JP2012065713A - Method for removing abnormal heartbeat and trend of electrocardiogram data, autonomic nerve monitor device, and septicemia sideration alarm device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for removing the abnormal heartbeat and trend in electrocardiogram data capable of utilizing a change in heartbeat fluctuation in proper diagnosis by automatically removing the abnormal heartbeat and the trend from the heartbeat interval data row obtained from an electrocardiogram, an autonomic nerve monitor device, and a septicemia sideration alarm device utilizable in the diagnosis of the sideration of septicemia.SOLUTION: The CPU 32 of the autonomic nerve monitor device acquires an RRI time series xfrom the electrocardiogram data. The CPU 32 determines a flag time series Q containing a trend function and a normal heartbeat so that the probability density distribution, which is related to a case that a flag wherein qin the residual series of the RRI time series xand the trend function thereof shows normality is formed, most becomes normal distribution to remove the abnormal heartbeat and the trend to acquire RRI time series data used in HRV analysis. The CPU 32 performs HRV analysis on the basis of the RRI time series data wherein the abnormal heartbeat and the trend are removed.

Description

  The present invention relates to a method for removing abnormal heartbeats and trends from electrocardiogram data, an autonomic nerve monitor device, and a sepsis onset warning device, and more particularly, an abnormality in electrocardiogram data that can automatically remove abnormal heartbeats and trends in electrocardiogram data. The present invention relates to a heartbeat and trend removal method, an autonomic nerve monitor device, and a sepsis onset warning device.

  As shown in FIG. 1, the heart rate represents the number of times the heart beats in a certain time (usually 1 minute), and is the most important measure for grasping the patient's condition. Specifically, the heart rate is defined as the reciprocal of the average value of the interval between beats multiplied by a fixed time unit. While the heart rate is derived from the average of the interval between beats (RR interval; RRI), the variation in the interval between beats is called HRV (heart rate variability). The heart rate changes by the action of the autonomic nerve in response to internal and external stimuli. For this reason, the greater the HRV (heart rate variability), the better the response to internal and external stimuli (ie, health). HRV (heart rate variability) is also used as a measure for measuring athletic ability, and it is considered that information that cannot be understood only by heart rate can be obtained.

  Therefore, autonomic nerve activity can be monitored by analyzing HRV (heart rate variability). The HRV analysis method is mainly classified into a time domain analysis method (Non-patent document 1), a frequency domain analysis method (spectrum analysis), and a nonlinear and chaos / fractal analysis method (Non-patent document 1).

  The time domain analysis method includes a tone entropy method (Non-patent document 2), and the nonlinear and chaos / fractal analysis methods include a detrended fluctuation analysis (DFA) method (Non-patent document 3).

  Frequency domain analysis methods include FFT method (Non-patent document 4), autoregressive model method (Non-patent document 5), maximum entropy method (Non-patent document 6), Complex demodulation (CD) method (Non-patent document 7), etc. There is. In the frequency domain analysis, the high frequency band from 0.14 to 0.4 Hz (High Frequency; HF), the low frequency band from 0.04 to 0.14 Hz (Low Frequency; LF), from 0.003 to 0.04 Hz. The VLF (Very Low Frequency) and the ULF (Ultra Low Frequency) band of 0.003 Hz or less are important. In general, the LF component represents sympathetic nerve activity and parasympathetic nerve activity, and the HF component represents parasympathetic nerve activity (Non-patent Document 8). Furthermore, it is said that a body temperature adjustment system etc. influences a VLF component and a ULF component (nonpatent literature 10).

  By the way, abnormal heartbeats such as arrhythmia are caused by abnormalities in the heart. For this reason, it is unrelated to autonomic nerve activity. For this reason, it is necessary to remove abnormal heartbeats in the HRV analysis. If abnormal heartbeats are not removed from the electrocardiogram, it will be a factor for excessively estimating HRV (heart rate fluctuation).

  In order to detect an abnormal heartbeat, there is a method in which a heartbeat that differs by a certain amount or more compared to a typical ECG (electrocardiogram) waveform is regarded as an abnormal heartbeat (referred to as Prior Art 1). This method is adopted in many devices such as electrocardiographs that have been commercialized, but it cannot be applied unless the abnormality is so large that a change in the ECG waveform appears.

  In addition, when the heart rate interval (RRI) follows a normal distribution, the statistical property (3σ rule) in which 99.6% of the samples fall within the mean ± standard deviation is used, and those that do not fall within this range are regarded as abnormal values. There is. There is also a method of performing statistical properties (2σ rule) in which 95.4% of the samples fall within the mean ± standard deviation when the heart rate interval (RRI) follows a normal distribution instead of the 3σ rule (referred to as Prior Art 2).

However, in order to specify a normal (or abnormal) RRI by the 3σ rule or the 2σ rule, an average / standard deviation with respect to a normal value is necessary.
However, since normal data is not specified at that time, the average and standard deviation including abnormal values are generally substituted. For this reason, when there are many abnormal values, the abnormal values also exist normally, and the abnormal values cannot be removed.

  Furthermore, if the electrocardiogram data has a long-cycle component, that is, a trend, the abnormal value determination principle collapses. Unless the abnormal value is accurately removed, the reliability of the subsequent HRV analysis (especially frequency domain analysis) is lowered. For this reason, conventionally, it is common to visually specify and remove abnormal heartbeats or to re-measure an electrocardiogram.

  In particular, many patients who are admitted to the ICU (hereinafter referred to as ICU patients) have many chronic illnesses, and abnormal heartbeats such as arrhythmia due to a disorder are extremely large. For this reason, electrocardiograms of ICU patients are often not usable in normal HRV analysis.

  By the way, the diagnosis of sepsis is currently performed by a blood test after a doctor suspects the onset of a patient through subjective observation or the like. When bacteria are detected by the blood test, by specifying the bacteria, an antibiotic to be used is determined and treatment is started.

  In the treatment of sepsis, administration of antibiotics has been proposed within 1 hour after the patient has developed sepsis so as not to make the sepsis severe. However, since the diagnosis of sepsis is largely based on the subjectivity of doctors, there is a problem that the detection of sepsis tends to be delayed.

  Therefore, it is known that when a patient develops sepsis, the autonomic nerve is modulated and HRV (heart rate variability) is reduced (Non-patent Documents 9 and 10). However, at present, a problem has been pointed out that HRV (heart rate variability) cannot be specified from an electrocardiogram in real time.

  The reason is presumed that in ICU patients, abnormal heartbeats such as arrhythmia are extremely large as described above, and thus it is difficult to remove abnormal heartbeats.

Junichiro Hayano, “Cardiovascular Diseases and Autonomic Nervous Function”, 2nd edition, Medical School, 2001, p.71-109, E. Oida, Tee. Moritani, Wai. Gecko, “Tone Entropy Analysis on Cardiac Recovery After Dynamic Exercise”, Jay. A.P.P.L. P.E.I.S.I.O.L. Vol.82,1997, p. 1794-1801 (E. Oida, T.Moritani, and Y.Yamori, “Tone-entropy analysis on cardiac recovery after dynamic exercise”, J. Appl. Physiol., Vol. 82, 1997 , p.1794-1801) Sea. K. Pen, S. Harbin, H. E. Stanley, A. El. Goldberger. “Quantification of Scaling Exposures and Crossover Phenomena is Non-Stationary Heartbeat Time Series”, Sea H.O.S., Vol. 5, No. 1,1995, p.82-87 (CK Peng, S. Havlin, HE Stanley, AL Goldberger, “Quantification of scaling exponents and crossover phenomena is nonstationary heartbeat time series”, Chaos, Vol. 5, No. 1,1995, p.82-87) Jai. W. Coolie, and J, Turkey, “An Algorithm for the Machine Calculation of Complex Fourier Series, Materialistics of Computation”, Vol. 19,1965, p.297-301 (JW Cooley, and JW Turkey, “An Algorithm for the machine calculation of complex Fourier series, Mathematics of Computation, Vol. 19,1965, p.297-301) Etch. Akaike, “Power Spectrum Estimate Through Auto Regressive Model Fitting”. "N.N.N.INST.ST.I.T.I.S.Timas", Vol. 21,1969, p.407-419 (H. Akaike, “Power spectrum estimation through autoregressive model fitting, Ann Inst Statist Math ”, Vol. 21,1969, p.407-419) A. Mariani, M. Pagani, F. Lombardi, and S. Serutti, “Cardiovascular neural regulation Explore in the frequency domain”, Circulation, Vol. 84, No. 2,1991, p.482-492 (A. Malliani, M. Pagani, F. Lombardi, and S. Cerutti, “Cardiovascular Neural Regulation Explored in the Frequency Domain”, Circulation, Vol. 84, No. 2,1991, p.482-492) Pee. Bloomfield, “Fourier Analysis of Time Series, Introduction”, John Wiley and Sons, 1976, p. 118-150 (P. Bloomfield, “Fourier Analysis of Time Series, An Introduction”, John Wiley & Sons, 1976, p. 118-150) S. Axelrod, Di. Gordon, F. A. Ubel, Di. Sea. Shannon, A. Sea. Burger, and Earl. Jay. Cohen, “Power Spectrum Analysis of Heart Rate Fractionation, Accentive Probe of Beat to Beat Cardiovascular Control”, Science, Vol. 213,1981, p.220-222 (S. Akselrod, D. Gordon , FA Ubel, DC Shannon, AC Barger, and RJ Cohen ”Power Spectrum Analysis of Heart Rate Fluctuation: A Quantitative Probe of Beat-to-Beat Cardiovascular Control”, Science, Vol. 213, 1981, p.220-222) Takeshi Moriguchi, Hiroyuki Hirasawa, Adult Oda, Yoshihisa Tateishi, “Study on the prediction of septic shock onset by heart rate variability (HRV) analysis”, Japanese Clinical, Vol. 62, No. 12, 2004, p. 2285-2290 Task Force of the European Society of Cardiology The North American Society of Pacing Electrophysiology: Heart Rate Variability; Standards of Measurement, Physiological Interpretation, and Clinical Youth, Circulation, Vol. 93 , No. 5,1966, p1043-1065 (Task Force of the European Society of Cardiology the North American Society of Pacing Electrophysiology: Heart Rate Variability, Standards of Measurement, Physiological Interpretation, and Clinical Use, Circulation, Vol. 93, 1966, No. 5, p1043-1065)

  The first object of the present invention is to automatically remove abnormal heartbeats and trends such as arrhythmia from a heartbeat interval data string obtained from an electrocardiogram, and to determine changes in HRV (heart rate variability) obtained as a result of HRV analysis. An object of the present invention is to provide a method for removing abnormal heartbeats and trends in electrocardiogram data and an autonomic nerve monitoring apparatus that can be used for appropriate diagnosis.

  The second object of the present invention is to automatically remove abnormal heartbeats and trends such as arrhythmia from the heartbeat interval data string obtained from the electrocardiogram, and to change the HRV (heart rate fluctuation) obtained as a result of the HRV analysis. Is to provide a sepsis onset warning device that can be used for diagnosis of the onset of sepsis.

  In order to achieve the above object, the invention according to claim 1 is characterized in that, when an abnormal heartbeat and a trend are removed from time series data of RR intervals representing a beat-to-beat interval included in the electrocardiogram data, The gist is a method for removing abnormal heartbeats and trends from electrocardiogram data, wherein HRV (heartbeat fluctuation) data is calculated after removing abnormal heartbeats and trends so that the density distribution is close to a normal distribution.

The invention of claim 2 is a first step of acquiring time series data of RR intervals (hereinafter referred to as RRI time series _xn ) representing the interval between beats from electrocardiogram data, and f (t _n ; θ). A trend function representing a long-cycle component of the electrocardiogram data, t as time, θ as a parameter of the trend function f, q _n as a flag indicating normality or abnormality of the RRI time series x _n , Q = When (q ₁ , q ₂ , q ₃ ,... Q _n ) is a flag time series, residual series x _n −f (t _n ; θ), n = 1, 2,. . . Among them, the flag time series Q including the trend function f (t _n ; θ) and the normal heartbeat is determined so that the probability density distribution for a case where a flag indicating that _n is normal is established is closest to the normal distribution. The gist of the method for removing abnormal heartbeats and trends from electrocardiogram data, comprising a second step of calculating HRV (heartbeat fluctuation) data after removing abnormal heartbeats and trends.

  According to a third aspect of the present invention, in the second aspect, in the second step, the flag time series Q is fixed, the trend function that minimizes the average error of the residual series is determined, and the residual series is obtained. And a step B for updating the residual series by removing the abnormal heartbeat by determining the flag time series Q so that the residual series obtained in the A step has the most normality. And

  According to a fourth aspect of the present invention, in the third aspect, in the second step, when the flag time series Q determined in the B step changes differently from the flag time series Q fixed in the A step, A Returning to the step, the flag time series Q is updated to the changed flag time series Q and the process of the A step is performed. Thereafter, until the flag time series Q determined in the B step does not change, the A step and the B step It is characterized by repeating.

  According to a fifth aspect of the present invention, in the step B, in the step B, when determining the flag time series Q so that the residual series obtained in the step A has the most normality, an abnormality included in the residual series With regard to the heartbeat, an abnormal value likelihood is calculated so as to include an abnormal heartbeat and a heartbeat continuous to the abnormal heartbeat, and the abnormal value and the heartbeat continuous to the abnormal heartbeat are removed based on the abnormal value likelihood.

The invention of claim 6 is an acquisition means for acquiring electrocardiogram data, and RRI time-series acquisition for acquiring time-series data of RR intervals (hereinafter referred to as RRI time-series _xn ) representing the interval between beats from the electrocardiogram data. Means, f (t _n ; θ) is a trend function representing a component of a long period of the electrocardiogram data, t is time, θ is a parameter of the trend function f, and q _n is an RRI time series x _n Is a flag representing normality or abnormality of Q, and when Q = (q ₁ , q ₂ , q ₃ ,... Q _n ) is a flag time series, residual series x _n −f (t _n ; θ), n = 1, 2,. . . Among them, the flag time series Q including the trend function f (t _n ; θ) and the normal heartbeat is determined so that the probability density distribution for a case where a flag indicating that _n is normal is established is closest to the normal distribution. By removing the abnormal heart rate and the trend, the removing means for extracting the RRI time series data used for the analysis of the HRV analysis, and the HRV (heart rate fluctuation) for the RRI time series data from which the abnormal heart rate and the trend are removed. The gist of the present invention is an autonomic nerve monitor device including HRV analysis means for performing analysis.

  According to a seventh aspect of the present invention, in the sixth aspect, the removing means fixes the flag time series Q, determines a trend function that minimizes an average error of the residual series, and obtains the residual series. And a second process for removing the abnormal heartbeat and updating the residual series by determining a flag time series Q so that the residual series obtained by the first processing means has the most normality. Means.

  The invention according to claim 8 is the invention according to claim 7, wherein the removing means changes the flag time series Q determined by the second processing means different from the flag time series Q fixed by the first processing means. Returns to the first processing means, updates the flag time series Q to the changed flag time series Q and performs the processing of the first processing means, and thereafter the flag time series Q determined by the second processing means changes. The processing of the first processing means and the processing of the second processing means are repeated until not.

  According to a ninth aspect of the present invention, in the eighth aspect, in the second processing means, when determining the flag time series Q so that the residual series obtained by the first processing means has the most normality, the residual series And calculating an abnormal value likelihood so as to include an abnormal heartbeat and a heartbeat continuous to the abnormal heartbeat, and removing the abnormal value and the heartbeat continuous to the abnormal heartbeat according to the abnormal value likelihood. To do.

  According to a tenth aspect of the present invention, the autonomic nerve monitoring device according to any one of the sixth to ninth aspects is a sepsis onset warning device, and an HRV (heart rate variability) is determined based on an analysis result of the HRV analysis means. Detection means for detecting the minimum value in the time series of data, and warning that there is a risk of developing sepsis if the detection means detects the minimum value in the time series of the HRV (heart rate variability) data The gist of the present invention is a sepsis onset warning device comprising a warning means.

  According to the method of the present invention, it is possible to automatically remove abnormal heartbeats and trends such as arrhythmia from a heartbeat interval data string obtained from an electrocardiogram, and to use a change in HRV (heart rate variability) for appropriate diagnosis. it can.

  In addition, according to the autonomic nerve monitor apparatus of the present invention, abnormal heartbeats and trends such as arrhythmia are automatically removed from the heartbeat interval data string obtained from the electrocardiogram, and a change in HRV (heart rate variability) is appropriately diagnosed. It is possible to provide an autonomic nerve monitor device that can be used for the above.

  Further, according to the sepsis onset warning device of the present invention, abnormal heartbeats such as arrhythmia and trends are automatically removed from the heartbeat interval data string obtained from the electrocardiogram, and changes in HRV (heart rate variability) are diagnosed. It is possible to provide a sepsis onset warning device that can be used in the future.

Explanatory drawing of heart rate variability. (A) is an explanatory diagram of heart rate variability including an abnormal heart rate, (b) is an explanatory diagram of heart rate variability from which a trend and an abnormal heart rate have been removed. Explanatory drawing of a trend function. The block circuit diagram of an autonomic nerve monitor apparatus. The flowchart in the case of taking out heart rate variability data from electrocardiogram data. (A) is a diagram of time-series data of heartbeats detected by an electrocardiogram waveform recorder, (b) is a diagram of time-series data of RRI (RR interval) obtained as a normal heartbeat according to the present invention using (a). (C) is a figure of time series data of RRI (RR interval) which removed a trend from (b), and (d) is a figure of a power spectrum which carried out HRV analysis of (c). (A) is a diagram of time-series data of heartbeats detected by a conventional electrocardiogram waveform recorder, (b) is a diagram of time-series data of RRI (RR interval) of a comparative example using (a), ( (c) is a time-series data diagram of RRI (RR interval) obtained as a normal heartbeat that falls within MEAN ± 2σ, and (d) is a power spectrum diagram obtained by HRV analysis of (c). The figure showing the time change of HRV (heart rate variability) of case 1 in which septic shock of a patient who developed sepsis occurred. The figure showing the time change of HRV (heart rate variability) of case 2 in which septic shock occurred in a patient who developed sepsis. The figure showing the time change of HRV (heart rate variability) of case 3 in which septic shock occurred in a patient who developed sepsis. The figure showing the time change of HRV (heart rate variability) of case 4 in which septic shock of a patient who developed sepsis occurred. The figure showing the time change of HRV (heart rate variability) of case 5 in which septic shock of a patient who developed sepsis occurred.

A method for removing abnormal heartbeats and trends from electrocardiogram data and an autonomic nerve monitoring apparatus according to an embodiment of the present invention will be described below with reference to FIGS.
<Configuration>
FIG. 4 shows an overall view of the autonomic nerve monitoring apparatus.

  As shown in FIG. 4, the autonomic nerve monitor device 10 includes an electrocardiogram waveform recorder 20 and an electrocardiogram waveform analyzer 30. The electrocardiogram waveform recorder 20 acquires an electrocardiogram waveform signal via an electrode (not shown) attached to the subject and an A / D conversion circuit (not shown), and continuously outputs the electrocardiogram waveform signal to the electrocardiogram waveform analyzer 30. To do. The electrocardiogram waveform recorder 20 corresponds to an acquisition unit. The electrocardiogram waveform analysis apparatus 30 includes a database 32 including a CPU 32, a ROM 34, a RAM 36, and a storage device. The ROM 34 stores an abnormal heartbeat, trend removal, and HRV analysis processing program. The RAM 36 is a working memory when the CPU 32 performs various processes.

The CPU 32 corresponds to RRI time series acquisition means, removal means, HRV analysis means, first processing means, and second processing means.
The electrocardiogram waveform analysis device 30 performs processing based on the abnormal heartbeat, trend removal, and HRV analysis processing program on the ECG waveform signal input via the interface 40 and stores the processing result in the database 38.

  A display 42 and a warning device 44 as warning means are connected to the electrocardiogram waveform analysis device 30. The display 42 is composed of a liquid crystal display device or the like. The warning device 44 includes a display lamp or a buzzer.

<Action>
Next, the operation of the autonomic nerve monitor device 10 will be described with reference to the flowchart of FIG.
In S100 of FIG. 5, the CPU 32 of the electrocardiogram waveform analyzer 30 detects the peak of the waveform from the electrocardiogram data (ECG waveform) input from the electrocardiogram waveform recorder 20 every predetermined time according to the abnormal heartbeat, trend removal and HRV analysis processing program. To detect an R point and a heartbeat interval (hereinafter abbreviated as RR interval). In the present embodiment, the predetermined time is one minute, but is not limited. The electrocardiogram data (ECG waveform) is an electrocardiogram waveform signal.

That is, the CPU 32 detects the time t _n when the R wave of the electrocardiogram waveform signal occurred in S100 of FIG. 5 and calculates the RR interval to thereby calculate time series data of the RR interval representing the beat-to-beat interval ( RRI time series x _n ) is obtained. The RRI time series _xn is stored in the database 38. The step of obtaining the RRI time series _xn corresponds to the first step.

<Principle of removal of abnormal heartbeat and trend in ECG data>
Here, the principle of removing abnormal heartbeats and trends will be described.
RRI time series _{x n} is, the RR interval at time _{t n} is the time series and _{x n.} In the present embodiment, if the RRI time series x _n is normal, q _n = 1, and if it is abnormal, q _n = 0, and these flags are combined, and the flag time series Q = (q ₁ , q ₂ , q ₃ ,... Q _n ). The value taken by q _n may be other values as long as the RR interval x _n is normal or can be distinguished, and is not limited to 0 and 1.

A function representing a trend representing a long-cycle component of the electrocardiogram data (hereinafter referred to as a trend function) is represented by f (t; θ) with the parameter θ (see FIG. 3).
In general, in the HRV analysis, frequency components of about 0.04 Hz or higher are targeted, and frequencies below that may be considered as trends. For example, when analyzing every minute, a frequency of 0.04 Hz means that a minimum value of about 5 appears. Since the lowest polynomial order having five extreme values is 6, a polynomial of about 6th order may be used.

Here, since the fluctuation of a pure phenomenon follows a normal distribution, the pure heart rate fluctuation derived from autonomic nerves, that is, HRV (heart rate fluctuation) also follows the normal distribution.
That is, if the trend function f (t _n ; θ) and the flag time series Q representing the normal heartbeat are correct, the residual series x _n− f (t _n ; θ), n = 1, 2,. . . such that q _n = 1 follows a normal distribution.

That is, the residual series x _n −f (t _n ; θ), n = 1, 2,. . . The trend function f (t _n ; θ) and the flag time series Q representing the normal heartbeat are determined so that the probability density distribution of such that q _n = 1 is closest to the normal distribution.

FIG. 2A shows an RRI time series including trends and abnormal heartbeats. FIG. 2B shows an RRI time series from which trends and abnormal heartbeats have been removed.
Thus, the obtained residual series x _n −f (t _n ; θ), n = 1, 2,. . . such that q _n = 1 is an RR interval sequence derived from an autonomic nerve from which a trend or abnormal heartbeat has been removed. The RR interval sequence derived from the autonomic nerve from which the trend and abnormal heartbeat have been removed can be suitably used for HRV analysis.

Here, the sample sequences x ₁ , x ₂ ,. . . x _N is an example of a normality measure of of Rashi which follow a normal distribution, using 3,4-order central moment m _3, m ₄ the formula with (1). The normality scale is not limited to the following formula (1).

これは、正規分布では、分散を１に規格化した際の３次中心モーメントｍ_３は０、４次中心モーメントｍ_４は３になり、それから離れるほど正規性が低いと判断されることを利用している。ここで、ｍ_３／σ^３は、歪度であり、分布の非対称性の度合いを示す。又、ｍ_４／σ^４は、尖度である。ただし、ａ_３，ａ_４はともに正の実数であり、３，４次中心モーメントｍ_３，ｍ_４の重みを規定する。

This is based on the fact that in the normal distribution, the third-order central moment m ₃ when the variance is normalized to 1 is 0, and the _fourth- order central moment m ₄ is 3, and the normality is judged to be lower as the distance from the third is smaller. is doing. Here, m ₃ / σ ³ is the skewness and indicates the degree of asymmetry of the distribution. M ₄ / σ ⁴ is kurtosis. However, both a ₃ and a ₄ are positive real numbers, and define the weights of the _third and fourth order central moments m ₃ and m ₄ .

Specifically, Q = (0,0,0, ..., 0), (0,0,0, ..., 1),. . . , (1, 1, 1,..., 1), x _n and f (t _n ; θ), n = 1, 2,. . . such that q _n = 1 over to determine the theta to minimize the average error, residual sequence _{x n -f (t n; θ} ) normalized measure _g of _{({x n -f (t n} ; θ) | N = 1,..., N; q _n = 1}) is maximized. However, since this method requires ^2N regularity evaluations, it cannot be solved in real time.

For this reason, it carries out in the following order as an alternative method.
(1) Therefore, as the A step, first, the flag time series Q is fixed, and the trend function f (t _n ; θ) is determined so that the average error is minimized.

(2) Next, the trend function f (t _n ; θ) is fixed as the B step, and Q is determined so that the residual series has the most normality, that is, the normality measure g is maximized. To do. Thereafter, the process is repeated until there is no change in Q. The processes (1) and (2) correspond to the second step.

<Specific processing>
In the present embodiment, the CPU 32 performs the following processing in S200 of FIG. 5 based on the above principle.

(S200-1)
Q = (1, 1, 1,..., 1) is fixed as the initial value of the flag time series Q.
(S200-2)
A trend function f (t; θ) that minimizes the average error is determined.

(S200-3)
Next, RRI time series y _k = x _k −f (t _k ; θ) from which the trend is removed is _obtained .
The flag time series Q is determined and updated by the following method so that the RRI time series y _k has the most normality.

<Determination method of flag time series Q>
Using all y _k , the outlier probability z _k

を求める。

Ask for.

However, w shows a weight. If a certain heart rate is abnormal, the weight w has a high probability that the subsequent first and second beats are also abnormal heart rates. For example, when the error | y _k | is large, the abnormal value likelihood of the k-th heartbeat increases, even if | y _k-1 | and | y _k-2 | are large. Since the influence degree decrease with decreasing past, the weight of the past n as w _n, may be a formula (2).

In this embodiment, w ₀ = 1, w ₁ = 0.5, w ₂ = 0.2, w ₃ = w ₄ =. . . = 0. W ₀ is a numerical value of the weight of the abnormal heartbeat. w ₁ and w ₂ are numerical values of the weights of the first and second beats following the abnormal heartbeat. In this way, the weight of the abnormal heartbeat is set larger than the other heartbeats, and the weights of the first and second beats following the abnormal heartbeat are set to be gradually reduced. Note that the values of w ₀ = 1, w ₁ = 0.5, and w ₂ = 0.2 are examples, and other positive numerical values may be used. Further, instead of the above equation (2), z _k may be a more complicated function such as the function z _k = h (y _k , y _k−1 , y _k−2 ,...).

Next, the values are regarded as abnormal values in descending order of z _k , and the normality scale g of the remaining series from which these abnormal values are removed is evaluated. Then, let Q be the set of samples that maximizes the normality measure g.
(S200-4)
If the acquired Q is different from the fixed Q, that is, if there is a change in Q, the process returns to S200-2. The change in Q is determined by comparing the Q acquired in the process before the repetition with the Q obtained in the current process. If there is no change in Q, the process of S200-4 is terminated.

The CPU 32 removes the trend and abnormal heartbeat from the RRI time series input from the electrocardiogram waveform recorder 20 by performing the processes of S200-1 to S200-4.
Next, as shown in S300 of FIG. 5, the CPU 32 performs HRV analysis on the RRI time series from which the trend and abnormal heartbeat are removed. The analysis method of HRV (heart rate variability) is performed by a known method described in Non-Patent Document 1 to Non-Patent Document 7.

  An actual example in which abnormal heartbeats and trends are actually removed by the autonomic nerve monitor device 10 is shown in FIG. FIG. 6A is a diagram of time series data of the subject's electrocardiogram and the detected heartbeat. FIG. 6B is a diagram of time series data of RRI (RR interval) obtained based on the electrocardiogram data of FIG. In FIG. 6B, the solid curve represents the trend. In FIG. 6B, a black circle represents a normal heartbeat, and a white circle represents an abnormal heartbeat. FIG. 6C is a diagram of time series data of RRI (RR interval) after the trend and abnormal heartbeat are removed from the time series data of FIG. 6B. FIG. 6 (d) shows the power spectrum obtained by performing spectrum analysis as HRV analysis on the RRI time series after removing the trend and abnormal heartbeat shown in FIG. 6 (c). ing.

  On the other hand, comparative examples in which abnormal heartbeats and trends are removed using the time series data of heartbeats in FIG. 7A are shown in FIGS. 7B to 7D. FIG. 7A shows the same time series data as FIG. 6A, and the same electrocardiogram is used for comparison.

  FIG. 7B is a diagram of time series data of RRI (RR interval) estimated by the method of the prior art 1 described in the background art section using the time series data of FIG. 7A. FIG. 7C is a diagram of time series data of RRI (RR interval) in which abnormal heartbeats are identified and abnormal heartbeats are removed by the method of the prior art 2, that is, the 2σ rule. In FIG.7 (c), a white circle is a heartbeat specified as an abnormal heartbeat. In FIG. 7C, a black circle is a heartbeat specified as a normal heartbeat. In FIG. 7C of the comparative example, the trend is not removed, and since it is within the average ± 2σ, there is a heartbeat from which an abnormal value is not removed. FIG. 7D shows a power spectrum obtained by performing spectrum analysis as HRV analysis on the RRI time series after removing the trend and abnormal heart rate shown in FIG. 7C. ing.

  In FIG. 7 (d), unlike the example of FIG. 6 (d), HRV (heart rate variability) was overestimated, and the peak observed in FIG. 6 (d) could not be confirmed. As a result, according to this method and this apparatus, when spectrum analysis is performed as HRV analysis, the accuracy of spectrum analysis can be improved.

<Explanation of sepsis onset warning device>
Here, sepsis will be described. Sepsis is a systemic inflammatory response syndrome caused by bacterial infection of the blood. When sepsis becomes severe, it causes organ dysfunction, decreased organ circulation, and decreased blood pressure. Furthermore, when it becomes severe, important organs such as lung, kidney, heart, and liver become dysfunctional after septic shock, and 40 to 60% die within 30 days. In particular, in the Intensive Care Unit (ICU), since various drains, catheters, and sensors including infusions and respirators are inserted into the patient's body, there is a high risk of infection.

  When sepsis develops, cytokines increase in the blood of the patient. Therefore, the onset of sepsis is detected by monitoring cytokines constantly or periodically. Cytokines are proteins secreted from cells of the immune system, and in particular, many are related to immunity and inflammation. However, constant monitoring of cytokines is not preferred because it can be a new source of infection. Moreover, there is a problem that the regular monitor takes too much time. Furthermore, cytokine detection has the problem of cost and time.

  Therefore, the electrocardiogram data of the ICU patient was acquired using the autonomic nerve monitor device 10 of the above embodiment, and the processing of S100 to S300 in FIG. 5 was executed to perform an analysis test of HRV (heart rate variability). 8 to 12 are HRV (heart rate variability) charts of cases of sepsis in each patient.

  In addition, about the onset of sepsis, it was performed with the consent of the patient admitted to the Gifu University School of Medicine hospital advanced lifesaving treatment center. It was also conducted with the approval of the Ethics Committee for Medical Research, Gifu University Graduate School of Medicine. Of vital data (electrocardiogram data) collected from a biological information monitor (Philips MP70) attached to a patient from the time of admission to the ICU, the electrocardiogram 606 days measured by the second lead was used. The measurement sampling frequency is 500 Hz.

  In the HRV analysis performed in S300 of FIG. 5, the RRI time series for 1 minute is resampled at equal intervals, then the power spectrum by the AR model (autoregressive model) is estimated, and the LF component and the HF component are estimated. , That is, a value obtained by integrating a frequency of 0.04 Hz to 0.4 Hz was used as an estimated value of HRV (heart rate fluctuation).

  The example shown in FIG. 8 shows the result of analyzing HRV (heart rate variability) from the time of ICU admission to 9 days after patient I (male, 65 years old, disease name: burn). In the figure, the horizontal axis represents the number of days (hours) elapsed since admission, and the vertical axis represents the HRV (heart rate variability) of the analysis result. Patient A had an extremely decreased urine volume around 23:00 on the 4th day, and treatment was suspected due to a sudden rise in blood pressure around 14:00 on the 5th day. However, HRV (heart rate variability) was first It started to decrease around 15:00 on the 4th day, and a downwardly convex pattern (hereinafter referred to as a V-shaped pattern) that started to increase around 4 o'clock on the 5th day was recognized. Before this V-shaped pattern appears, the minimum value of HRV (heart rate fluctuation) is not observed. At these times, no medical procedure that affects HRV (heart rate variability) is performed. Also in the following example, before the V-shaped pattern is recognized, no medical treatment that affects HRV (heart rate variability) is performed.

  The example shown in FIG. 9 shows the HRV transition from hospitalization to death for patient B (male, 67 years old, disease name: sinus bradycardia). In this patient B, an increase in the pulse respiration rate was confirmed around 18:00 on the fourth day. From the X-ray examination at 14:00 on day 5, patient B was diagnosed with septic shock due to pneumonia. In this example, the nutritional status of the patient B was poor and had weakened from the beginning of hospitalization, and similar characteristics appeared not only in the V-shaped pattern before the onset of septic shock but also immediately before death. Before the V-shaped pattern before the onset of septic shock appears, the minimum value of HRV (heart rate variability) is not observed.

  The example shown in FIG. 10 shows the result of analyzing HRV (heart rate variability) from the time of admission to the ICU to 9 days after the patient C (male, 65 years old, disease name: ruptured aneurysm). Also in the example shown in FIG. 10, a V-shaped pattern having a minimum value of HRV (heart rate variability) appeared before the onset of sepsis.

  The example shown in FIG. 11 shows the result of analyzing HRV (heart rate variability) from the 20th day to the 29th day of ICU admission for the patient c. In FIG. 11 as well, a V-shaped pattern having a minimum value of HRV (heart rate variability) appeared in the afternoon of 23rd before the day of 25th when septic shock appeared.

  The example shown in FIG. 12 shows the result of analyzing the HRV (heart rate variability) from the 8th day to the 18th day of ICU admission for the patient D (male, 23 years old, disease name: burn). Also in FIG. 12, a V-shaped pattern having a minimum value of HRV (heart rate variability) appeared in the afternoon of the 13th day before midnight on the 14th day when the septic shock appeared.

  As described above, it is presumed that this HRV (heart rate variability) V-shaped pattern is one time course until septic shock. Before reaching such septic shock, the appearance of a V-shaped pattern in HRV (heart rate variability) was first observed in this study.

  As described above, the RRI time series was estimated from the electrocardiogram of the biological information monitor, and the RRI time series from which the abnormal heartbeat and the trend were removed by S200 was obtained. Thereafter, HRV analysis was performed. As described above, as a result of application to 4 patients with septicemia, it was found that the characteristics of the V-shaped pattern of HRV (heart rate variability) were shown in 5 of 4 patients immediately before the onset of septic shock. That is, when HRV (heart rate variability) gradually decreases and HRV (heart rate variability) reaches the minimum value so far, and HRV (heart rate variability) starts to rise, after observing the minimum value, 1 It was found to cause septic shock on the order of ~ 3 days.

  Here, the cause of the HRV (heart rate variability) reduction of the V-shaped pattern is that the corticotropin (ACTH) is not secreted sufficiently from the pituitary gland, or the corticosteroid cannot be produced, and thus the adrenal cortex is not produced. It is possible that the stimulation hormone is not transmitted normally and does not fulfill its original role. On the other hand, the cause of the change from HRV (heart rate variability) decrease to increase has not been clarified at the present time.

<Sepsis warning device>
A case where the autonomic nerve monitor device 10 is used as a sepsis onset warning device will be described. The present embodiment has been made based on the above findings.

  In this case, a sepsis onset warning program is stored in the ROM 34. In the sepsis onset warning program, the sepsis onset warning process is performed after the various processes (S100 to S300) described in the abnormal heartbeat, trend removal and HRV analysis processing program.

  In the present embodiment, the HRV analysis of the CPU 32 is performed on the frequency axis. Also, the same result as above is obtained in the HRV analysis on the time axis. The HRV analysis is not limited to the HRV analysis on the frequency axis or the time axis, and may be performed by other HRV analysis.

  In the sepsis onset warning process, the CPU 32 detects a change in HRV (heart rate variability) obtained from the HRV analysis. Specifically, it is determined whether the first derivative of HRV (heart rate variability) is zero and the second derivative is positive. When the first derivative of HRV (heart rate variability) is zero and the second derivative is positive, the CPU 32 determines that the HRV (heart rate variability) is a minimum value and associates it with the determined time as a minimum value. And stored in the database 38. Subsequently, the CPU 32 compares the minimum value stored in the database 38 with the latest minimum value stored in the previous process. Here, the most recent value may be an empirical value, but is preferably within several hours. In this comparison, the CPU 32 determines that the current minimum value is the minimum value of the local minimum values so far, and the CPU 32 performs warning control on the warning device 44. That is, if the warning device 44 is a display lamp, the display lamp is turned on or blinked to warn the patient that there is a risk of developing septic shock. Also, if the warning device 44 is a buzzer, it is sounded to warn the patient that there is a risk of developing septic shock.

In the present embodiment, the CPU 32 corresponds to RRI time series acquisition means, removal means, HRV analysis means, first processing means, second processing means, and detection means.
The sepsis onset warning device configured as described above can issue warnings at times P1 to P13 in FIGS. 8 to 12, respectively, when applied to the examples of FIGS. In this case, in each of the examples of FIGS. 8 to 12, it is possible to give a prior warning that septic shock will occur at the time of P2, P5, P7, P10, P13, that is, at least half a day before the occurrence of septic shock. . Thus, the patient can be treated for sepsis before septic shock occurs.

The effects exhibited by this embodiment will be described below.
(1) In the method for removing abnormal heartbeats and trends from electrocardiogram data according to the present embodiment, when removing a trend from time series data of RR intervals representing intervals between beats included in the electrocardiogram data, the probability density after removal After removing the abnormal heart rate and trend so that the distribution is close to the normal distribution, HRV (heart rate fluctuation) is calculated. As a result, abnormal heartbeats such as arrhythmia and trends are automatically removed from the heartbeat interval data string obtained from the electrocardiogram, and changes in heartbeat variability obtained as a result of HRV analysis can be used for appropriate diagnosis. .

(2) In the method for removing abnormal heartbeats and trends from the electrocardiogram data according to the present embodiment, as a first step, time series data of RR intervals (hereinafter referred to as RRI time series _xn ) representing the interval between beats from the electrocardiogram data. To get. As a second step, the residual series x _n −f (t _n ; θ), n = 1, 2,. . . Among them, the flag time series Q including the trend function f (t _n ; θ) and the normal heartbeat is determined so that the probability density distribution for the case where the flag indicating that q _n is normal is established is closest to the normal distribution. Thus, after removing the abnormal heart rate and trend, HRV (heart rate fluctuation) data is calculated. As a result, the effect (1) can be easily realized.

(3) The autonomic nerve monitor apparatus of this embodiment includes an electrocardiogram waveform recorder 20 (acquisition means) that acquires electrocardiogram data. Then, the CPU 32 of the electrocardiogram waveform analyzer 30 acquires RRI time series _xn from the electrocardiogram data as RRI time series acquisition means. Further, the CPU 32 uses residual series x _n −f (t _n ; θ), n = 1, 2,. . . Among them, the flag time series Q including the trend function f (t _n ; θ) and the normal heartbeat is determined so that the probability density distribution for a case where a flag indicating that _n is normal is established is closest to the normal distribution. Thus, after removing the abnormal heart rate and trend, HRV (heart rate fluctuation) data is calculated. Further, the CPU 32 performs HRV analysis of RRI time series data from which abnormal heartbeats and trends have been removed as HRV analysis means.

  As a result, abnormal heartbeats such as arrhythmia and trends are automatically removed from information obtained from the electrocardiogram, and changes in HRV (heart rate variability) obtained as a result of HRV analysis can be used for appropriate diagnosis. An autonomic nerve monitor device can be provided.

  (4) In the sepsis onset warning device of the present embodiment, the CPU 32 detects a minimum value in the time series of HRV (heart rate variability) data based on the HRV analysis result as a detection means. The sepsis onset warning device uses the warning device 44 as a warning means, and warns that there is a risk of developing sepsis when the CPU 32 detects a minimum value in the time series of HRV (heart rate variability) data. As a result, abnormal heartbeats and trends such as arrhythmia are automatically removed from information obtained from the electrocardiogram, and changes in HRV (heart rate variability) obtained as a result of HRV analysis can be used for diagnosis of the onset of sepsis.

In addition, the said embodiment can also be changed and comprised as follows.
In the above-described embodiment, the autonomic nerve monitor device is configured to function as a sepsis onset warning device. However, the autonomic nerve monitor device can be used not only as a sepsis onset warning device but also as an autonomic nerve monitor device. In this case, in the above-described embodiment, this can be realized by adopting a configuration in which the CPU 32 performs processing up to HRV analysis.

  In the above embodiment, the normality measure employs equation (1), but the log likelihood for the maximum likelihood estimate

を使用してもよい、又、他の正規性尺度として、正規性の検定法を利用してもよい。

May be used, and a normality test may be used as another normality measure.

  The warning means is not limited to a lighting lamp or a buzzer, but may be a warning transmitter. In this case, when the warning transmitter transmits a warning signal via wireless or wired control under the control of the CPU 32, the wired or wireless receiver provided in the nurse station or the portable wireless receiver held by the doctor receives it. In this case, a receiver or the like that has received the warning signal notifies that there is a risk of developing septic shock based on the warning signal.

10 ... Autonomic nerve monitor device, 20 ... ECG waveform recorder,
30 ... ECG waveform analyzer,
32 ... CPU (RRI time series acquisition means, removal means, HRV analysis means, first processing means, second processing means and detection means), 38 ... database, 44 ... warning device (warning means).

Claims (10)

  When removing abnormal heartbeats and trends from time series data of RR intervals representing beat-to-beat intervals included in ECG data, abnormal heartbeats and trends were removed so that the probability density distribution after removal was close to the normal distribution Thereafter, HRV (heart rate variability) data is calculated, and a method for removing abnormal heart rate and trend from electrocardiogram data. A first step of acquiring time series data of an RR interval (hereinafter referred to as RRI time series _xn ) representing an interval between beats from electrocardiogram data;
f (t _n ; θ) is a trend function representing a long-period component of the electrocardiogram data, t is time, θ is a parameter of the trend function f, and q _n is normal of the RRI time series x _n or Assuming that the flag represents an abnormality and Q = (q ₁ , q ₂ , q ₃ ,... Q _n ) is the flag time series, the residual series x _n −f (t _n ; θ), n = 1,2 ,. . . Among them, the flag time series Q including the trend function f (t _n ; θ) and the normal heartbeat is determined so that the probability density distribution for a case where a flag indicating that _n is normal is established is closest to the normal distribution. A method of removing abnormal heartbeats and trends from electrocardiogram data, comprising: a second step of calculating HRV (heartbeat fluctuation) data after removing abnormal heartbeats and trends. The second step includes
A step of fixing the flag time series Q, determining a trend function that minimizes the average error of the residual series, and obtaining the residual series;
A B step of removing an abnormal heart beat and updating the residual series by determining a flag time series Q so that the residual series obtained in the A step has the most normality. Item 3. A method for removing abnormal heartbeats and trends from electrocardiogram data according to Item 2. In the second step,
When the flag time series Q determined in the B step changes differently from the flag time series Q fixed in the A step, the process returns to the A step and the flag time series Q is updated to the changed flag time series Q. 4. The abnormal heartbeat of the electrocardiogram data according to claim 3, wherein the A step is performed, and thereafter, the A step and the B step are repeated until the flag time series Q determined in the B step does not change. And trend removal method.   In the B step, when the flag time series Q is determined so that the residual series obtained in the A step has the most normality, the abnormal heart beat and the abnormal heart beat continuous with respect to the abnormal heart beat included in the residual series. 5. The abnormal heart rate and trend removal of the electrocardiogram data according to claim 4, wherein an abnormal value likelihood is calculated so as to include a heart beat, and an abnormal value and a heart beat continuous to the abnormal heart beat are removed based on the abnormal value likelihood. Method. Acquisition means for acquiring electrocardiogram data;
RRI time series acquisition means for acquiring time series data of RR intervals (hereinafter referred to as RRI time series _xn ) representing beat-to-beat intervals from the electrocardiogram data;
f (t _n ; θ) is a trend function representing a long-period component of the electrocardiogram data, t is time, θ is a parameter of the trend function f, and q _n is normal of the RRI time series x _n or Assuming that the flag represents an abnormality and Q = (q ₁ , q ₂ , q ₃ ,... Q _n ) is the flag time series, the residual series x _n −f (t _n ; θ), n = 1,2 ,. . . Among them, the flag time series Q including the trend function f (t _n ; θ) and the normal heartbeat is determined so that the probability density distribution for a case where a flag indicating that _n is normal is established is closest to the normal distribution. Removing means for extracting RRI time series data used for HRV analysis after removing abnormal heartbeats and trends,
An autonomic nerve monitor apparatus comprising: HRV analysis means for performing HRV (heart rate variability) analysis on RRI time-series data from which the abnormal heartbeat and trend are removed. The removing means includes
First processing means for fixing the flag time series Q, determining a trend function that minimizes an average error of the residual series, and obtaining the residual series;
Including a second processing means for removing the abnormal heartbeat and updating the residual series by determining the flag time series Q so that the residual series obtained by the first processing means has the most normality. The autonomic nerve monitor apparatus according to claim 6. The removing means includes
When the flag time series Q determined by the second processing means changes differently from the flag time series Q fixed by the first processing means, the flag time series Q is returned to the first processing means to change the flag time series Q. The first processing means and the second processing means are processed until the flag time series Q determined by the second processing means does not change. The autonomic nerve monitor apparatus according to claim 7, wherein:   In the second processing means, when determining the flag time series Q so that the residual series obtained by the first processing means has the most normality, the abnormal heartbeat and the abnormal heartbeat are related to the abnormal heartbeat included in the residual series. 9. The autonomic nerve monitor apparatus according to claim 8, wherein an abnormal value likelihood is calculated so as to include a heartbeat continuous with a heartbeat, and an abnormal value and a heartbeat continuous with the abnormal heartbeat are removed based on the abnormal value likelihood. The autonomic nerve monitor device according to any one of claims 6 to 9, is a sepsis onset warning device,
Detecting means for detecting a minimum value in a time series of HRV (heart rate variability) data according to an analysis result of the HRV analyzing means;
A sepsis onset warning device characterized by comprising warning means for warning that there is a risk of developing sepsis when the detecting means detects a minimum value in the time series of the HRV (heart rate variability) data.

Images