CN105342556A - Tumor signal extraction method based on integrated empirical mode decomposition - Google Patents

Abstract

The invention relates to a tumor signal extraction method based on integrated empirical mode decomposition. According to the method, a plurality of antennae, distributed on the surface of a breast phantom uniformly at intervals, are adopted. The method includes: enabling one of an antenna array to transmit signals and the rest antennae to receive reflection signals from the inside of the breast, setting that the reflection signals received by one specific antenna add equal-length normally-distributed white noise to original signals for multiple times, performing signal decomposition after the white noise is added every time, continuing signal decomposition on the remainders, and stopping decomposition until the remainders are in a monotonous trend so as to obtain a decomposition result; subjecting each intrinsic mode component to ensemble average according to a principle that a statistical average of an uncorrelated random sequence is zero, computing correlation coefficients of the intrinsic mode components and the original signals, and extracting the intrinsic mode components with large correlation coefficients for processing so as to obtain ultra-wideband tumor signals. The tumor signal extraction method based on integrated empirical mode decomposition is simple and capable of obtaining more information so as to make breast imaging more accurate.

Description

A kind of tumor signal extracting method based on integrated empirical mode decomposition

Technical field

The invention belongs to signal processing technology field, relate to a kind of microwave imagery formation method.

Background technology

Breast tumor is the malignancy disease that women's sickness rate is the highest, and mortality rate occupies first of women's mortality of malignant tumors.The diagnosis of early stage breast tumor all has decisive significance for the survival rate at a specified future date improving mastotic treatment rate and patient undoubtedly.The detection method of breast carcinoma of early stage conventional at present comprises mammography, ultrasonograph technology, computed tomography, nmr imaging technique, thermal imaging detection etc., but all multi-methods all exist certain shortcoming, as low, costly etc. in produced radiation injury, image contrast to human body.The principle of ultra broadband Electromagnetic Wave Detection breast carcinoma is that different biological tissue is different to electromagnetic absorption, reflection and transmissison characteristic, and the electromagnetic field produced when the pulse signal of antenna transmission is propagated in mammary gland tissue can reflect the abundant information of malignant tissue.Simultaneously ultra-wideband microwave signal has that radiant power is low, the target information amount of carrying large, provides grade to locate and the advantage such as testing cost is lower, can as the conventional means of early stage breast cancer.Therefore exploitation ultra-wideband microwave breast image imaging algorithm is necessary.Make imaging process easier, imaging results is more accurate.Whether can accurately extract ultra-wideband microwave tumor signal, be the prerequisite more accurately realizing the imaging of ultra-wideband microwave breast image.

Summary of the invention

The object of this invention is to provide a kind of process easier, abundant information can be obtained, breast image imaging tumor signal extracting method more accurately can be made.Technical scheme of the present invention is as follows:

Based on a tumor signal extracting method for integrated empirical mode decomposition, the antenna that the method adopts is the antenna that several intervals are evenly distributed on that breast imitates surface, comprises the following steps:

1) one of them antenna transmission signal in aerial array, all the other antennas receive the reflected signal from internal mammary, if the reflected signal that certain antenna receives is primary signal y (t);

2) in primary signal y (t), repeatedly add the white noise n of the normal distribution of equal length _it (), obtains y _i(t)=y (t)+n _i(t); Wherein, y _it () is the signal after adding white noise i-th time, often add a white noise, all will carry out following signal decomposition operation:

A. y is determined _it all Local Extremum of (), ask for y by cubic spline function _ithe coenvelope u of (t) _i1(t) and lower envelope u _i2the local mean value of (t) $m_i\left(t\right)=\frac{1}{2}\left(x_{i1}\right(t)+x_{i2}(t\left)\right);$

B. new signal h can be obtained by deducting local mean value from the signal adding white noise _i(t)=y _i(t)-m _i(t);

C. h is judged _it whether () be intrinsic mode function (IMF): judge whether h (t) meets following two conditions: in (1) each intrinsic mode function, the number of its zero crossing is equal with the number of extreme point or differ from most one; (2) at h _it, in the arbitrary local signal be truncated in (), the envelope that its maximum point is formed and the result that both envelopes that its minimum point is formed are averaging are zero; If do not met, then in primary signal y (t), newly add a white noise obtain new y _it (), repeats step a)-b), until decompose the h obtained _it () meets the condition of intrinsic mode function;

If d. met, then h _it () is for decomposing first the natural mode of vibration component c obtained _i1=h _it (), obtains remainder r simultaneously _i1(t)=y _i(t)-c _i1;

3) according to step 2) signal decomposition method, to step 2) the remainder r that obtains _i1t () carries out same signal decomposition, obtain second natural mode of vibration component c _i2and remainder r _i2(t);

4) r is judged _i2t whether () be monotonic trend, if r _i2t () is in monotonic trend, then decompose stopping, obtaining decomposition result c _i1, c _i2and remainder r _i2(t), otherwise, continue according to step 2) signal decomposition method to r _i2t () decomposes, until remainder is monotonic trend decompose stopping, now, obtain decomposition result: multiple natural mode of vibration component c _i1(t), c _i2(t), c _i3(t) ..., c _ij(t) and remainder r _ij(t), wherein, c _ija jth natural mode of vibration component of gained is decomposed t () expression adds white noise i-th time after;

5) average statistical of uncorrelated random sequence is utilized to be the principle of zero, by each natural mode of vibration component c _ijt () carries out ensemble average, repeatedly add the impact of white noise on true IMF to offset, final decomposition result is:

6) calculate the correlation coefficient of each natural mode of vibration component and primary signal, natural mode of vibration component extraction larger for correlation coefficient is out processed, obtains the ultra-broadband signal carrying out breast imaging.

Accompanying drawing explanation

Fig. 1 organize models and antenna structure view

Fig. 2 antenna A6 sends out scheming containing tumor signal and IMF1-IMF13 and residual term r13 thereof of A11 receipts

The all signals of Fig. 3 extract the tumor confocal imaging figure of tumor response signal gained after decomposing

Detailed description of the invention

Detect in breast tumor experiment, by linear array antenna cover on breast model at ultra-wideband microwave.Breast model and linear antenna array are as shown in Figure 1.Breast model is followed successively by air layer, petrolatum layer, skin layer, fat deposit and breastbone layer from top to bottom, wherein comprises a tumor and gland tissue in fatty tissue.Linear antenna array is close to skin and is arranged on breast model.Aerial array comprises 12 antennas.Tumoral character signal extraction adopts antenna A1 ~ A12 alternately to measure tumor echo reflection signal as dual-mode antenna, and carries out signal processing to many group echo-signals, extracts tumoral character target signal, accurate positioning tumor position.For the requirement meeting imaging resolution and investigation depth adopts the single order that mid frequency is 5GHz, bandwidth is 10GHz to lead gaussian signal, signal waveform as shown in Figure 2.Specific implementation process is as follows:

1. antenna transmission signal, all the other antennas receive the reflected signal from internal mammary.Scan the every bit of imaging region, the signal obtained is represented by y (t).The white noise n of the normal distribution of equal length is repeatedly added in signal y (t) _i(t), namely

y _i(t)＝y(t)+n _i(t)(1)

In formula, y _it () is the signal after adding white noise i-th time.

2. determine y _it all Local Extremum of (), ask for y by cubic spline function _ithe coenvelope u of (t) _i1(t) and lower envelope u _i2the local mean value of (t) $m_i\left(t\right)=\frac{1}{2}\left(u_{i1}\right(t)+u_{i2}(t\left)\right);$

3. can obtain new signal h by deducting local mean value from the signal adding white noise _i(t)=y _i(t)-m _i(t)

4. judge h _it whether () be intrinsic mode function (IMF): judge whether h (t) meets following two conditions: in (1) each intrinsic mode function, the number of its zero crossing is equal with the number of extreme point or differ from most one; (2) at h _it, in the arbitrary local signal be truncated in (), the envelope that its maximum point is formed and the result that both envelopes that its minimum point is formed are averaging are zero; If do not met, then in primary signal y (t), newly add a white noise obtain new y _it (), repeats step 2-3, until decompose the h obtained _it () meets the condition of intrinsic mode function;

5. according to the signal decomposition method of step 2-4, to the remainder r obtained _i1t () carries out same signal decomposition, obtain second natural mode of vibration component c _i2and remainder r _i2(t)

6. judge r _i2t whether () be monotonic trend, if r _i2t () is in monotonic trend, then decompose stopping, obtaining decomposition result c _i1, c _i2and remainder r _i2(t), otherwise, continue signal decomposition method according to step 2-5 to r _i2t () decomposes, until remainder is monotonic trend decompose stopping, now, obtain decomposition result: multiple natural mode of vibration component c _i1(t), c _i2(t), c _i3(t) ..., c _ij(t) and remainder r _ij(t), wherein, c _ija jth natural mode of vibration component of gained is decomposed t () expression adds white noise i-th time after

7. utilize the average statistical of uncorrelated random sequence to be the principle of zero, by each natural mode of vibration component c _ijt () carries out ensemble average, repeatedly add the impact of white noise on true IMF to offset, final decomposition result is: in formula, N refers to the number adding white noise sequence.

8. carried out by all signals obtained selecting white noise coefficient to be 2.0 when EEMD decomposes, cycle-index is 50, obtains the IMF1 to IMF13 (i.e. C1-C13) after decomposing.Be illustrated in figure 2 antenna A6 and send ultra-wideband microwave, the signal original graph comprising tumor information received by antenna A11 and decompose the signal graph of IMF1 to IMF13 and the remainder r13 thereof obtained through EEMD.Calculate the correlation coefficient of each IMF and primary signal, obtain result as shown in table 1.The correlation coefficient that can obtain IMF7 and IMF8 is comparatively large, is added by two signals, is reconstructed into a new signal.This signal is the signal containing tumor response.

9. pair useful signal confocal imaging algorithm that process decomposition extracts carries out imaging, can obtain the breast image comprising tumor information.

For the effectiveness of the breast imaging algorithm that checking the present invention proposes, the breast tissue model shown in Fig. 1 is adopted to detect.Moulded dimension is set as that the aerial array be made up of 12 antennas is arranged on skin layer, and for close to true acquisition environment, set up skin, fat, rib successively, wherein body of gland and tumor are present in fat deposit, and aerial array is distributed in skin surface.Diameter is that the tumor of 4mm is positioned at (100,50).Carried out by all signals obtained selecting white noise coefficient to be 2.0 when EEMD decomposes, cycle-index is 50, obtains the IMF1 to IMF13 after decomposing.As Fig. 2 for there being a tumor time A6 launch, the reception waveform of A11 and the IMF component map obtained after decomposing thereof.To all IMF obtained, calculate the correlation coefficient of itself and primary signal, extract the IMF that correlation coefficient is larger.Here the correlation coefficient of IMF7 and IMF8 is comparatively large, therefore extracts IMF7 and IMF8 and reconstructs a new signal.Here extract the signal after IMF7 and IMF8 addition, achieve the extraction of tumor scattered signal.All effective signal use confocal imaging algorithms that obtains are carried out being processed into picture, the breast imaging figure comprising tumor more clearly can be obtained, as shown in Figure 3.Demonstrate the effectiveness of this algorithm.

Correlation coefficient between table 1 each IMF and primary signal

Claims (1)

1., based on a tumor signal extracting method for integrated empirical mode decomposition, the antenna that the method adopts is the antenna that several intervals are evenly distributed on that breast imitates surface, comprises the following steps:

1) one of them antenna transmission signal in aerial array, all the other antennas receive the reflected signal from internal mammary, if the reflected signal that certain antenna receives is primary signal y (t);

2) in primary signal y (t), repeatedly add the white noise n of the normal distribution of equal length _it (), obtains y _i(t)=y (t)+n _i(t); Wherein, y _it () is the signal after adding white noise i-th time, often add a white noise, all will carry out following signal decomposition operation:

A. y is determined _it all Local Extremum of (), ask for y by cubic spline function _ithe coenvelope u of (t) _i1(t) and lower envelope u _i2the local mean value of (t) $m_i\left(t\right)=\frac{1}{2}(u_{i1}\left(t\right)+u_{i2}\left(t\right));$

B. new signal h can be obtained by deducting local mean value from the signal adding white noise _i(t)=y _i(t)-m _i(t);

C. h is judged _it whether () be intrinsic mode function (IMF): judge whether h (t) meets following two conditions: in (1) each intrinsic mode function, the number of its zero crossing is equal with the number of extreme point or differ from most one; (2) at h _it, in the arbitrary local signal be truncated in (), the envelope that its maximum point is formed and the result that both envelopes that its minimum point is formed are averaging are zero; If do not met, then in primary signal y (t), newly add a white noise obtain new y _it (), repeats step a) and b), until decompose the h obtained _it () meets the condition of intrinsic mode function;

If d. met, then h _it () is for decomposing first the natural mode of vibration component c obtained _i1=h _it (), obtains remainder r simultaneously _i1(t)=y _i(t)-c _i1;

3) according to step 2) signal decomposition method, to step 2) the remainder r that obtains _i1t () carries out same signal decomposition, obtain second natural mode of vibration component c _i2and remainder r _i2(t);

4) r is judged _i2t whether () be monotonic trend, if r _i2t () is in monotonic trend, then decompose stopping, obtaining decomposition result c _i1, c _i2and remainder r _i2(t), otherwise, continue according to step 2) signal decomposition method to r _i2t () decomposes, until remainder is monotonic trend decompose stopping, now, obtain decomposition result: multiple natural mode of vibration component c _i1(t), c _i2(t), c _i3(t) ..., c _ij(t) and remainder r _ij(t), wherein, c _ija jth natural mode of vibration component of gained is decomposed t () expression adds white noise i-th time after;

5) average statistical of uncorrelated random sequence is utilized to be the principle of zero, by each natural mode of vibration component c _ijt () carries out ensemble average, repeatedly add the impact of white noise on true IMF to offset, final decomposition result is:

6) calculate the correlation coefficient of each natural mode of vibration component and primary signal, natural mode of vibration component extraction larger for correlation coefficient is out processed, obtains the ultra-broadband signal carrying out breast imaging.