CN101879058B - Method for segmenting intracranial pressure signal beat by beat - Google Patents

Abstract

The invention provides a method for segmenting an intracranial pressure signal beat by beat. In the method, a difference vector between points is taken as a basic characteristic with shifting and rotating invariance, so the influence of baseline drift of the intracranial pressure signal can be overcome. Meanwhile, logarithm polar coordinate transformation is performed on the difference vector to measure the similarity of waveforms, and the measurement is sensitive to the morphological characteristics of adjacent waveforms, can capture the global profile information of the waveforms and has robustness to waveform vibration. Furthermore, different weights are given to the characteristics of the waveforms before and after a point to be detected so as to reduce the influence of an interference point at the beat tail of the intracranial pressure signal on identification and further improve the identification rate; and the interference point in the intracranial pressure signal is further removed by setting a proper threshold value so as to avoid confusion of segmentation beat by beat. The method can accurately segment the intracranial pressure signal beat by beat, and is favorable for improving detection and analysis capability of intracranial pressure analysis equipment.

Description

A kind of beat-to-beat division method of intracranial pressure signal

Technical field

The present invention relates to intracranial pressure and detect automatically and analysis technical field, particularly a kind of beat-to-beat division method of intracranial pressure signal based on wave character extraction and template matching.

Background technology

Intracranial hypertension is the common cause that causes the intracranial disease death, grasps the level and the quantitative Diagnosis of patient's intracranial pressure in time, exactly, is a vital step of clinical treatment.Intracranial hypertension can cause a series of physiological function disorder and pathological change, shows typical case such as katzeniammer, vomiting, papilloedema performance, serious intracranial hypertension also can concurrent pulmonary edema etc. complication; Also can form compressing or destroy hypothalamus and cause the autonomic nervous dysfunction because of cerebral hernia, and threat to life at short notice, be that neural Medicine and Surgery disease causes main causes of death.For the clinician, accurately know patient's intracranial pressure height and intracranial pressure variation tendency, for judge the state of an illness, instruct treatment, rescue life and cure after check judge it all is very important.

Intracranial pressure signal fluctuates along with heartthrob, this is that each of heart is put out and caused arteriodilating result, therefore intracranial pressure signal is a kind of paracycle of propagating in the mode of continuous fluctuation of a signal by shooting, continuously by shooting in the starting point of each beat representing the arrival of an intracranial pressure signal composition during the heartbeat.One section intracranial pressure signal as shown in Figure 2, its mid point p ₁, p ₂, p ₃Be the beat starting point.Can see from Fig. 2, although being with heartthrob, intracranial pressure signal changes paracycle, its waveform morphology is but very irregular, go back the high-frequency rhythmicity fluctuation of simultaneous in the signal waveform, and also there is the drift that rises and falls in the baseline of signal, this mainly is because intracranial pressure signal is subjected to the influence of many-sided factor, and for example the active motion effects of whole body blood vessel and cerebrovascular intracranial pressure, and intracranial pressure is to the change of venous pressure also unusual sensitivity etc.Simultaneously, because the influence of venous pressure, feel suffocated, firmly also can cause that the intracranial pressure signal generation obviously fluctuates accordingly, particularly actions such as cough, sneeze can cause that more intracranial pressure signal produces corresponding violent shake, these fluctuate and shake can form violent interfering signal, shown in the latter half of intracranial pressure signal among Fig. 3.Because these complicated factors, normal beat by computer identification intracranial pressure signal is relatively more difficult, therefore never occur one and well adopt computer implemented beat-to-beat division method of intracranial pressure signal, intracranial pressure checkout equipment clinically all selects to calculate the intracranial pressure meansigma methods mostly as clinical indices; And cutting apart by shooting of intracranial pressure signal can only be by manually cutting apart behind the artificial visually examine, under the huge situation of intracranial pressure signal data volume, this artificial dividing method is not only very consuming time, and because different clinicists' viewpoint difference to some extent, segmentation standard is uneven, causes cutting apart that efficient is low, accuracy is difficult to guarantee.

Yet intracranial pressure signal is gone back other important informations of carrier except that its average.For example, intracranial pressure signal during a heartbeat comprises 2～3 crests that amplitude is bigger usually, there is research to point out, the ratio of the amplitude of wherein preceding 2 crests can be predicted the degree of intracranial hypertension, the wave slope of these preceding 2 crest lines can react the compliance of hydrocrania system, and all reactions to some extent in the intracranial pressure waveform of the compliance of the self-discipline of cerebral blood flow and cerebrospinal fluid system.Therefore, intracranial pressure signal is cut apart by shooting, determined normal beat in the intracranial pressure signal, and then extract useful information from the normal signal beat, medical field is significant ahead of the curve.

Summary of the invention

At the prior art above shortcomings, the technical problem that the inventive method solved provides the computer implemented beat-to-beat division method of intracranial pressure signal of a kind of employing, to reduce intracranial pressure clinicist's workload, that improves intracranial pressure signal cuts apart efficient and accuracy by shooting.This method is applied to possess in the cutting techniques by shooting of intracranial pressure analytical equipment to continuous intracranial pressure signal of computer function, helps to improve the detection and the analysis ability of intracranial pressure analytical equipment.

The object of the present invention is achieved like this: a kind of beat-to-beat division method of intracranial pressure signal, the intracranial pressure signal that ICP (monitor intracranial pressure monitor) is gathered is imported computer, carry out low-pass filtering and sampling pretreatment by computer, then intracranial pressure signal is cut apart by shooting, the concrete steps of cutting apart by shooting comprise:

A) set up K mutually different template signal; Each template signal is the segment signal of each one-period before and after beat starting point in the intracranial pressure signal of having discerned, and this segment signal is a N sampled point by sampling or interpolation processing; Wherein, K 〉=2, the span of N is 100～1000;

B) set up N sampled point in each template signal respectively with respect to the normalization log-polar of its beat starting point;

C) for intracranial pressure signal to be measured, extracting duration from its starting point is t ₀Signal segment as detection segment; Then detection segment is carried out autocorrelation analysis, the interval in the auto-correlation function of calculating detection segment between every adjacent two local maximums is got the approximate cycle of the meansigma methods of described interval as detection segment; Wherein, t ₀Span be 30～90s;

D) calculate detection segment rise in from the beginning to ε the doubly approximate cycle place all local minimum points; Wherein, the span of ε is 1.2～1.6;

E) extract the characteristic area of each local minimum point; The characteristic area of each local minimum point is the segment signal in each approximate cycle before and after this local minimum point in the intracranial pressure signal to be measured, and this segment signal is a N sampled point by sampling or interpolation processing;

F) set up N sampled point in the characteristic area of each local minimum point respectively with respect to the normalization log-polar of its local minimum point;

G) calculate the characteristic area of each local minimum point and each template signal cross-correlation coefficient respectively, with the similarity of the maximum in the cross-correlation coefficient of the characteristic area of each local minimum point and each template signal as this local minimum point based on the normalization log-polar; The computing formula of described cross-correlation coefficient is:

And,

Wherein, P _{I, k}For current ε in the detection segment doubly is similar to the cycle with the characteristic area of interior i local minimum point and the cross-correlation coefficient of k template signal; (β _{I, n}, γ _{I, n}) be in the detection segment the doubly approximate cycle of current ε with the normalization log-polar of n sampled point in the characteristic area of interior i local minimum point with respect to this local minimum point, β _{I, n}Be normalization utmost point footpath, γ _{I, n}Be polar angle;

Be in k the template signal n sampled point with respect to the normalization log-polar of its beat starting point, α _{K, n}Be normalization utmost point footpath,

Be polar angle; K ∈ 1,2 ..., K}, n ∈ 1,2 ..., N}; W _pThe expression weight, the span of its weight coefficient λ is 0＜λ＜1;

H) relatively draw the local minimum point of doubly approximate cycle of current ε in the detection segment with interior similarity maximum, and with the similarity and the pre-set threshold C of this local minimum point ₀Compare; If its similarity is greater than threshold value C ₀, judge that promptly this local minimum point is a beat starting point; Wherein, threshold value C ₀Span be 0.1～0.4;

I) be starting point with the doubly approximate cycle of current ε in the detection segment with a local minimum point of interior similarity maximum, calculate thereafter the doubly approximate cycle of ε with interior all local minimums point; Repeating step e then)～i), judge beat starting points all in the detection segment thus;

J) in intracranial pressure signal to be measured, be starting point with last beat starting point position in the current detection section, extracting thereafter, duration is t ₀Signal segment as new detection segment; And, with the meansigma methods of last 3 normal cycle intervals in the current detection section the approximate cycle as new detection segment; Repeating step d then)～j), judge beat starting points all in the intracranial pressure signal to be measured thus;

Described normal cycle interval is meant the interval that is no more than 1.5 times between adjacent two beat starting points and is not less than 0.5 times of current approximate cycle duration;

K) intracranial pressure signal to be measured is cut apart by shooting, stored and show intracranial pressure signal beat by beat segmentation result to be measured.

In the above-mentioned steps, described step b) is specially:

B1) set up N sampled point in each template signal respectively with respect to Descartes's relative coordinate of its beat starting point, and carry out the average normalized; The computing formula of average normalized is as follows:

${{\mathrm{\ρ}}_{k,n}}^{\text{'}}=\frac{{\mathrm{\ρ}}_{k,n}}{{\stackrel{\‾}{\mathrm{\ρ}}}_{k,n}}=\frac{\sqrt{{x_{k,n}}^2+{y_{k,n}}^2}}{\frac{1}{N-1}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{{x_{k,n}}^2+{y_{k,n}}^2}};$

θ _{K, n}'=θ _{K, n}, and θ _{K, n}' ∈ (π, π];

Wherein, (x _{K, n}, y _{K, n}) be in k the template signal n sampled point with respect to Descartes's relative coordinate of its beat starting point, (ρ _{K, n}, θ _{K, n}) be and (x _{K, n}, y _{K, n}) corresponding polar coordinate; (ρ _{K, n}', θ _{K, n}') be (ρ _{K, n}, θ _{K, n}) polar coordinate after the average normalized; K ∈ 1,2 ..., K}, n ∈ 1,2 ..., N};

B2) according to step b1) polar coordinate after the average normalized of gained, respectively the sampled point of the N in each template signal is projected the log-polar territory, and carry out normalized, obtain N sampled point in each template signal with respect to the normalization log-polar of its beat starting point; The computing formula of normalized is as follows:

${\mathrm{\α}}_{k,n}=\frac{{\mathrm{\ξ}}_{k,n}-{\mathrm{\ξ}}_{k,\min }}{{\mathrm{\ξ}}_{k,\max }-{\mathrm{\ξ}}_{k,\min }},$

Wherein,

Be in k the template signal n sampled point with respect to the normalization log-polar of its beat starting point, α _{K, n}Be normalization utmost point footpath,

Be polar angle; (ξ _{K, n}, ψ _{K, n}) be the log-polar of n sampled point correspondence after throwing in k the template signal, utmost point footpath ξ _{K, n}=log ρ _{K, n}', polar angle ψ _{K, n}=θ _{K, n}'; K ∈ 1,2 ..., K}, n ∈ 1,2 ..., N}; ξ _{K, max}And ξ _{K, min}Be respectively each sampled point in k the template signal maximum and the minima in utmost point footpath in the corresponding log-polar after throwing.

In the above-mentioned steps, described step f) is specially:

F1) set up N sampled point in the characteristic area of each local minimum point respectively with respect to Descartes's relative coordinate of this local minimum point, and carry out the average normalized; The computing formula of average normalized is as follows:

${{\mathrm{\ρ}}_{i,n}}^{\text{'}}=\frac{{\mathrm{\ρ}}_{i,n}}{{\stackrel{\‾}{\mathrm{\ρ}}}_{i,n}}=\frac{\sqrt{{x_{i,n}}^2+{y_{i,n}}^2}}{\frac{1}{N-1}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{{x_{i,n}}^2+{y_{i,n}}^2}};$

θ _{I, n}'=θ _{I, n}, and θ _{I, n}' ∈ (π, π];

Wherein, (x _{I, n}, y _{I, n}) be in the detection segment the doubly approximate cycle of current ε with the Descartes relative coordinate of n sampled point in the characteristic area of interior i local minimum point with respect to this local minimum point, (ρ _{I, n}, θ _{I, n}) be and (x _{I, n}, y _{I, n}) corresponding polar coordinate; (ρ _{I, n}', θ _{I, n}') be (ρ _{I, n}, θ _{I, n}) polar coordinate after the average normalized; N ∈ 1,2 ..., N};

F2) according to step f1) polar coordinate after the average normalized of gained, respectively the sampled point of the N in the characteristic area of each local minimum point is projected the log-polar territory, and carry out normalized, obtain N sampled point in the characteristic area of each local minimum point with respect to the normalization log-polar of this local minimum point; The computing formula of normalized is as follows:

${\mathrm{\β}}_{i,n}=\frac{{\mathrm{\ξ}}_{i,n}-{\mathrm{\ξ}}_{i,\min }}{{\mathrm{\ξ}}_{i,\max }-{\mathrm{\ξ}}_{i,\min }},$ γ _i，n＝ψ _i，n；

Wherein, (β _{I, n}, γ _{I, n}) be in the detection segment the doubly approximate cycle of current ε with the normalization log-polar of n sampled point in the characteristic area of interior i local minimum point with respect to this local minimum point, β _{I, n}Be normalization utmost point footpath, γ _{I, n}Be polar angle; (ξ _{I, n}, ψ _{I, n}) being the log-polar of doubly approximate cycle of current ε in the detection segment with n sampled point correspondence after throwing in the characteristic area of interior i local minimum point, the utmost point is ξ directly _{I, n}=log ρ _{I, n}', polar angle ψ _{I, n}=θ _{I, n}'; N ∈ 1,2 ..., N}; ξ _{I, max}And ξ _{I, min}Be respectively each sampled point in the characteristic area of i local minimum point in the detection segment maximum and the minima in utmost point footpath in the corresponding log-polar after throwing.

In such scheme, the cut-off frequency of described low-pass filtering is 20～50Hz; The pretreated sample frequency of described pre-sampling is 125～1000Hz.

As further optimization, the preferred value of described N is 200; Described t ₀Preferred value be 60s; The preferred value of described ε is 1.5; The preferred value of weight coefficient λ is 0.8 in the described step g), described step h) middle threshold value C ₀Preferred value be 0.20.

Compared to existing technology, the present invention has following beneficial effect:

1, realized that computer to the cutting apart by shooting of intracranial pressure signal, not only greatly reduces intracranial pressure clinicist's workload, also overcome the error of lattice manually, that has improved intracranial pressure signal cuts apart efficient and accuracy by shooting.

2, as foundation characteristic, this foundation characteristic has translation and rotational invariance to the inventive method, can overcome the influence of the baseline drift of intracranial pressure signal with between points difference vector.

3, difference vector is carried out the similarity that log-polar conversion is measured waveform, this tolerance can be caught the overall profile information of waveform simultaneously again to contiguous waveform morphology feature-sensitive, and shake has robustness to waveform.

4, only with the local minimum point in the intracranial pressure signal as identification point, ignore calculating and identification to non local minimum point, simplified the data computation amount in the testing process greatly, further improved the robustness of identification.

5, respectively to giving different weights with some wave character afterwards before the local minimum point to be measured, the weight of wave character before weakening a little, strengthen the some weight of wave character afterwards, thereby weaken of the influence of the high-frequency rhythmicity fluctuation of intracranial pressure signal beat afterbody noise spot, improve the recognition accuracy of beat starting point the identification of beat starting point.

6, by appropriate thresholds is set, can further effectively get rid of noise spot, further improve the recognition accuracy of beat starting point.

Description of drawings

Fig. 1 is the FB(flow block) of the inventive method;

Fig. 2 is intracranial pressure signal example waveform figure;

Fig. 3 is the intracranial pressure signal example waveform figure that contains jammr band;

Fig. 4 is the cartesian coordinate mapping sketch map of a template signal;

Fig. 5 is sampled point a in the template signal shown in Figure 4 _{K, n}Normalized mapping sketch map in the log-polar territory;

Fig. 6 is template signal A among the embodiment ₆Oscillogram;

Fig. 7 is the oscillogram of preceding 10 seconds signals in the first detection segment of intracranial pressure signal to be measured among the embodiment;

Fig. 8 is local minimum point s in the signal shown in Figure 7 ₁, s ₂And s ₃The position;

Fig. 9 is beat starting point s in the signal shown in Figure 7 ₃And local minimum point s ₄, s ₅And s ₆The position;

Figure 10 is beat starting point s in the signal shown in Figure 7 ₃₃And local minimum point s ₃₄, s ₃₅, s ₃₆, s ₃₇, s ₃₈And s ₃₉The position;

Figure 11 is each beat starting point position in the signal shown in Figure 7;

Figure 12 is the similarity scattergram of each local minimum point in the signal shown in Figure 7.

The specific embodiment

Below in conjunction with drawings and Examples technical scheme of the present invention is described further:

The present invention proposes a kind of beat-to-beat division method of intracranial pressure signal that carries out analysis-by-synthesis in conjunction with the waveform profiles of intracranial pressure signal.Corresponding by shooting the heartbeat of intracranial pressure signal, and the inherent driving mechanism of each beat is identical all is the result that the combined effect of arteriogram, vein ripple, cerebrospinal fluid etc. drives, and the waveform of adjacent beat has similarity; If can be measured and mated to this similarity, just can find the point similar to the beat starting point, realize that the beat of intracranial pressure signal is cut apart.The present invention extracts the point in the intracranial pressure signal and the relative position relation of other point on its place waveform, and by its distribution characteristics in the log-polar territory of tolerance, measures the similarity between these points and the beat starting point; Simultaneously, the similarity measurement of point with point is converted into the similarity matching degree of waveform behind log-polar transform at a place measured, the tolerance after the conversion can be caught the overall profile information of waveform simultaneously again to contiguous waveform morphology feature-sensitive.The inventive method is applied to have the identification treatment facility (as possessing intracranial pressure analyser, intracranial pressure analytical system of computer function etc.) of calculation processing units such as microprocessor, in conjunction with intracranial pressure signal local form structure and overall profile information are discerned, just can judge the position of beat starting point accurately, thereby realize accurately cutting apart by shooting intracranial pressure signal.

The present invention adopts ICP (monitor intracranial pressure monitor) to gather intracranial pressure signal, these signals are by the digital signal after the A/D conversion (sample frequency of A/D conversion is 400Hz), with these signal input computers, carry out low-pass filtering and pre-sampling processing, the cut-off frequency of its filtering is 20～50Hz, and pre-sample frequency is between 125～1000Hz; Set up template by computer then, treat the intrinsic pressure signal of laterocranium and handle, and then intracranial pressure signal is cut apart by shooting by coupling.The FB(flow block) that computer is cut apart by shooting carries out as shown in Figure 1 successively as follows:

I, set up template signal, and the tolerance feature of beat starting point in the template signal:

A) set up template signal:

Under Different Individual, different condition, the cycle of the intracranial pressure signal that is collected, amplitude and waveform profiles all are not quite similar, and therefore should take into full account these factors when setting up template, set up K mutually different template signal, K 〉=2.The process of setting up of template signal is: choose the mutually different intracranial pressure signal of a plurality of waveform profiles, and its parameter such as cycle, amplitude and beat starting point separately all is retrieved as known conditions by manual detection identification or other existing detection means of identification in advance, is convenient to set up template signal.The intracranial pressure signal that these are chosen, should contain the practice waveform profiles of common several intracranial pressure signals clinically as far as possible, its cycle is between 0.43～1.5 second, to make these template signals can be used in the identification heart rate range at 40～140 times/minute intracranial pressure signal as far as possible.Gather above-mentioned all kinds of intracranial pressure signal by ICP (monitor intracranial pressure monitor), the input computer is chosen K beat starting point after carrying out low-pass filtering and pre-sampling processing then from the intracranial pressure signal that this has discerned, and wherein k beat starting point is designated as O arbitrarily _k, k ∈ 1,2 ..., K}.Because intracranial pressure signal is a quasi-periodic signal, the signal segment in two cycles is enough to embody near the waveform profiles information of beat starting point, therefore considers from the angle that improves robustness, extracts beat starting point O _kLast cycle and the back signal segment of one-period

As the length range of setting up template signal.Yet for different intracranial pressure signals, its cycle is not quite similar, thereby the sampling number in the two periodic signal sections of being extracted is also inconsistent; In order to set up unified template standard, need by sampling again or again interpolation processing be a fixed N sampled point with the length of each template signal is unified.For signal segment

, promptly calculate the sampling number N that wherein after the sampling pretreatment, is comprised _k, if N _kGreater than unifying length N then to signal segment

Sample again, if N _kLess than unifying length N then to signal segment

Carry out interpolation again, its length adjustment is a N sampled point the most at last, forms template signal A _kBy above-mentioned steps K the signal segment of choosing handled, can be set up K template signal, the length of each template signal is N sampled point.The size of N has determined the precision of later stage identification to a certain extent, takes into account the consideration of accuracy of identification and robustness, and the suitable span of N is 100～1000.

B) set up N sampled point in each template signal respectively with respect to the normalization log-polar of its beat starting point:

Because influence of various factors, the waveform profiles of each beat can not fit like a glove in the intracranial pressure signal, therefore can only discern the beat starting point by the similarity matching degree that compares waveform morphology.There is bigger difference between the waveform morphology of waveform morphology that the beat starting point is contiguous and non-beat starting point vicinity, if can set up a kind of metric relation, make tolerance responsive more to contiguous waveform morphology feature, with regard to easier beat starting point and non-beat starting point are significantly distinguished, reached recognition objective.The present invention is mapped to the intracranial pressure signal of gathering in the log-polar territory, allow the identification point in the intracranial pressure signal and the relative position relation of other point on its place waveform present the logarithm Changing Pattern, by other distribution characteristics in the log-polar territory of the identification point in the tolerance intracranial pressure signal with respect to its place waveform, embody the sensitivity characteristic of identification point with its logarithm Changing Pattern, and then realize coupling identification beat starting point in the intracranial pressure signal to its contiguous waveform morphology.The log-polar territory can be changed with the cartesian coordinate system mutual mapping.If the log-polar territory be (ξ, ψ), itself and cartesian coordinate system (x, transformational relation y) is as follows:

$\mathrm{\ξ}=\log \mathrm{\ρ}=\log \sqrt{x^2+y^2};$

Wherein, (ρ, θ) be cartesian coordinate system (utmost point footpath ξ promptly represents the logarithm value of distance between points in the log-polar territory for x, y) pairing polar coordinate, in the log-polar territory span of polar angle ψ be (π, π].

Concrete processing mode of the present invention is, for template signal A _k, in order to measure and calculation template signal A _kMiddle beat starting point O _kWith the relative position relation of N sampled point, this N sampled point is projected with beat starting point O _kIn the cartesian coordinate system for initial point, set up the Descartes relative coordinate of each sampled point, measure each sampled point and beat starting point O with Descartes's relative coordinate with respect to this beat starting point _kDifference vector; The size of difference vector only with beat starting point O _kRelevant with the relative position relation between its distribution characteristics point, and with beat starting point O _kThe baseline of last cycle and back one-period signal waveform is irrelevant, therefore with between points difference vector as foundation characteristic, make foundation characteristic have translation and rotational invariance, this characteristic can overcome the influence of the baseline drift of intracranial pressure signal.A _kIn each sampled point with respect to its beat starting point O _kDescartes's relative coordinate, need carry out the average normalized, mainly be that the length of the represented difference vector of Descartes's relative coordinate is carried out the average normalized, and keep the direction of difference vector constant; The reason of Chu Liing is because the rhythmicity of intracranial pressure signal high frequency fluctuation interference ratio is more like this, and interferential quantity of each beat high frequency and position have nothing in common with each other, thereby form the distinctive personal characteristics information of each beat, can weaken template signal A by the average normalized _kThis personal characteristics information, simultaneously can be so that the common feature of the beat starting point that wherein contains periphery waveform profiles is kept.

Then, again according to the difference vector after the average normalized, with template signal A _kN sampled point be mapped in the log-polar territory, obtain the log-polar of sampled point; The log-polar of this N sampled point has directly reflected itself and beat starting point O _kBetween position relation, and distribute and be logarithmic parabola and change, by the log-polar of N sampled point of tolerance, this tolerance is to beat starting point O _kContiguous waveform morphology feature-sensitive, the while can be caught the overall profile information of waveform again.At last, again sampled point in the template signal is carried out again normalized with respect to the log-polar of its beat starting point, obtain the normalization log-polar, further eliminate individual difference wherein.Sampled point with respect to the normalization log-polar of its beat starting point is in the note template signal

For example, template signal A _kIn n sampled point a _{K, n}, n ∈ 1,2 ..., N} is with beat starting point O _kFor the Descartes's relative coordinate in the cartesian coordinate system of initial point is (x _{K, n}, y _{K, n}), corresponding polar coordinate are (ρ _{K, n}, θ _{K, n}), as shown in Figure 4; It is carried out the average normalized, that is:

${{\mathrm{\ρ}}_{k,n}}^{\text{'}}=\frac{{\mathrm{\ρ}}_{k,n}}{{\stackrel{\‾}{\mathrm{\ρ}}}_{k,n}}=\frac{\sqrt{{x_{k,n}}^2+{y_{k,n}}^2}}{\frac{1}{N-1}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{{x_{k,n}}^2+{y_{k,n}}^2}};$

θ _{K, n}'=θ _{K, n}, and θ _{K, n}' ∈ (π, π];

(ρ _{K, n}', θ _{K, n}') then be (ρ _{K, n}, θ _{K, n}) polar coordinate after the average normalized; Again by (ρ _{K, n}', θ _{K, n}') be mapped to after the log-polar territory, obtain sampled point a _{K, n}With respect to beat starting point O _kLog-polar (ξ _{K, n}, ψ _{K, n}), wherein, utmost point footpath ξ _{K, n}=log ρ _{K, n}', polar angle ψ _{K, n}=θ _{K, n}'=θ _{K, n}, and ψ _{K, n}∈ (π, π]; Obtain template signal A thus _kIn each sampled point with respect to beat starting point O _kLog-polar after, calculate the maximum ξ in the footpath of the utmost point wherein _{K, max}With minima ξ _{K, min}, between normalized to 0～1, utmost point footpath with each sampled point log-polar, keep polar angle constant, more specifically for sampled point a _{K, n}, be:

${\mathrm{\α}}_{k,n}=\frac{{\mathrm{\ξ}}_{k,n}-{\mathrm{\ξ}}_{k,\min }}{{\mathrm{\ξ}}_{k,\max }-{\mathrm{\ξ}}_{k,\min }},$

Then be template signal A _kIn n sampled point a _{K, n}With respect to beat starting point O _kThe normalization log-polar, α _{K, n}∈ [0,1], As shown in Figure 5.

Can set up respectively by above-mentioned steps that sampled point is stored in it in memory device of computer or intracranial pressure analytical equipment, as the match-on criterion of beat starting point in the intracranial pressure signal to be measured with respect to the normalization log-polar of its beat starting point in each template signal.So far, test preparation is finished, next can carry out the testing procedure of intracranial pressure signal to be measured.

Beat starting point in II, the identification intracranial pressure signal to be measured:

Intracranial pressure signal to be measured is imported computer again and is carried out low-pass filtering and sampling pretreatment, in order to dividing processing also by the ICP (monitor intracranial pressure monitor) collection.Each beat starting point in the intracranial pressure signal to be measured all should be a local minimum point, if only calculate as identification point with each local minimum point in the detection segment, can avoid obvious non-beat starting points a large amount of in the signal is discerned, simplify the data computation amount in the testing process greatly, can further improve the robustness of identification.In each beat, the local minimum point except that actual beat starting point is noise spot, judges that local minimum point the most similar to the beat starting point of template signal in the single beat is the actual beat starting point in this beat.But before definite beat starting point, the cycle duration of single beat can't be judged accurately, therefore needs one to judge duration, can determine to comprise at least a beat starting point in this judgement duration, can not surpass 2 beat durations again, with the accuracy that guarantees as far as possible to judge.We are used as the judgement duration benchmark of single beat in the measured signal with one " approximate cycle ".

But in the practical operation, each is variant probably for the speed of different period hearts rate, and the variation of heart rate directly causes the variation in intracranial pressure signal cycle, therefore in the whole section intracranial pressure signal of gathering to be measured, may have mutually different beat of cycle; If the cycle difference between the beat of difference place is excessive, but this difference is discerned with the same approximate cycle, certainly will cause recognition result to have bigger error.For this reason, the present invention has adopted intracranial pressure signal to be measured with segmented mode by detecting processing, each sectional duration is set at 30～90s, and is excessive to avoid in the single split cycle difference between the different beats, thereby identification error is controlled in the limited scope.

Take all factors into consideration above-mentioned factor, the present invention is as follows to the identifying of beat starting point in the intracranial pressure signal to be measured:

C) determine the first detection segment of intracranial pressure signal to be measured:

For the first detection segment of intracranial pressure signal to be measured, be that to extract duration from the starting point of intracranial pressure signal to be measured be t ₀Signal segment as detection segment; Then detection segment is carried out autocorrelation analysis, the interval in the auto-correlation function of calculating detection segment between every adjacent two local maximums is got the approximate cycle of the meansigma methods of described interval as detection segment; Wherein, t ₀Span be 30～90s.

Calculate the approximate cycle of detection segment, can adopt this area auto-correlation function commonly used to find the solution, detection segment is carried out autocorrelation analysis, calculate the interval between every adjacent two local maximums in its auto-correlation function, get the approximate cycle of the meansigma methods of described interval as detection segment.For example, for detection segment

Its signal value is the function of time, is designated as S (t), then detection segment

Auto-correlation function R _S(τ) be:

$R_S\left(t\right)=\underset{-\∞}{\overset{+\∞}{\∫}}S\left(t\right)S(t+\mathrm{\ι})\mathrm{dt},$

Calculate its auto-correlation function R _SA pairing L τZhi is designated as τ when (τ) getting local maximum _l, l ∈ 1,2 ..., L}, then detection segment

The approximate cycle

For:

${\stackrel{\‾}{T}}_s=\frac{1}{L-1}\underset{l=2}{\overset{L}{\mathrm{\Σ}}}({\mathrm{\ι}}_l-{\mathrm{\ι}}_{l-1}).$

D) determine the local minimum point in the first judgement duration in the detection segment:

In a detection segment, the duration of some beat might be greater than the length in pre-above-mentioned approximate cycle.Determine to comprise a beat starting point in order to guarantee to judge in the duration for one, the present invention is a benchmark with the length in approximate cycle, gets the judgement duration of ε times (ε＞1) approximate cycle as reality; The value of ε can not be excessive, and its span is 1.2～1.6, caused wherein comprising the beat starting point of 2 reality to avoid judging the duration that duration has surpassed 2 beats, and then produced the omission situation.

For the first judgement duration of detection segment, then be all local minimum points that calculate the cycle that doubly is similar to from section start to ε in the detection segment, judge in order to carrying out follow-up detection.Calculating local minimum point can adopt this area to use certain methods always.For example, can calculate the difference in magnitude between each neighbouring sample point,, judge that then this sampled point is the local minimum point if the difference in magnitude between a certain sampled point and its forward and backward neighbouring sample point all is not more than zero.Also can utilize method of derivation, detection segment is carried out derivative operation, obtain the extreme point of detection segment upper derivate for " 0 ", judge further that again these extreme points are maximum point or minimum point, minimum point wherein is the local minimum point of intracranial pressure signal.

E) characteristic area of extraction local minimum point:

For allow each local minimum point can be respectively with template signal in the beat starting point carry out corresponding coupling and tolerance, need to extract the characteristic area of each local minimum point.The concrete grammar that extracts is, from the waveform profiles of intracranial pressure signal to be measured, extract the signal segment in each each approximate cycle of local minimum point front and back, utilize and the step a) similar methods, be N sampled point (consistent) with the length of each signal segment of extracting is unified with the sampling number in the template signal, so that mate and compare, thereby form the characteristic area of each local minimum point with template signal.For example, detection segment

The approximate cycle is By calculating detection segment In the doubly approximate cycle of current ε with interior I local minimum point, wherein i local minimum point is s _i, i ∈ 1,2 ..., I}.From intracranial pressure signal to be measured, extract local minimum point s _iThe signal segment in last approximate cycle and one approximate cycle of back

Calculate the pre-sampling number N that wherein comprises _i, if N _iGreater than unifying length N then to signal segment Sample again, if N _iLess than unifying length N then to signal segment Carry out interpolation again, its length adjustment is a N sampled point the most at last, forms local minimum point s _iCharacteristic area S _iBy above-mentioned steps, from intracranial pressure signal to be measured, extract detection segment

In the doubly approximate cycle of current ε put characteristic of correspondence district separately with an interior I local minimum.

F) set up N sampled point in the characteristic area of each local minimum point respectively with respect to the normalization log-polar of its local minimum point;

Correspondingly, similar to step b), set up in the characteristic area of each local minimum point the Descartes relative coordinate of N sampled point respectively with respect to its local minimum point, then the length of the represented difference vector of Descartes's relative coordinate is carried out the average normalized, keep the direction of difference vector constant; According to the difference vector after the average normalized, N sampled point in the characteristic area of each local minimum point is mapped in the log-polar territory again, obtains its log-polar, obtain the normalization log-polar by further normalized at last.Sampled point is (beta, gamma) with respect to the normalization log-polar of its local minimum point in the note characteristic area.

For detection segment

In the doubly approximate cycle of current ε with interior i local minimum point s _iCharacteristic area S _i, i ∈ 1,2 ..., and I}, wherein N sampled point projects with local minimum point s _iIn the cartesian coordinate system for initial point, set up each sampled point with respect to s _iDescartes's relative coordinate, measure each sampled point and local minimum point s with Descartes's relative coordinate _iDifference vector; Wherein, characteristic area S _iIn n sampled point s _{I, n}, n ∈ 1,2 ..., and N}, it is with local minimum point s _iFor the Descartes's relative coordinate in the cartesian coordinate system of initial point is (x _{I, n}, y _{I, n}), corresponding polar coordinate are (ρ _{I, n}, θ _{I, n}), it is carried out the average normalized, that is:

${{\mathrm{\ρ}}_{i,n}}^{\text{'}}=\frac{{\mathrm{\ρ}}_{i,n}}{{\stackrel{\‾}{\mathrm{\ρ}}}_{i,n}}=\frac{\sqrt{{x_{i,n}}^2+{y_{i,n}}^2}}{\frac{1}{N-1}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{{x_{i,n}}^2+{y_{i,n}}^2}};$

θ _{I, n}'=θ _{I, n}, and θ _{I, n}' ∈ (π, π];

(ρ _{I, n}', θ _{I, n}') be (ρ _{I, n}, θ _{I, n}) polar coordinate after the average normalized; Again by (ρ _{I, n}', θ _{I, n}') be mapped to after the log-polar territory, obtain sampled point s _{I, n}With respect to local minimum point s _iLog-polar (ξ _{I, n}, ψ _{I, n}), utmost point footpath ξ _{I, n}=log ρ _{I, n}', polar angle ψ _{I, n}=θ _{I, n}'=θ _{I, n}Obtain local minimum point s thus _iCharacteristic area S _iIn each sampled point put s with respect to local minimum _iLog-polar after, the maximum and the minima that calculate the footpath of the utmost point wherein are respectively ξ _{I, max}And ξ _{I, min}, sampled point s then _{I, n}With respect to local minimum point s _iNormalization log-polar (β _{I, n}, γ _{I, n}) satisfy:

${\mathrm{\β}}_{i,n}=\frac{{\mathrm{\ξ}}_{i,n}-{\mathrm{\ξ}}_{i,\min }}{{\mathrm{\ξ}}_{i,\max }-{\mathrm{\ξ}}_{i,\min }},$ γ _i，n＝ψ _i，n；

After normalized, β _{I, n}∈ [0,1], γ _{I, n}∈ (π, π].Can set up in the detection segment the doubly approximate cycle of current ε respectively with the normalization log-polar of the sampled point of the N in the characteristic area of interior each local minimum point by above-mentioned steps with respect to its local minimum point.

G) calculate the similarity of doubly approximate cycle of current ε in the detection segment respectively with interior each local minimum point:

In cycle, having only a local minimum point is real beat starting point at each beat of detection segment, and this local minimum point should be the highest with the similarity matching degree of beat starting point in the template signal.So, be incorporated herein " similarity " this notion, by calculating the similarity of local minimum point, the similarity matching degree of beat starting point in local minimum point and the template signal is described; The similarity of local minimum point is big more, represents that then the similarity matching degree of beat starting point in this local minimum point and the template signal is high more, and this local minimum point might be the actual beat starting point of detection segment more just.The present invention adopts the cross-correlation coefficient of the characteristic area of local minimum point and template signal to measure the similarity of each local minimum point in the detection segment, concrete processing mode is: the normalization log-polar of setting up based on step b) and step f), calculate the characteristic area of each local minimum point and the cross-correlation coefficient of each template signal respectively, with the similarity of the maximum in the cross-correlation coefficient of the characteristic area of each local minimum point and each template signal, thereby obtain the similarity of each local minimum point as this local minimum point.

For example, detection segment In the doubly approximate cycle of current ε with interior i local minimum point s _iCharacteristic area S _i, i ∈ 1,2 ..., and I}, n sampled point in the N of this characteristic area sampled point is s _{I, n}, n ∈ 1,2 ..., N}, s _{I, n}With respect to local minimum point s _iThe normalization log-polar be (β _{I, n}, γ _{I, n}); Simultaneously, k template signal A _k, k ∈ 1,2 ..., and K}, the beat starting point in this template signal is O _k, n sampled point in its N sampled point is a _{K, n}, n ∈ 1,2 ..., N}, a _{K, n}With respect to beat starting point O _kThe normalization log-polar be

Local minimum point s then _iCharacteristic area S _iWith template signal A _kCross-correlation coefficient P _{I, k}For:

Wherein,

Wherein, n ∈ 1,2 ..., N}; W _pThe expression weight, the span of its weight coefficient λ is 0＜λ＜1; Because characteristic area S _iN sampled point in, local minimum point s _iWith the utmost point electrical path length of normalization log-polar of himself be 0, therefore, actual only have N-1 not to be that 0 inner product summation is averaged, so the coefficient before the sum term is

Thus, can obtain detection segment In the doubly approximate cycle of current ε with interior i local minimum point s _iCharacteristic area S _iCross-correlation coefficient P with beat starting point in each template signal _{I, 1}, P _{I, 2}, P _{I, 2}... P _{I, K}With P _{I, 1}, P _{I, 2}, P _{I, 2}... P _{I, K}In maximum put s as local minimum _iSimilarity C _i, measure detection segment with this

In the doubly approximate cycle of current ε with interior i local minimum point s _iSimilarity matching degree with beat starting point in the template signal.Can see from the cross-correlation calculation formula, at characteristic area S _iIn, local minimum point s _iSampled point before (

Perhaps

) and template signal A _kThe cross-correlation weight be λ (less than 1), local minimum point s _iSampled point afterwards (

) and template signal A _kThe cross-correlation weight be 1, its objective is for relative reduction local minimum point s _iThe matching properties of waveform profiles before.If local minimum point s _iMinimum point is disturbed in a high-frequency rhythmicity fluctuation that is intracranial pressure signal beat afterbody, at a s _iMust also there be a bit of High-frequency Interference shake waveform afterwards, at this moment s _iWaveform profiles is relative better with the template signal coupling before the point, s _iPoint back waveform profiles is relative relatively poor with the template signal coupling because of the relation of High-frequency Interference shake waveform; But s _iThe matching properties of waveform profiles is after weight coefficient λ reduction, again with s before the point _iMinimum point s is disturbed in the matching properties summation of some back waveform profiles, average _iCharacteristic area S _iWith template signal A _kCross-correlation coefficient then correspondingly reduce the corresponding local minimum point s that reduced simultaneously _iSimilarity.If local minimum point s _iBe a beat starting point in the intracranial pressure signal to be measured, even there is a bit of High-frequency Interference shake waveform before, but this moment s _iWaveform profiles is relative relatively poor with the template signal coupling because of the relation of High-frequency Interference shake waveform before the point, s _iPoint back waveform profiles is relative better with the template signal coupling; Even weight coefficient λ's s has weakened _iThe matching properties of waveform profiles before the point, and s _iThe matching properties of some back waveform profiles is kept, thereby allows local minimum point s _iCharacteristic area S _iWith template signal A _kCross-correlation coefficient maintain higher value, correspondingly guaranteed local minimum point s _iSimilarity higher relatively.Because the setting of different weights, local minimum point s relatively weakens _iThe matching properties of waveform profiles before, make the fluctuation of actual beat starting point and high-frequency rhythmicity disturb the similarity difference of smallest point more obvious, weaken of the influence of the high-frequency rhythmicity fluctuation of intracranial pressure signal beat afterbody noise spot, help further to improve discrimination and cut apart accuracy rate by shooting the identification of beat starting point.Weight coefficient λ is in the span of 0＜λ＜1, and value is more little, and actual beat starting point is just obvious more with the similarity difference of disturbing smallest point.

One by one the doubly approximate cycle of current ε in the detection segment is carried out cross-correlation analysis with interior each local minimum point by this step, obtain the similarity of each local minimum point.

H) judge in the detection segment that the doubly approximate cycle of current ε is with interior beat starting point:

In each beat of detection segment, the local minimum point except that actual beat starting point is noise spot, should be got rid of in identifying.Noise spot is to produce owing to influence that intracranial pressure signal is subjected to many-sided factor, these noise spots can be divided into weak jamming point and violent two kinds of noise spots from the identification angle.The weak jamming point, be near some the local minimum points the beat starting point, this noise spot may be because the active motion effects of whole body blood vessel and cerebrovascular intracranial pressure, make and exist in the intracranial pressure signal high-frequency rhythmicity fluctuation to form, also might be that (as the detection probe shake of intracranial pressure detector etc.) causes because the of short duration instability of detection signal, but the amplitude of this noise spot is less, the beat of unlikely destruction intracranial pressure signal, and this noise spot is also often little than beat starting point with the similarity of template signal, can be got rid of by comparing the similarity size.Violent noise spot, be outside the disturbing wave of whole body blood vessel and cerebrovascular motion generation, also owing to feel suffocated, exert oneself etc. to cause that intracranial pressure signal produces obviously fluctuates accordingly, perhaps actions such as cough, sneeze cause intracranial pressure signal acutely to be shaken, this fluctuation or violent shake have randomness, and amplitude is big, the persistent period is longer, forms one section violent disturbing wave; This violent disturbing wave just may cause superimposed interferential beat by serious destruction if overlap in the above intracranial pressure signal of a beat, and this being present in by the local minimum point in the destructive beat of violent disturbing wave is regarded as violent noise spot.If have so violent interference in one section intracranial pressure signal, useful information is also just destroyed in this segment signal, has in fact just lost the clinical identification meaning of intracranial pressure.Therefore, the present invention is by preestablishing a threshold value C ₀Beat starting point and violent noise spot are distinguished, avoid being caused cutting apart by shooting confusion for normal beat by the destructive beat flase drop survey of violent disturbing wave.

Concrete processing mode is, earlier by relatively obtaining in the detection segment the doubly approximate cycle of current ε with a local minimum point of interior similarity maximum, and other local minimum point except that this point all is regarded as weak jamming point and is got rid of; Then with the similarity and the pre-set threshold C of this local minimum point ₀Compare, if its similarity is greater than threshold value C ₀, judge that promptly this local minimum point is a beat starting point; If its similarity is less than threshold value C ₀, judge that then this local minimum point is a violent noise spot.For example, calculate detection segment In current ε in the doubly approximate cycle local minimum point of similarity maximum be s _i, its similarity is C _iWith C _iWith pre-set threshold C ₀Compare, if C _i≤ C ₀, then local minimum is put s _iBeing considered as violent noise spot excludes; If C _i＞C ₀, then judge local minimum point s _iBe the beat starting point.

In this step, threshold value C ₀Value be to get rid of the signals of violent noise spot, if threshold value C ₀Value is too small, then can cause the omission of violent noise spot; If threshold value C ₀Value is excessive, then the beat starting point of reality may be judged to be violent noise spot and get rid of in the lump.Usually, as the actual beat starting point of detection segment, its similarity is the highest may to reach 0.4; But under situation about existing than the large amplitude interfering signal, if the similarity of actual beat starting point is greater than 0.1 in the disturbed signal, can think that still the useful information in its beat is not destroyed fully, these useful informations still can be employed and accept clinically.Therefore, threshold value C ₀Span get 0.1～0.4 and be advisable threshold value C ₀The big more i.e. expression of value judges that the requirement of beat starting point is strict more.

I) all beat starting points in the judgement detection segment:

With the doubly approximate cycle of current ε in the detection segment be starting point with a local minimum point of interior similarity maximum, calculate thereafter the doubly approximate cycle of ε with interior all local minimums point; Repeating step e then)～i), judge beat starting points all in the detection segment thus.For example in current detection segment

In, with the local minimum point s of doubly approximate cycle of current ε with interior similarity maximum _iBe starting point, calculate detection segment

Mid point s _iThe local minimum point of ε similarity maximum in the doubly approximate cycle is judged its similarity and threshold value C then afterwards ₀Between size, thereby judge whether it is the beat starting point; Be starting point with this point again, calculate ε thereafter in the doubly approximate cycle local minimum point of similarity maximum further judge ... by parity of reasoning, calculates current detection segment piecemeal

In all beat starting points.

J) all beat starting points in the judgement intracranial pressure signal to be measured:

In intracranial pressure signal to be measured, be starting point with last beat starting point position in the current detection section, extracting thereafter, duration is t ₀Signal segment as new detection segment, prepare to detect the beat starting point in the new detection segment.But the beat cycle may there are differences in new detection segment and the current detection section, therefore needs to upgrade earlier the approximate cycle, to avoid occurring the bigger error of calculation.The method of upgrading the approximate cycle is, with the meansigma methods of last 3 normal cycle intervals in the current detection section approximate cycle as new detection segment; Described normal cycle interval is meant the interval that is no more than 1.5 times between adjacent two beat starting points and is not less than 0.5 times of current approximate cycle duration.Repeating step d then)～j), judge beat starting points all in the intracranial pressure signal to be measured thus.

III, intracranial pressure signal to be measured is cut apart by shooting:

K) be cut-point with each beat starting point in the intracranial pressure signal to be measured, intracranial pressure signal to be measured cut apart by shooting, and the intracranial pressure signal to be measured after will cutting apart shows and store, so that observe and carry out subsequent treatment by computer.

Further specify the detailed process that this employing inventive method realizes that intracranial pressure signal beat by beat is cut apart below by embodiment.

Embodiment:

In the present embodiment, by ICP (monitor intracranial pressure monitor) (Codman-Hakim, Johson ﹠ Johnson, the U.S.) gather intracranial pressure signal, these signals are that sample frequency is the digital signal of 400Hz, with these signal input computers, carry out low-pass filtering and sampling pretreatment, its wave filter adopts second order Butterworth low pass filter, and cut-off frequency is 25Hz, sample frequency is 125Hz, with the signal that obtains as intracranial pressure signal to be measured.Utilize the inventive method, this intracranial pressure signal to be measured is cut apart by shooting, cutting procedure is carried out as follows by computer by shooting:

At first, gather a plurality of cycle differences, the representative intracranial pressure signal of identification of common waveform profiles (parameters such as cycle, amplitude, beat starting point have all been discerned and known) clinically respectively by ICP (monitor intracranial pressure monitor), these signals also are that sample frequency is the digital signal of 400Hz, with these signals and relevant parameter input computer thereof, carry out low-pass filtering and sampling pretreatment, its wave filter adopts second order Butterworth low pass filter, and cut-off frequency is 25Hz, and sample frequency is 125Hz.Choose 20 beat starting points (getting K=20) the identification signal from above-mentioned each, extract each a last cycle of beat starting point and the back segment signal of one-period respectively; The 6th beat starting point O wherein ₆The segment signal of last cycle with back one-period is

Then by sampling again or the length of interpolation processing each segment signal that will extract is unified again is fixed 200 sampled points (getting N=200); For example, beat starting point O as calculated ₆A last cycle and the back segment signal of one-period

The sampling number that comprises after the middle sampling pretreatment is 212, and it is 200 sampled points that predetermined template signal is unified length, therefore with signal segment

Be sampled as 200 sampled points again, obtain template signal A ₆, its waveform profiles as shown in Figure 6; Obtain 20 template signals thus.Set up in each template signal the normalization log-polar of 200 sampled points according to the described method of step b) again with respect to its beat starting point.

After finishing about the preparation of template signal, then carry out the identification of beat starting point in the intracranial pressure signal to be measured.Determine earlier the first detection segment of intracranial pressure signal to be measured, extracting duration from the starting point of intracranial pressure signal to be measured is that the signal segment of 60s (is got t ₀=60s) as first detection segment

And obtain detection segment by autocorrelation analysis The approximate cycle

Detection segment In preceding 10 seconds waveform profiles as shown in Figure 7, as can be seen from Fig. 7, between 1000～1500 sampled points, have one section because the violent interfering signal that cough causes, cause around here several signal beat by more serious destruction.

Next, with 1.5 times of approximate cycles

For judging duration, calculate detection segment

In from section start extremely

All local minimum points at place; But because the waveform profiles of intracranial pressure signal to be measured in initial first approximate cycle is imperfect, therefore the local minimum point in initial first approximate cycle can't extract the complete signal in its last cycle, just can't utilize the inventive method to test, so the local minimum point of intracranial pressure signal to be measured in initial first approximate cycle cast out, obtained detection segment

In from section start extremely The place can have 3 as the local minimum point of identifying object, is respectively a s ₁, s ₂And s ₃, as shown in Figure 8.Then, extract the characteristic area of these 3 local minimum points respectively; With a s ₁Be example, extract s ₁The signal segment in last approximate cycle and one approximate cycle of back

Calculating the sampling number that wherein comprises after the sampling pretreatment is 192, less than predetermined 200 sampled points of unified length, therefore with signal segment

Carrying out interpolation processing is 200 sampled points, forms some s ₁Characteristic area S ₁Form some s in the same way respectively ₂, s ₃Characteristic area S ₂And S ₃Set up characteristic area S respectively according to the described method of step f) again ₁, S ₂And S ₃In 200 sampled points with respect to the normalization log-polar of local minimum point separately.Calculation level s ₁Characteristic area S ₁Cross-correlation coefficient (in the present embodiment, getting weight coefficient λ=0.8 when calculating cross-correlation coefficient) with 20 template signals obtains characteristic area S ₁20 cross-correlation coefficients in maximum be and template signal A ₆Cross-correlation coefficient P _1,6=0.22, promptly determine some s ₁Similarity C ₁=P _1,6=0.07; Equally, calculation level s ₂, s ₃Characteristic area S ₂And S ₃With the cross-correlation coefficient of 20 template signals, obtain characteristic area S respectively ₂20 cross-correlation coefficients in maximum be and template signal A ₆Cross-correlation coefficient P _2,6=0.16, S ₃20 cross-correlation coefficients in maximum also be and template signal A ₆Cross-correlation coefficient P _3,6=0.23, promptly determine some s ₂, s ₃Similarity be respectively C ₂=P _2,6=0.16, C ₃=P _3,6=0.23.Comparatively speaking, C ₁＜C ₂＜C ₃, put s among the three ₃Similarity bigger, with s ₃Similarity and pre-set threshold C ₀Compare C ₀Value is 0.20; Because C ₃=0.23＞C ₀, be not violent noise spot, thus local minimum point s ₃Be judged as detection segment A beat starting point.Next, with local minimum point s ₃Be starting point, calculate detection segment

Mid point s ₃Afterwards

Be respectively s with interior all local minimum points ₄, s ₅And s ₆, as shown in Figure 9; Then extract some s respectively ₄, s ₅And s ₆Characteristic area be S ₄, S ₅And S ₆, set up 200 sampled points in above-mentioned each characteristic area respectively with respect to the normalization log-polar of local minimum point separately according to the described method of step f) again, respectively calculation level s ₄, s ₅And s ₆Characteristic area S ₄, S ₅And S ₆With the cross-correlation coefficient of each template signal based on log-polar; By calculating some s ₄, s ₅And s ₆Characteristic area all with respect to template signal A ₆The cross-correlation coefficient maximum, promptly get s ₄, s ₅And s ₆Similarity be respectively: C ₄=P _4,6=0.07, C ₅=P _5,6=0.12 and C ₆=P _6,6=0.22.By relatively learning, the similarity size is C ₄＜C ₅＜C ₆Thereby the local minimum point of similarity maximum is s in current 1.5 times of approximate cycles ₆, and C ₆=0.22＞C ₀, promptly judge local minimum point s ₆Be detection segment

Another beat starting point.Then, again with local minimum point s ₆Be starting point, calculate detection segment

Mid point s ₆Afterwards

With interior all local minimum point s ₇, s ₈And s ₉Further judge ... judge beat starting point s ₃₃After, calculate detection segment

Mid point s ₃₃Afterwards

Be respectively s with interior all local minimum points ₃₄, s ₃₅, s ₃₆, s ₃₇, s ₃₈And s ₃₉, as shown in figure 10; Repeat above-mentioned steps once more and calculate the step of similarity, obtain s ₃₄, s ₃₅, s ₃₆, s ₃₇, s ₃₈And s ₃₉Similarity be respectively: C ₃₄=0.08, C ₃₅=0.11, C ₃₆=0.17, C ₃₇=0.14, C ₃₈=0.12 and C ₃₉=0.07, by relatively learning, the similarity size is C ₃₉＜C ₃₄＜C ₃₅＜C ₃₈＜C ₃₇＜C ₃₆Thereby the local minimum point of similarity maximum is s in current 1.5 times of approximate cycles ₃₆, but because C ₃₆=0.17＜C ₀, promptly judge local minimum point s ₃₆Be detection segment

In a violent noise spot, therefore will put s ₃₆Get rid of.Again with local minimum point s ₃₆Be starting point, calculate detection segment Mid point s ₃₆Afterwards

With interior all local minimum point s ₃₇, s ₃₈, s ₃₉, s ₄₀And s ₄₁Further judge ... recursion is determined detection segment thus

In 87 beat starting points, be respectively s ₂, s ₅, s ₈..., s ₂₇₅, s ₂₇₈, s ₂₈₆, s ₂₈₉And s ₂₉₂Also determine detection segment in addition

3 violent noise spots of middle existence are respectively s ₃₆, s ₄₁And s ₂₈₂Violent noise spot s ₃₆, s ₄₁Be positioned at beat starting point s ₃₃With s ₄₄Between, be because the noise spot that fluctuates that firmly causes; Violent noise spot s ₂₈₂Be positioned at beat starting point s ₂₇₈With s ₂₈₆Between.Wherein, detection segment Preceding 10 seconds with interior beat starting point as shown in figure 11, detection segment

Can see beat starting point s in 1000～1500 sampled points in the signal in preceding 10 seconds with the similarity scattergram of interior each local minimum point as shown in figure 12 in conjunction with Figure 11 and 12 ₃₃With s ₄₄Between local maximum point all be regarded as noise spot and exclude, therefore at a s ₃₃With s ₄₄Between do not have significant beat starting point.

Determine the first detection segment of intracranial pressure signal to be measured

In after all beat starting point, with detection segment In last beat starting point s ₂₈₂The position is a starting point, extracts some s from intracranial pressure signal to be measured ₂₈₂The signal segment of 60s duration is as new detection segment afterwards

Prepare to detect new detection segment

In the beat starting point.Upgrade the value in approximate cycle this moment, with detection segment

In the meansigma methods of last 3 normal cycle intervals as new detection segment

The approximate cycle; Because beat starting point s ₂₇₈With s ₂₈₆Between because of existing violent interfering signal to cause interval R therebetween ₈₇Surpass

So T ₈₇Be not the normal cycle interval, get beat starting point s ₂₇₅With s ₂₇₈Interval T ₈₆, beat starting point s ₂₈₆With s ₂₈₉Interval T ₈₈With beat starting point s ₂₈₉With s ₂₉₂Interval T ₈₉Three's meansigma methods is as new detection segment

Average period

${\stackrel{\‾}{T}}_s\left(2\right)=\frac{1}{3}\×(T_{86}+T_{88}+T_{89});$

Then, repeat above-mentioned steps, judge the detection segment of intracranial pressure signal to be measured

In all beat starting points.Recursion is judged the detection segment of intracranial pressure signal to be measured equally thus

All beat starting point in judging intracranial pressure signal to be measured.At last, be cut-point with each beat starting point in the intracranial pressure signal to be measured, by computer intracranial pressure signal to be measured is cut apart by shooting, and the intracranial pressure signal to be measured after will cutting apart shows and stores processing.

In order to assess the detection performance of the inventive method, we have made up a data base, and to play number of spots be 68132 to the beat of intracranial pressure signal to be measured among the data base, and the beat starting point of these intracranial pressure signals to be measured has been passed through the clinical expert manual markings.Utilize the inventive method that intracranial pressure signal to be measured among the data base to be measured is carried out the identification of beat starting point, the beat starting point with testing result and expert's labelling compares then, and then assesses detection performance of the present invention.We are made as fault-tolerant interval with 8ms before and after the beat starting point of manual markings, think that this detection is correct when being not more than 8ms by the point tolerance of starting auction of starting auction a little of detecting of the present invention and expert's manual markings that is:.The inventive method is 96.82% to the identification accuracy of these 68132 beat starting points, and specificity is 96.20%, satisfies the requirement of Clinical recognition.

The inventive method not only local messages such as the amplitude in the intracranial pressure signal, local minimum points as the reference factor, the waveform profiles that more combines intracranial pressure signal carries out analysis-by-synthesis, with between points difference vector as foundation characteristic, this foundation characteristic has translation and rotational invariance, can overcome the influence of the baseline drift of intracranial pressure signal; Difference vector is carried out the similarity that log-polar conversion is measured waveform, and this tolerance can be caught the overall profile information of waveform again to the waveform morphology feature-sensitive of identification point vicinity, simultaneously waveform shake and distortion is had robustness; In computational process also respectively to before the local minimum point to be measured with the point after wave character give different weights, the weight of wave character before weakening a little, strengthen the some weight of wave character afterwards, thereby weaken of the influence of intracranial pressure signal beat afterbody noise spot to identification, further improve the recognition accuracy of beat starting point, and then realized identification accurately intracranial pressure signal beat starting point; Simultaneously, also, can further effectively get rid of noise spot, help further improving the recognition accuracy of beat starting point by appropriate thresholds is set.Need benly be, in the cross-correlation coefficient that calculates local minimum point and template signal, if with weights W _pThe corresponding raising of value G doubly (G be the arbitrary value greater than 0), and do not change before the local minimum point with point after weight relationship (put preceding weight less than after weight), the while is correspondingly with threshold value C ₀Improve G doubly, then can not produce any influence result of calculation; Substantially identical by all schemes that numerical value is expanded or compression obtains like this with the present invention.

Explanation is at last, above embodiment is only unrestricted in order to technical scheme of the present invention to be described, although the present invention is had been described in detail with reference to preferred embodiment, those of ordinary skill in the art is to be understood that, can make amendment or be equal to replacement technical scheme of the present invention, and not breaking away from the aim and the scope of technical solution of the present invention, it all should be encompassed in the middle of the claim scope of the present invention.

Claims (9)

1. beat-to-beat division method of intracranial pressure signal, it is characterized in that the intracranial pressure signal input computer with ICP (monitor intracranial pressure monitor) is gathered carries out low-pass filtering and sampling pretreatment by computer, then intracranial pressure signal is cut apart by shooting, the concrete steps of cutting apart by shooting comprise:

A) set up K mutually different template signal; Each template signal is the segment signal of each one-period before and after beat starting point in the intracranial pressure signal of having discerned, and this segment signal is a N sampled point by sampling or interpolation processing; Wherein, K 〉=2, the span of N is 100～1000;

B) set up N sampled point in each template signal respectively with respect to the normalization log-polar of its beat starting point;

C) for intracranial pressure signal to be measured, extracting duration from its starting point is t ₀Signal segment as detection segment; Then detection segment is carried out autocorrelation analysis, the interval in the auto-correlation function of calculating detection segment between every adjacent two local maximums is got the approximate cycle of the meansigma methods of described interval as detection segment; Wherein, t ₀Span be 30～90s;

D) calculate in the detection segment from section start to ε doubly all local minimum points in approximate cycle; Wherein, the span of ε is 1.2～1.6;

E) extract the characteristic area of each local minimum point; The characteristic area of each local minimum point is the segment signal in each approximate cycle before and after this local minimum point in the intracranial pressure signal to be measured, and this segment signal is a N sampled point by sampling or interpolation processing;

F) set up N sampled point in the characteristic area of each local minimum point respectively with respect to the normalization log-polar of its local minimum point;

G) calculate the characteristic area of each local minimum point and each template signal cross-correlation coefficient respectively, with the similarity of the maximum in the cross-correlation coefficient of the characteristic area of each local minimum point and each template signal as this local minimum point based on the normalization log-polar; The computing formula of described cross-correlation coefficient is:

And,

Wherein, P _{I, k}For current ε in the detection segment doubly is similar to the cycle with the characteristic area of interior i local minimum point and the cross-correlation coefficient of k template signal; (β _{I, n}, γ _{I, n}) be in the detection segment the doubly approximate cycle of current ε with the normalization log-polar of n sampled point in the characteristic area of interior i local minimum point with respect to this local minimum point, β _{I, n}Be normalization utmost point footpath, γ _{I, n}Be polar angle;

Be in k the template signal n sampled point with respect to the normalization log-polar of its beat starting point, α _{K, n}Be normalization utmost point footpath,

Be polar angle; K ∈ 1,2 ..., K}, n ∈ 1,2 ..., N}; W _pThe expression weight, the span of its weight coefficient λ is 0＜λ＜1;

H) relatively draw the local minimum point of doubly approximate cycle of current ε in the detection segment with interior similarity maximum, and with the similarity and the pre-set threshold C of this local minimum point ₀Compare; If its similarity is greater than threshold value C ₀, judge that promptly this local minimum point is a beat starting point; Wherein, threshold value C ₀Span be 0.1～0.4;

I) be starting point with the doubly approximate cycle of current ε in the detection segment with a local minimum point of interior similarity maximum, calculate thereafter the doubly approximate cycle of ε with interior all local minimums point; Repeating step e then)～i), judge beat starting points all in the detection segment thus;

J) in intracranial pressure signal to be measured, be starting point with last beat starting point position in the current detection section, extracting thereafter, duration is t ₀Signal segment as new detection segment; And, with the meansigma methods of last 3 normal cycle intervals in the current detection section the approximate cycle as new detection segment; Repeating step d then)～j), judge beat starting points all in the intracranial pressure signal to be measured thus;

Described normal cycle interval is meant the interval that is no more than 1.5 times between adjacent two beat starting points and is not less than 0.5 times of current approximate cycle duration;

K) intracranial pressure signal to be measured is cut apart by shooting, stored and show intracranial pressure signal beat by beat segmentation result to be measured.

2. beat-to-beat division method of intracranial pressure signal according to claim 1 is characterized in that: described step b) is specially:

B1) set up N sampled point in each template signal respectively with respect to Descartes's relative coordinate of its beat starting point, and carry out the average normalized; The computing formula of average normalized is as follows:

${{\mathrm{\ρ}}_{k,n}}^{\′}=\frac{{\mathrm{\ρ}}_{k,n}}{{\stackrel{\‾}{\mathrm{\ρ}}}_{k,n}}=\frac{\sqrt{{x_{k,n}}^2+{y_{k,n}}^2}}{\frac{1}{N-1}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{{x_{k,n}}^2+{y_{k,n}}^2}};$

θ _{K, n}'=θ _{K, n}, and θ _{K, n}' ∈ (π, π];

Wherein, (x _{K, n}, y _{K, n}) be in k the template signal n sampled point with respect to Descartes's relative coordinate of its beat starting point, (ρ _{K, n}, θ _{K, n}) be and (x _{K, n}, y _{K, n}) corresponding polar coordinate; (ρ _{K, n}', θ _{K, n}') be (ρ _{K, n}, θ _{K, n}) polar coordinate after the average normalized; K ∈ 1,2 ..., K}, n ∈ 1,2 ..., N};

B2) according to step b1) polar coordinate after the average normalized of gained, respectively the sampled point of the N in each template signal is projected the log-polar territory, and carry out normalized, obtain N sampled point in each template signal with respect to the normalization log-polar of its beat starting point; The computing formula of normalized is as follows:

${\mathrm{\α}}_{k,n}=\frac{{\mathrm{\ξ}}_{k,n}-{\mathrm{\ξ}}_{k,\min }}{{\mathrm{\ξ}}_{k,\max }-{\mathrm{\ξ}}_{k,\min }},$

Wherein,

Be in k the template signal n sampled point with respect to the normalization log-polar of its beat starting point, α _{K, n}Be normalization utmost point footpath,

Be polar angle; (ξ _{K, n}, ψ _{K, n}) be the log-polar of n sampled point correspondence after throwing in k the template signal, utmost point footpath ξ _{K, n}=log ρ _{K, n}', polar angle ψ _{K, n}=θ _{K, n}'; K ∈ 1,2 ..., K}, n ∈ 1,2 ..., N}; ξ _{K, max}And ξ _{K, min}Be respectively each sampled point in k the template signal maximum and the minima in utmost point footpath in the corresponding log-polar after throwing.

3. beat-to-beat division method of intracranial pressure signal according to claim 1 is characterized in that: described step f) is specially:

F1) set up N sampled point in the characteristic area of each local minimum point respectively with respect to Descartes's relative coordinate of this local minimum point, and carry out the average normalized; The computing formula of average normalized is as follows:

${{\mathrm{\ρ}}_{i,n}}^{\′}=\frac{{\mathrm{\ρ}}_{i,n}}{{\stackrel{\‾}{\mathrm{\ρ}}}_{i,n}}=\frac{\sqrt{{x_{i,n}}^2+{y_{i,n}}^2}}{\frac{1}{N-1}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{{x_{i,n}}^2+{y_{i,n}}^2}};$

θ _{I, n}'=θ _{I, n}, and θ _{I, n}' ∈ (π, π];

Wherein, (x _{I, n}, y _{I, n}) be in the detection segment the doubly approximate cycle of current ε with the Descartes relative coordinate of n sampled point in the characteristic area of interior i local minimum point with respect to this local minimum point, (ρ _{I, n}, θ _{I, n}) be and (x _{I, n}, y _{I, n}) corresponding polar coordinate; (ρ _{I, n}', θ _{I, n}') be (ρ _{I, n}, θ _{I, n}) polar coordinate after the average normalized; N ∈ 1,2 ..., N};

F2) according to step f1) polar coordinate after the average normalized of gained, respectively the sampled point of the N in the characteristic area of each local minimum point is projected the log-polar territory, and carry out normalized, obtain N sampled point in the characteristic area of each local minimum point with respect to the normalization log-polar of this local minimum point; The computing formula of normalized is as follows:

${\mathrm{\β}}_{i,n}=\frac{{\mathrm{\ξ}}_{i,n}-{\mathrm{\ξ}}_{i,\min }}{{\mathrm{\ξ}}_{i,\max }-{\mathrm{\ξ}}_{i,\min }},$ γ _i，n＝ψ _i，n；

Wherein, (β _{I, n}, γ _{I, n}) be in the detection segment the doubly approximate cycle of current ε with the normalization log-polar of n sampled point in the characteristic area of interior i local minimum point with respect to this local minimum point, β _{I, n}Be normalization utmost point footpath, γ _{I, n}Be polar angle; (ξ _{I, n}, ψ _{I, n}) being the log-polar of doubly approximate cycle of current ε in the detection segment with n sampled point correspondence after throwing in the characteristic area of interior i local minimum point, the utmost point is ξ directly _{I, n}=log ρ _{I, n}', polar angle ψ _{I, n}=θ _{I, n}'; N ∈ 1,2 ..., N}; ξ _{I, max}And ξ _{I, min}Be respectively each sampled point in the characteristic area of i local minimum point in the detection segment maximum and the minima in utmost point footpath in the corresponding log-polar after throwing.

4. according to each described beat-to-beat division method of intracranial pressure signal in the claim 1～3, it is characterized in that: the scope of the cut-off frequency of described low-pass filtering is 20～50Hz.

5. according to each described beat-to-beat division method of intracranial pressure signal in the claim 1～3, it is characterized in that: the scope of the pretreated sample frequency of described pre-sampling is 125～1000Hz.

6. according to each described beat-to-beat division method of intracranial pressure signal in the claim 1～3, it is characterized in that: the preferred value of described N is 200.

7. according to each described beat-to-beat division method of intracranial pressure signal in the claim 1～3, it is characterized in that: described t ₀Preferred value be 60s.

8. according to each described beat-to-beat division method of intracranial pressure signal in the claim 1～3, it is characterized in that: the preferred value of described ε is 1.5.

9. according to each described beat-to-beat division method of intracranial pressure signal in the claim 1～3, it is characterized in that: the preferred value of weight coefficient λ is 0.8 in the described step g), described step h) middle threshold value C ₀Preferred value be 0.20.

Images