CN101637394A - Method for positioning and segmenting heart sound signal - Google Patents

Abstract

The invention relates to a method for positioning and segmenting heart sound signal and a device. The device comprises a preprocessing unit, an initial state determination unit, a calculation unit, afirst search unit, a second search unit and a segmentation unit, wherein the preprocessing unit comprises a filter functional module which is used for preprocessing the heart sound signal and filtering noise out of band; the initial state determination unit is used for determining a start cycle, the maximum cycle and an initial point; the calculation unit is used for calculating correlation coefficient of signals in two adjacent calculation cycles; the first search unit is used for searching correlation coefficient extremum 1 in the correlation coefficient; the second search unit is used for search correlation coefficient extremum 2 in the correlation coefficient extremum 1; and the segmentation unit is used for segmenting the heart sound signal. In the invention, the method for positioning and segmenting the heart sound signal and the device use a self-correlated coefficient method to position automatically and accurately.

Description

A kind of cardiechema signals location, segmentation method

Technical field

The present invention relates to the cardiechema signals process field, is a kind of cardiechema signals cycle localization method specifically.

Background technology

Ultrasonic multispectral development and utilization of reining in advanced auxiliary diagnosis instrument such as instrument and ECG (Electrocardiogram electrocardiogram) makes that the utilization of cardiechema signals is treated coldly.Yet the ultrasonic multispectral instrument of reining in, it costs an arm and a leg, and is not easy to popularize; Though the ECG signal but can not reflect the pathological information relevant with organic heart disease effectively to more effective with the diagnosis of blood circulation and blood tissues relevant disease.

Cardiechema signals has comprised the abundant information that can reflect the normal or pathology of heart; normal hear sounds comprises first heart sound (S1), second heart sound (S2), third heart sound (S3) and four hear sounds compositions of fourth heart sound (S4); wherein S1, S2 are the parts that can hear; S3, S4 intensity are very weak, can hear hardly.If cardiac function occurs unusual or pathological changes appears in cardiovascular; to comprise other the outer noise compositions except that S1, S2 in the hear sounds; important diagnostic information such as the noise of these appearance and distortion can reflect that cardiovascular disease still is difficult to produce clinical and pathological change symptom in the past with some.Comprised these abundant information in the cardiechema signals, in the field of detecting the type disease, cardiechema signals has unrivaled superiority.Simultaneously easy to detect, the noinvasive of cardiechema signals, cost are minimum, can be used as that heart disease detects, the conventional means of prevention.Yet traditional cardiac auscultation technology is carried out with people's ear audition, and this mode depends on the sensitivity of people's ear audition and hearer's subjective experience and judgement, and its effect has significant limitation.Develop a kind of digital hear sounds automatic analyzer, can be simple and easy, detect cardiechema signals easily, for medical personnel and patient provide effective reference information, satisfy hospital and patient's needs, have great social value and economic worth.

The research core that the numeral hear sounds is analyzed automatically is the automatic segmentation technology and the mode identification technology of cardiechema signals, at present the automatic segmentation technology of cardiechema signals mainly contains two kinds: a kind of is traditional to do with reference to the hear sounds segmentation algorithm by means of electrocardiosignal etc., and another kind is not by the segmentation algorithm of any signal.

Need by the corresponding time relationship of electrocardio QRS (electrocardio wave group) waveform and hear sounds, make a cardiac cycle by means of the cardiechema signals segmentation algorithm of electrocardiosignal or machcardiogram signal, carry out segmentation then for first kind according to the position of electrocardio QRS ripple.M.W.Groch proposes to utilize electrocardio as a reference, carries out the segmentation of cardiechema signals according to the method for hear sounds time domain specification; Lener proposes to utilize electrocardiosignal and the machcardiogram hear sounds segmentation algorithm as reference.

Second kind is exactly that LG Durand and H Liang etc. have proposed not segmentation algorithm by means of cardiac electrical hear sounds.LG Durand utilizes the main energy distributions of matching pursuit algorithm location cardiechema signals, and then hear sounds is carried out segmentation.H Liang then adopts wavelet decomposition and restructing algorithm that hear sounds is carried out segmentation, and at first the wavelet decomposition cardiechema signals is selected the reconstruct frequency band according to the frequency of s1, s2 then, to the calculated signals Shannon energy after the reconstruct, and then passes through the segmentation that the Shannon energy is realized hear sounds.The optimal wavelet threshold values de-noising algorithm that utilizes of a domestic Zhao Zhi proposition carries out the segmentation of the inherent character realization hear sounds of pretreatment, Hilbert transform extraction cardiechema signals envelope and cardiechema signals to cardiechema signals.

Traditional cardiechema signals segmentation algorithm weak point is: the first, need electrocardiosignal or machcardiogram signal for referencial use, increased the burden of software and hardware; The second, cardiechema signals itself is a kind of typical non-stationary signal, and traditional hear sounds segmentation algorithm is handled cardiechema signals as a kind of stationary signal, utilize its time domain and frequency domain character to position, and this way can produce than mistake; Three, traditional hear sounds segmentation algorithm is higher to the prescription of cardiechema signals, and to noise-sensitive.

Not by means of electrocardiosignal fragmentation technique shortcoming: the first, algorithm complex height, very consuming time.The second, noise is bigger to sectional influential effect; Three, can only locate the Position Approximate of s1, s2, lose s1, the information of s2 persistent period, can not judge the time range of systole and relaxing period, more can not accomplish the accurate location in cardiechema signals cycle.

Summary of the invention

At above-mentioned shortcoming, the invention provides a kind of utilize that the autocorrelation coefficient method realizes, can accurately locate cardiechema signals and sectional method automatically.

A kind of cardiechema signals location, segmentation method comprise the steps:

1) to the cardiechema signals pretreatment, promptly remove after making an uproar at least greater than the cardiechema signals in two hear sounds cycles; Need that promptly the primary cardiechema signals that takes out is carried out filtering and (can adopt low-pass filtering, the removal high-frequency noise), denoising (can adopt small echo denoising method to remove specific noise such as respiratory murmur, partial frictional sound) is handled, remove after making an uproar at least greater than the cardiechema signals in two hear sounds cycles, for step, the device of back provides purified cardiechema signals.

2) the original state determining unit is determined start cycle, maximum cycle and starting point, cycle rule of thumb is worth, multiply by sampling rate by the experience hear sounds cycle determines, computing cycle with start cycle as initial value, the determining of starting point generally can adopt first of cardiechema signals after the pretreatment as starting point; This experience hear sounds cycle is an empirical data, for example heart beating is 70 times under people's normal condition, heart beat cycle is exactly about 14ms, general heart beating scope can be considered 50 times to 150 times, cycle is exactly 20ms to about the 6ms like this, 8k converts according to sample rate, and the cycle of calculating with sampling number is between 160 o'clock to 48 o'clock.This is to realize easily.

3) calculate the correlation coefficient that writes down adjacent two computing cycle signals; Get the cardiechema signals of a computing cycle after the starting point and adjacent thereafter length and be the cardiechema signals data of a computing cycle and carry out related operation and storage.

4) slide backward the calculating starting point, if judgement reaches end point then enter the 5th) step, otherwise return the 3rd) step; The step-length that starting point is slided is adjusted according to computational accuracy and computational complexity, generally can get 1, and the definite of end point deducts the computing cycle acquisition by cardiechema signals length.

5) search correlation coefficient extreme value 1 in above-mentioned correlation coefficient is promptly searched for maximum.

6) increase computing cycle according to step-length, if computing cycle then enters the 8th greater than maximum cycle) step, otherwise return the 3rd) step; The step-length that is increased in this step can be adjusted according to computational accuracy and system's operational capability, can obtain the highest computational accuracy when getting 1, but can bring maximum amount of calculation.

7) correlation coefficient extreme value 2 in above-mentioned correlation coefficient extreme value 1, extreme value 2 pairing computing cycles are the hear sounds cycle;

8) be the cardiechema signals of complete one-period from corresponding the core one section cardiechema signals of sound Cycle Length of starting points that calculates of 2 of correlation coefficient extreme values.

The described the 3rd) step is that the calculating correlation coefficient is to calculate according to following formula:

$r=\frac{\mathrm{n\Σxy}-\mathrm{\Σx\Σy}}{\sqrt{\mathrm{n\Σ}{x}^2-{\left(\mathrm{\Σx}\right)}^2}\sqrt{n{\mathrm{\Σy}}^2-{\left(\mathrm{\Σy}\right)}^2}}$

Wherein, r is a correlation coefficient,

N is counting of being correlated with, the promptly above-mentioned the 6th) computing cycle in the step,

X is the above-mentioned the 3rd) in first section heart sound data of adjacent two computing cycle signals,

Y is the above-mentioned the 3rd) in second section heart sound data of adjacent two computing cycle signals,

I, j refer to the sequence number of data in two sections hear sounds respectively, scope from 1 to n.

The follow-up needs of the correlation coefficient that calculates are used, so a memory element is set, are used to write down the correlation coefficient that above-mentioned process calculates, and this is to realize easily.

Like this, through pretreated cardiechema signals is purified cardiechema signals, can obtain its correlation coefficient through calculating then, has just obtained the accurate hear sounds cycle through twice search again, and then carry out segmentation according to the hear sounds cycle that obtains, realized purpose of the present invention.

A kind of cardiechema signals location, sectioning comprise:

One pretreatment unit comprises the filter function module, is used for cardiechema signals is carried out pretreatment the filter bag external noise;

One original state determining unit is used for determining start cycle, maximum cycle and starting point;

One computing unit is used to calculate the correlation coefficient of adjacent two computing cycle signals;

One first search unit is used in above-mentioned correlation coefficient search correlation coefficient extreme value 1;

One second search unit is used in above-mentioned correlation coefficient extreme value 1 correlation coefficient extreme value 2;

One segmenting unit is used for cardiechema signals is carried out segmentation.

Described filter function module is the noise removal function module, and cardiechema signals is carried out filtering (can adopt low-pass filtering, remove high-frequency noise), denoising (can adopt small echo denoising method to remove specific noise such as respiratory murmur, partial frictional sound).

Preferably, cardiechema signals provided by the invention location, sectioning also comprise a memory element, are used to write down the correlation coefficient through calculating.

Cardiechema signals provided by the invention location, segmentation method and device have following advantage:

1. algorithm complex is low, practical, obtains the hear sounds cycle of needs very fast, is easy to use under embedded environment, does not need cardiechema signals is carried out envelope extraction, and algorithm complex has been simplified in pretreatment such as Shannon energy calculating greatly.

2. locate segmentation precision height (being up to 1 sampling point), and precision is adjustable according to applied environment.

3. capacity of resisting disturbance is strong.

4. be easy to realize, need not complicated at a high speed electronic circuit and just can realize easily.

5. need be by any reference signal such as electrocardiosignal, machcardiogram signal.

6. the different cycles signal had adaptivity, for sorting algorithm provides accurate reference.

Description of drawings

Fig. 1 is the flow chart of the embodiment of the invention;

Fig. 2 a is original hear sounds design sketch in the embodiment of the invention (second section splitting of heart sounds);

Fig. 2 b is later first section hear sounds design sketch of segmentation in the embodiment of the invention;

Fig. 2 c is later second section hear sounds design sketch of segmentation in the embodiment of the invention;

Fig. 2 d is later the 3rd section hear sounds design sketch of segmentation in the embodiment of the invention.

The specific embodiment

A kind of cardiechema signals location, segmentation method, this cardiechema signals comprises 15 hear sounds cycles, as shown in Figure 1, is achieved in that

At first, start from step S1 above-mentioned cardiechema signals is carried out pretreatment, i.e. filtering filters out high-frequency noise, by frequency limitation at 1.5K.

Enter step S2 then and determine start cycle, maximum cycle, get at 1500 and do the initial cycle, do maximum cycle at 3000, and be starting point with first point of this section cardiechema signals.

Entering S3 step then does relevantly with first cycle (beginning the back one-period at interval from starting point) and second period (the one-period interval of first all after date), calculate correlation coefficient according to following formula:

$r=\frac{\mathrm{n\Σxy}-\mathrm{\Σx\Σy}}{\sqrt{\mathrm{n\Σ}{x}^2-{\left(\mathrm{\Σx}\right)}^2}\sqrt{n{\mathrm{\Σy}}^2-{\left(\mathrm{\Σy}\right)}^2}}$

Wherein, r is a correlation coefficient,

N is counting of being correlated with, the computing cycle during promptly S6 goes on foot,

X is first section heart sound data of adjacent two computing cycle signals among the above-mentioned S3,

Y is second section heart sound data of adjacent two computing cycle signals among the above-mentioned S3,

I, j refer to the sequence number of data in two sections hear sounds respectively, scope from 1 to n.

And the correlation coefficient that calculates is stored in the memory element, for future use.

Entering S4 step then slides backward the calculating starting point, judges whether to reach end point (end point is got 31000 points), if reach end point then enter the S5 step, otherwise returns the S3 step; The step-length that starting point is slided is adjusted according to computational accuracy and computational complexity, generally can get 1, and the definite of end point deducts the computing cycle acquisition by cardiechema signals length.

Enter the S5 step then, promptly first search unit is searched for correlation coefficient extreme value 1 in above-mentioned correlation coefficient, promptly searches for maximum.

And then enter the S6 step, and increase computing cycle according to step-length, if computing cycle then enters the S7 step greater than maximum cycle (getting 3000), otherwise return the S3 step; The step-length that is increased in this step can be adjusted according to computational accuracy and system's operational capability, can obtain the highest computational accuracy when getting 1, but can bring maximum amount of calculation.

Enter S7 step again, i.e. second search unit correlation coefficient extreme value 2 in above-mentioned correlation coefficient extreme value 1, the pairing computing cycle of extreme value 2 (being 2279) is the hear sounds cycle.

Entering S8 step at last carries out segmentation, according to above-mentioned definite hear sounds cycle to the segmentation of this section hear sounds, the result is as shown in Figure 2.Wherein, Fig. 2 a is the original cardiechema signals with splitting of second heart sound feature, and Fig. 2 b, c, d are first three section cardiechema signals adjacent after the segmentation, by the contrast of Fig. 2 a and Fig. 2 b, c, d as can be seen, segmentation algorithm is effectively estimated the hear sounds cycle, and has been carried out effective segmentation.

Claims (7)

1. cardiechema signals location, segmentation method is characterized in that comprising the steps:

1) to the cardiechema signals pretreatment, promptly remove after making an uproar at least greater than the cardiechema signals in two hear sounds cycles;

2) the original state determining unit is determined start cycle, maximum cycle and starting point;

3) calculate the correlation coefficient that writes down adjacent two computing cycle signals;

4) slide backward the calculating starting point, if judgement reaches end point then enter the 5th) step, otherwise return the 3rd) step;

5) search correlation coefficient extreme value 1 in above-mentioned correlation coefficient;

6) increase computing cycle according to step-length, if computing cycle then enters the 8th greater than maximum cycle) step, otherwise return the 3rd) step;

7) correlation coefficient extreme value 2 in above-mentioned correlation coefficient extreme value 1, extreme value 2 pairing computing cycles are the hear sounds cycle;

8) be the cardiechema signals of complete one-period from corresponding the core one section cardiechema signals of sound Cycle Length of starting points that calculates of 2 of correlation coefficient extreme values.

2. cardiechema signals according to claim 1 location, segmentation method is characterized in that: the described the 2nd) step determines that start cycle and maximum cycle are achieved in that rule of thumb value, multiply by sampling rate by the experience hear sounds cycle and determines.

3. cardiechema signals according to claim 1 and 2 location, segmentation method is characterized in that: the described the 3rd) step is that the calculating correlation coefficient is to calculate according to following formula:

$r=\frac{\mathrm{n\Σxy}-\mathrm{\Σx\Σy}}{\sqrt{\mathrm{n\Σ}{x}^2-{\left(\mathrm{\Σx}\right)}^2}\sqrt{\mathrm{n\Σ}{y}^2-{\left(\mathrm{\Σy}\right)}^2}}$

Wherein, r is a correlation coefficient,

N is counting of being correlated with, the promptly above-mentioned the 6th) computing cycle in the step,

X is the above-mentioned the 3rd) in first section heart sound data of adjacent two computing cycle signals,

Y is the above-mentioned the 3rd) in second section heart sound data of adjacent two computing cycle signals,

I, j refer to the sequence number of data in two sections hear sounds respectively, scope from 1 to n.

4. cardiechema signals according to claim 3 location, segmentation method is characterized in that: a memory element is set, is used to write down the correlation coefficient through calculating.

5. cardiechema signals location, sectioning is characterized in that comprising:

One pretreatment unit comprises the filter function module, is used for cardiechema signals is carried out pretreatment the filter bag external noise;

One original state determining unit is used for determining start cycle, maximum cycle and starting point;

One computing unit is used to calculate the correlation coefficient of adjacent two computing cycle signals;

One first search unit is used in above-mentioned correlation coefficient search correlation coefficient extreme value 1;

One second search unit is used in above-mentioned correlation coefficient extreme value 1 correlation coefficient extreme value 2;

One segmenting unit is used for cardiechema signals is carried out segmentation.

6. cardiechema signals according to claim 5 location, sectioning, it is characterized in that: described filter function module is the noise removal function module.

7. cardiechema signals according to claim 5 location, sectioning is characterized in that: also comprise a memory element, be used to write down the correlation coefficient through calculating.

Images