12.

[0043] The RRI calculation unit 101 generates RRI data based on the R-wave data input from the communication unit 13. The RRI calculation unit 101 calculates, as an RRI variable, the time interval between the falling times of two temporally adjacent rectangular pulses from the R-wave data as shown in Fig. 2(b), and generates RRI data by arranging the calculated RRI variables in time series.

[0044] The detection unit 102 detects the mixing of artifacts (noise) into the RRI data. When the mixing of artifacts into the RRI data is detected, the correction unit 103 performs correction to eliminate the influence of the artifacts from the RRI data. The detection method and the correction method will be described later.

[0045] The detection unit 104 detects a predetermined biological state from the corrected RRI data. The detection method in the detection unit 104 may be a known detection method. For example, when the predetermined biological state is a sign of an epileptic seizure, the method described in Japanese Patent Application Laid-Open No. 2015-112423 can be adopted. Also, for example, when the predetermined biological state is an apnea state, the method described in Japanese Patent Application Laid-Open No. 2016-214491 can be adopted.

[0046] When the detection unit 104 detects a specific biological state, the processing unit 10 inputs a signal indicating that the specific biological state has been detected to the communication unit 13, and instructs to output the signal to a predetermined transmission destination. The predetermined transmission destination is a display or a speaker (not shown) connected to the arithmetic unit 1, and the processing unit 10 may notify that the specific biological state has been detected by them. Also, for example, the predetermined transmission destination is a predetermined address stored in association with the subject P, and the communication terminal associated with the subject P may be notified that the specific biological state has been detected.

[0047] <Detection method for PVC artifacts> The MSPC-SSA method is used for detecting PVC artifacts in the detector 102. MSPC-SSA is a method that introduces the monitoring index used in the multivariate statistical process control (MSPC) of the multivariate anomaly detection method into the singular spectrum analysis (SSA) of the anomaly detection method.

[0048] In MSPC, principal component analysis (PCA) is used for feature extraction. PCA is a method for reducing the dimensionality of data, and it creates new variables called principal components through linear combinations of multiple variables.

[0049] Figure 7A is a diagram showing the relationship between two HRV index variables (variable 1, variable 2) and two principal components (first principal component, second principal component), and Figure 7B is a diagram for explaining the relationship between the two HRV index variables, two principal components, and the T2 statistic. As the HRV index variables, two can be selected from the above HRV indexes. For example, meanNN and LF can be selected.

[0050] In PCA, one direction in the space representing the HRV index data is set as the first principal component, and one direction in the space orthogonal to the first principal component is set as the second principal component. By repeating this procedure, multiple principal components are set. Here, the first principal component is set in the direction where the variance of the principal component scores is the largest. Note that the "principal component score" is the coordinate on the principal component axis, that is, the value obtained by projecting the HRV index data into the space spanned by the principal component. In this case, the values obtained by projecting the HRV index data onto the first principal component and the second principal component become the principal component scores.

[0051] When using PCA, the control limit CL1 shown in Fig. 7A can be set. However, it is not easy to determine the deviation from this control limit CL1 in a multi-dimensional space. Therefore, in the method for detecting PVC artifacts according to this embodiment, MSPC that uses Hotelling's T2 statistic and Q statistic is adopted.

[0052] Hotelling's T2 statistic is the Mahalanobis distance between the sample and the origin in the subspace spanned by the principal components, and is calculated from the principal component scores obtained by principal component analysis. Hotelling's T2 statistic is used for anomaly detection in the subspace spanned by the principal components. In the Hotelling's T2 statistic, since the distance from the origin is normalized for each principal component direction, the control limit CL1 shown in Fig. 7A can be transformed into a circular control limit CL2 as shown in Fig. 7B.

[0053] The Q statistic is the squared distance between the sample and the subspace spanned by the principal components, and is calculated from the principal component scores obtained by principal component analysis. The Q statistic is used for anomaly detection in the orthogonal complement subspace of the subspace spanned by the principal components.

[0054] In MSPC, the Hotelling's T2 statistic and the Q statistic are monitored simultaneously, and when either one exceeds the control limit, it is determined as an anomaly. That is, the control limits of the Hotelling's T2 statistic and the Q statistic are used as the thresholds for determining whether or not the RRI column contains outliers.

[0055] SSA is a method for creating a Hankel matrix from univariate time series data and performing anomaly detection using the features extracted by singular value decomposition. For the RRI time series data x = {x1, x2,..., xT}, the Hankel matrix is shown by Equation (1) in Fig. 8A. Each element of the RRI time series data x is RRI data, and the subscript indicates the measurement order. T indicates the total number of measured RRI data.

[0056] Each row yi (i = 1, …, n) of the Hankel matrix is a partial RRI column. m is the number of RRI time series data constituting the partial RRI column, and m < T. In MSPC-SSA, regarding the partial RRI column yi as a sample, the control limits T2￣ and Q￣ of the Hotelling's T2 statistic and the Q statistic, respectively, which are used as thresholds for determining whether or not the partial RRI column yi contains outliers, are determined.

[0057] FIG. 9 is a flowchart showing an example of an algorithm for constructing a model used by the detection unit 102 to detect PVC artifacts using MSPC-SSA. In the flowchart of FIG. 9, n represents the total number of RRI columns used for model construction, and the variable I represents the number of the RRI column being processed.

[0058] FIG. 8B is a diagram for specifically explaining the construction of the model shown in FIG. 9. In FIG. 8B, each frame shows RRI time series data, which are arranged in time series in the order from left to right. Here, an example where the number m of RRI time series data constituting the partial RRI column is 4 (m = 4) is shown. The frames with solid lines represent the RRI data constituting the partial RRI column, and the frames with dotted lines represent the RRI data not constituting the partial RRI column.

[0059] Referring to FIG. 9, first, the variable I is incremented by 1 (step S101), and the I-th RRI column of the Hankel matrix shown in Equation (1) of FIG. 8A is created (step S103). Steps S101 and S103 are repeatedly executed (step S106) until the total number n of RRI columns used for model construction is reached (NO in step S105).

[0060] In the example of FIG. 8B, by repeating steps S101 and S103, an RRI column consisting of x1 to x4 as the first RRI column, an RRI column consisting of x2 to x5 as the second RRI column, …, and an RRI column consisting of x(m - 3) to xm as the n-th RRI column are created.

[0061] When creating the columns of the Hankel matrix for the total number n of RRI columns used for model construction (YES in step S105), all the columns from the 1st to the nth are combined into one matrix (step S107), and the Hankel matrix X is created by normalizing it so that the mean is 0 and the variance is 1 (step S109).

[0062] For the Hankel matrix X obtained in step S109, by performing singular value decomposition to calculate the principal components and the left singular vectors (step S111), the control limits T2￣ (threshold α1) and Q￣ (threshold β1) of the Hotelling's T2 statistic and the Q statistic respectively are determined (step S113).

[0063] In MSPC - SSA, for each partial RRI column yi of the model (hereinafter referred to as the normal model) constructed according to the flowchart of FIG. 9, whether at least one of the Hotelling's T2 statistic (value α) or the Q statistic (value β) exceeds the control limit (thresholds α1, β1) determines whether it is normal or abnormal, that is, whether there is a PVC artifact. When the partial RRI column yi is determined to be abnormal, it is considered that any of the elements included in the partial RRI column yi contains a PVC artifact. That is, when at least one of the values α and β of a plurality of consecutive partial RRI columns exceeds the control limit (thresholds α1, β1), it is considered that the elements commonly included in them contain a PVC artifact.

[0064] Using FIG. 8B, a method for specifying the range containing the PVC artifact will be described. Sporadic PVC affects the values of 2 beats of the RRI data. In FIG. 8B, the hatched frame shows the RRI time - series data affected by the PVC artifact. Specifically, an example in which the RRI data x4 and x5 are affected by the PVC artifact is shown.

[0065] In the example of FIG. 8A, in the normal model, each partial RRI sequence consists of a specified number (4 in the example of FIG. 8A) of consecutive RRI time series data. From the first to the nth, for each partial RRI sequence, the first RRI time series data is shifted one by one along the time series. Therefore, at least one of the RRI data x4 and x5 affected by the PVC artifact appears in the partial RRI sequences from the first to the fifth.

[0066] That is, two beats affected by PVC are included in m - 1 consecutive (the second to the fourth in FIG. 8A) partial RRI sequences. Therefore, in the PVC artifact detection method according to the present embodiment, when m - 1 consecutive partial RRI sequences are determined to be abnormal, it is determined that the PVC artifact has been detected.

[0067] Also, only one beat of the PVC artifact is included in the partial RRI sequences before and after the above m - 1 partial RRI sequences (the first and fifth in FIG. 8A). If these two partial RRI sequences before and after are also determined to be abnormal, up to m + 1 partial RRI sequences (the first to the fifth in FIG. 8A) are continuously determined to be abnormal.

[0068] In MSPC - SSA, a threshold τ1 of the number τ of partial RRI sequences in which at least one of the values α and β continuously exceeds the control limit is specified in advance. When the number τ is equal to or greater than the threshold τ1, it is determined that the target partial RRI sequence contains a PVC artifact. The threshold τ1 depends on the length m of the partial RRI sequence, that is, the number of columns m of the Hankel matrix.

[0069] If the last partial RRI sequence determined to be abnormal is yj = {xj,..., xj + m - 1}, if the values α and β of the partial RRI sequence continuously exceed the control limit for m + 1, the first beat of the PVC artifact is x^j - 1. If it continuously exceeds the control limit for m - 1, the first beat of the PVC artifact is x^j. That is, either of these two beats coincides with the first beat of the PVC artifact. Therefore, in MSPC - SSA, it is determined that x^j - 1 and x^j contain the PVC artifact.

[0070] FIG. 10 is a flowchart showing an algorithm for the detection unit 102 to detect PVC artifacts in real time using MSPC-SSA. In the algorithm of FIG. 10, it is assumed that the PVC detection model has already been created by the process of FIG. 9. The variable t represents the number at which the RRI data is acquired, and the maximum value of the number of measurements of the RRI data is set as MAX. The RRI data x_t represents the t-th RRI data, and the partial RRI column y_t represents an m-dimensional partial RRI column whose first element is x_t.

[0071] Referring to FIG. 10, first, the counter τ and the variable t are initialized (step S201). Next, the t-th RRI is newly measured and recorded in the buffer as RRI data according to the FIFO (First In First Out) method (step S203).

[0072] Next, from the buffer, the RRI data from x_(t - m + 1) to x_(t - 1) measured previously is read out, and the newly obtained RRI data x_t is added to create a partial RRI column y_m+1 with x_t as the final element (step S205). Then, the Hotelling's T2 statistic (value α) and the Q statistic (value β) of the partial RRI column y_m+1 are calculated (step S207).

[0073] If at least one of the values α and β exceeds the control limit (threshold values α1 and β1) (YES in step S209), the partial RRI column y_m+1 may contain PVC artifacts. Therefore, in this case, the counter τ is incremented by 1 (step S211).

[0074] If neither of the values α and β exceeds the management limits (thresholds α1 and β1) (NO in step S209), it is highly likely that the partial RRI sequence y_m+1 does not contain PVC artifacts. In this case, if the counter τ has reached the threshold τ1 (YES in step S215), up to the partial RRI sequence y_m, a number corresponding to the threshold τ1 of partial RRI sequences that may contain PVC artifacts are consecutive, and the possibility of the partial RRI sequence y_m+1 containing PVC artifacts has disappeared. That is, in the example of FIG. 8B, the threshold τ1 is 5, and the partial RRI sequence y_m+1 corresponds to the partial RRI sequence y6. Therefore, in this case, it is determined that the RRI data x_(t-m) and the RRI data x_(t-m+1) contain PVC artifacts (step S217).

[0075] Note that when the counter τ is less than the threshold τ1 (NO in step S215), until it is determined that PVC artifacts are included, partial RRI sequences in which at least one of the values α and β exceeds the management limits (thresholds α1 and β1) are not consecutive. That is, PVC artifacts are not included. Therefore, in this case, the counter τ is initialized (step S219).

[0076] Thereafter, the variable t is incremented to measure the next RRI data (step S213), and the above processing is repeated until the variable t reaches the maximum value MAX (NO in step S221) to detect PVC artifacts. When the variable t reaches the maximum value MAX of the number of measurements (YES in step S221), a series of processing ends.

[0077] <RRI Correction Method> In the RRI data correction method in the correction unit 103, SAE (sparse autoencoder) and DAE (denoising autoencoder), which is a neural network for noise removal, are used in combination. AE (autoencoder) is a dimensionality reduction method and a feature extraction method using a neural network. Since AE makes the output as equal as possible to the input, learning is performed to minimize the reconstruction error between the input and the output.

[0078] At this time, if the activation function is the identity mapping in the intermediate layer and the output layer, the AE coincides with the principal component analysis (PCA), which is a typical dimensionality reduction method. In order to obtain the dimensionally compressed features, the number of units in the hidden layer should be smaller than the number of input variables. However, by introducing a regularization term, the number of units in the hidden layer can be made larger than the number of input variables. This method is called SAE.

[0079] The DAE is a neural network with the same structure as the AE. During learning, noise is added to the input, and learning is performed to obtain the same output as the input before the noise is added. Since learning is performed with noise added, the DAE is considered to have noise removal ability.

[0080] In order to remove PVC artifacts by the DAE, the data passed to the input layer and the output layer during the learning of the AE is (noise-free) RRI data, and furthermore, noise caused by PVC is passed to the input layer. As a result, the DAE used to remove PVC artifacts is learned to remove noise caused by PVC from the input RRI data and output RRI data that does not contain noise caused by PVC. The correction method for removing noise by the DAE is called DAE-RM (DAE-based RRI modification).

[0081] When removing PVC artifacts included in the RRI data by the DAE, the RRI data to be corrected by the DAE is input. The RRI data input to the DAE is the range to be corrected among the RRI data continuous in time series. The range to be corrected is the RRI data changed by the PVC artifacts detected by the above detection method and the RRI data of a preset correction width before and after it. That is, detecting PVC artifacts according to the algorithm shown in FIG. 10 is detecting the range to be corrected.

[0082] FIG. 8C is a diagram showing an outline of the configuration of the DAE and an outline of data input to and output from the DAE. Referring to FIG. 8C, when data x1, x2, … xn are input to the DAE, the DAE applies an activation function to the input data in the intermediate layer and the output layer to output output variables x1^, x2^, … xn^. Since the DAE has been learned as described above, when the input variables x1, x2, … xn include noise caused by PVC, the DAE outputs output variables x1^, x2^, … xn^ from which the noise has been removed.

[0083] The number of input variables x1, x2, … xn is RRI data in the RRI data series that includes at least noise caused by PVC. Preferably, it further includes RRI data of a specified number before and after the RRI data including noise. This is because RRI data before and after the beats changed by the PVC artifact is required to perform correction by the DAE. The RRI data series input to these DAEs is set as the correction target range in the RRI data series. In the example of FIG. 8B, the RRI data including noise is two beats of data x4 and x5, and data x3 to x6 with one beat added before and after each of them are set as the correction target range. At this time, as shown in FIG. 8C, data x3 to x6 are input to the DAE and are the targets for correction.

[0084] FIG. 11 is a flowchart showing a correction algorithm in which the correction unit 103 removes PVC artifacts from RRI data using DAE-RM. As a premise for this correction, it is assumed that the DAE has been learned as described above. The variable t represents the number at which the RRI data was acquired, and the maximum value is MAX. The RRI data x_t indicates the t-th RRI data, y represents the RRI data in the range for which correction is to be performed, and y_mean indicates the average of each component of y. T represents the number of RRI data changed by the PVC artifact, and when T = 2, the PVC is single-occurrence. P is a parameter representing the correction width by the DAE. The DAE corrects P beats (for example, one beat before and after) before and after T beats (for example, two beats).

[0085] Referring to FIG. 11, first, the variable t is initialized (step S301). Next, the t-th RRI is newly measured and recorded in the buffer as RRI data according to the FIFO method (step S303). When a new RRI is measured, the detection process of FIG. 10 is executed in the detection unit 102 to determine the presence or absence of PVC artifacts.

[0086] As a result of the process according to the algorithm of FIG. 10 in the detection unit 102, if it is detected that the PVC artifact is mixed in the RRI data (YES in step S305), then, the measurement of the RRI data from the (t + 1)-th to the (t + T + P - 1)-th is waited for (step S307). When the measurement is completed and these RRI data are stored in the buffer according to the FIFO method, from the buffer, the measured RRI data from x_(t - P) to x_(t - 1) are read out, and a (T + 2P)-dimensional partial RRI column y is created (step S309). Thereby, a T-beat determined to include the PVC artifact and a correction range of P beats before and after it are created as the partial RRI column y.

[0087] The average value y_mean of the partial RRI column y obtained in step S309 is calculated, and the partial RRI column y is centered using the difference from the average value y_mean (step S311). The centered partial RRI column y obtained in step S311 is input to the DAE to obtain the output partial RRI column y^ (step S313). Thereby, the influence of the PVC artifact is removed from the centered partial RRI column y. Then, the average value y_mean obtained in step S311 is added to the obtained partial RRI column y^ to restore it to the partial RRI column y^ of the original size (step S315).

[0088] The difference between the sum of the original partial RRI column y obtained in step S309 and the sum of the corrected partial RRI column y^ obtained in step S315 is added to the last element of the partial RRI column y (step S317). Thereby, the total elapsed time in this range before and after the correction is made to match.

[0089] Thereafter, the partial RRI series y is set as the corrected partial RRI series ŷ (step S319), and the variable t is incremented to measure the next RRI data (step S321). Also, even when it is not detected that PVC artifacts are mixed in the RRI data measured in step S303 (NO in step S305), the variable t is incremented to measure the next RRI data (step S321). Then, until the variable t reaches the maximum value MAX (NO in step S323), the RRI data is corrected so as to exclude PVC artifacts from the RRI data by repeating the above processing. When the variable t reaches the maximum value MAX of the number of measurements (YES in step S323), the series of processes is terminated.

[0090] [Example 1]

[0091] The inventors applied MSPC-SSA and DAE-RM to the actual RRI data with added PVC artifacts, and evaluated the correction effect of DAE-RM against the mixing of PVC artifacts from the results. FIG. 12 is a diagram showing the attributes of subjects A to R. The subjects A to R shown in the figure are composed of five men aged from 26 to 45 years old (average 33.8 years old, standard deviation 7.7) and thirteen women aged from 20 to 50 years old (average 35.8 years old, standard deviation 7.7), and all of them have been diagnosed as not having arrhythmia.

[0092] Some of the RRI data measured from subjects A to R contained obvious artifacts, and these artifacts were removed because they hindered the evaluation. 166 partial RRI series y were created from subjects A to R after removing the obvious artifacts, and the total time of the partial RRI series y was 375 hours. In this example, artificial PVC artifacts were added to these partial RRI series y at random positions, and DAE-RM was applied to the ranges detected as abnormal by MSPC-SSA.

[0093] Note that the partial RRI sequence y obtained from subject A was used for the model construction of MSPC-SSA. The parameters of MSPC-SSA were determined using the partial RRI sequence y obtained from subjects E and F. As a result, the length m of the partial RRI sequence, that is, the number of columns of the Hankel matrix, was m = 6. The control limit was set so that 99% of the normal model construction data was determined to be normal, and the number of principal components was set so that the cumulative contribution rate exceeded 90%.

[0094] As performance evaluation indicators of SSA, sensitivity (SEN) and false positive rate (FPrate) were used. The index SEN represents the ratio of PVCs detected by MSPC-SSA among the artificial PVC artifacts added to the RRI data. The index FPrate represents the number of false detections per unit time (1 hour), that is, the number of times MSPC-SSA determined that a PVC artifact was included in the partial RRI sequence y that does not contain a PVC artifact.

[0095] Next, the partial RRI sequence y obtained from subject B was used for the learning of DAE. The sigmoid function and the identity mapping were used as the activation functions for the intermediate layer and the output layer, respectively. The parameters of DAE were determined using the partial RRI sequence y obtained from subjects C and D. The number of elements to be corrected by DAE was 4, the number of units in the intermediate layer was 20, and the maximum number of trials was 2000. That is, in the correction process shown in the flowchart of Fig. 11, T = 2 and P = 1. Also, sparsity was introduced using the L2 regularization term when determining the parameters. Therefore, almost all the weights connecting the units are 0.

[0096] For verification, all the partial RRI sequences y obtained from subjects G to R were used. Note that for the creation and correction of PVC-RRI data, in order to prevent the correction performance from depending on the position of the added PVC artifact, five trials were conducted and the average value was used as the result. The result of adding PVC artifacts to the partial RRI sequences y of subjects M to R and applying MSPC-SSA was SEN: 94.9% FPrate: 1.20 [times / hour] It was.

[0097] FIG. 13A is a diagram showing RRI data obtained from subject M, where the solid line represents the measured RRI data (original RRI), the thick dotted line represents the RRI with PVC artifact added (PVC-RRI), and the thin dotted line represents the corrected RRI (corrected RRI) obtained by correcting PVC-RRI with DAE-RM. FIGS. 13B, 13C), FIGS. 14A to 14C, and FIGS. 15A to 15C are each HRV index calculated from the RRI data obtained from subject M in FIG. 13A. FIG. 13B shows meanNN, FIG. 13C shows SDNN, FIG. 14A shows Total Power, FIG. 14B shows RMSSD, FIG. 14C shows NN50, FIG. 15A shows LF, FIG. 15B shows HF, and FIG. 15C shows LF / HF. In each figure, the solid line represents the HRV index calculated from the original RRI, the thick dotted line represents the HRV index calculated from PVC-RRI, and the thin dotted line represents the HRV index calculated from the corrected RRI.

[0098] In these figures, the thin dotted line generally overlaps with the solid line, and there is no significant visible deviation. That is, it can be seen that the corrected RRI (FIG. 13A) and the HRV indices calculated from the corrected RRI (FIGS. 13B, C, FIGS. 14A to 14C, and FIGS. 15A to 15C) all follow the original RRI or the HRV indices calculated from the original RRI with high accuracy.

[0099] Note that as an index of the degree of follow-up, the improvement rate I1 of the corrected RRI with respect to the root mean squared error (RMSE) between the original RRI and PVC-RRI, and the improvement rates I2 to I9 in each HRV index (meanNN, SDNN, Total Power, RMSSD, NN50, LF, HF, LF / HF) are as follows. I1 = 72.4% I2 = 57.3% I3 = 85.2% I4 = 86.4% I5 = 91.2% I6 = 60.9% I7 = 61.6% I8 = 71.7% I9 = 76.3%

[0100] From the verification results shown in the above first embodiment, it was verified that the DAE-RM can exclude the influence of PVC artifacts from the RRI data with high precision. As a result, it was verified that the DAE-RM can reduce the influence of PVC on HRV analysis.

[0101] [Example 2]

[0102] Next, the inventors applied each of a DAE with a sigmoid function as the activation function and a DAE with ReLU (Rectified Linear Unit) to the RRI data of subjects M to R shown in FIG. 12, and evaluated the effect of the activation function of the intermediate layer of the DAE applied for correction. FIG. 16A shows the results when a DAE with a sigmoid function as the activation function was applied, and FIG. 16B shows the results when a DAE with ReLU as the activation function was applied.

[0103] In this verification, it was evaluated by the degree of follow-up of the corrected RRI to the original RRI, and the root mean squared error (RMSE) between the original RRI and the corrected RRI was used as an index of the degree of follow-up. The smaller the RMSE, the higher the degree of follow-up, that is, the higher the correction accuracy. In FIGS. 16A and 16B, the vertical axis represents the reduction rate of RMSE, and the horizontal axis represents the number of nodes in the intermediate layer of the DAE. In this verification, the number of input variables to the DAE is set to 4.

[0104] From the comparison of FIGS. 16A and 16B, when ReLU is used as the activation function of the intermediate layer (FIG. 16B), the overall RMSE is smaller regardless of the number of nodes in the intermediate layer than when the sigmoid function is used (FIG. 16A). From this, it was verified that when performing correction to exclude the influence of PVC artifacts, using ReLU for the activation function of the DAE improves the correction accuracy more than the sigmoid function shown in FIG. 16B.

[0105] [Example 3]

[0106] Next, the inventors evaluated the effect of the input variables to the DAE when applying and correcting the RRI data of subjects M to R shown in FIG. 12 using the DAE. In the third embodiment, a DAE in which the activation function of the intermediate layer is ReLU is applied, and FIG. 17A shows the result when the number of input variables is 6, and FIG. 17B shows the result when the number of input variables is 8. Also in these figures, the vertical axis represents the reduction rate of RMSE, and the horizontal axis represents the number of nodes in the intermediate layer of the DAE.

[0107] From the comparison of FIGS. 16B, 17A, and 17B, the smaller the number of input variables (FIG. 16B), the smaller the overall RMSE regardless of the number of nodes in the intermediate layer compared to the case where the number is larger (FIGS. 17A and 17B). From this, it was verified that the smaller the input variables of the DAE, the higher the correction accuracy. That is, it was verified that the smaller the number of RRI data input to the DAE as the correction target range in the RRI data, the higher the correction accuracy.

[0108] [Second Embodiment]

[0109] In the first embodiment, the correction unit 103 employs DAE-RM using the DAE as a correction method, but the correction method is not limited to only DAE-RM. Other regression methods may be used as the correction method. Examples of regression methods other than DAE include PLS (partial least squares) and LW-PLS (locally weighted PLS).

[0110] PLS is a widely used linear regression method, for example, disclosed in Paul Geladi and Bruce R. Kowalski. Partial leastsquares regression: a tutorial. Analytica Chimica Acta, Vol. 185, pp. 1 - 17, 1986. Using this method, the problem of multicollinearity can be avoided by using latent variables fewer than the input variables.

[0111] LW-PLS is a method that extends PLS. For example, it is disclosed in Sanghong Kim, et al. Development of soft-sensor using locally weighted pls with adaptive similarity measure. Chemometrics and Intelligent Laboratory Systems, Vol. 124, pp. 43 - 49, 2013. In this method, a local PLS model is constructed using the samples in the already obtained dataset and the similarity between the query as weights, and using the samples that best represent the input-output relationship in the query.

[0112] [Example 4]

[0113] The inventors applied DAE, PLS, and LW-PLS respectively to the RRI data in the correction target range among the RRI data of subjects M to R shown in FIG. 12 for correction, and evaluated the correction method by comparing the corrected RRI data and the HRV indices obtained from the corrected RRI data.

[0114] Figures 18A, B, and C show the RRI data, meanNN, and SDNN after correction by applying DAE, PLS, and LW-PLS respectively. Figures 19A, B, and C show Total Powe, RMSSD, and NN50 respectively. Figures 20A, B, and C show LF, HF, and LF / HF respectively. Also in these figures, the vertical axis represents the reduction rate of RMSE, and the horizontal axis represents the method used.

[0115] From Figures 18A to 20C, it is shown that RRI data is corrected using any of DAE, PLS, and LW-PLS as the correction method. However, when comparing them, the RMSE of DAE is the smallest in any index. That is, it is verified that any method can perform correction, and in particular, DAE is suitable for correction.

[0116] [Third Embodiment]

[0117] In the above embodiments, the case where the artifact (noise) mixed in the RRI data is a PVC artifact has been described. As another artifact that can be mixed in the RRI data, a missed detection of the R wave can be cited. In this case, the value of the RRI data becomes about twice the normal value. Therefore, in this case, the detection unit 102 sets a value about twice the value of the normal RRI data as a threshold, and when there is RRI data exceeding the threshold, it can detect that an artifact due to a missed detection of the R wave (hereinafter, an R-wave dropout artifact) is mixed in. The detection unit 102 uses the RRI data exceeding the above threshold as the RRI data in the detection target range.

[0118] Even when the artifact mixed in the RRI data is an R-wave dropout artifact, similar to the first embodiment, the correction unit 103 inputs the RRI data in the detection target range to the DAE and obtains the corrected RRI data as the output. Therefore, it is not limited to the case where the artifact (noise) mixed in the RRI data is a PVC artifact, and for example, even an artifact due to other factors such as an R-wave dropout artifact can be corrected using a similar correction method, and RRI data from which the influence of the artifact has been eliminated can be obtained.

[0119] In this case, in order to remove the R-wave dropout artifact by the DAE, the data input during the learning of the AE is RRI data, and the added noise is the noise caused by the R-wave dropout. Thereby, the DAE used to remove the R-wave dropout artifact is learned to remove the noise caused by the R-wave dropout from the input RRI data and output RRI data that does not include the noise caused by the R-wave dropout.

[0120] [Example 5]

[0121] The inventors generated test RRI data assuming that the R wave could not be detected for one minute from the RRI data of subjects M to R shown in FIG. 12, and applied DAE (ReLUDAE) with the activation function being ReLU, DAE (typicalDAE) with the activation function being the sigmoid function, PLS, and LW-PLS to this test RRI data for correction respectively, and evaluated the correction effect when the artifact was an R-wave dropout artifact by comparing the corrected RRI data and the HRV indexes obtained from the corrected RRI data.

[0122] Figures 21A, B, C show the RRI data, meanNN, and SDNN after correction by applying ReLUDAE, typicalDAE, PLS, and LW-PLS respectively. Figures 22A, B, C show Total Power, RMSSD, and NN50 respectively. Figures 23A, B, C show LF, HF, and LF / HF respectively. Also in these figures, the vertical axis represents the reduction rate of RMSE, and the horizontal axis represents the method used.

[0123] From FIGS. 21A to 23C, it is shown that for correction methods other than ReLUDAE, the RRI data is corrected in all cases. On the other hand, no correction effect is seen for ReLUDAE. Therefore, it was verified that for R-wave dropout artifacts, typicalDAE, that is, DAE using the sigmoid function as the activation function, is more suitable for correction than ReLUDAE. Also, although methods other than DAE have a correction effect, it was verified that DAE is particularly suitable for correction.

[0124] [Fourth Embodiment]

[0125] From the first embodiment and the third embodiment, in the arithmetic unit 1, the DAE used for correction may be switched between ReLUDAE and typicalDAE depending on whether the artifact included in the RRI series is a PVC artifact or an R-wave dropout artifact.

[0126] As shown in FIG. 4, since the PVC artifact is mixed between adjacent R waves, when the PVC artifact is mixed, the measured RRI becomes smaller than the normal RRI. On the other hand, since one or more R waves are missing in the R wave dropout artifact, when the R wave dropout artifact is mixed, the measured RRI becomes larger than the normal RRI.

[0127] Therefore, in the fourth embodiment, the detection unit 102 includes the discrimination unit 105 shown in FIG. 3. The discrimination unit 105 stores in advance, as a threshold value, the value of the RRI that is the boundary between the PVC artifact and the R wave dropout artifact, and by comparing it with the measured RRI, determines whether the detected artifact is a PVC artifact or an R wave dropout artifact, and discriminates the type of the artifact.

[0128] Also, in the fourth embodiment, as shown in FIG. 3, the correction unit 103 prepares ReLUDAE (31) and typical DAE (32) in advance, and switches between ReLUDAE and typical DAE according to the type of the artifact discriminated by the detection unit 102.

[0129] As a result, since an optimal correction method is adopted according to the detected artifact, corrected RRI data with high accuracy and free from the influence of the artifact can be obtained. As a result, even when multiple types of artifacts are included, the accuracy of HRV analysis can be improved.

[0130] [Fifth Embodiment] The factor that reduces the reliability of R wave detection is not limited to PVC. Another example of a factor that reduces the reliability of R wave detection is premature atrial contraction (PAC). PAC is a type of premature contraction in which excitation occurs in a location called the atrium or other upper chamber prior to excitation of the sinoatrial node. PAC is also a type of arrhythmia that can occur in healthy individuals, just like PVC. Even when PAC occurs, an artifact is mixed into the RRI, which becomes a factor that reduces the reliability of R wave detection.

[0131] [Example 6]

[0132] The inventors also evaluated the correction effect for PAC artifacts in the same way as for PVC artifacts. As shown in FIG. 24, it can be seen that only the RRI before the occurrence of PAC changes in the electrocardiogram waveform where PAC occurred. To reflect this feature, the inventors randomly selected one beat of the RRI data, decreased the value of the RRI, and used it for verification. Also, it was assumed that the P wave due to PAC occurs only after the T wave, and the one with the RRI value decreased to satisfy this assumption was used for verification. Specifically, PAC artifacts as shown in FIG. 25 were added to the electrocardiogram waveform and used. The size L of the PAC artifact was set to a random value within the range where the RRI does not become shorter than the normal QT interval. Also, it was assumed that PAC does not occur continuously in the electrocardiogram waveform.

[0133] The middle layer of the neural network used for correction was set as the sigmoid function, and the number of units in the middle layer was 8. Also, the activation function of the middle layer was ReLU, and the activation function of the output layer was the identity mapping. Also, for the detection of PAC by the autoencoder (AE), an electrocardiogram waveform of 8 beats was used. Also, the number of parameters of the DAE, which is the number of elements of the RRI to be corrected, was 4, and the number of parameters of the DAE, which is the unit of the middle layer, was 2. Note that AE is a neural network for dimensionality compression and feature extraction.

[0134] In FIGS. 26B to 28C, each index obtained from the original RRI data, the RRI data with artificial artifacts added, and the RRI data after correction by DAE-RM shown in 26A is shown. For each index shown in FIGS. 26B to 28C, anomaly detection by AE was applied, and the original RRI data and the RRI data after correction by DAE-RM, and each index obtained from the original RRI data and each index obtained from the RRI data after correction by DAE-RM were compared.

[0135] As a result, the improvement rate of the original RRI data and the corrected RRI data was 27.4%. Also, it was 0.1% for meanNN, 24.8% for SDNN, 22.7% for Total Power, 70.8% for RSMMD, 34.0% for NN50, 22.7% for LF, 28.2% for HF, and 30.0% for LF / HF. That is, it can be seen that both the corrected RRI and the HRV indices calculated from the corrected RRI closely follow the original RRI or the HRV indices calculated from the original RRI with high accuracy.

[0136] From this result, it was verified that, similar to the PVC artifact, the PAC artifact can be excluded from the RRI data with higher accuracy by DAE-RM. Therefore, it was verified that the influence of PAC on HRV analysis can be reduced in the same way as PVC by DAE-RM.

[0137] [3. Addendum]

[0138] In addition, according to another aspect of the present embodiment, the following arithmetic device is included. That is, (1) according to an embodiment, the arithmetic device is an arithmetic device that corrects RRI data, which is data on the intervals between adjacent R waves of an electrocardiogram signal, and detects a correction target range including RRI data having noise among the RRI data obtained in time series, and corrects the RRI data in the correction target range.

[0139] (2) The arithmetic device according to (1), wherein the correcting includes correcting the RRI data in the correction target range by inputting the RRI data in the correction target range into PLS (partial least squares).

[0140] (3) The said detecting includes at least one of: detecting the range to be corrected including RRI data where noise is caused by PVC; detecting the range to be corrected including RRI data where noise is caused by PAC; and detecting the range to be corrected including RRI data where noise is caused by the absence of R-wave detection. The arithmetic device according to (1) or (2).

[0141] The present invention is not limited to the above-described embodiments, and various modifications are possible.

Explanation of Signs

[0142] 1 Arithmetic device 2 Heart rate monitor 10 Processing unit 12 Memory 13 Communication unit 21 Electrodes 31 ReLUDAE 32 typicalDAE 100 Detection device 101 RRI calculation unit 102 Detection unit 103 Correction unit 104 Detection section 105 Discrimination unit 121 Program

Claims

1. An arithmetic unit for correcting RRI data, which is data on the intervals between adjacent R waves of an electrocardiogram signal, comprising: detecting a correction target range including RRI data having noise among the RRI data obtained in time series; correcting the RRI data in the correction target range by inputting the RRI data in the correction target range into a DAE (denoising autoencoder); wherein said detecting includes detecting the correction target range including RRI data in which the noise is due to premature ventricular contraction (PVC); the activation function of the DAE is a ReLU (Rectified Linear Unit) or a sigmoid function Arithmetic unit.

2. An arithmetic unit for correcting RRI data, which is data on the intervals between adjacent R waves of an electrocardiogram signal, comprising: detecting a correction target range including RRI data having noise among the RRI data obtained in time series; correcting the RRI data in the correction target range by inputting the RRI data in the correction target range into a DAE (denoising autoencoder); wherein said detecting includes detecting the correction target range including RRI data in which the noise is caused by missing detection of R waves; the activation function of the DAE is a sigmoid function Arithmetic unit.

3. An arithmetic unit for correcting RRI data, which is data on the intervals between adjacent R waves of an electrocardiogram signal, comprising: detecting a correction target range including RRI data having noise among the RRI data obtained in time series; correcting the RRI data in the correction target range by inputting the RRI data in the correction target range into a DAE (denoising autoencoder); wherein said detecting includes detecting the correction target range including RRI data in which the noise is due to premature atrial contraction (PAC); the activation function of the DAE is a ReLU (Rectified Linear Unit) Arithmetic unit.

4. An arithmetic unit for correcting RRI data, which is data on the intervals between adjacent R waves of an electrocardiogram signal, comprising: detecting a correction target range including RRI data having noise among the RRI data obtained in time series; By inputting the RRI data within the correction target range into a DAE (denoising autoencoder), the RRI data within the correction target range is corrected. When correcting the RRI data within the correction target range, if the detected correction target range is a correction target range including RRI data where the noise is due to premature ventricular contraction (PVC), the activation function of the DAE is ReLU (Rectified Linear Unit). If it is a correction target range including RRI data where the noise is caused by the missing detection of the R wave, the activation function of the DAE is the sigmoid function. The activation function of the DAE is switched as described above. An arithmetic unit. **Claim 5** The number of RRI data input to the DAE when correcting the RRI data within the correction target range matches the number of RRI data included in the correction target range. The arithmetic unit according to any one of Claims 1 to 4. **Claim 6** The correction target range includes the RRI data having the noise and a specified number of RRI data before and after the RRI data having the noise. The arithmetic unit according to any one of Claims 1 to 5. **Claim 7** Equipped with the arithmetic unit according to any one of Claims 1 to 6, A specific biological state is detected based on the RRI data obtained in a time series including the RRI data within the correction target range. A detection device. **Claim 8** An arithmetic method for correcting RRI data, which is data on the intervals between adjacent R waves of an electrocardiogram signal, comprising: a step of detecting a correction target range including RRI data having noise among the RRI data obtained in a time series; and a step of correcting the RRI data within the correction target range by inputting the RRI data within the correction target range into a DAE. The detecting step includes any one of the following (i) to (iii). An arithmetic method. (i) includes a step of detecting the correction target range including RRI data where the noise is due to premature ventricular contraction (PVC), and the activation function of the DAE is ReLU (Rectified Linear Unit) or the sigmoid function. including a step of detecting the range to be corrected including RRI data caused by noise due to missing detection of R waves, wherein the activation function of the DAE is a sigmoid function including a step of detecting the range to be corrected including RRI data caused by noise due to premature atrial contraction (PAC), wherein the activation function of the DAE is ReLU (Rectified Linear Unit)

9. A program for causing a computer to function as an arithmetic unit that corrects RRI data, which is data on the intervals between adjacent R waves of an electrocardiogram signal, wherein the computer is caused to detect a range to be corrected including RRI data having noise among the RRI data obtained in time series; correct the RRI data in the range to be corrected by inputting the RRI data in the range to be corrected into a DAE; and the step of detecting includes any one of the following (i) to (iii), a computer program including a step of detecting the range to be corrected including RRI data caused by noise due to premature ventricular contraction (PVC), wherein the activation function of the DAE is ReLU (Rectified Linear Unit) or a sigmoid function including a step of detecting the range to be corrected including RRI data caused by noise due to missing detection of R waves, wherein the activation function of the DAE is a sigmoid function including a step of detecting the range to be corrected including RRI data caused by noise due to premature atrial contraction (PAC), wherein the activation function of the DAE is ReLU (Rectified Linear Unit)

Images