CN103637821A - Hepatic fibrosis degree Fisher identification method based on ultrasonic radio frequency (RF) time sequence - Google Patents

Abstract

The invention belongs to the technical field of ultrasonic medicine and particularly relates to a hepatic fibrosis degree Fisher identification method based on an ultrasonic radio frequency (RF) time sequence. The method includes that first, a broadband ultrasonic linear array probe is used for scanning a living hepatic tissue, and multi-frame ultrasonic echo RF signals are recorded; a certain frame of B type pattern is demodulated and displayed, a region of interest (ROI) is selected, all frame RF signals of each point in the ROI are taken to form the ultrasonic RF time sequence; SMR fractal dimension mean values and a plurality of characteristic parameters of the RF time sequence in the ROI are extracted, the 7 features are extracted for a large amount of normal and fibrosis hepatic samples, Fisher distinguishing rate of each feature parameter is collected and obtained, the normalization Fisher distinguishing rate is used as the weight of a new input sample corresponding feature parameter, and the absolute value of each feature parameter of a new sample is weighted and summed to obtain Score-fisher. The hepatic fibrosis degree Fisher identification method based on the ultrasonic RF time sequence is brought up for the first time and provides quantification reference for clinical hepatic fibrosis diagnosis and dynamic monitoring.

Description

Based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods

Technical field

The present invention relates to ultrasound medicine research field, particularly a kind of based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods.

Background technology

In Asia, hepatic fibrosis is a kind of common hepatic disease, and it is chronic hepatic injury to the key link of liver cirrhosis progress, the only stage which must be passed by of developing into liver cirrhosis for various chronic hepatopathys.Now think that the treatment of hepatic fibrosis process still has reverse to normal possibility, liver cirrhosis is no.But extent of liver fibrosis lacks quantitative, noninvasive diagnosis and monitoring means, also Estimating curative effect in time when treatment.In clinical medicine, liver biopsy is the golden standard of current liver fibrosis diagnosis, but liver biopsy is to have the inspection of wound property, should not repeat, and easily cause other complication.Ultrasonic tissue surely levy with its noinvasive, simple and easy, the advantage such as can repeat and on liver fibrosis diagnosis, be subject to extensive concern.

The Data Source that traditional hepatic tissue is levied surely mainly contains two kinds: ultrasonic B figure and ultrasonic echo radio frequency rf signal.Hepatic tissue based on ultrasonic B image is levied the textural characteristics of main extraction image surely, then with the classification accuracy of the graders such as SVM, neutral net, sensitivity, specificity etc., classifying quality is assessed.Ultrasonic B Tu Dingzheng exists following a few point defect: (1) adopts briliancy modulation to show amplitude information, has abandoned abundant phase information, makes portion of tissue information dropout; (2) check result is subject to the impact that instrument model, instrument arrange etc., and concordance is poor.Based on ultrasonic echo radio frequency rf signal, mainly to same frame back scattering echo RF signal, use ultrasonic spectrum parametric method to carry out tissue characterization, its specific practice is to adopt wideband ultrasonic probe scanning hepatic tissue to obtain ultrasonic echo radio frequency rf signal, demodulation shows a certain frame, and intercepting obtains region of interest ROI.Acoustic beam in region of interest is carried out to spectrum analysis one by one, extract spectrum signature parameter, find out the characteristic parameter that can effectively distinguish normal and pathological changes hepatic tissue.The method needs depth attenuation to compensate, and the damping capacity of organizing of different lesions degree is discrepant, and individual variation also can make acoustic wave propagation path different simultaneously, and then has a strong impact on its nicety of grading.

In above-mentioned ultrasonic tissue is levied surely, be to extract the feature that can distinguish normal structure and undesired tissue mostly, according to feature, distribute sample is classified, be only qualitative analysis, lack the quantitative analysis to degree of hepatic fibrosis.

Summary of the invention

Main purpose of the present invention is that the shortcoming that overcomes prior art is with not enough, provide a kind of based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods, the RF time series of obtaining in the conventional ultrasound diagnosis of the method utilization, carry out feature extraction, and utilize Fisher method of discrimination to carry out quantitative analysis to degree of hepatic fibrosis, measurable auxiliary information is provided to the identification of degree of hepatic fibrosis, does not rely on doctor's experience, recognition result is more objective.

Object of the present invention realizes by following technical scheme: based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods, comprise the following steps:

(1) training stage: choose a large amount of normal hepatocytes and fibrosis liver sample data proceeds as follows, obtain the weight of each characteristic parameter:

(1-1) scanning live body hepatic tissue, obtains multiframe ultrasonic echo radio frequency rf signal;

(1-2) the Type B figure of each frame ultrasonic echo radio frequency rf signal of difference demodulation;

(1-3) on each Type B figure, choose region of interest ROI;

(1-4) every bit in ROI is got to its all frame RF signals and form ultrasonic RF time series;

(1-5) extract ultrasonic RF seasonal effect in time series SMR fractal dimension average and several spectrum signature parameters, obtain characteristic vector;

(1-6), to a large amount of normal hepatocytes and the above-mentioned characteristic vector of fibrosis liver sample extraction, statistics obtains average and the variance of each characteristic parameter in normal hepatocytes and fibrosis liver characteristic vector;

(1-7) calculate the Fisher differentiation rate of each characteristic parameter, normalized then, obtains the weight of each characteristic parameter;

(2) cognitive phase: for new input sample to be identified, according to step (1-1)-(1-5), obtain characteristic vector, the weight of each characteristic parameter obtaining with step (1-7), the absolute value of each characteristic parameter of new samples is weighted to summation and obtains Fisher scoring, according to the score value of its scoring, realize the identification of degree of hepatic fibrosis.

Concrete, in described step (1-5), the method for extracting ultrasonic RF seasonal effect in time series SMR fractal dimension average is as follows:

(1-5-1-1) establish region of interest ROI size for M * N, the every bit in ROI is got to its all frame RF signals, thereby obtain M * N RF time series;

(1-5-1-2) normalization amplitude is:

$f_i^{\mathrm{nor}}=\frac{f_i^{\′}}{\frac{1}{N_t}\underset{j=1}{\overset{N_t}{\mathrm{\Σ}}}\left|f_j^{\′}\right|},f_i^{\′}=f_i-\frac{1}{N_t}\underset{j=1}{\overset{N_t}{\mathrm{\Σ}}}{f}_j,i=\mathrm{1,2},\·\·\·,N_t.;$

Wherein: f _ibe the amplitude that in a RF time series, i is ordered, N _ttotal number for RF time series sampled point;

(1-5-1-3) signal calculated total length L _s, ask for successively the difference of two adjacent normalization amplitudes, to absolute difference, summation obtains;

(1-5-1-4) ask for the arc distance R between amplitude maximum and amplitude smallest point;

(1-5-1-5) obtain SMR fractal dimension:

(1-5-1-6) calculate respectively each RF seasonal effect in time series SMR fractal dimension, then summation, then divided by M * N, obtains SMR fractal dimension average.

Preferably, in described step (1-5), extracting following 6 spectrum signature parameters, is respectively the spectrum energy sum S of four frequency ranges ₁, S ₂, S ₃, S ₄with spectrum slope S lope, spectrum intercept Intercept, calculation procedure is:

First each RF time series f (n) is converted to obtain to frequency spectrum F (k) as FFT, ask in every frame ROI institute's spectral magnitude a little with, then divided by M * N, obtain the spectrum mean F of every frame _ave(k), to F _ave(k) as normalized, obtain normalization frequency spectrum, that is:

$\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|=\left|F_{\mathrm{ave}}\left(k\right)\right|/\max \left(\left|F_{\mathrm{ave}}\left(k\right)\right|\right);$

Normalization frequency spectrum is carried out to the sane matching of straight line, obtain composing slope S lope, spectrum intercept Intercept;

According to formula below, ask for S ₁, S ₂, S ₃, S ₄:

${\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="HDA0000424137850000023.png"\; state="\{\{state\}\}">S</figure-callout>}}_1=\underset{k=1}{\overset{\frac{L}{8}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|,{\mathrm{<figure-callout\; id="2"\; label="S"\; filenames="HDA0000424137850000022.png"\; state="\{\{state\}\}">S</figure-callout>}}_2=\underset{k=\frac{L}{8}+1}{\overset{\frac{\mathrm{<figure-callout\; id="4"\; label="L"\; filenames="HDA0000424137850000022.png,HDA0000424137850000023.png"\; state="\{\{state\}\}">L</figure-callout>}}{4}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|$

${\mathrm{<figure-callout\; id="3"\; label="S"\; filenames="HDA0000424137850000023.png"\; state="\{\{state\}\}">S</figure-callout>}}_3=\underset{k=\frac{\mathrm{<figure-callout\; id="4"\; label="L"\; filenames="HDA0000424137850000022.png,HDA0000424137850000023.png"\; state="\{\{state\}\}">L</figure-callout>}}{4}+1}{\overset{\frac{3}{4}L}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|,{\mathrm{<figure-callout\; id="4"\; label="S"\; filenames="HDA0000424137850000022.png,HDA0000424137850000023.png"\; state="\{\{state\}\}">S</figure-callout>}}_4=\underset{k=\frac{3\mathrm{<figure-callout\; id="4"\; label="L"\; filenames="HDA0000424137850000022.png,HDA0000424137850000023.png"\; state="\{\{state\}\}">L</figure-callout>}}{4}+1}{\overset{\frac{\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="HDA0000424137850000022.png"\; state="\{\{state\}\}">L</figure-callout>}}{2}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|$

Wherein, L is RF time series totalframes.

Concrete, in described step (1-7), the step of calculating the Fisher differentiation rate of each characteristic parameter is:

$w_i=\frac{({\mathrm{\&mu;}}_{i-\mathrm{NOR}}-{\mathrm{\&mu;}}_{i-\mathrm{CIR}})}{{\mathrm{\&sigma;}}_{i-\mathrm{<figure-callout\; id="2"\; label="NOR"\; filenames="HDA0000424137850000022.png"\; state="\{\{state\}\}">NOR</figure-callout>}}^2+{\mathrm{\&sigma;}}_{i-\mathrm{<figure-callout\; id="2"\; label="CIR"\; filenames="HDA0000424137850000022.png"\; state="\{\{state\}\}">CIR</figure-callout>}}^2};$

Wherein, w _irepresent the Fisher differentiation rate of i characteristic parameter, μ _i-CIRfor i characteristic parameter average of fibrosis liver sample, μ _i-NORfor i characteristic parameter average of normal hepatocytes sample, σ _i-NORand σ _i-CIRbe respectively the standard deviation of i characteristic parameter of normal hepatocytes and fibrosis liver sample.

Further, the step in described step (7), the Fisher differentiation rate of each characteristic parameter being normalized is:

${\hat{w}}_i=\frac{w_i}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i}$

Wherein,

it is the weight of i characteristic parameter.

In described step (2), the step that obtains Fisher scoring is:

If the weight of i characteristic parameter is in the new samples extracting, i characteristic parameter is ε _i, Fisher scoring Score-fisher is:

$\mathrm{Score}-\mathrm{fisher}=\underset{i}{\mathrm{\&Sigma;}}{\hat{w}}_i\left|{\mathrm{\&epsiv;}}_i\right|.$

Compared with prior art, tool has the following advantages and beneficial effect in the present invention:

1, the present invention proposes to mark based on RF seasonal effect in time series degree of hepatic fibrosis first, based on scoring, degree of hepatic fibrosis is identified, can be on the basis of great amount of samples data, set up the corresponding relation of mark and degree of hepatic fibrosis, thereby provide quantitative reference for clinical liver fibrosis diagnosis and dynamic monitoring.

2, in the present invention when carrying out feature extraction based on ultrasonic RF time series, therefore because ultrasonic RF time series comes from the same degree of depth, do not need depth attenuation to compensate, thereby can improve, surely levy precision.

3, the present invention based on ultrasonic RF time series can in conventional ultrasound diagnosis, obtain, can not produce extra expenses.

4, the present invention proposes Fisher scoring does not need doctor to possess relevant background knowledge, and the result obtaining is more objective, more meets clinical practice reality.

Accompanying drawing explanation

Fig. 1 is the flow chart of the inventive method;

Fig. 2 is hepatic tissue Type B figure;

Fig. 3 is the ultrasonic RF time series chart of hepatic tissue region of interest ROI;

Fig. 4 is ultrasonic RF seasonal effect in time series normalization spectrogram.

The specific embodiment

Below in conjunction with embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited to this.

Embodiment 1

As shown in Figure 1, the present embodiment based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods details are as follows.

S1, build ultrasonic RF time series.

S1.1 is used the hepatic tissue region under the wideband linear array ultrasonic probe scanning liver peplos of Sonix TOUCH that Canadian Ultrasonix company produces and mid frequency 6.6MHz, record 128 frame ultrasonic echo radio frequency rf signals.

S1.2 carries out demodulation and shows its ultrasonic B figure (as shown in Figure 2) the 1st frame data.

It is 50 * 20 ROI that S1.3 chooses size on ultrasonic B figure, and intercepting obtains the ultrasonic echo radio frequency rf signal of 1000 points in ROI, gets its all frame data obtain 1000 ultrasonic RF time serieses to every in ROI, and sequence chart as shown in Figure 3.

S2, characteristic parameter extraction.

S2.1 the present embodiment has adopted 30 normal hepatocytes samples and 36 hepatic fibrosis samples.For 1000 in each sample ROI ultrasonic RF time serieses, calculate each seasonal effect in time series SMR fractal dimension, be then averaging the SMR fractal dimension average that obtains this sample.Detailed process is:

Normalization amplitude is:

$f_i^{\mathrm{nor}}=\frac{f_i^{\&prime;}}{\frac{1}{N_t}\underset{j=1}{\overset{N_t}{\mathrm{\&Sigma;}}}\left|f_j^{\&prime;}\right|},f_i^{\&prime;}=f_i-\frac{1}{N_t}\underset{j=1}{\overset{N_t}{\mathrm{\&Sigma;}}}{f}_j,i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,N_t.;$

Wherein: f _ibe the amplitude that in a RF time series, i is ordered, N _ttotal number for RF time series sampled point;

Signal calculated total length L _s, ask for successively the difference of two adjacent normalization amplitudes, to absolute difference, summation obtains signal total length L _s;

Ask for the arc distance R between amplitude maximum and amplitude smallest point;

Obtain SMR fractal dimension:

Calculate respectively each RF seasonal effect in time series SMR fractal dimension, then summation, then divided by M * N, obtains SMR fractal dimension average.

S2.2 is to 1000 of a sample ultrasonic RF time serieses, with length 1024, do FFT conversion respectively, then ask for the spectrum mean of every frame, divided by maximum spectrum average, obtain size [0,1] normalization frequency spectrum, use Matlab to carry out the sane matching of straight line to normalization frequency spectrum, obtain slope S lope, the intercept Intercept of this sample frequency spectrum.Meanwhile, calculate four frequency range spectrum energy sum S of this sample ₁, S ₂, S ₃, S ₄.Calculation procedure is:

First each RF time series f (n) is converted to obtain to frequency spectrum F (k) as FFT, ask in every frame ROI institute's spectral magnitude a little with, then divided by M * N, obtain the spectrum mean F of every frame _ave(k), to F _ave(k) as normalized, obtain normalization frequency spectrum, that is:

$\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|=\left|F_{\mathrm{ave}}\left(k\right)\right|/\max \left(\left|F_{\mathrm{ave}}\left(k\right)\right|\right);$

Spectrogram as shown in Figure 4, carries out the sane matching of straight line to normalization frequency spectrum, obtains composing slope S lope, spectrum intercept Intercept;

According to formula below, ask for S ₁, S ₂, S ₃, S ₄:

${\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="HDA0000424137850000023.png"\; state="\{\{state\}\}">S</figure-callout>}}_1=\underset{k=1}{\overset{\frac{L}{8}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|,{\mathrm{<figure-callout\; id="2"\; label="S"\; filenames="HDA0000424137850000022.png"\; state="\{\{state\}\}">S</figure-callout>}}_2=\underset{k=\frac{L}{8}+1}{\overset{\frac{\mathrm{<figure-callout\; id="4"\; label="L"\; filenames="HDA0000424137850000022.png,HDA0000424137850000023.png"\; state="\{\{state\}\}">L</figure-callout>}}{4}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|$

${\mathrm{<figure-callout\; id="3"\; label="S"\; filenames="HDA0000424137850000023.png"\; state="\{\{state\}\}">S</figure-callout>}}_3=\underset{k=\frac{\mathrm{<figure-callout\; id="4"\; label="L"\; filenames="HDA0000424137850000022.png,HDA0000424137850000023.png"\; state="\{\{state\}\}">L</figure-callout>}}{4}+1}{\overset{\frac{3}{4}L}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|,{\mathrm{<figure-callout\; id="4"\; label="S"\; filenames="HDA0000424137850000022.png,HDA0000424137850000023.png"\; state="\{\{state\}\}">S</figure-callout>}}_4=\underset{k=\frac{3\mathrm{<figure-callout\; id="4"\; label="L"\; filenames="HDA0000424137850000022.png,HDA0000424137850000023.png"\; state="\{\{state\}\}">L</figure-callout>}}{4}+1}{\overset{\frac{\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="HDA0000424137850000022.png"\; state="\{\{state\}\}">L</figure-callout>}}{2}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|$

Wherein, L is RF time series totalframes.

Thereby obtain based on ultrasonic RF seasonal effect in time series characteristic vector ε _i.

S3, the scoring of Fisher point system.

S3.1 is to above-mentioned normal hepatocytes sample and fibrosis liver sample extraction SMR fractal dimension average and 6 spectrum parameter attributes.Statistics obtains average and the variance of each characteristic parameter in normal hepatocytes sample

and average and the variance of each characteristic parameter in fibrosis liver sample

S3.2 calculates the Fisher differentiation rate of each characteristic parameter, and step is:

$w_i=\frac{|{\mathrm{\&mu;}}_{i-\mathrm{NOR}}-{\mathrm{\&mu;}}_{i-\mathrm{CIR}}|}{{\mathrm{\&sigma;}}_{i-\mathrm{NOR}}^2+{\mathrm{\&sigma;}}_{i-\mathrm{<figure-callout\; id="2"\; label="CIR"\; filenames="HDA0000424137850000022.png"\; state="\{\{state\}\}">CIR</figure-callout>}}^2}$

S3.3 is normalized the Fisher differentiation rate of each characteristic parameter, and step is:

${\hat{w}}_i=\frac{w_i}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i}$

Wherein,

it is the weight of i characteristic parameter.

The Fisher weights result of this example is as shown in table 1:

The weights of each characteristic parameter of table 1

S4 marks to new input sample.

S4.1, according to above-mentioned steps, first obtains its each characteristic parameter for new input sample calculation.

S4.2, according to the weight of each characteristic parameter obtaining, is weighted summation to the absolute value of each characteristic parameter of new samples and obtains Fisher scoring, realizes the identification of degree of hepatic fibrosis according to the score value of its scoring.It is to be negative value for fear of scoring that characteristic parameter is taken absolute value.

Such as a SMR fractal dimension FD=1.06, slope S lope=-0.1264, intercept Intercept=0.1287, S ₁=7.6986, S ₂=2.5977, S ₃=1.4557, S ₄=0.8832 normal hepatocytes sample, its Fisher point system scoring is:

Score-fisher=0.1632×|1.06|+0.2597×|-0.1264|+0.3359×|0.1287|+0.0394×|7.6986|

+0.0394×|2.5977|+0.0653×|1.4557|+0.0971×|0.8832|

=0.83553856

And for example, SMR fractal dimension FD=1.06, slope S lope=-0.0483, intercept Intercept=0.0498, S ₁=3.9795, S ₂=1.0305, S ₃=0.559, S ₄=0.3134 hepatic fibrosis sample, its Fisher point system scoring is:

Score-fisher=0.1632×|1.06|+0.2597×|-0.0483|+0.3359×|0.0498|+0.0394×|3.9795|

+0.0394×|1.0305|+0.0653×|0.559|+0.0971×|0.3134|

=0.46659117

In actual applications, can add up the Fisher score of great amount of samples data, thereby set up the corresponding relation of mark and degree of hepatic fibrosis, set up a classification chart, thereby can after calculating score, can directly obtain the degree of hepatic fibrosis of current sample to be identified, thereby avoid relying on the own experience of doctor, carry out subjective judgment, make recognition result more objective, more accurate.

Above-described embodiment is preferably embodiment of the present invention; but embodiments of the present invention are not restricted to the described embodiments; other any do not deviate from change, the modification done under spirit of the present invention and principle, substitutes, combination, simplify; all should be equivalent substitute mode, within being included in protection scope of the present invention.

Claims (6)

1. based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods, it is characterized in that, comprise the following steps:

(1) training stage: choose a large amount of normal hepatocytes and fibrosis liver sample data proceeds as follows, obtain the weight of each characteristic parameter:

(1-1) scanning live body hepatic tissue, obtains multiframe ultrasonic echo radio frequency rf signal;

(1-2) the Type B figure of each frame ultrasonic echo radio frequency rf signal of difference demodulation;

(1-3) on each Type B figure, choose region of interest ROI;

(1-4) every bit in ROI is got to its all frame RF signals and form ultrasonic RF time series;

(1-5) extract ultrasonic RF seasonal effect in time series SMR fractal dimension average and several spectrum signature parameters, obtain characteristic vector;

(1-6), to a large amount of normal hepatocytes and the above-mentioned characteristic vector of fibrosis liver sample extraction, statistics obtains average and the variance of each characteristic parameter in normal hepatocytes and fibrosis liver characteristic vector;

(1-7) calculate the Fisher differentiation rate of each characteristic parameter, normalized then, obtains the weight of each characteristic parameter;

(2) cognitive phase: for new input sample to be identified, according to step (1-1)-(1-5), obtain characteristic vector, the weight of each characteristic parameter obtaining with step (1-7), the absolute value of each characteristic parameter of new samples is weighted to summation and obtains Fisher scoring, according to the score value of its scoring, realize the identification of degree of hepatic fibrosis.

2. according to claim 1ly based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods, it is characterized in that, in described step (1-5), the method for extracting ultrasonic RF seasonal effect in time series SMR fractal dimension average is as follows:

(1-5-1-1) establish region of interest ROI size for M * N, the every bit in ROI is got to its all frame RF signals, thereby obtain M * N RF time series;

(1-5-1-2) normalization amplitude is:

$f_i^{\mathrm{nor}}=\frac{f_i^{\&prime;}}{\frac{1}{N_t}\underset{j=1}{\overset{N_t}{\mathrm{\&Sigma;}}}\left|f_j^{\&prime;}\right|},f_i^{\&prime;}=f_i-\frac{1}{N_t}\underset{j=1}{\overset{N_t}{\mathrm{\&Sigma;}}}{f}_j,i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,N_t.;$

Wherein: f _ibe the amplitude that in a RF time series, i is ordered, N _ttotal number for RF time series sampled point;

(1-5-1-3) signal calculated total length L _s, ask for successively the difference of two adjacent normalization amplitudes, to absolute difference, summation obtains;

(1-5-1-4) ask for the arc distance R between amplitude maximum and amplitude smallest point;

(1-5-1-5) obtain SMR fractal dimension:

(1-5-1-6) calculate respectively each RF seasonal effect in time series SMR fractal dimension, then summation, then divided by M * N, obtains SMR fractal dimension average.

3. according to claim 1ly based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods, it is characterized in that, in described step (1-5), extract following 6 spectrum signature parameters, is respectively the spectrum energy sum S of four frequency ranges ₁, S ₂, S ₃, S ₄with spectrum slope S lope, spectrum intercept Intercept, calculation procedure is:

First each RF time series f (n) is converted to obtain to frequency spectrum as FFT, ask in every frame ROI institute's spectral magnitude a little with, then divided by M * N, obtain the spectrum mean F of every frame _ave(k), to F _ave(k) as normalized, obtain normalization frequency spectrum, that is:

$\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|=\left|F_{\mathrm{ave}}\left(k\right)\right|/\max \left(\right|F_{\mathrm{ave}}\left(k\right)\left|\right);$

Normalization frequency spectrum is carried out to the sane matching of straight line, obtain composing slope S lope, spectrum intercept Intercept;

According to formula below, ask for S ₁, S ₂, S ₃, S ₄:

$S_1=\underset{k=1}{\overset{\frac{L}{8}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|,S_2=\underset{k=\frac{L}{8}+1}{\overset{\frac{L}{4}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|$

$S_3=\underset{k=\frac{L}{4}+1}{\overset{\frac{3}{4}L}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|,S_4=\underset{k=\frac{3L}{4}+1}{\overset{\frac{L}{2}}{\mathrm{\&Sigma;}}}\left|{\hat{F}}_{\mathrm{ave}}\left(k\right)\right|$

Wherein, L is RF time series totalframes.

4. according to claim 1ly based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods, it is characterized in that, in described step (1-7), the step of calculating the Fisher differentiation rate of each characteristic parameter is:

$w_i=\frac{|{\mathrm{\&mu;}}_{i-\mathrm{NOR}}-{\mathrm{\&mu;}}_{i-\mathrm{CIR}}|}{{\mathrm{\&sigma;}}_{i-\mathrm{NOR}}^2+{\mathrm{\&sigma;}}_{i-\mathrm{CIR}}^2};$

Wherein, w _irepresent the Fisher differentiation rate of i characteristic parameter, μ _i-CIRfor i characteristic parameter average of fibrosis liver sample, μ _i-NORfor i characteristic parameter average of normal hepatocytes sample, σ _i-NORand σ _i-CIRbe respectively the standard deviation of i characteristic parameter of normal hepatocytes and fibrosis liver sample.

5. according to claim 4ly based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods, it is characterized in that, the step in described step (7), the Fisher differentiation rate of each characteristic parameter being normalized is:

${\hat{w}}_i=\frac{w_i}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i}$

Wherein,

it is the weight of i characteristic parameter.

6. according to claim 1ly based on ultrasonic RF seasonal effect in time series degree of hepatic fibrosis Fisher recognition methods, it is characterized in that, in described step (2), the step that obtains Fisher scoring is:

If the weight of i characteristic parameter is

in the new samples extracting, i characteristic parameter is ε _i, Fisher scoring Score-fisher is:

$\mathrm{Score}-\mathrm{fisher}=\underset{i}{\mathrm{\&Sigma;}}{\hat{w}}_i\left|{\mathrm{\&epsiv;}}_i\right|.$

Images