CN107358156A - The feature extracting method of Ultrasonic tissue characterization based on Hilbert-Huang transform - Google Patents

The feature extracting method of Ultrasonic tissue characterization based on Hilbert-Huang transform Download PDF

Info

Publication number
CN107358156A
CN107358156A CN201710417075.3A CN201710417075A CN107358156A CN 107358156 A CN107358156 A CN 107358156A CN 201710417075 A CN201710417075 A CN 201710417075A CN 107358156 A CN107358156 A CN 107358156A
Authority
CN
China
Prior art keywords
imf
hilbert
frequency
feature
energy
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201710417075.3A
Other languages
Chinese (zh)
Other versions
CN107358156B (en
Inventor
林春漪
曹文雄
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201710417075.3A priority Critical patent/CN107358156B/en
Publication of CN107358156A publication Critical patent/CN107358156A/en
Application granted granted Critical
Publication of CN107358156B publication Critical patent/CN107358156B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Signal Processing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Complex Calculations (AREA)

Abstract

The invention discloses a kind of feature extracting method of the Ultrasonic tissue characterization based on Hilbert-Huang transform, comprise the following steps：The ultrasonic echo RF signal multiframes of tissue samples are gathered, region of interest is selected, constructs ultrasonic RF time serieses；Using HHT algorithms, feature extraction is carried out to ultrasonic RF time serieses, obtains the sampling feature vectors based on time, frequency and energy relationship；Using rankfeatures algorithms, Feature Selection and Fusion Features are carried out, calculate sample characteristics fusion index, using this feature fusion index as the feature for levying tissue surely.The difference that Fusion Features index proposed by the invention can effectively reflect between different classes of tissue samples, suitable for the feature extraction field of Ultrasonic tissue characterization.

Description

The feature extracting method of Ultrasonic tissue characterization based on Hilbert-Huang transform
Technical field
The present invention relates to Ultrasonic tissue characterization technical field, more particularly to a kind of ultrasound based on Hilbert-Huang transform The feature extracting method of tissue characterization.
Background technology
Not yet it is apparent from because ultrasonic wave enters after biological tissue the mechanism interacted therewith, people can only be by carrying Take ultrasonic echo information and make explanations and to reach indirectly identification institutional framework, composition, the purpose of state, so as to promote people to enter The research of row Ultrasonic tissue characterization feature extraction.
Existing Ultrasonic tissue characterization feature extracting method mainly has three classes：Based on ultrasonic B figure, based on back scattering acoustic beam RF signals and based on ultrasonic RF time series analyses.In research based on ultrasonic B figure, Type B gradation of image point is mainly utilized The texture of cloth is special, the method use ultrasound image grayscale, therefore easily by imaging parameters such as diasonograph, TGC adjustment Deng influence；Method based on back scattering acoustic beam RF signals is, it is necessary to carry out depth attenuation's compensation, and different tissue attenuation abilities are present Difference, while individual difference can also make it that acoustic wave propagation path is different, so as to have a strong impact on the effect of such method.And based on super The method of sound RF time serieses, RF time serieses derive from the RF data of the same depth different frame of same position, to a certain degree On reduce the influence of individual difference.
Traditional frequency spectrum analysis method is all based on greatly Fast Fourier Transform (FFT) (FFT) and wavelet transformation (WT).Due to depositing In spectral leakage and fence effect, there is larger error in fft analysis non-integer harmonics, though can be compared with by window function and interpolation algorithm Good elimination spectral leakage and fence effect, but these algorithms are often to reduce frequency resolution as cost.In WT, due to high pass It is not complete preferable wave filter with low pass resolution filter, serious crossover phenomenon is present between each frequency band, therefore very Difficulty realizes the strict division of each frequency band.Above method essence is a kind of theory based on base function expansion, signal analysis knot Fruit is largely dependent upon the experience of designer.
Compared with traditional Time-Frequency Analysis Method, and Hilbert-Huang transform (Hilbert-Huang Transform, referred to as HHT) as a kind of relatively emerging Time-Frequency Analysis Method, have in nonlinear and non local boundary value problem is handled more obvious excellent Gesture, a series of component for decompositing different frequencies that HHT can be adaptive according to the characteristics of signal itself, i.e. intrinsic mode function (IMF), the interference of other non-self factors such as selection without considering basic function or window function, obtained result can be preferably The characteristics of reflecting signal itself.
The content of the invention
The shortcomings that it is an object of the invention to overcome prior art and deficiency, there is provided one kind is based on Hilbert-Huang transform Ultrasonic tissue characterization feature extracting method, the Fusion Features index proposed can effectively reflect different classes of tissue sample Difference between this, suitable for the feature extraction field of Ultrasonic tissue characterization.
The purpose of the present invention is realized by following technical scheme：
A kind of feature extracting method available for Ultrasonic tissue characterization based on Hilbert-Huang transform, including following step Suddenly：
S1：The ultrasonic echo RF signal multiframes of tissue samples are gathered, ROI is selected, constructs ultrasonic RF time serieses；
S2：Using HHT algorithms, feature extraction is carried out to ultrasonic RF time serieses, obtains closing based on time, frequency and energy The sampling feature vectors of system；
S3：Using rankfeatures algorithms, Feature Selection and Fusion Features are carried out, calculate sample characteristics fusion index, Using this feature fusion index as the feature for levying tissue surely.
Preferably, in the step S1, the detailed process for constructing ultrasonic RF time serieses is：
S1-1：Tissue is scanned, obtains its ultrasonic echo RF signal multiframes；
S1-2：Demodulate certain frame ultrasonic echo RF signal and show Type B figure；
S1-3：The ROI that size is M × N is chosen on Type B figure；
S1-4：Its preceding L frames RF signal is taken to construct the ultrasonic RF time serieses that M × N number of length is L to every in ROI.
Preferably, in the step S2, using HHT algorithms, the method for calculating sampling feature vectors is：
S2-1：A time series x (t) in this ROI is sampled, enters row set empirical mode decomposition, obtains n IMF points Amount；
S2-2：Based on IMF component extraction temporal signatures：
IMF zero passages are counted 5：IMF1-ZCs、IMF2-ZCs、IMF3-ZCs、IMF4-ZCs、IMF5-ZCs；
IMF variances 5：IMF1-Var、IMF2-Var、IMF3-Var、IMF4-Var、IMF5-Var；
IMF variance contribution ratios 5 IMF1-VarR, IMF2-VarR, IMF3-VarR, IMF4-VarR, IMF5-VarR；
S2-3：To time series x (t) n rank IMF, Hilbert spectral analysis is carried out respectively；
Specifically, step S2-3 includes：
Step a：Hilbert conversion is carried out to an IMF component c (t)P is Cauchy master in formula Value, then seek c (t) analytic signalIt is expressed as polar formWherein amplitude FunctionEnergy functionPhase function C (t) instantaneous frequency can be calculated by phase function
Step b：Hilbert conversion is carried out according to step a respectively to x (t) n ranks IMF, tectonic knot signal, is expressed as Polar form, x (t) is finally reconstructed, is obtainedIt is omitted here Residual components r (t), Re represent to take real part；
Step c：ClaimFor Hilbert amplitude spectrums, abbreviation Hilbert spectrums, utilize Hilbert is composed to time integral, obtains Hilbert marginal spectrums In formulaRepresent kth rank IMF components ck(t) Hilbert marginal spectrums；
S2-4：Based on magnitude function and energy function extraction time domain-energy feature：
5 IMF mean intensities：IMF1-AvgA, IMF2-AvgA, IMF3-AvgA, IMF4-AvgA, IMF5-AvgA,
5 IMF energy：IMF1-Egy、IMF2-Egy、IMF3-Egy、IMF4-Egy、IMF5-Egy；
S2-5：Based on instantaneous frequency distilling time domain-frequecy characteristic：
5 IMF highest frequencies：IMF1-MaxF、IMF2-MaxF、IMF3-MaxF、IMF4-MaxF、IMF5-MaxF；
S2-6：Energy-frequecy characteristic is extracted based on Hilbert marginal spectrums：
5 IMF mean center frequencies：IMF1-MCF、IMF2-MCF、IMF3-MCF、IMF4-MCF、IMF5-MCF；1 x (t) mean center frequency Orig-MCF；
S2-7：Frequency domain-energy feature is extracted based on Hilbert marginal spectrums：
X (t) Hilbert marginal spectrum entropys Orig-EgyS；
X (t) normalization Hilbert marginal spectrum low-frequency ranges energy, middle low-frequency range energy, medium-high frequency section energy, high band Energy Orig-MargS1, Orig-MargS2, Orig-MargS3, Orig-MargS4；
S2-8：Frequency domain-energy curve fit characteristic is extracted based on Hilbert marginal spectrums：
X (t) normalization Hilbert marginal spectrum fitting a straight lines slope, intercept：O-MLFSlope、O-MLFInterp；
Exponent of polynomial function curve fit slope, an intercept：O-MEFSlope、O-MEFInterp；
7 six rank curve matching coefficients：O-MSOFa0、O-MSOFa1、O-MSOFa2、O-MSOFa3、O-MSOFa4、O- MSOFa5、O-MSOFa6；
S2-9：Extraction time-frequency domain-energy feature is composed based on Hilbert：
X (t) Hilbert spectrum gray-scale statistical histogram statistical nature --- average TFImgMean, variance TFImgSD, Degree of bias TFImgSkew, kurtosis TFImgKurto；
S2-10：M × N bar time serieses x in separately sampled ROIi(t), wherein i=0,1,2 ..., M × N-1, weight Step S2-1~S2-9 is performed again, obtains M × N number of characteristic vector corresponding to M × N bar time serieses, finally, then by this M × N number of Characteristic vector is averaging to obtain a characteristic vector, i.e. sampling feature vectors.
Preferably, n IMF component, n >=5 are obtained in step S2-1.
Specifically, the specific processes of step S2-1 include：
Step a：Gaussian sequence is added into x (t), obtains X (t)=x (t)+ω (t)；
Step b：X (t) all local maximums and local minimum are determined, using cubic spline interpolation, are obtained upper and lower Envelope bmaxAnd b (t)min(t) average value, is calculatedObtain h1(t)=X (t)-m1(t)；
Step c：If h1(t) intrinsic mode functions IMF conditions are met, then h1(t) be exactly X (t) the first IMF components；It is no Then repeat step b, h1(t) as new X (t), m is calculated11(t), then h is judged11(t)=h1(t)-m11(t) whether IMF is met Condition, such as it is unsatisfactory for, circulates k times, until h1k(t)=h1(k-1)(t)-m1k(t) IMF conditions are met.Remember c1(t)=h1k(t) it is X (t) single order IMF components, now, residual components r1(t)=X (t)-c1(t)；
Step d：By r1(t) as new X (t), repeat step a~c, obtain X (t) second meets IMF conditions Component c2(t), circulation performs step a~c, until rn(t) monotonic function is turned into, circulation terminates, so as to obtain signal X (t) N (n >=5) rank IMF components set；
Step e：Step a~d repeats n times, adds different Gaussian sequence ω every timei(t), decomposition obtains The set of N group IMF components, then in N group IMF components, the component of corresponding exponent number is averagingAs most Whole IMF components, ck(t) the k rank IMF components after EEMD is decomposed, obtained to x (t) are represented.
Specifically, step S2-2 includes：
Step a：For the IMF components c that length is Li(t), counting the method that its zero passage is counted out is：As 1≤j≤L, If c (j) * c (j-1)<0, then a zero crossing is designated as, traversal [1, L] finds out IMF components ci(t) all zero crossings, that is, obtain ci(t) zero passage is counted out, thus it is possible to obtain preceding 5 rank IMF zero passage points IMF1-ZCs, IMF2-ZCs, IMF3-ZCs, IMF4-ZCs、IMF5-ZCs；
Step b：According to formula5 rank IMF variance before calculating IMF1-Var, IMF2-Var, IMF3-Var, IMF4-Var, IMF5-Var, according still further to formula5 rank IMF variance contribution ratio IMF1-VarR, IMF2-VarR, IMF3- before calculating VarR、IMF4-VarR、IMF5-VarR。
Preferably, in step S2-4, for time series x (t) n ranks IMF by step S2-3 step a handle with Afterwards, the magnitude function a that n length is L can be obtainedi(t), energy function ei(t), (i=0,1,2 ..., n-1), according to formula5 rank IMF mean intensity before calculating, further according to formulaBefore calculating 5 rank IMF energy.
Preferably, in step S2-5, for time series x (t) n ranks IMF by step S2-3 step a handle with Afterwards, the instantaneous frequency function f that n length is L can be obtainedi(t), (i=0,1,2 ..., n-1), according to formula5 rank IMF highest frequency before calculating.
Preferably, in step S2-6, step S2-3 step a~c processing is passed through for time series x (t) n ranks IMF After, the n Hilbert marginal spectrums h for including m frequencies points can be obtainedi(f), and x (t) counts comprising m frequency Hilbert marginal spectrum h (f), respectively according to formulaWith Wherein hi(fj) represent i-th of IMF component Hilbert marginal spectrums j-th of Frequency point amplitude, fjRepresent j-th of frequency Point, 5 rank IMF mean center frequency and x (t) mean center frequency before calculating.
Preferably, step S2-7 includes step：
Step a：According to Shannon entropy formulaCalculate x (t) Hilbert marginal spectrums EntropyWherein
Step b：X (t) marginal spectrum h (f) is normalized, obtained
Step c：By ho(f) it is divided into 4 frequency bands：Low-frequency rangeMiddle low-frequency rangeMedium-high frequency sectionAnd high bandRespectively according to formula WithCalculate x (t) normalization Hilbert limits Compose low-frequency range, middle low-frequency range, medium-high frequency section and high band energy.
Preferably, normalization Hilbert marginal spectrum h of the least square method to x (t) is used in step S2-8o(f) carry out straight Line is fitted to obtain O-MLFSlope and O-MLFInterp, carries out an exponent of polynomial function curve and is fitted to obtain O- MEFSlope and O-MEFInterp, carry out six rank curve matchings and obtain O-MSOFa0, O-MSOFa1, O-MSOFa2, O- MSOFa3、O-MSOFa4、O-MSOFa5、O-MSOFa6。
Preferably, step S2-9 is specifically included：
Step a：For time series x (t) n ranks IMF by step S2-3 step a~c processing after, can obtain X (t) Hilbert amplitude spectrum H (f, t), using time t as transverse axis, frequency f is the longitudinal axis, controls gray scale using amplitude, obtains the time Sequence x (t) Hilbert time spectrum frequency gray level images；
Step b：It is every on image using the gray level of image as transverse axis again by x (t) Hilbert time spectrum frequency gray level images The number that one gray level occurs obtains time series x (t) Hilbert time spectrum frequency gray level image histograms h as the longitudinal axisi, Wherein i=0,1,2 ..., l-1.hiThe number that i-th of gray-level pixels of grey level histogram occur is represented, l represents time-frequency gray scale The number of greyscale levels of image；
Step c：Formula is utilized respectively again WithCalculate x's (t) Hilbert composes average, variance, the skewness and kurtosis of gray-scale statistical histogram.
Preferably, in step S3, rankfeatures algorithms use MATLAB softwares tool box function, the algorithm according to Classification separation criterion is ranked up to key feature, specifically uses function
[IDX, Z]=rankfeatures (X, Group),
Feature is ranked up using the independence interpretational criteria of binary classification, input X is a matrix, and each row represent One observation vector and line number corresponds to original characteristic；Input Group includes class label, and output IDX is one by X The list that row subscript is formed, subscript list arrange from high to low according to the conspicuousness of feature in matrix X, and output Z is usage criteria Afterwards, weight corresponding to each feature.
Preferably, sample characteristics fusion index is calculated using rankfeatures algorithms in step S3, specific step is：
S3-1：The characteristic matrix that all sampling feature vectors are formed is assigned to X, all sample labels are formed Vector assignment is input to [IDX, Z]=rankfeatures (X, Group) function and handled, obtain including 56 to Group The feature weight vector Z of component；
S3-2：M larger feature of selection weight is weighted summation, obtains sample characteristics fusion index.
The present invention compared with prior art, has the following advantages that and beneficial effect：
1st, the present invention handles ultrasonic RF time serieses using Hilbert-Huang transform, from time, frequency and energy relationship Angle extraction feature, the taken HHT of invention is the Time-Frequency Analysis Method more suitable for non-stationary, Analysis of nonlinear signals, The characteristic information amount of extraction is more rich.
2nd, the present invention carries out Feature Selection using rankfeatures algorithms, calculates the weight of each feature, selects weight Larger feature is weighted summation, construction feature fusion index, and Fusion Features index proposed by the invention can reflect not Difference between generic tissue samples.
3rd, the present invention is based on ultrasonic RF time serieses, because the signal of ultrasonic RF time serieses is derived from same position, reduces The error that ultrasonic wave propagation path difference is brought, and depth attenuation's compensation is not needed, so as to improve determine sign precision；To tissue Whether particular/special requirement is uniformly had no, and application is more extensive.
4th, the ultrasonic RF time serieses that the present invention is based on can obtain in conventional ultrasound diagnosis, will not produce extra set Standby hardware costs.
Brief description of the drawings
Fig. 1 is the flow chart of the method for embodiment 2；
Fig. 2 is the control group of embodiment 2 and treatment group's tumor of breast region pathology figure：Fig. 2 (a) control groups；Fig. 2 (b) is treated Group；
Fig. 3 is the control group of embodiment 2 and treatment group's tumor of breast region Type B figure：Fig. 3 (a) control groups；Fig. 3 (b) is treated Group；
Fig. 4 is the rat mammary gland RF time series distribution maps of embodiment 2.
Embodiment
With reference to embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are unlimited In this.
Embodiment 1
A kind of feature extracting method of the Ultrasonic tissue characterization based on Hilbert-Huang transform, comprises the following steps：
S1：The ultrasonic echo RF signal multiframes of collection tissue samples, selection area-of-interest (Region OfInterest, Abbreviation ROI), construct ultrasonic RF time serieses；
S2：Using HHT algorithms, feature extraction is carried out to ultrasonic RF time serieses, obtains closing based on time, frequency and energy The sampling feature vectors of system；
S3：Using rankfeatures algorithms, Feature Selection and Fusion Features are carried out, calculate sample characteristics fusion index, Using this feature fusion index as the feature for levying tissue surely.
In the step S1, the detailed process for constructing ultrasonic RF time serieses is：
S1-1：Tissue is scanned, obtains its ultrasonic echo RF signal multiframes；
S1-2：Demodulate certain frame ultrasonic echo RF signal and show Type B figure；
S1-3：The ROI that size is M × N is chosen on Type B figure；
S1-4：Its preceding L frames RF signal is taken to construct the ultrasonic RF time serieses that M × N number of length is L to every in ROI.
In the step S2, using HHT algorithms, the method for calculating sampling feature vectors is：
S2-1：A time series x (t) in this ROI is sampled, enters row set empirical mode decomposition (Ensemble Empirical Mode Decomposition, abbreviation EEMD), the individual IMF components of n (n >=5) are obtained, specific process is：
Step a：Gaussian sequence is added into x (t), obtains X (t)=x (t)+ω (t)；
Step b：X (t) all local maximums and local minimum are determined, using cubic spline interpolation, are obtained upper and lower Envelope bmaxAnd b (t)min(t) average value, is calculatedObtain h1(t)=X (t)-m1(t)；
Step c：If h1(t) intrinsic mode functions (IntrinsicModeFunction, abbreviation IMF) condition is met, then h1(t) be exactly X (t) the first IMF components；Otherwise repeat step b, h1(t) as new X (t), m is calculated11(t), then judge h11(t)=h1(t)-m11(t) whether meet IMF conditions, be such as unsatisfactory for, circulate k times, until h1k(t)=h1(k-1)(t)-m1k(t) Meet IMF conditions.Remember c1(t)=h1k(t) the single order IMF components for being X (t), now, residual components r1(t)=X (t)-c1(t)；
Step d：By r1(t) as new X (t), repeat step a~c, obtain X (t) second meets IMF conditions Component c2(t), circulation performs step a~c, until rn(t) monotonic function is turned into, circulation terminates, so as to obtain signal X (t) N (n >=5) rank IMF components set；
Step e：Step a~d repeats n times, adds different Gaussian sequence ω every timei(t), decomposition obtains The set of N group IMF components, then in N group IMF components, the component of corresponding exponent number is averagingAs final IMF components, ck(t) the k rank IMF components after EEMD is decomposed, obtained to x (t) are represented.
S2-2：Extract temporal signatures (being based on IMF components)：IMF zero passages points (5) IMF1-ZCs, IMF2-ZCs, IMF3-ZCs, IMF4-ZCs, IMF5-ZCs, IMF variances (5) IMF1-Var, IMF2-Var, IMF3-Var, IMF4-Var, IMF5-Var, IMF variance contribution ratio (5) IMF1-VarR, IMF2-VarR, IMF3-VarR, IMF4-VarR, IMF5-VarR, Detailed process is：
Step a：For the IMF components c that length is Li(t), counting the method that its zero passage is counted out is：As 1≤j≤L, If c (j) * c (j-1)<0, then a zero crossing is designated as, traversal [1, L] finds out IMF components ci(t) all zero crossings, that is, obtain ci(t) zero passage is counted out, thus it is possible to obtain preceding 5 rank IMF zero passage points IMF1-ZCs, IMF2-ZCs, IMF3-ZCs, IMF4-ZCs、IMF5-ZCs；
Step b：According to formula5 rank IMF variance IMF1- before calculating Var, IMF2-Var, IMF3-Var, IMF4-Var, IMF5-Var, according still further to formulaMeter 5 rank IMF variance contribution ratio IMF1-VarR, IMF2-VarR, IMF3-VarR, IMF4-VarR, IMF5-VarR before calculation；
S2-3：To time series x (t) n rank IMF, Hilbert spectral analysis is carried out respectively (HilbertSpectrumAnalysis, abbreviation HSA), detailed process is as follows：
Step a：Hilbert conversion is carried out to an IMF component c (t)P is Cauchy master in formula Value, then seek c (t) analytic signalIt is expressed as polar form z (t)=a (t) eiθ(t), wherein amplitude FunctionEnergy functionPhase functionBy Phase function can calculate c (t) instantaneous frequency
Step b：Hilbert conversion is carried out according to step a respectively to x (t) n ranks IMF, tectonic knot signal, is expressed as Polar form, x (t) is finally reconstructed, is obtainedIt is omitted here Residual components r (t), Re represent to take real part；
Step c：ClaimFor Hilbert amplitude spectrums, abbreviation Hilbert spectrums, utilize Hilbert is composed to time integral, obtains Hilbert marginal spectrumsFormula InRepresent kth rank IMF components ck(t) Hilbert marginal spectrums.
S2-4：Extract time domain-energy feature (based on magnitude function and energy function)：IMF mean intensities (5) IMF1- AvgA, IMF2-AvgA, IMF3-AvgA, IMF4-AvgA, IMF5-AvgA, IMF energy (5) IMF1-Egy, IMF2-Egy, IMF3-Egy, IMF4-Egy, IMF5-Egy, specific method are：
For time series x (t) n ranks IMF by step S2-3 step a processing after, can obtain n length is L magnitude function ai(t), energy function ei(t), (i=0,1,2 ..., n-1), according to formula 5 rank IMF mean intensity before calculating, further according to formula5 rank IMF energy before calculating.
S2-5：Extract time domain-frequecy characteristic (being based on instantaneous frequency)：IMF highest frequencies (5) IMF1-MaxF, IMF2- MaxF, IMF3-MaxF, IMF4-MaxF, IMF5-MaxF, specific method are：
For time series x (t) n ranks IMF by step S2-3 step a processing after, it is L that can obtain n length Instantaneous frequency function fi(t), (i=0,1,2 ..., n-1), according to formula 5 rank IMF highest frequency before calculating.
S2-6：Extract energy-frequecy characteristic (being based on Hilbert marginal spectrums)：IMF mean centers frequency (5) IMF1- MCF, IMF2-MCF, IMF3-MCF, IMF4-MCF, IMF5-MCF, x (t) mean centers frequency (1) Orig-MCF, specific side Method is：
For time series x (t) n ranks IMF by step S2-3 step a~c processing after, can obtain n wrap The Hilbert marginal spectrums h of the points of frequency containing mi(f), and x (t) the Hilbert marginal spectrum h (f) to be counted comprising m frequency, Respectively according to formulaWithWherein hi(fj) represent The amplitude of j-th of Frequency point of the Hilbert marginal spectrums of i-th of IMF component, fjJ-th of Frequency point is represented, 5 rank IMF before calculating Mean center frequency and x (t) mean center frequency.
S2-7：Extract frequency domain-energy feature (being based on Hilbert marginal spectrums)：X (t) Hilbert marginal spectrums entropy (1) Orig-EgyS, x (t) normalization Hilbert marginal spectrum low-frequency ranges energy, middle low-frequency range energy, medium-high frequency section energy, high frequency Duan Nengliang (4) Orig-MargS1, Orig-MargS2, Orig-MargS3, Orig-MargS4, is comprised the following steps that：
Step a：According to Shannon entropy formulaCalculate x (t) Hilbert marginal spectrums EntropyWherein
Step b：X (t) marginal spectrum h (f) is normalized, obtained
Step c：By ho(f) it is divided into 4 frequency bands：Low-frequency rangeMiddle low-frequency rangeMedium-high frequency sectionAnd high bandRespectively according to formula WithCalculate x (t) normalization Hilbert limits Compose low-frequency range, middle low-frequency range, medium-high frequency section and high band energy.
S2-8：Extract frequency domain-energy curve fit characteristic (being based on Hilbert marginal spectrums)：X (t) normalization Hilbert Marginal spectrum fitting a straight line slope, intercept (2) O-MLFSlope, O-MLFInterp, an exponent of polynomial function curve fitting Slope, intercept (2) O-MEFSlope, O-MEFInterp, six rank curve matching coefficient (7) O-MSOFa0, O-MSOFa1, O-MSOFa2, O-MSOFa3, O-MSOFa4, O-MSOFa5, O-MSOFa6, specific method are：
Using normalization Hilbert marginal spectrum h of the least square method to x (t)o(f) carry out fitting a straight line and obtain O- MLFSlope and O-MLFInterp, carry out an exponent of polynomial function curve and be fitted to obtain O-MEFSlope and O- MEFInterp, carry out six rank curve matchings and obtain O-MSOFa0, O-MSOFa1, O-MSOFa2, O-MSOFa3, O-MSOFa4, O- MSOFa5、O-MSOFa6。
S2-9：Extract time-frequency domain-energy feature (being composed based on Hilbert)：X (t) Hilbert spectrum gray-scale statisticals are straight The statistical nature of square figure --- average, variance, the degree of bias, kurtosis (4) TFImgMean, TFImgSD, TFImgSkew, TFImgKurto, comprise the following steps that：
Step a：For time series x (t) n ranks IMF by step S2-3 step a~c processing after, can obtain X (t) Hilbert amplitude spectrum H (f, t), using time t as transverse axis, frequency f is the longitudinal axis, controls gray scale using amplitude, obtains the time Sequence x (t) Hilbert time spectrum frequency gray level images；
Step b：It is every on image using the gray level of image as transverse axis again by x (t) Hilbert time spectrum frequency gray level images The number that one gray level occurs obtains time series x (t) Hilbert time spectrum frequency gray level image histograms h as the longitudinal axisi, Wherein i=0,1,2 ..., l-1.hiThe number that i-th of gray-level pixels of grey level histogram occur is represented, l represents time-frequency gray scale The number of greyscale levels of image；
Step c：Formula is utilized respectively again WithCalculate x's (t) Hilbert composes average, variance, the skewness and kurtosis of gray-scale statistical histogram.
S2-10：M × N bar time serieses x in separately sampled ROIi(t), wherein i=0,1,2 ..., M × N-1, weight Step S2-1~S2-9 is performed again, obtains M × N number of characteristic vector corresponding to M × N bar time serieses, finally, then by this M × N number of Characteristic vector is averaging to obtain a characteristic vector, i.e. sampling feature vectors.
In the step S3, rankfeatures algorithms use the tool box function of MATLAB softwares, and the algorithm is according to class Not Fen Li criterion key feature is ranked up, specifically use function
[IDX, Z]=rankfeatures (X, Group),
Feature is ranked up using the independence interpretational criteria of binary classification, input X is a matrix, and each row represent One observation vector and line number corresponds to original characteristic；Input Group includes class label, and output IDX is one by X The list that row subscript is formed, subscript list arrange from high to low according to the conspicuousness of feature in matrix X, and output Z is usage criteria Afterwards, weight corresponding to each feature.Acquiescence uses joint variance evaluation double sample t test criterions.
Sample characteristics fusion index is calculated using rankfeatures algorithms, specific step is：
S3-1：The characteristic matrix that all sampling feature vectors are formed is assigned to X, all sample labels are formed Vector assignment is input to [IDX, Z]=rankfeatures (X, Group) function and handled, obtain including 56 to Group The feature weight vector Z of component；
S3-2：Select the larger m (m of weight<56) individual feature is weighted summation, obtains sample characteristics fusion index.
Embodiment 2
The present embodiment in addition to following characteristics other structures with embodiment 1：
As shown in figure 1, the present embodiment is a kind of feature extraction side of the Ultrasonic tissue characterization based on Hilbert-Huang transform Method is applied to the detection of breast cancer tissue's chemotherapy tissue microstructure change, comprises the following steps：
Step 1：Construct ultrasonic RF time serieses
The Sonix TOUCH and centre frequency 6.6MHz wideband linear array produced using Canadian Ultrasonix companies is surpassed Sonic probe scans the thoracic cavity hypodermis of female BAl BIc/C nude mices, records multiple frames of ultrasonic echo RF signals.This experiment uses 27 4 To 6 week old female BAl BIc/C nude mices as experimental subjects.Human breast cancer cell line Bcap-37 is cultivated first, added 10% hyclone and 100 units/ml ask streptomysin, are placed in the environment that temperature is 37 DEG C, CO2 concentration is 5%, then in nothing Will about 4*10 under the conditions of bacterium8The thoracic cavity that individual MCF-7 cells are inoculated into nude mice is subcutaneous.In medical angle, these nude mice models can Researched and analysed as indiscriminate sample.Breast cancer animal model is divided into treatment when tumour growth is to diameter about 5mm Group and control group, quantity are respectively 14 and 13, and treatment group nude mice carries out chemotherapy with 3mg/kg taxol daily, and compares Group is only given physiological saline.Fig. 2 is control group and treatment group's tumor of breast region pathology figure, it is observed that treatment group tumors are thin The change for the tissue microstructures such as born of the same parents' density substantially reduces, the gap between cell increases；
S101：Scanning experiment nude mice breast tumor tissues, obtain its ultrasonic echo RF signals；
S102：Choose the demodulation of the 100th frame ultrasonic echo RF signals and show Type B figure, Fig. 3 is that control group and treatment group's mammary gland swell Knurl region Type B figure, contrasts two width Type B figures, the image difference unobvious of tumor region, and None- identified tissue microstructure changes Become；
S103：The ROI of 20 × 70 sizes is selected on Type B figure；
S104：Its preceding 256 frame echo-signal is taken to each point in ROI, that is, the length for obtaining 20 × 70 points is 256 Ultrasonic RF time serieses, Fig. 4 are ultrasonic RF time serieses distribution maps；
Step 2：Utilize HHT algorithms, experiment with computing nude mice sampling feature vectors
According to the method for calculating sampling feature vectors, each experiment nude mice sample ROI time series progress feature is carried Take, can obtain including the sampling feature vectors of 56 components；
Step 3：Using rankfeatures algorithms, nude mice sample characteristics fusion index is calculated
S301：27 nude mices are calculated with sampling feature vectors respectively, and by the transposition square of 27 × 56 characteristic matrix Battle array is assigned to X, and the vector assignment that 27 sample labels are formed utilizes formula [IDX, Z]=rankfeatures to Group (X, Group), the weight vectors of 56 features are calculated；
S302：16 larger features of weight are weighted summation before selection, by calculating, in the present embodiment, and control group The Fusion Features index of 14 samples is respectively：0.8334,1.2672,0.9598,16.4901,9.1725,1.0033, 3.0090,1.2001,0.8029,1.6267,1.2665,2.0368,1.3963,1.3334；The feature of 13 samples for the treatment of group Fusion index is respectively：1.6160,1.5381,0.7148,0.7618,0.7112,1.0661,0.8636,1.5183, 0.9405,1.0340,0.8058,0.8384,0.6947.
The average of the Fusion Features index of control group is 3.0284, variance 19.6270；The Fusion Features index for the treatment of group Average be 1.0079, variance 0.1119.
It can thus be seen that the average of control group Fusion Features index is more than the average for the treatment of group's Fusion Features index, it is right It is more than the variance for the treatment of group's Fusion Features index according to group variance of Fusion Features index, the result shows the treatment Jing Guo chemotherapy The Fusion Features index of group sample is less than control group sample, and fluctuates smaller.Illustrate that Fusion Features index can be used for distinguishing The tissue that micro-structural changes, thus can be as the feature of tissue characterization.
Above-described embodiment is the preferable embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment Limitation, other any Spirit Essences without departing from the present invention with made under principle change, modification, replacement, combine, simplification, Equivalent substitute mode is should be, is included within protection scope of the present invention.

Claims (10)

1. the feature extracting method of the Ultrasonic tissue characterization based on Hilbert-Huang transform, it is characterised in that including following step Suddenly：
S1：The ultrasonic echo RF signal multiframes of tissue samples are gathered, ROI is selected, constructs ultrasonic RF time serieses；
S2：Using HHT algorithms, feature extraction is carried out to ultrasonic RF time serieses, obtained based on time, frequency and energy relationship Sampling feature vectors；
S3：Using rankfeatures algorithms, Feature Selection and Fusion Features are carried out, sample characteristics fusion index are calculated, this Fusion Features index is as the feature for levying tissue surely.
2. feature extracting method according to claim 1, it is characterised in that in step S1, construct ultrasonic RF time serieses Detailed process be：
S1-1：Tissue is scanned, obtains its ultrasonic echo RF signal multiframes；
S1-2：Demodulate certain frame ultrasonic echo RF signal and show Type B figure；
S1-3：The ROI that size is M × N is chosen on Type B figure；
S1-4：Its preceding L frames RF signal is taken to construct the ultrasonic RF time serieses that M × N number of length is L to every in ROI.
3. feature extracting method according to claim 2, it is characterised in that in step S2, using HHT algorithms, calculate sample Eigen vector method be：
S2-1：A time series x (t) in this ROI is sampled, enters row set empirical mode decomposition, obtains n IMF component；
S2-2：Based on IMF component extraction temporal signatures：
IMF zero passages are counted 5：IMF1-ZCs、IMF2-ZCs、IMF3-ZCs、IMF4-ZCs、IMF5-ZCs；
IMF variances 5：IMF1-Var、IMF2-Var、IMF3-Var、IMF4-Var、IMF5-Var；
IMF variance contribution ratios 5 IMF1-VarR, IMF2-VarR, IMF3-VarR, IMF4-VarR, IMF5-VarR；
S2-3：To time series x (t) n rank IMF, Hilbert spectral analysis is carried out respectively；
Specifically, step S2-3 includes：
Step a：Hilbert conversion is carried out to an IMF component c (t)P is Cauchy's principal value in formula, so C (t) analytic signal is sought afterwardsIt is expressed as polar form z (t)=a (t) eiθ(t), wherein amplitude letter NumberEnergy functionPhase functionBy Phase function can calculate c (t) instantaneous frequency
Step b：Hilbert conversion is carried out according to step a respectively to x (t) n ranks IMF, tectonic knot signal, represents that poling is sat Mark form, x (t) is finally reconstructed, is obtainedIt is omitted here residue Component r (t), Re represent to take real part；
Step c：ClaimFor Hilbert amplitude spectrums, abbreviation Hilbert spectrums, Hilbert is utilized Spectrum obtains Hilbert marginal spectrums to time integralIn formulaRepresent kth rank IMF components ck(t) Hilbert marginal spectrums；
S2-4：Based on magnitude function and energy function extraction time domain-energy feature：
5 IMF mean intensities：IMF1-AvgA、IMF2-AvgA、IMF3-AvgA、IMF4-AvgA、IMF5-AvgA；
5 IMF energy：IMF1-Egy、IMF2-Egy、IMF3-Egy、IMF4-Egy、IMF5-Egy；
S2-5：Based on instantaneous frequency distilling time domain-frequecy characteristic：
5 IMF highest frequencies：IMF1-MaxF、IMF2-MaxF、IMF3-MaxF、IMF4-MaxF、IMF5-MaxF；
S2-6：Energy-frequecy characteristic is extracted based on Hilbert marginal spectrums：
5 IMF mean center frequencies：IMF1-MCF、IMF2-MCF、IMF3-MCF、IMF4-MCF、IMF5-MCF；1 x (t) is flat Equal centre frequency Orig-MCF；
S2-7：Frequency domain-energy feature is extracted based on Hilbert marginal spectrums：
X (t) Hilbert marginal spectrum entropys Orig-EgyS；
X (t) normalization Hilbert marginal spectrum low-frequency ranges energy, middle low-frequency range energy, medium-high frequency section energy, high band energy Orig-MargS1、Orig-MargS2、Orig-MargS3、Orig-MargS4；
S2-8：Frequency domain-energy curve fit characteristic is extracted based on Hilbert marginal spectrums：
X (t) normalization Hilbert marginal spectrum fitting a straight lines slope, intercept：O-MLFSlope、O-MLFInterp；
Exponent of polynomial function curve fit slope, an intercept：O-MEFSlope、O-MEFInterp；
7 six rank curve matching coefficients：O-MSOFa0、O-MSOFa1、O-MSOFa2、O-MSOFa3、O-MSOFa4、O- MSOFa5、O-MSOFa6；
S2-9：Extraction time-frequency domain-energy feature is composed based on Hilbert：
Statistical nature --- average TFImgMean, variance TFImgSD, the degree of bias of x (t) Hilbert spectrum gray-scale statistical histograms TFImgSkew, kurtosis TFImgKurto；
S2-10：M × N bar time serieses x in separately sampled ROIi(t), wherein i=0,1,2 ..., M × N-1, repetition are held Row step S2-1~S2-9, M × N number of characteristic vector corresponding to M × N bar time serieses is obtained, finally, then by this M × N number of feature Vector is averaging and obtains a characteristic vector, i.e. sampling feature vectors.
4. feature extracting method according to claim 3, it is characterised in that obtain n IMF component in step S2-1, n >= 5。
5. feature extracting method according to claim 3, it is characterised in that the specific processes of step S2-1 include：
Step a：Gaussian sequence is added into x (t), obtains X (t)=x (t)+ω (t)；
Step b：X (t) all local maximums and local minimum are determined, using cubic spline interpolation, obtain upper and lower envelope Line bmaxAnd b (t)min(t) average value, is calculatedObtain h1(t)=X (t)-m1(t)；
Step c：If h1(t) intrinsic mode functions IMF conditions are met, then h1(t) be exactly X (t) the first IMF components；Otherwise weigh Multiple step b, h1(t) as new X (t), m is calculated11(t), then h is judged11(t)=h1(t)-m11(t) IMF bars whether are met Part, such as it is unsatisfactory for, circulates k times, until h1k(t)=h1(k-1)(t)-m1k(t) IMF conditions are met；Remember c1(t)=h1k(t) it is X (t) single order IMF components, now, residual components r1(t)=X (t)-c1(t)；
Step d：By r1(t) the component c for meeting IMF conditions as new X (t), repeat step a~c, obtain X (t) second2 (t), circulation performs step a~c, until rn(t) monotonic function is turned into, circulation terminates, so as to obtain signal X (t) n (n >=5) set of rank IMF components；
Step e：Step a~d repeats n times, adds different Gaussian sequence ω every timei(t), decompose and obtain N groups The set of IMF components, then in N group IMF components, the component of corresponding exponent number is averagingAs final IMF components, ck(t) the k rank IMF components after EEMD is decomposed, obtained to x (t) are represented.
6. feature extracting method according to claim 3, it is characterised in that step S2-2 includes：
Step a：For the IMF components c that length is Li(t), counting the method that its zero passage is counted out is：As 1≤j≤L, if c (j)*c(j-1)<0, then a zero crossing is designated as, traversal [1, L] finds out IMF components ci(t) all zero crossings, that is, obtain ci (t) zero passage is counted out, thus it is possible to obtain preceding 5 rank IMF zero passage points IMF1-ZCs, IMF2-ZCs, IMF3-ZCs, IMF4-ZCs、IMF5-ZCs；
Step b：According to formula5 rank IMF variance IMF1- before calculating Var, IMF2-Var, IMF3-Var, IMF4-Var, IMF5-Var, according still further to formulaMeter 5 rank IMF variance contribution ratio IMF1-VarR, IMF2-VarR, IMF3-VarR, IMF4-VarR, IMF5-VarR before calculation.
7. feature extracting method according to claim 3, it is characterised in that step S2-4 specifically has to step S2-8：
In step S2-4, for time series x (t) n ranks IMF by step S2-3 step a processing after, n can be obtained Individual length is L magnitude function ai(t), energy function ei(t), (i=0,1,2 ..., n-1), according to formula5 rank IMF mean intensity before calculating, further according to formulaBefore calculating 5 rank IMF energy；
In step S2-5, for time series x (t) n ranks IMF by step S2-3 step a processing after, n can be obtained Individual length is L instantaneous frequency function fi(t), (i=0,1,2 ..., n-1), according to formula5 rank IMF highest frequency before calculating；
In step S2-6, for time series x (t) n ranks IMF by step S2-3 step a~c processing after, can obtain The Hilbert marginal spectrums h for including m frequencies points to ni(f), and x (t) the Hilbert comprising m frequency points it is marginal H (f) is composed, respectively according to formulaWithWherein hi (fj) represent i-th of IMF component Hilbert marginal spectrums j-th of Frequency point amplitude, fjJ-th of Frequency point is represented, is calculated Preceding 5 rank IMF mean center frequency and x (t) mean center frequency；
Step S2-7 includes step：
Step a：According to Shannon entropy formulaCalculate x (t) Hilbert marginal spectrum entropysWherein
Step b：X (t) marginal spectrum h (f) is normalized, obtained
Step c：By ho(f) it is divided into 4 frequency bands：Low-frequency rangeMiddle low-frequency rangeMedium-high frequency sectionAnd high bandRespectively according to formula WithCalculate x (t) normalization Hilbert limits Compose low-frequency range, middle low-frequency range, medium-high frequency section and high band energy；
Normalization Hilbert marginal spectrum h of the least square method to x (t) is used in step S2-8o(f) carry out fitting a straight line and obtain O- MLFSlope and O-MLFInterp, carry out an exponent of polynomial function curve and be fitted to obtain O-MEFSlope and O- MEFInterp, carry out six rank curve matchings and obtain O-MSOFa0, O-MSOFa1, O-MSOFa2, O-MSOFa3, O-MSOFa4, O- MSOFa5、O-MSOFa6。
8. feature extracting method according to claim 3, it is characterised in that step S2-9 is specifically included：
Step a：For time series x (t) n ranks IMF by step S2-3 step a~c processing after, x (t) can be obtained Hilbert amplitude spectrum H (f, t), using time t as transverse axis, frequency f is the longitudinal axis, using amplitude control gray scale, obtain time series X (t) Hilbert time spectrum frequency gray level images；
Step b：Again by x (t) Hilbert time spectrum frequency gray level images, using the gray level of image as transverse axis, on image each The number that gray level occurs obtains time series x (t) Hilbert time spectrum frequency gray level image histograms h as the longitudinal axisi, wherein I=0,1,2 ..., l-1, hiThe number that i-th of gray-level pixels of grey level histogram occur is represented, l represents time-frequency gray level image Number of greyscale levels；
Step c：Formula is utilized respectively again WithCalculate x's (t) Hilbert composes average, variance, the skewness and kurtosis of gray-scale statistical histogram.
9. feature extracting method according to claim 1, it is characterised in that in step S3, rankfeatures algorithms are adopted With the tool box function of MATLAB softwares, the algorithm is ranked up to key feature according to classification separation criterion, specifically used Function
[IDX, Z]=rankfeatures (X, Group),
Feature is ranked up using the independence interpretational criteria of binary classification, input X is a matrix, and each row represent one Observation vector and line number corresponds to original characteristic；Input Group includes class label, and output IDX is one by under X row The list formed is marked, subscript list arranges from high to low according to the conspicuousness of feature in matrix X, after output Z is usage criteria, respectively Weight corresponding to individual feature.
10. feature extracting method according to claim 1, it is characterised in that calculated in step S3 using rankfeatures Method calculates sample characteristics fusion index, and specific step is：
S3-1：The characteristic matrix that all sampling feature vectors are formed is assigned to X, the vector that all sample labels are formed Group is assigned to, [IDX, Z]=rankfeatures (X, Group) function is input to and is handled, obtains including multiple components Feature weight vector Z；
S3-2：M larger feature of selection weight is weighted summation, obtains sample characteristics fusion index.
CN201710417075.3A 2017-06-06 2017-06-06 Feature extraction method for ultrasonic tissue characterization based on Hilbert-Huang transform Expired - Fee Related CN107358156B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710417075.3A CN107358156B (en) 2017-06-06 2017-06-06 Feature extraction method for ultrasonic tissue characterization based on Hilbert-Huang transform

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710417075.3A CN107358156B (en) 2017-06-06 2017-06-06 Feature extraction method for ultrasonic tissue characterization based on Hilbert-Huang transform

Publications (2)

Publication Number Publication Date
CN107358156A true CN107358156A (en) 2017-11-17
CN107358156B CN107358156B (en) 2020-05-19

Family

ID=60271035

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710417075.3A Expired - Fee Related CN107358156B (en) 2017-06-06 2017-06-06 Feature extraction method for ultrasonic tissue characterization based on Hilbert-Huang transform

Country Status (1)

Country Link
CN (1) CN107358156B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108614926A (en) * 2018-04-12 2018-10-02 西安交通大学 A kind of modal parameters discrimination method being combined with Hilbert-Huang transform based on manifold learning
CN109214092A (en) * 2018-09-11 2019-01-15 吉林大学 The four component borehole strain data exception extraction methods based on Hilbert-Huang transform
CN110032585A (en) * 2019-04-02 2019-07-19 北京科技大学 A kind of time series bilayer symbolism method and device
CN113191232A (en) * 2021-04-21 2021-07-30 西安交通大学 Electro-hydrostatic actuator fault identification method based on multi-mode homologous features and XGboost model
CN113205000A (en) * 2021-04-07 2021-08-03 江苏师范大学 Modulation mode identification method based on Cauchy Score polar coordinate diagram
CN114509604A (en) * 2022-04-18 2022-05-17 国网江西省电力有限公司电力科学研究院 GIS shell transient state ground potential rise waveform spectrum analysis method and system
CN116028882A (en) * 2023-03-29 2023-04-28 深圳市傲天科技股份有限公司 User labeling and classifying method, device, equipment and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102855623A (en) * 2012-07-19 2013-01-02 哈尔滨工业大学 Method for measuring myocardium ultrasonic angiography image physiological parameters based on empirical mode decomposition (EMD)
CN105030279A (en) * 2015-06-24 2015-11-11 华南理工大学 Ultrasonic RF (radio frequency) time sequence-based tissue characterization method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102855623A (en) * 2012-07-19 2013-01-02 哈尔滨工业大学 Method for measuring myocardium ultrasonic angiography image physiological parameters based on empirical mode decomposition (EMD)
CN105030279A (en) * 2015-06-24 2015-11-11 华南理工大学 Ultrasonic RF (radio frequency) time sequence-based tissue characterization method

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108614926A (en) * 2018-04-12 2018-10-02 西安交通大学 A kind of modal parameters discrimination method being combined with Hilbert-Huang transform based on manifold learning
CN109214092A (en) * 2018-09-11 2019-01-15 吉林大学 The four component borehole strain data exception extraction methods based on Hilbert-Huang transform
CN110032585A (en) * 2019-04-02 2019-07-19 北京科技大学 A kind of time series bilayer symbolism method and device
CN113205000A (en) * 2021-04-07 2021-08-03 江苏师范大学 Modulation mode identification method based on Cauchy Score polar coordinate diagram
CN113191232A (en) * 2021-04-21 2021-07-30 西安交通大学 Electro-hydrostatic actuator fault identification method based on multi-mode homologous features and XGboost model
CN114509604A (en) * 2022-04-18 2022-05-17 国网江西省电力有限公司电力科学研究院 GIS shell transient state ground potential rise waveform spectrum analysis method and system
CN116028882A (en) * 2023-03-29 2023-04-28 深圳市傲天科技股份有限公司 User labeling and classifying method, device, equipment and storage medium
CN116028882B (en) * 2023-03-29 2023-06-02 深圳市傲天科技股份有限公司 User labeling and classifying method, device, equipment and storage medium

Also Published As

Publication number Publication date
CN107358156B (en) 2020-05-19

Similar Documents

Publication Publication Date Title
CN107358156A (en) The feature extracting method of Ultrasonic tissue characterization based on Hilbert-Huang transform
Ren et al. Machine learning driven approach towards the quality assessment of fresh fruits using non-invasive sensing
CN105030279B (en) A kind of tissue characterization method based on ultrasonic radio frequency time series
Zhou et al. High spatial–temporal resolution reconstruction of plane-wave ultrasound images with a multichannel multiscale convolutional neural network
CN101667289B (en) Retinal image segmentation method based on NSCT feature extraction and supervised classification
CN109784242A (en) EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks
CN107346541B (en) Tissue characterization method based on ultrasonic radio frequency time series wavelet analysis
Besson et al. A sparse reconstruction framework for Fourier-based plane-wave imaging
CN104240203A (en) Medical ultrasound image denoising method based on wavelet transform and quick bilateral filtering
CN106446566A (en) Elderly cognitive function classification method based on random forest
CN109589114A (en) Myoelectricity noise-eliminating method based on CEEMD and interval threshold
Lu et al. Reconstruction for diverging-wave imaging using deep convolutional neural networks
Guan et al. Multifractal correlation characteristic for radar detecting low-observable target in sea clutter
CN103479398A (en) Method of detecting hepatic tissue microstructure based on ultrasonic radio frequency flow analysis
CN102012466A (en) Method for measuring noise of digital X-ray imaging system
CN106983524A (en) A kind of parameter and its measuring method for reflecting that biological tissue is abnormal
Ergün et al. Classification of carotid artery stenosis of patients with diabetes by neural network and logistic regression
CN109061591A (en) A kind of time-frequency line-spectrum detection method based on sequential cluster
CN109965905B (en) Contrast region detection imaging method based on deep learning
Lei et al. Fully complex-valued gated recurrent neural network for ultrasound imaging
Cheng et al. Quantitative evaluation of liver fibrosis based on ultrasound radio frequency signals: An animal experimental study
CN114492519A (en) Lung ultrasonic special sign B-line identification and classification method based on ultrasonic echo radio frequency signals
CN103099642B (en) Ultrasonic blood flow signal quality real-time analysis method
CN108986909A (en) Soft tissue elasticity and viscoelastic characterizations&#39; method and device based on ultrasonic elastograph imaging
Sultan et al. Estimation of Cortical Bone Strength Using CNN-based Regression Model

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200519