CN102488517A - Method and device for detecting burst suppression state in brain signal - Google Patents

Abstract

The invention discloses a revolution quantitative analysis method, which is used for detecting a burst suppression wave in a brain signal and comprises the following steps of: collecting the brain signal; removing artifacts and noise in the brain signal; carrying out phase space reconstruction of the brain signal, the artifacts and the noise of which are removed, to obtain a revolution picture; quantitatively analyzing the revolution picture, and calculating a revolution ratio (RR) of the revolution picture; and setting a threshold, judging a burst suppression state of the signal according to the RR, and calculating a burst suppression ratio (BSR). The method disclosed by the invention is capable of judging the burst suppression state in the brain signal more accurately.

Description

A kind of method and device that detects outburst inhibitory state in the EEG signals

Technical field

The present invention relates to the detection method and the device of outburst inhibition ripple in the EEG signals.

Background technology

In recent years; The outburst that increasing research begins to pay close attention in the continuous EEG signals suppresses ripple; This ripple is a kind of special brain wave patterns that often occurs; Its waveform shows as the outburst state of high-amplitude high frequency on form and the inhibitory state of low amplitude value alternately occurs, and occurs with immesurable aperiodic pattern.Present existing brain electricity analytical system is mainly the EEG spectrum analysis.Because outburst suppresses the particularity and the unstable state characteristic of ripple, conventional frequency spectrum analysis method is difficult to detect accurately outburst and suppresses ripple.

Outburst the earliest suppresses the detection of waveform and adopts frequency spectrum analysis method (to please refer to KoskinenM; Sarkela M; SeppanenT; Jantti V, Suominen K.Characterizing EEG burst suppression with spectral analysis methods.Proceedings of the 1999 Finnish Signal Processing Symposium, 1999:132-136); Outburst suppresses signal and has comprised the change of amplitude and the vibration of different frequency, has adopted analysis of spectrum to extract the frequency and the power features of signal in the paper.But frequency spectrum analysis method has tangible limitation, and it has only analyzed the frequency of signal; Characteristics such as power; Be the analytical method of EEG signals being regarded as linear system, and EEG signals itself have very strong nonlinear characteristic, this can impact for the accuracy of judging; And being difficult to be given in the index (outburst suppresses) on the time scale based on the method for spectrum analysis, this also is one of limitation of spectrum analysis.

Non-linear energy factors (NLEO) is that Plotkin and Swamy at first proposes and be applied to analyze outburst in the EEG signals by Agarwal etc. to suppress waveform and (please refer to Sarkela M, Mustola S, Seppanen T; Koskinen M; Lepola P, Suomi-nen K, Juvonen T; Tolvanen-Laakso H, Jantti V.Automatic analysis and monitoring of burst suppression inanesthesia.J ClinMonit 2002; 17:125-134).Non-linear energy factors method is the time-domain analysis method; It can capture frequency and amplitude information, suppresses the variation characteristic of brain electricity analytical through the amplitude on the computation time yardstick based on the outburst of NLEO, compares with pre-set threshold; Signal is divided into detonator signal; Suppress signal, and artefact, and then calculate the inhibition time of appearance in a minute and the ratio of total time.

But because non-linear energy factors analytical method is based on the signal amplitude characteristic, choosing the variation meeting of amplitude of its threshold value is relatively more responsive.Because people's individuality difference; And the amplitude difference of all ages and classes stage EEG signals opposite sex; Cause result of calculation to lose efficacy based on amplitude information easily as the analytical method of threshold value, therefore, non-linear energy factors suppresses to exist certain limitation on the ripple detecting outburst.

In recent years, increasing nonlinear dynamic analysis begins to be applied to the detection of EEG signals.Wherein using is exactly the entropy index analysis more widely, comprise the Ke Ermogenuofu entropy (Kolmogorov Entropy, K2), state entropy, reaction entropy, approximate entropy etc.The major defect of these nonlinear dynamic analysis is to the dependency of data length and to the sensitivity of noise.EEG signals inherent instability and the various noises that comprised will directly influence the result of these methods.

Find by retrieval; One Chinese patent application 03137747.5 uses the nonlinear dynamic analysis method to come EEG signals are handled; What but its method mainly adopted correlation dimension, complexity and approximate entropy comprehensively provides a non-linear index; And be used for monitoring in real time the nonlinear trend judgement of brain electricity, consider that outburst suppresses the detection of ripple.

Find through retrieval again, Chinese patent ZL 200310122416.2, name is called: a kind of brain electricity fluctuation signal analysis equipment, this technology provides a kind of brain function detection method, and its main theoretical basis is a Frequency Spectral Theory.

Summary of the invention

The present invention is directed to the deficiency of prior art, the outburst that provides quantized analytical method of a kind of cycle and device to be used for detecting in the brain electricity suppresses ripple, can judge the outburst inhibitory state in the EEG signals comparatively accurately.In order to reach above-mentioned target, technical scheme of the present invention is:

A kind of method that detects outburst inhibitory state in the EEG signals may further comprise the steps:

Step 101 is gathered EEG signals;

Step 102 is removed artefact and noise in the EEG signals;

Step 103 is carried out phase space reconfiguration to the EEG signals behind removal artefact and the noise, obtains one-period figure;

Step 104 is carried out quantitative analysis to periodogram, calculates its cycle ratio RR;

Step 105, setting threshold is judged the outburst inhibitory state of signal according to cycle ratio RR, and calculates outburst rejection ratio BSR.

Remove the artefact of primary signal in the said step 102, adopt baseline filtration and Kalman's adaptive autoregressive model to combine and carry out filtering.

Said baseline filter method passes through each 50 point before and after each sampled point, and 101 points are made even all altogether, and from sampling point value, deduct the meansigma methods of being tried to achieve, and to remove baseline drift, its computing formula is following:

$x\_\mathrm{baseline}\left(j\right)=x\left(j\right)-\frac{1}{101}\underset{k=j-50}{\overset{j+50}{\mathrm{\Σ}}}x\left(k\right)$

In the formula, x (j) is that each sampled point must be worth, and x_baseline (j) is the signal value that filters each sampled point that obtains afterwards through baseline; Said Kalman's adaptive autoregressive model is a parameter of utilizing Kalman filter estimation self-adaptive autoregression model, utilizes forecast error to detect the high-amplitude noise; One of them has the p rank adaptive autoregressive models (AAR) of time-varying parameter with following formal description EEG signal y _t

y _t＝a _1，ty _t-1+a _2，ty _t-2+…+a _p，ty _t-p+v _t

In the formula, v _tBe that average is 0, variance is σ ²White-noise process, and v _tWith y _tUncorrelated, promptly covariance function is 0, y _T-i, i=1,2 ... P is at p the timed sample sequence of t before the moment, a _{I, t}, i=1,2 ... P is the auto-regressive parameter of dependence time; Forecast error (residual error) is the part that can not use the AAR model to characterize in the EEG signal, high-amplitude noise just,

$e_t={\hat{y}}_t-a_{1,t}{y}_{t-1}-a_{2,t}{y}_{t-2}-...-a_{p,t}{y}_{t-p}$

Be the EEG signals of actual measurement, utilize

Deduct a _{1, t}y _T-1+ a _{2, t}y _T-2+ ... + a _{P, t}y _T-pThe e that is obtained _tBe forecast error, i.e. the high-amplitude noise.

Said step 103 through the time-delay method carry out phase space reconfiguration, preset time sequence x ₁, x ₂... X _L, utilize the reconstruct on embedded space of time-delay coordinate method to embed vectorial X _k=(x _k, x _K+t..., x _{K+ (m-1) τ}), computing cycle figure then

I, j=1 ... N, in the formula, N is state vector X _iNumber, ε differentiates distance, || || expression L ₂Norm; Said step 104 statistical analysis periodogram catercorner length distributes, computing cycle ratio R R, and its computing formula: $\mathrm{RR}=\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="HDA0000119447290000012.png,HDA0000119447290000021.png"\; state="\{\{state\}\}">N</figure-callout>}}^2}\underset{i,j=1}{\overset{N}{\mathrm{\&Sigma;}}}{R}_{i,j}.$

A kind of device that detects outburst inhibitory state in the EEG signals comprises

Eeg signal acquisition equipment is used to gather original EEG signals;

A/D converter will be simulated the accurate digital signal that changes into of EEG signals;

A/D converter connection processing device, processor is used for removing the artefact of EEG signals, then the EEG signals behind removal artefact and the noise is carried out phase space reconfiguration, obtains one-period figure; Periodogram is carried out quantitative analysis, calculate its cycle ratio RR; Last setting threshold is judged the outburst inhibitory state of signal according to cycle ratio RR, and calculates outburst rejection ratio BSR.

Said processor is removed the artefact of primary signal, adopts baseline filtration and Kalman's adaptive autoregressive model to combine and carries out filtering.

Pass through when said processor adopting baseline is filtering each 50 point before and after each sampled point, 101 points are made even all altogether, and from sampling point value, deduct the meansigma methods of being tried to achieve, and to remove baseline drift, its computing formula is following:

$x\_\mathrm{baseline}\left(j\right)=x\left(j\right)-\frac{1}{101}\underset{k=j-50}{\overset{j+50}{\mathrm{\&Sigma;}}}x\left(k\right)$

In the formula, x (j) is that each sampled point must be worth, and x_baseline (j) is the signal value that filters each sampled point that obtains afterwards through baseline.

Said processor adopting Kalman's adaptive autoregressive model utilizes the parameter of Kalman filter estimation self-adaptive autoregression model; Utilize forecast error to detect the high-amplitude noise, one of them has the p rank adaptive autoregressive models (AAR) of time-varying parameter with following formal description EEG signal y _t

y _t＝a _1，ty _t-1+a _2，ty _t-2+…+a _p，ty _t-p+v _t

In the formula, v _tBe that average is 0, variance is σ ²White-noise process, and v _tWith y _tUncorrelated, promptly covariance function is 0, y _T-i, i=1,2 ... P is at p the timed sample sequence of t before the moment, a _{I, t}, i=1,2 ... P is the auto-regressive parameter of dependence time; Forecast error (residual error) is the part that can not use the AAR model to characterize in the EEG signal, high-amplitude noise just,

$e_t={\hat{y}}_t-a_{1,t}{y}_{t-1}-a_{2,t}{y}_{t-2}-...-a_{p,t}{y}_{t-p}$

Be the EEG signals of actual measurement, utilize

Deduct a _{1, t}y _T-1+ a _{2, t}y _T-2+ ... + a _{P, t}y _T-pThe e that is obtained _tBe forecast error, i.e. the high-amplitude noise.

Said processor through the time-delay method carry out phase space reconfiguration, preset time sequence x ₁, x ₂X _L, utilize the reconstruct on embedded space of time-delay coordinate method to embed vectorial X _k=(x _k, x _K+t..., x _{K+ (m-1) τ}), computing cycle figure then

I, j=1 ... N, in the formula, N is state vector X _iNumber, ε differentiates distance, || || expression L ₂Norm.

Said processor statistical analysis periodogram catercorner length distributes, computing cycle ratio R R, and its computing formula: $\mathrm{RR}=\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="HDA0000119447290000012.png,HDA0000119447290000021.png"\; state="\{\{state\}\}">N</figure-callout>}}^2}\underset{i,j=1}{\overset{N}{\mathrm{\&Sigma;}}}{R}_{i,j}.$

A kind of periodic quantity fractional analysis method and apparatus provided by the invention is used for detecting outburst and suppresses ripple, obtains periodogram, computing cycle ratio R R with break out rejection ratio BSR, can judge the outburst inhibitory state in the EEG signals comparatively accurately.

Description of drawings

Fig. 1 is a workflow sketch map of the present invention.

Fig. 2 A comprises the primary EEG signals that break out inhibitory state.

Fig. 2 B carries out the EEG signals that obtain after baseline filters to the primary signal among Fig. 2 A.

Fig. 2 C carries out the filtering of Kalman's adaptive autoregressive model to the EEG signals that obtain among Fig. 2 B to handle the EEG signals that obtain afterwards.

Fig. 3 A is to outburst, respectively gets 10 seconds EEG signals figure (being 3000 data points under the sample frequency 100HZ) under inhibition and the normal condition.

Fig. 3 B is outburst, the seasonal effect in time series periodogram that inhibition and normal condition are formed.

Fig. 4 demonstration utilizes period map method that the EEG signals of Fig. 2 C are analyzed resulting result.

Fig. 5 A shows that includes only the eeg data that outburst suppresses data.

Fig. 5 B demonstration utilizes the non-linear energy factors method result that analysis obtains to Fig. 5 a-signal.

Fig. 5 C shows the result who utilizes period map method that Fig. 5 a-signal is analyzed.

Fig. 6 is the schematic block diagram of collection analysis brain electricity of the present invention.

The specific embodiment

Fig. 1 is a workflow diagram of the present invention, at first is step 101, gathers EEG signals.The collection of EEG signals is accomplished by eeg signal acquisition equipment; The special Ag/AgCl electrode of general employing; Perhaps existing proprietary electrode for encephalograms medicated cap; Cooperate eeg amplifier to obtain EEG signals, convert the brain electric analoging signal that obtains to digital signal through A/D converter then, and with the digital signal input processor.

Step 102 is taken out artefact and noise in the EEG signals.The artefact of removing original EEG signals is a very important step, and the bandwidth of EEG signals is generally (0-100Hz), has comprised the composition of various frequencies; But HFS receives the interference of myoelectricity easily; Also have electrocardio simultaneously, interference such as eye electricity, these interferential amplitudes all are greater than brain.If do not remove the artefact of other frequency ranges, can cause the analysis result of many mistakes, therefore, in algorithm of the present invention, the EEG signals that extracted are selected the signal of 0.5-30Hz frequency band.Be one section like Fig. 2 A and comprise the primary EEG signals that outburst suppresses ripple.At first adopt baseline filtration and Kalman's adaptive autoregressive model to combine and carry out filtering.

Wherein, baseline filters and is equivalent to a high-pass filtering, here; The baseline filter method passes through each 50 point before and after each sampled point, and 101 points are made even all altogether, and from sampling point value, deduct the meansigma methods of being tried to achieve; To remove baseline drift, its computing formula is following:

$x\_\mathrm{baseline}\left(j\right)=x\left(j\right)-\frac{1}{101}\underset{k=j-50}{\overset{j+50}{\mathrm{\&Sigma;}}}x\left(k\right)$

In the formula, x (j) is that each sampled point must be worth, and x_baseline (j) is the signal value that filters each sampled point that obtains afterwards through baseline.Fig. 2 B is the EEG signals after baseline filters.

Kalman's adaptive autoregressive model is a parameter of utilizing Kalman filter estimation self-adaptive autoregression model, utilizes the change procedure of forecast error to detect of short duration high-amplitude noises such as electromyographic signal.

P rank adaptive autoregressive models (AAR) that have time-varying parameter are with following formal description EEG signal y _t

y _t＝a _1，ty _t-1+a _2，ty _t-2+…+a _p，ty _t-p+v _t

In the formula, v _tBe that average is 0, variance is σ ²White-noise process, and v _tWith y _tUncorrelated, promptly covariance function is 0.y _T-i, i=1,2 ... P is at p the timed sample sequence of t before the moment, a _{I, t}, i=1,2 ... P is the auto-regressive parameter of dependence time, change in time, so the application card Thalmann filter is estimated auto-regressive parameter.

Utilize the Kalman Filter Estimation device to calculate forecast error (being residual error again).Residual error is the part that can not use the AAR model to characterize in the EEG signal, just is about to the high-amplitude noises such as myoelectricity that remove:

$e_t={\hat{y}}_t-a_{1,t}{y}_{t-1}-a_{2,t}{y}_{t-2}-...-a_{p,t}{y}_{t-p}$

is the EEG signals of actual measurement.Utilize

Deduct a _{1, t}y _T-1+ a _{2, t}y _T-2+ ... + a _{P, t}y _T-pThe e that is obtained _tBe forecast error, i.e. the high-amplitude noise.This process is promptly earlier estimated to such an extent that the AAR parameter is described EEG signal y through Kalman filter _t, obtain EEG signals through actual measurement then and deduct y _tObtain residual error, remove residual error and be equivalent to remove afterwards the artefact in the EEG signals.

Through removing the purpose that residual error realizes removing the high-amplitude artefact, Fig. 2 C carries out the filtering of Kalman's adaptive autoregressive model to the EEG signals that obtain among Fig. 2 B to handle the EEG signals that obtain afterwards.

Step 103 after removing the artefact of EEG signals, carried out phase space reconfiguration to new EEG signals, preset time sequence x ₁, x ₂... X _L, utilize the reconstruct on embedded space of time-delay coordinate method to embed vector:

X _k＝(x _k，x _k+t，…，x _k+(m-1)τ)

In the formula, X _k=(x _k, x _K+t..., x _{K+ (m-1) τ}), τ is called time delay, and m is called the embedding dimension.Utilize the phase space observation procedure of propositions such as Eckmann to come analysis state X _iRecursiveness.Track on the m dimension phase space representes to adopt following formula to handle in the recurrence of one 2 dimension space:

Wherein, N is state vector X _iNumber, ε differentiates distance, || || expression L ₂Norm.As the i moment and j state vector X constantly _iAnd X _jSpace length R during less than ε _{I, j}=1, then be called periodic state, represent by stain in the drawings.In this example; Like Fig. 3 A, the EEG signals (sample rate is 1000 data points under 100Hz) that each state was got 10 seconds are used for setting up phase space, and dimension k and delay time T are confirmed by nearest-neighbor method and cross-correlation function respectively; In this instance; K=3, τ=1, current; Differentiation does not have unified standard apart from choosing of ε; In this algorithm, selected threshold ε is

(σ is the seasonal effect in time series standard variance, and m is the phase space dimension).In phase space, i can represent with stain or white point through one 2 dimension square formation constantly and at j periodic state constantly, obtain periodogram like this.Shown in Fig. 3 B, carry out periodogram analysis to the time series that one section outburst, inhibition and three states of normal condition are formed, the frame in the lower left corner is pairing to be the periodogram under the inhibitory state; Wherein stain is densely distributed, and inhibitory state demonstrates certain rules property, and middle boxes is pairing to be the periodogram under the outburst state; Wherein stain distributes sparse; Be illustrated in that periodic state seldom appears in time series under the outburst state, the frame in the upper right corner is pairing to be the periodogram under the normal condition, and wherein stain is evenly distributed; Similar with the periodogram distribution of white noise, there is not certain rules property.It is thus clear that, can distinguish outburst and suppress and normal condition based on the different distributions of cyclic graph mid point under the different conditions.

Distribution character for periodic state point among the further analytical cycle figure; Thereby better portray the seasonal effect in time series dynamics; The periodogram diagonal structure is studied, and statistical analysis periodogram catercorner length distributes, and proposes periodogram quantitative analysis method.Adopt cycle ratio RR here, as shown in the formula:

$\mathrm{RR}=\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="HDA0000119447290000012.png,HDA0000119447290000021.png"\; state="\{\{state\}\}">N</figure-callout>}}^2}\underset{i,j=1}{\overset{N}{\mathrm{\&Sigma;}}}{R}_{i,j}$

The density of stain among the indication cycle figure, promptly periodic state point with the ratio of possible state.

Cycle ratio RR for obtaining confirms threshold range according to RR values different under three kinds of states then, and here threshold setting is 0.1.When the RR value greater than 0.2 the time, think inhibitory state, time through calculating 1 fen inhibitory stage in the clock time and the ratio of total time obtain outburst rejection ratio BSR.Fig. 2 A, 2B, 2C have provided EEG signals that write down 16 minutes, wherein show as tangible outburst inhibitory state from about the 9th minute to the 12nd minute brain electricity.Be to have extracted three segment datas in the EEG signals among Fig. 3 B, be respectively inhibitory stage, outbreak period, normal brain activity electricity, represent to obtain the recurrence figure of two dimension through recurrence.Because the amplitude of period map method and signal does not concern directly that threshold value is also insensitive to the variation of signal amplitude.Fig. 4 utilizes period map method that the EEG signals among Fig. 2 C are carried out the result that unitary analysis obtains, and can find out, RR value desired value under normal condition and outburst inhibitory state has obvious variation.BS is the time period diagram through outburst that obtains after the RR analysis of threshold and inhibition, 0 expression inhibitory state, and 1 expression outburst state can be found out outburst and the inhibition stage when threshold value gets 0.1, can distinguished accurately.Outburst rejection ratio BSR be through statistics inhibitory stage time span and total time length ratio obtain.

For the reliability of further proving period figure method for outburst inhibition monitoring, one section signal (shown in Fig. 5 A) that the outburst inhibition is concentrated through manual construction utilizes period map method analysis then, and in this segment signal, data length is 80 seconds.The outburst that Fig. 5 B is based on the NLEO method suppresses to judge that the outburst that Fig. 5 C is based on RR suppresses to judge.Can know by Fig. 5 A; It is straight line that NLEO carries out the BS that threshold process obtains in analysis, is accurately to distinguish outbreak period and inhibitory stage, can be known by Fig. 5 B; The BS that period map method obtains can distinguish outbreak period and inhibitory stage clearly, can find out that NLEO suppresses to judge inefficacy to outburst.

Fig. 6 be one based on this analysis method, can be exclusively used in the sketch map that check and analysis outbursts suppresses the brain electricity analytical monitoring system of ripple.At first utilize electrode to gather EEG signals, convert digital signal to through analog/digital converter then, digital filter is for further processing to the signal that collects, and artefact is wherein removed.Then utilize phase space reconfiguration to set up the phase space vector, cycle ratio RR is tried to achieve in the cycle quantitative analysis of utilization, is judging that outburst suppresses setting threshold, tries to achieve and confirms that type numerical value BSR also shows intuitively.The design's method and scheme can be integrated in existing brain electric system, perhaps are used for independent detection and use.

Claims (10)

1. a method that detects outburst inhibitory state in the EEG signals is characterized in that, may further comprise the steps:

Step 101 is gathered EEG signals;

Step 102 is removed artefact and noise in the EEG signals;

Step 103 is carried out phase space reconfiguration to the EEG signals behind removal artefact and the noise, obtains one-period figure;

Step 104 is carried out quantitative analysis to periodogram, calculates its cycle ratio RR;

Step 105, setting threshold is judged the outburst inhibitory state of signal according to cycle ratio RR, and calculates outburst rejection ratio BSR.

2. a kind of method that detects outburst inhibitory state in the EEG signals according to claim 1 is characterized in that, removes the artefact of primary signal in the said step 102, adopts baseline filtration and Kalman's adaptive autoregressive model to combine and carries out filtering.

3. a kind of method that detects outburst inhibitory state in the EEG signals according to claim 2; It is characterized in that; Said baseline filter method passes through each 50 point before and after each sampled point, and 101 points are made even all altogether, and from sampling point value, deduct the meansigma methods of being tried to achieve; To remove baseline drift, its computing formula is following:

$x\_\mathrm{baseline}\left(j\right)=x\left(j\right)-\frac{1}{101}\underset{k=j-50}{\overset{j+50}{\mathrm{\&Sigma;}}}x\left(k\right)$

In the formula, x (j) is that each sampled point must be worth, and x_baseline (j) is the signal value that filters each sampled point that obtains afterwards through baseline; Said Kalman's adaptive autoregressive model is a parameter of utilizing Kalman filter estimation self-adaptive autoregression model, utilizes forecast error to detect the high-amplitude noise; One of them has the p rank adaptive autoregressive models (AAR) of time-varying parameter with following formal description EEG signal y _t

y _t＝a _1，ty _t-1+a _2，ty _t-2+…+a _p，ty _t-p+v _t

In the formula, v _tBe that average is 0, variance is σ ²White-noise process, and v _tWith y _tUncorrelated, promptly covariance function is 0, y _T-i, i=1,2 ... P is at p the timed sample sequence of t before the moment, a _{I, t}, i=1,2 ... P is the auto-regressive parameter of dependence time; Forecast error (residual error) is the part that can not use the AAR model to characterize in the EEG signal, high-amplitude noise just,

$e_t={\hat{y}}_t-a_{1,t}{y}_{t-1}-a_{2,t}{y}_{t-2}-...-a_{p,t}{y}_{t-p}$

Be the EEG signals of actual measurement, utilize

Deduct a _{1, t}y _T-1+ a _{2, t}y _T-2+ ... + a _{P, t}y _T-pThe e that is obtained _tBe forecast error, i.e. the high-amplitude noise.

4. a kind of method that detects in the EEG signals outburst inhibitory state according to claim 3 is characterized in that said step 103 is carried out phase space reconfiguration through the method for time-delay, preset time sequence x ₁, x ₂X _L, utilize the reconstruct on embedded space of time-delay coordinate method to embed vectorial X _k=(x _k, x _K+t..., x _{K+ (m-1) τ}), computing cycle figure then I, j=1 ... N, in the formula, N is state vector X _iNumber, ε differentiates distance, || || expression L ₂Norm; Said step 104 statistical analysis periodogram catercorner length distributes, computing cycle ratio R R, and its computing formula: $\mathrm{RR}=\frac{1}{N^2}\underset{i,j=1}{\overset{N}{\mathrm{\&Sigma;}}}{R}_{i,j}.$

5. a device that detects outburst inhibitory state in the EEG signals is characterized in that, comprises

Eeg signal acquisition equipment is used to gather original EEG signals;

A/D converter will be simulated the accurate digital signal that changes into of EEG signals;

A/D converter connection processing device, processor is used for removing the artefact of EEG signals, then the EEG signals behind removal artefact and the noise is carried out phase space reconfiguration, obtains one-period figure; Periodogram is carried out quantitative analysis, calculate its cycle ratio RR; Last setting threshold is judged the outburst inhibitory state of signal according to cycle ratio RR, and calculates outburst rejection ratio BSR.

6. a kind of device that detects outburst inhibitory state in the EEG signals according to claim 5 is characterized in that said processor is removed the artefact of primary signal, adopts baseline filtration and Kalman's adaptive autoregressive model to combine and carries out filtering.

7. a kind of device that detects outburst inhibitory state in the EEG signals according to claim 6; It is characterized in that; Pass through when said processor adopting baseline is filtering each 50 point before and after each sampled point, 101 points are made even all altogether, and from sampling point value, deduct the meansigma methods of being tried to achieve; To remove baseline drift, its computing formula is following:

$x\_\mathrm{baseline}\left(j\right)=x\left(j\right)-\frac{1}{101}\underset{k=j-50}{\overset{j+50}{\mathrm{\&Sigma;}}}x\left(k\right)$

In the formula, x (j) is that each sampled point must be worth, and x_baseline (j) is the signal value that filters each sampled point that obtains afterwards through baseline.

8. according to claim 6 or 7 described a kind of devices that detect outburst inhibitory state in the EEG signals; It is characterized in that; Said processor adopting Kalman's adaptive autoregressive model utilizes the parameter of Kalman filter estimation self-adaptive autoregression model; Utilize forecast error to detect the high-amplitude noise, one of them has the p rank adaptive autoregressive models (AAR) of time-varying parameter with following formal description EEG signal y _t

y _t＝a _1，ty _t-1+a _2，ty _t-2+…+a _p，ty _t-p+v _t

In the formula, v _tBe that average is 0, variance is σ ²White-noise process, and v _tWith y _tUncorrelated, promptly covariance function is 0, y _T-i, i=1,2 ... P is at p the timed sample sequence of t before the moment, a _{I, t}, i=1,2 ... P is the auto-regressive parameter of dependence time; Forecast error (residual error) is the part that can not use the AAR model to characterize in the EEG signal, high-amplitude noise just,

$e_t={\hat{y}}_t-a_{1,t}{y}_{t-1}-a_{2,t}{y}_{t-2}-...-a_{p,t}{y}_{t-p}$

Be the EEG signals of actual measurement, utilize

Deduct a _{1, t}y _T-1+ a _{2, t}y _T-2+ ... + a _{P, t}y _T-pThe e that is obtained _tBe forecast error, i.e. the high-amplitude noise.

9. according to claim 5 or 6 described a kind of devices that detect in the EEG signals outburst inhibitory state, it is characterized in that said processor carries out phase space reconfiguration through the method for time-delay, preset time sequence x ₁, x ₂X _L, utilize the reconstruct on embedded space of time-delay coordinate method to embed vectorial X _k=(x _k, x _K+t..., x _{K+ (m-1) τ}), computing cycle figure then I, j=1 ... N, in the formula, N is state vector X _iNumber, ε differentiates distance, || || expression L ₂Norm.

10. a kind of device that detects outburst inhibitory state in the EEG signals according to claim 9 is characterized in that said processor statistical analysis periodogram catercorner length distributes, computing cycle ratio R R, and its computing formula: $\mathrm{RR}=\frac{1}{N^2}\underset{i,j=1}{\overset{N}{\mathrm{\&Sigma;}}}{R}_{i,j}.$

Images