JP2009279384A - Signal processing method, signal processing apparatus, and pulse photometer using the same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a signal processing method for separating signals to obtain a signal component and an artifact component if large artifacts such as movement of limbs, sobbing, shaking, and coughing are mixed in; and a pulse photometer using the same. <P>SOLUTION: In the signal processing method for separating a pulse wave signal and an artifact signal from two measuring signals extracted from the same medium, vectors of the two measuring signals are separated into pulse wave signal vectors and artifact signal vectors by a separation matrix according to a norm ratio in a stable section of the pulse wave signal and a successively-compensated norm ratio in an artifact section. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

The present invention relates to signal processing for extracting two common signals extracted from a single medium at the same time to extract a common signal component, and more particularly, a pulse used in the medical field, particularly for diagnosis of the circulatory system. The present invention relates to improvement of signal processing in a photometer.

Various methods have been proposed for separating a signal component and a noise component from two signals extracted almost simultaneously from one medium.
In general, they are processed in the frequency domain or the time domain.
Even in the medical field, a device that measures the pulse waveform and pulse rate, which is called a photoelectric plethysmograph, a concentration measurement of light-absorbing substances in blood, a device for measuring oxygen saturation SpO2, carbon monoxide hemoglobin, Met hemoglobin, etc. A device for measuring the concentration of special hemoglobin, a device for measuring the concentration of injected dye, and the like are known as pulse photometers.
In particular, the oxygen saturation SpO2 measuring device is called a pulse oximeter.

The principle of the pulse photometer is that pulse wave data obtained by transmitting or reflecting light of multiple wavelengths with different absorption to the target substance to the living tissue and continuously measuring the amount of transmitted or reflected light. The concentration of the target substance is obtained from the signal.
If noise is mixed in the pulse wave data, the correct concentration cannot be calculated, and there is a risk of mishandling.
Conventionally, in a pulse photometer, in order to reduce noise, a method of dividing a frequency band and paying attention to a signal component or taking a correlation between two signals has been proposed.
However, these methods have a problem that analysis takes time.

Therefore, the applicant of the present invention is different in Patent No. 32701717 (see Patent Document 1).
Draw a graph with the ordinate and abscissa representing the magnitude of each of the two pulse wave signals obtained from transmitted light by irradiating light of one wavelength to the living tissue, and then calculating its regression line, based on the slope of the regression line Thus, it has been proposed to obtain the oxygen saturation or the concentration of light-absorbing substance in arterial blood.
According to the present invention, measurement accuracy can be improved and power consumption can be reduced.
However, in order to obtain the regression line or its inclination using a large amount of sampling data for pulse wave signals of each wavelength, a lot of calculation processing is still required.

Furthermore, the present applicant uses frequency analysis in the prior application (see Patent Document 2).
In the analysis, instead of extracting the pulse wave signal itself as in the prior art, the fundamental frequency of the pulse wave signal is obtained, and in order to improve the accuracy, the pulse wave signal is used using a filter using the harmonic frequency. A method of filtering is proposed.

In particular, pulse oximeters are expected to be used in various ways at home, and there are a variety of artifacts, requiring higher artifact resistance than SpO2 used in hospitals.
If artifacts are mixed, the measurement system is disturbed and SpO2 may be displayed erroneously.
The typical artifact is body movement, which is caused by the movement of the probe worn by the person to be measured, movement of the optical path between the light source and the light receiving surface, or deformation of the tissue due to the force applied to the living tissue. .
Especially in newborns and toddlers, there are many artifacts, movement of limbs, crying, trembling,
Cough is a source of artifacts.
Japanese Patent No. 32701717 JP 2003-135434 A

An object (object) of the present invention is to provide a signal processing method, a signal processing apparatus, and a signal processing method capable of measuring SpO2 more accurately even when large artifacts such as movements of limbs, crying, tremors, and coughs are mixed. It is to provide a pulse photometer used.

In order to solve the above problems, a signal processing method for separating a pulse wave signal and an artifact signal from two measurement signals extracted from the same medium, wherein a measurement signal vector based on the two measurement signals is separated into a matrix. Therefore, when separating into a pulse wave signal vector and an artifact signal vector, the pulse wave signal is separated according to a norm ratio of a stable section of the pulse wave signal and a sequentially corrected norm ratio of the artifact section. (Claim 1)
The successive correction norm ratio is obtained by the following equation (8). (Claim 2)

 Further, when the successive correction norm ratio becomes a predetermined condition, the successive correction norm ratio is set to a predetermined value.
It is characterized by replacing. (Claim 3)

3. A biological signal processing apparatus, wherein a measurement signal is processed by the signal processing method according to claim 1 or 2. (Claim 4)
In addition, the biological signal processing apparatus converts two kinds of light emitted from the light emitting means to light transmitted through or reflected from the biological tissue into an electric signal, and sequentially converts the artifact signal component using the corrected norm ratio. A pulse photometer characterized in that a pulse wave signal is obtained by removal and at least one of oxygen saturation, special hemoglobin concentration, or injected dye concentration in arterial blood is calculated. (Claim 5)
Further, when the successive correction norm ratio becomes a predetermined condition, the successive correction norm ratio is set to a predetermined value.
It is characterized by replacing. (Claim 6)

According to the configuration of claims 1 to 6, a signal processing method, a signal processing apparatus, and a signal processing method that can more accurately separate signal components even when large artifacts such as movements of limbs, crying, tremors, and coughs are mixed The used pulse photometer can be realized.

In describing an embodiment of the present invention, the principle will be described by taking a pulse oximeter for measuring arterial blood oxygen saturation as an example.
The technique of the present invention is not limited to a pulse oximeter, and a blood photoabsorbing substance such as special hemoglobin (carbon monoxide hemoglobin, Met hemoglobin, etc.) or a dye injected into blood is used by the principle of pulse photometry. Applicable to measuring device (pulse photometry).

The configuration of the pulse oximeter for measuring arterial blood oxygen saturation is as shown in FIG. 1, which is a schematic configuration block diagram.
The light-emitting elements 1 and 2 that emit light of different wavelengths are driven by the drive circuit 3 so as to emit light alternately.
The light used for the light-emitting elements 1 and 2 is infrared light that is less affected by arterial oxygen saturation (
For example, 940 [nm]), red light (eg, 660 with high sensitivity to changes in arterial oxygen saturation)
[nm]) is good.

Light emitted from the light emitting elements 1 and 2 is transmitted through the living tissue 4, received by the photodiode 5, and converted into an electrical signal.
In FIG. 1, transmitted light is received, but reflected light may be received.
These converted signals are amplified by the amplifier 6 and distributed by the multiplexer 7 to the filters 8-1 and 8-2 corresponding to the respective optical wavelengths.
The signals distributed to the filters are filtered by the filters 8-1 and 8-2 to reduce noise components, and are digitized by the A / D converter 9.

Each signal sequence corresponding to digitized infrared light and red light forms a respective pulse wave signal.
Each digitized signal sequence is input to the processing unit 10, processed by a program stored in the ROM 12, the oxygen saturation SpO 2 is measured, and the value is displayed on the display unit 11.

First, measurement of the change in absorbance (light attenuation) of a light-absorbing substance in blood will be described with reference to FIG.
2A and 2B are pulse wave data in which the light emitted from the light emitting elements 1 and 2 is transmitted through the living tissue 4 and received by the photodiode 5 and converted into an electrical signal. (a) shows the case of red light, and (b) shows the infrared light.
In FIG. 2A, assuming that the horizontal axis is time and the vertical axis is the light reception output, the light reception output from the photodiode 5 is obtained by superimposing the direct current component (R ′) and the pulsation component (ΔR ′) of red light. It has a waveform.
In FIG. 2 (b), when the horizontal axis is time and the vertical axis is the received light output, the received light output from the photodiode 5 includes the direct current component (IR ′) and the pulsation component (ΔIR ′) of infrared light. The waveform is superimposed.

The measurement waveforms in FIGS. 2A and 2B each include an artifact (noise) due to body movement, and there is a frequency band common to the frequency component of the artifact and the frequency component of the signal. It was difficult to accurately grasp the signal component by removing the artifact component.

In the example of FIG. 2, the artifact is small, but in the following description, a large artifact corresponding to a large artifact such as movement of limbs, crying, trembling, coughing, etc. is artificially added and described based on a measurement example.

FIG. 3 is an observation signal measured with a two-wavelength probe attached to the fingertip of a healthy male.
The upper row is infrared light (IR), and the lower row is red light (R).
The wavelengths are 940 nm (infrared light) and 660 nm (red light), respectively.
The black line section shown in the upper part of FIG. 3 indicates a tapping section in which artifacts are artificially added,
The operation of repeatedly hitting the desk surface quickly (for example, about 3 Hz) with the fingertip on the probe mounting side is performed.

In the example of FIG. 3, the pulse wave cannot be visually recognized in the artifact section in both the infrared light and red light observation signals.
The observation signal is normalized with each direct-current transmitted light component, then filtered with a 6th-order Butterworth filter with a bandwidth of 0.5 to 5 Hz, and the sampling interval is 16 ms.

FIG. 4 is a correlation diagram of the observation signals of the infrared light and the red light shown in FIG.
The horizontal axis is infrared light, and the vertical axis is red light.
The slope in FIG. 4 is given by the norm ratio of infrared light and red light in the artifact interval of the observation signal.
From the inclination, the norm ratio “φN” of the artifact is known.

Next, signal processing for separating a pulse wave signal and noise (artifact signal) will be described with reference to FIG.
FIG. 5 is a diagram obtained by modeling the example of FIG. 3 for signal processing.
The pulse wave signal p at time tn is transmitted to the IR terminal with a coefficient of 1, and is transmitted to the R terminal with a coefficient φS.
Similarly, the artifact signal n at time tn is transmitted to the IR terminal with a coefficient of 1 and R
The signal is transmitted to the terminal with a coefficient φN.
φS and φN are the pulse wave signal φS: = Rp, tn / IRp, tn and the artifact signal φN: = Rn, tn / IRn, tn, respectively.
If the observation time is expanded to tn: tn + k, p, n, IR, and R become vectors.

In the following, bold is a vector.
The subscript pulse represents a pulse wave stable section, and the subscript noise represents an artifact section.
The stable section of the pulse wave and the artifact section are different in time and section length.

Next, FIG. 6 shows the results of three types of simulations in which the relationship between the amplitude of the pulse wave signal (pulse) and the amplitude of the artifact signal (Artifact) is changed.
6A, 6B, and 6C, the horizontal axis represents the IR signal, and the vertical axis represents the R signal.
In the simulation, the pulse wave was a sawtooth wave, and the artifact was a sine wave.
The relationship between the amplitude of the pulse wave signal and the amplitude of the artifact signal is shown in FIG.
lse) and the amplitude of the artifact signal (Artifact) are (0.25: 0.
75).
In FIG. 6B, the amplitude of the pulse wave signal (pulse) and the amplitude of the artifact signal (Artifact) are (0.33: 0.66), respectively.
In FIG. 6C, the amplitude of the pulse wave signal (pulse) and the amplitude of the artifact signal (Artifact) are (0.5: 0.5), respectively.
The norm ratio φS = 0.55 expressed by Equation 1 above, and the true value of the norm ratio expressed by Equation 2 above.

Is 1.
The correlation diagrams shown in (a) to (c) of FIG. 6 are all parallelograms, and artifacts>
In the case of a pulse wave, the inclination of the short side coincides with φS, and the inclination of the long side coincides with φN.
If artifact <pulse wave, the slope of the long side coincides with φS, and the slope of the short side coincides with φN.

The dotted lines on the correlation diagrams of FIGS. 6A to 6C are the “norm ratio” of the observation signal with the inclination of “φN”. The solid line is the “corrected norm ratio” to be proposed later as a reference example.

Shows the slope.
"Corrected norm ratio" is the slope of the long side of the parallelogram "true value of norm ratio",

Is consistent with
It can be seen that the difference between the “norm ratio” and the “true value of the norm ratio” increases in order as the amplitudes of the pulse wave and the artifact approach.

Next, separation by the norm ratio between the pulse wave signal and the artifact signal will be described.
The observation signal vector [IR R] T is separated into a pulse wave signal vector p and an artifact signal vector n by a separation matrix S.
The “T” on the shoulder represents transposition.

From the following equation (1), the mixing matrix M is determined by defining “φS” and “φN”.
The separation matrix S is an inverse matrix of M shown in Expression (2).
Here, “φS” is the norm ratio of the observation signal in the stable section of the pulse wave.
If the remaining “φN” can be estimated, M having [1 φS] T and [1 φN] T as basis vectors is determined.

S = M−1 may be used.
“ΦN” is the norm ratio of the observation signal in the artifact section.
In the stable pulse wave section, “φNpulse” = 1.

Here, || || 2 represents a two-dimensional Euclidean norm. Subscript pulse = tl: l
+ N noise = tj: j + k indicates a stable section and an artifact section of the pulse wave, respectively.
The time and the number of samples are different in the pulse wave, the stable section, and the artifact section.
The advantage of obtaining “φS” and “φN” from the norm ratio is that it is robust against sporadic noise from the definition of Euclidean norm.

The observation signal is a combined vector of a pulse wave vector and an artifact vector having different inclinations.
Therefore, even in the artifact section, because the pulse wave is superimposed,

And the true value of the norm ratio,

Diverge.
Therefore, as a reference example, the “corrected norm ratio” is used as a means of correcting the deviation.

There is a method of using.
This correction method is based on the premise that the pulse wave superimposed on the artifact section has the same amplitude as the stable section.
The correction is performed by the following equation (4).

Is referred to as the “corrected norm ratio” of the artifact section, and “φN” is referred to as the norm ratio of the section.
In the artifact section, the pulse wave signal is often buried and difficult to observe.
Therefore, take the norm in the stable section of the pulse wave signal,

Is the pulse wave amplitude.
The amplitude of the artifact signal is the norm of the artifact interval.

It is.
Here, since the number of sample points in the pulse wave section and the artifact section are different, it is divided by the square root of the number of sample points in each section.
The absolute value takes into account that the inside of the root sign becomes negative due to observation noise other than artifacts.

The result of separating the observation signal by the separation matrix S by using the “corrected norm ratio” as a reference example will be described with reference to FIG.
FIG. 7 is a diagram showing a waveform obtained by separating the observation signal by the separation matrix S. As shown in FIG.
The sampling interval in FIG. 7 is 16 ms, and the filter is a 6th-order Butterworth filter with a bandwidth of 0.5 to 5 Hz.
The horizontal line at the top of the waveform in FIG. 7 indicates the artifact section.
FIG. 7A shows a pulse wave separated by “norm ratio” and “φS”, and FIG.
It is a pulse wave separated by “corrected norm ratio” and “φS”.

In the former, a large amplitude needle-like artifact is conspicuous.
In the latter, the pulse wave can be clearly recognized, and in particular, the pulse wave after 20 seconds is more clearly separated.
On the other hand, the artifact increases in the section of 12.5 seconds to 18.5 seconds.

FIG. 8 is a diagram showing a result of performing a moving average of 17 points on the artifact section on the pulse wave obtained by separating the observation signal shown in FIG.
FIG. 8A shows a pulse wave separated by “norm ratio” and “φS”, and FIG.
It is a pulse wave separated by “corrected norm ratio” and “φS”.
It can be seen that the separation based on the “corrected norm ratio” reduces the needle-like artifacts of large amplitude compared to the separation based on the norm ratio “φN”.
In the separation based on the “corrected norm ratio”, artifacts are reduced in a section of 17 to 23 seconds, and a smooth pulse wave shape is shown. Further, even in the section of 30 to 35 seconds, a smooth pulse wave shape is shown by the separation based on the “corrected norm ratio”.

Since pulse waves and artifacts are considered independent, if the separation is appropriate, the correlation diagram approaches a square or rectangle.
FIG. 9 is a diagram showing the correlation after separation.
FIG. 9A shows the correlation between the pulse wave and artifact separated by “norm ratio” and “φs”.
b) Correlation between pulse wave and artifact separated by “corrected norm ratio” and “φs”, (c)
Is the correlation between the pulse wave and artifact obtained by moving average of 17 points of the pulse wave after separation with “norm ratio” and “φs”, (d) is the pulse wave after separation with “corrected norm ratio” and “φs” The correlation of the pulse wave and the artifact which performed the moving average of 17 points | pieces is shown. From (a), it can be seen that (b) and (d) are properly separated from (c).

In observation, since the “true value of the norm ratio” is unknown, it is not known how close the “corrected norm ratio” is to the “true value of the norm”.
Therefore, the result of separation by the “corrected norm ratio” is evaluated by the evaluation function H of the following equation (6). [Σ] is a variance-covariance matrix of the separated pulse wave vector p and artifact vector n.
H is the ratio of the absolute value of trace [Σ] and the diagonal element 2Σ12.
If the correction is appropriate, [Σ] approaches a diagonal matrix. The smaller the value of H, the higher the independence of p and n.

FIG. 10 shows the result of evaluating the goodness of separation of the pulse wave and the artifact by the evaluation function H.
The upper part of FIG. 10 shows the case where the separation is performed based on the “norm ratio”, and the lower part shows the case where the separation is performed based on the “corrected norm ratio”.
In each case, the evaluation results when the filter bandwidth is 0.5 to 5 Hz, the sixth-order Butterworth filter, and the moving average of 17 points after the processing of the filter are shown.
It can be seen that the diagonal separation is improved when the separation is performed by the “corrected norm ratio”, with H = 0.0042 and H = 0.0492, which are smaller values.

The observation waveform of FIG. 3 used for explaining the “corrected norm ratio” as the above reference example artificially adds an artifact, so that the “corrected norm ratio” is a fixed value.
However, actual artifacts usually change over time, in which case
It is “the true value of the norm ratio”

Changes, but it is a "corrected norm ratio"

Is a fixed value, unable to follow the change, and the quality of the separation is also reduced.
The present invention relates to a technique for improving the point that the “corrected norm ratio”, which is the reference example described above, is not sufficient for artifacts that change with time.

Therefore, the amount of artifact leakage into the pulse wave in the model of FIG. 5 will be examined.
A true mixing matrix

And the inverse matrix is defined as a separation matrix S.

と前記分離マトリクスＳのパラメータ値が異なれば、脈波にアーチファクトが漏れ込む。
漏れ込み比率Ｌは、式（７）で評価できる。
例えば、脈波とアーチファクトの振幅をそれぞれ１とし、

The true mixing matrix

If the parameter values of the separation matrix S are different, an artifact leaks into the pulse wave.
The leakage ratio L can be evaluated by Expression (7).
For example, the amplitude of the pulse wave and the artifact is set to 1,

After mixing the pulse wave and artifact with

When the separation is performed at, the difference between the “norm ratio” and the “true norm ratio” of only 5% leaks 16% of the artifact amplitude to the pulse wave side, and this leakage amount is not small.
In general, the “true norm ratio” changes with time, whereas the “corrected norm ratio” is a constant value in the artifact section and cannot follow the time change of the “true norm ratio”, so some countermeasure is required. become.

As described above, since the “norm ratio” and the “true norm ratio” are deviated due to the pulse wave being superimposed also in the artifact section, in the present invention, as means for correcting the deviation, "Sequential correction norm ratio"

Propose.
This correction method is based on the premise that the amplitude of the pulse wave superimposed on the artifact section is the same as the amplitude of the pulse wave in the stable section.
The correction is sequentially performed for each sample point by the following equation (8). this,

Is referred to as a successively-compensated-norm-ratio at the sampling time J of the artifact interval.
It is difficult to observe the pulse wave in the artifact section because it is buried in the artifact. The pulse wave amplitude takes the norm of the stable section of the pulse wave,

And
On the other hand, the artifact amplitude at point J takes k points before and after J, and a total norm of 2k + 1 points.

Is the artifact amplitude at time J.
Here, because the number of samples in the pulse wave section and the artifact section are different,

Is the square root of the number of samples in the interval,

Divided by. The absolute value considers the case where the root sign is negative due to observation noise other than artifacts.

Using the “sequential correction norm ratio” given by equation (8) above, the observation signal is separated into matrices
The result of separation by S is shown in FIG. 11 together with the case where “norm ratio” and “corrected norm ratio” which is a reference example are used.
FIG. 11A shows the case where the artifact section is separated by “sequential correction norm ratio”.
FIG. 11B shows the case where the artifact section is separated by the “corrected norm ratio”.
FIG. 11C shows the case where the artifact section is separated by the “norm ratio”.

In FIG. 11, comparing the pulse waves in the section of 5 to 20 seconds, the artifact amplitude is smaller than that of B and C in the separation of A by “sequential correction norm ratio”.
In particular, there is a difference in amplitude at an artifact around 20 seconds, and the presence of pulse waves of B and C can be visually recognized at around 8 seconds, but in A, it is clearly separated as a pulse wave that is well connected to the front and back.

Compared with B and C, the shape of the pulse wave is clearly separated at the base before the artifact around 17 seconds.
Furthermore, the steep artifact rising immediately before 20 seconds is clearly separated as compared with B and C.

FIG. 12 is a diagram showing artifacts separated by three types (A, B, C) of separation methods.
From FIG. 12, it can be understood that the artifact amplitude is separated in the order of C>B> A.
Therefore, it can be understood that the “sequential correction norm ratio” of the present invention is corrected to a more appropriate value and superior to the “norm ratio” and the “corrected norm ratio” which is the reference example.

In observation, since the “true value of the norm ratio” is unknown, it is not known how close the “corrected norm ratio” is to the “true value of the norm”.
Therefore, the result of separation by the “corrected norm ratio” is evaluated by the evaluation function H of the above equation (6). [Σ] is a variance-covariance matrix of the separated pulse wave vector p and artifact vector n.
H is the ratio of the absolute value of trace [Σ] and the diagonal element 2Σ12.
If the correction is appropriate, [Σ] approaches a diagonal matrix. The smaller the value of H, the higher the independence of p and n.

The result is shown in FIG.
The result of evaluation by the evaluation function H indicates that the smaller the value of H, the better the separation.
It can be understood that the result evaluated by the evaluation function H is also separated in the order of C>B> A, similarly to the result of FIG.

  Further, the separation matrix S is “sequential correction norm ratio”.

  Depending on the value taken by, it may be ill-conditioned, and vibration and spikes may occur.
  The following invention relates to an improved technique for alleviating the adverse conditions of the separation matrix S.

・ Instability of separation matrix and noise amplification factor
  (1) Unstable area
  In the inverse problem, as a general property, the solution becomes unstable if the separation matrix is ill-conditioned.
  FIG. 14 (c) shows the number of samples k = 10 obtained from the observed waveforms shown in FIGS. 14 (a) and 14 (b).
Φ _N The norm ratio Sφ _S It is.
  As shown in FIG. 14 (c), Sφ _N The ratio has changed greatly.
  The bar in Fig. 14 (c) indicates the φ obtained in the stable pulse wave interval (1 to 3.5 seconds). _S And φ _S = 0.5357
The
  Bar wire and Sφ _N Near the intersection of Sφ _N ≒ φ _S The separation matrix is a region where vibrations and spikes occur under adverse conditions.
  Generally, φ of healthy person _S Is about 0.55.

  (2) Eigenvector and eigenvalue of separation matrix
  If there is observation noise in the observation signal, abnormalities such as an increase in amplitude due to separation may occur.
The This anomaly occurs when the separation matrix is ill-conditioned.
Toll and eigenvalue are related.

  In order to examine adverse conditions, the separation matrix S is expressed by equation (9).

  ε is a small value. The value of ε continuously changes from linearly independent to linearly dependent. If ε> 0
For example, the column vector of S is linearly independent. If ε = 0, it is a linearly dependent and bad matrix.

 ε ≒ 0
  When this happens, an abnormality occurs.
  In the two-wavelength SpO2, the separation matrix S is 2 × 2, two eigenvectors V and two
It has an eigenvalue λ.
  In Expression (10), G is a signal source vector such as a pulse wave, and O is an observation signal vector. Isolated matto
The eigenvectors and eigenvalues of Rix S are given by equations (11) and (12). V ₁ And V ₂ Is standardized
It is an eigenvector. If the observed signal O is equal to the eigenvector V, the matrix is fixed.
Equation (13) is established from the relationship between the existence vector and the eigenvalue. The norm of the separated signal source vector G
Can be expressed by equation (14).

  If ε <1, λ is a large value.
  FIG. 15 shows the noise expansion coefficient obtained from the equation (14).
  φ _N Φ _S It is expressed as a ratio. φ _S  Was 0.55. The noise enhancement factor is φ _N / Φ _S  = 1.2
About 15 times more, φ _N / Φ _S  = 1.3, which is about 10 times. φ _N / Φ _S  = 1 is infinite.

  (3) Noise expansion direction and noise expansion coefficient
  FIG. 15 shows the noise expansion coefficient. Here, the noise expansion direction is confirmed. In addition to the observation signal
As the norm 1 noise, sin of amplitude 1 was added to the IR observation signal, and cos was added to the R observation signal.
Independent random noise N (0.1) was studied. Φ of separation matrix _S = 0.55, φ _N / Φ
_S  FIG. 16 shows the simulation result when = 1.3.

  The center circle in FIG. 16 (a) is a unit circle formed by sin and cos. The alternate long and short dash line is φ _N / Φ _S = 1.3 and
This is the eigenvector direction of the separation matrix.
  The ellipse is the locus drawn by the separated signal when the unit circle signal is separated by the separation matrix.
The The amplitude of the observation signal increases in the eigenvector direction. Its amplitude is the length of the ellipse's long axis vertex and origin, φ _N / Φ _S = 1.3 is about 10 times. In FIG. _N / Φ _S = 1.3 shows about 10 times the result
I agree.

  Similarly, the circular portion in FIG. 16B is the added random noise. Its peak amplitude is almost a circle
It is. A one-dot chain line is an eigenvector of the separation matrix. Noise spread even in the case of random noise
The large direction is the maximum in the eigenvector direction of the separation matrix. Its noise amplification factor is φ _N /
φ _S = 1.3, which is about 10 times.

・ Reduction of adverse conditions by gating
  (1) Processing flow
  Propose gating as a mitigation measure for adverse conditions.
  FIG. 17 shows a flow of processing for processing parameters related to the artifact.
  Gating opens the gate under adverse conditions, "sequential correction norm ratio"

  The

Replace with
  here,

Is φ _S  It shall be determined in advance with a value away from.
  The threshold for adverse conditions is the upper limit of ζ _u And lower limit ζ _l   Set with. ζ is φ _S Ζ _u And ζ _l  Double
The coefficient to be determined is determined in FIG.
  For example, ζ _u = 1.3, ζ _l = 1. This is the target coefficient with 1 in between.

  If so, set the separation matrix parameter to φ _S When

  And if,

  If the parameter _S When

 And "Sequential correction norm ratio after gating"

  Is written.

  (2) Selection of ζ
  For example, ζ = φ _N / Φ _S If you choose = 1.3, the noise enhancement factor is about 10. If the signal-to-noise ratio (SNR) of the observed signal can be secured to 1/100 (-40dB), the desired SNR after separation is about 1/10 (-20dB).
I will.
  Here, the increasing observation noise is a vector directed around the eigenvector direction of the mixing matrix.
It is. As the ellipse moves away from the direction of the eigenvector, the noise expansion coefficient decreases as shown by the ellipse in FIG.
.

  (3) Parameter values used in the stable pulse wave section
  Gating

  If so, set the separation matrix parameter to φ _S  When

 And
  Φ for tissues without blood _N Is about 1, so here

 Was set to 1.

Is selected, the separation matrix is approximated by equation (15), and the pulse wave in the stable section is projected as it is.

・ Evaluation of evaluation function and correction
  "True value of norm ratio"

  Is unknown.
  Therefore,

I don't know how close I was to Therefore, "Sequential correction norm ratio after gating"

  The result separated in step (5) is evaluated by the evaluation function H in equation (20). [Σ] is the separation shown in equation (16)
Is a variance covariance matrix of the pulse wave vector p and the artifact vector n.
  H is trace [Σ] and diagonal element 2Σ ₁₂ = 2Σ ₂₁ Is the ratio of the absolute values of. The correction is appropriate
If there is, [Σ] approaches a diagonal matrix. The smaller the value of H, the higher the independence of p and n
It is separation.

  FIG. 18 shows the result of separating the pulse wave from the observed waveforms of (a) and (b) of FIG.
  The start point of the artifact section is marked with | and the end point with>.
  A in FIG. 18 is “sequential correction norm ratio” without gating.

  And k = 10. Increase in amplitude and spikes occur in bad conditions
is doing.
  B in FIG. 18 indicates the gating coefficient ζ _u = 1.3, ζ _l = 0.7, k = 10
Norm ratio "

  This is the result of separation at. Spikes remain. Spike is ζ _u > 1.3 and ζ _l It occurs outside the threshold of <0.7. These remaining spikes will be described later.
  18C shows the gating coefficient ζ _u = 1.3, ζ _l = 0.7, k = 60
There is no spike. Sequential compensation for the first, second, and third sections of the artifact
Artifacts are greatly reduced by separation by positive norm ratio. Evaluation function value H = 0 without adjustment.
0382 was obtained.
  FIG. 18D shows the norm ratio φ of the entire observation section. _N This is the result of separation at. Evaluation function value H = 0.27
81.

  In FIG. 19, k = 60 and three types of ζ _u And ζ _l  Gating with the combination of
This is the result of wave separation.
  ζ _u And ζ _l  It can be seen that the occurrence of spikes increases as the interval of _u = 1.2 and ζ
_l At = 0.8, there are no spikes.
  If the threshold width is increased too much, gating is frequently performed, and an appropriate "sequential correction norm ratio"
Since the number of sections where the is used decreases, the quality of the separation decreases.

・ Consideration of remaining spikes
  In FIG. 18B, spikes remain in several places after gating.
  Therefore, the nature of spikes around 6 seconds is examined.
  FIG. 20 shows the pulse wave between 5.7 and 6.2 seconds in B of FIG.
Rum ratio "

  The time axis is expanding.
  A in FIG. 20 is a pulse wave separated by gating at k = 10, and there are a plurality of remaining spikes before 5.8 seconds and around 5.9 seconds.
  The solid line in FIG. 20B shows the “sequentially corrected norm ratio after gating” performed by gating.

  Is the trend. ○ indicates “sequential correction norm ratio” without gating

  Is the trend.
  On the vertical axis of the trend, the vertex for displaying the vicinity of the threshold is cut off.
  A one-dot chain line indicates a threshold value at which gating is performed. The threshold is ζ _u = 1.3, ζ _l = 0.7
.
  Gating is executed in the section where the ○ mark is within the threshold (for example, around 5.8 seconds to 5.9 seconds before)
Yes.
 “Sequential correction norm ratio after gating” was determined in advance for that section.

Has been replaced.

  A straight line portion of value 1 seen in the vicinity of 5.8 to 5.9 seconds in B of FIG. 20 is the section.
 Remaining spy where the trend of “Sequential corrected norm ratio after gating” is outside the threshold
This occurs at 5.8 seconds before and around 5.9 seconds.
  From this result, it is confirmed that gating is correctly executed within the threshold value specified in the initial setting.
I understand.

  C in FIG. 20 is a pulse wave separated at k = 17. The spike has disappeared.
  In FIG. 20D, the trend of “sequential correction norm ratio” is all above the threshold value, so
No ringing has been performed.
  The remaining spike occurs outside the threshold specified in the initial setting. These spikes are observed
The amplitude increases because the noise vector is in the same direction as the eigenvector direction of the separation matrix.
It is a spike that occurred.

・ Elimination of adverse conditions by regularization parameters
  As shown in Equation (18), a regularization parameter is added as an additional term to the separation matrix and
Λ, which is a small value, is adjusted so that the evaluation function value is minimized. Iterative operation to determine λ
Is necessary and the adjustment is difficult.

  FIG. 21 shows the result of mitigating adverse conditions using regularization parameters.
  The observed waveforms used are (a) and (b) in FIG.
  The upper part (a) of FIG. 21 is an example in which the adverse condition is relaxed at λ = 0.05 and k = 10. Evaluation function at λ = 0.05
The numerical value was the smallest at H = 0.0042. However, many spikes remain.
  (B) in the lower part of FIG. 21 is an example in which the adverse condition is relaxed at λ = 0.455 and k = 10. Spike disappears
However, the separation of the pulse wave is not good, and the evaluation function value is also bad as H = 0.888.

・ Example of processing of pulse wave buried in tapping artifact
  The pulse wave buried in the tapping artifact is the same parameter as B in FIG.
  In FIG. 18C, the gating coefficient is k = 60, ζ. _u = 1.3, ζ _l Separated using = 0.7
An example is shown in FIG.

  The upper part (a) of FIG. 22 shows the IR observation waveform. Pulse wave images are buried in artifacts and identified
difficult.
  The middle (b) and the bottom (c) of FIG. 22 are pulse waves separated by the successively corrected norm ratio by gating.
is there. In the middle (b), the pulse wave image is well separated in the tapping section, and spikes are generated.
Not in. The lower part (c) is the result of moving average of 17 points considering the period length of the pulse wave.
It is clearer than that. The evaluation function values are = 0.1302 and H = 0.0253.
  From the above description, the artifact is determined by the successive correction norm ratio for performing gating according to the present invention.
It is clear that the pulse wave can be separated without specifying the segment interval.

It is a block diagram of the configuration of a pulse oximeter for measuring arterial oxygen saturation. It is a figure which shows the example of a measurement of the fluctuation | variation of the light absorbency (light extinction degree) of the light absorption substance in the blood. This is an observation signal measured by wearing a two-wavelength probe on the fingertip of a healthy male and artificially mixing artifacts. It is a figure which shows the correlation of the observation signal of infrared light and red light shown in FIG. It is the figure which modeled the example of FIG. 3 for signal processing. It is a figure which shows the result of having performed three types of simulations which changed the relationship between the amplitude of a pulse wave signal (pulse) and the amplitude of an artifact signal (Artifact). It is a figure which shows the waveform which isolate | separated the observation signal by the separation matrix S. It is a figure which shows the result of having performed the moving average of 17 points | pieces to the pulse wave which isolate | separated the observation signal shown in FIG. It is a figure which shows the correlation after isolation | separation. It is a figure which shows the result of having evaluated the goodness of separation of a pulse wave and an artifact with evaluation function H. It is a figure which shows the result of having isolate | separated the observation signal using "sequential correction norm ratio", "norm ratio", and "correction norm ratio". It is a figure which shows the artifact isolate | separated by the separation method by "sequential correction norm ratio", "norm ratio", and "correction norm ratio". It is a figure which shows the result evaluated with the evaluation function H. It is a figure which shows the trend of the successive correction | amendment norm ratio in the sample number k = 10 obtained from the observed waveform of IR and R. FIG. It is a figure which shows the relationship between a noise expansion direction and a noise expansion coefficient. Φ of separation matrix _S = 0.55, φ _N / Φ _S  Simulation when = 1.3 FIG. It is a figure which shows the flow of a process which processes the parameter regarding an artifact. . It is a figure which shows the result of having isolate | separated the pulse wave from the observed waveform of (a) and (b) of FIG. Results of separating pulse waves with different parameters from observed waveforms in (a) and (b) of FIG. FIG. The time axis that is the trend of the pulse wave and “sequential correction norm ratio after gating” FIG. It is a figure which shows the result of relaxation of the unfavorable condition by the regularization parameter of this invention. It is a figure which shows another example of a process of the pulse wave buried in the tapping artifact.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Light emitting element 2 Light emitting element 3 Drive circuit 4 Living body tissue 5 Photodiode 6 Converter 7 Multiplexer 8 Filter 9 A / D converter 10 Processing part 11 Display part 12 ROM
13 RAM

Claims (6)

A signal processing method for separating a pulse wave signal and an artifact signal from two measurement signals extracted from the same medium,
When separating the measurement signal vector from the two measurement signals into a pulse wave signal vector and an artifact signal vector by a separation matrix, the norm ratio of the stable section of the pulse wave signal and the sequentially corrected norm ratio of the artifact section A signal processing method characterized by separating. The signal separation method according to claim 1, wherein the successive correction norm ratio is obtained by the following Equation 4.

 When the successive correction norm ratio becomes a predetermined condition, the successive correction norm ratio is set as a predetermined value.
The signal processing method according to claim 2, wherein the signal processing method is changed.
A biological signal processing apparatus, wherein a measurement signal is processed by the signal processing method according to claim 1 . The biological signal processing device converts two kinds of light emitted from the light emitting means, which are transmitted or reflected through the biological tissue, into electrical signals, and sequentially removes the artifact signal component using the corrected norm ratio. A pulse photometer that obtains a pulse wave signal and calculates at least one of oxygen saturation, special hemoglobin concentration, or injected dye concentration in arterial blood.
 When the successive correction norm ratio becomes a predetermined condition, the successive correction norm ratio is set as a predetermined value.
The pulse photometer according to claim 5, wherein the pulse photometer is replaced.

Images