CN103942785A - PET and CT image lung tumor segmenting method based on graph cut - Google Patents

Abstract

The invention discloses a PET and CT image lung tumor segmenting method based on graph cut. The method includes the steps that first, PET image data acquisition and CT image data acquisition are conducted, a PET image is sampled, affine alignment is carried out on the PET image and a CT image, and accordingly pixels on the PET image and pixels on the CT image are made to be in one-to-one correspondence; seed point calibration is conducted on tumor locations and non-tumor locations of the images; a tumor golden standard of tumors is obtained with the help and supervision of clinical oncologists; through PET information extraction and CT information extraction, a lung tumor is segmented and tested by confluence analysis of the extracted information of the PET image and the CT image through a graph cut algorithm, and consequently a final testing result can be obtained.

Description

The lung neoplasm dividing method of a kind of PET cutting based on figure and CT image

Technical field

The lung neoplasm dividing method that the present invention relates to a kind of PET cutting based on figure and CT image, belongs to Biologic Medical Image process field, the optimized dividing method of robotization that the method for utilizing figure to cut (graph cut) is carried out lung tumors.

Background technology

At present, the today going from bad to worse at environment, lung neoplasm has become the main cause of harm humans health.The treatment of lung tumors needs accurate knub position, size and shape, and therefore the accurate lung neoplasm of lung neoplasm is cut apart becomes a popular research topic.PET-CT, as quantitative molecule-structure imaging technology, has been widely used in the formulation of tumor analysis and oncotherapy scheme.The optimization that Graph cut figure cuts is provided by metabolic information and the CT(X computer on line tomography of the main human body in conjunction with utilizing PET (Positron Emission Computed Tomography) image to provide of algorithmic technique) information of the anatomical structure that provides, segmentation problem is changed into the problem (energy minimization problem) of energy minimization, thereby lung neoplasm is automatically split from PET and CT image.

At present, many experts both domestic and external have used several different methods to cut apart lung tumors, for example: Threshold segmentation, algorithm of region growing, machine learning algorithm.These methods all adopt single mode to cut apart lung neoplasm substantially, and accuracy and reliability are not high.

Summary of the invention

Object: in order to overcome the deficiencies in the prior art, the invention provides the lung neoplasm dividing method of a kind of PET cutting based on figure and CT image, by position and the size of the clear and definite tumour of scientific research methods, assist a physician the disease of tumour is judged in advance, help clinical tumor expert to provide better therapeutic scheme for patient.

Technical scheme: for solving the problems of the technologies described above, the technical solution used in the present invention is:

A lung neoplasm dividing method for the PET cutting based on figure and CT image, is characterized in that, comprises the following steps:

(1) PET image is carried out to up-sampling, and PET and CT image are carried out to affine registration, make the point on PET and CT image corresponding one by one;

(2) cut algorithm needs according to figure, the Seed Points of manual marked tumor region and non-tumor region on the PET merging and CT image;

(3) to the medical image extracting, determine the goldstandard of tumour, tumor region is demarcated;

(4), to PET and CT image filter filtering, smoothed image retains the information on border simultaneously;

(5) maximal value and the minimum value of the focus standard uptake value SUV on calculating PET image;

(6) cut according to figure the energy function proposing in algorithm, calculate the energy function value of the upper each pixel of PET and CT;

(7) cut algorithm according to figure, for PET and CT image are set up two figures, and two width figures are connected to d-link with one connect;

(8) computational context cost function;

(9) algorithm cutting with figure is cut apart tumour on PET and CT image, and the result of cutting apart and goldstandard are compared, and uses weighing criteria to quantize testing result.

In described step (1), PET image is carried out to up-sampling for PET image is carried out to linear up-sampling.

In described step (6), energy function comprises: PET energy function E _pET, CT energy function E _cT, and the punishment PET context cost function E different with CT information _context, total energy function is as follows: E (f)=E _pET(f)+E _cT(f)+E _context(f); F is the mark of distributing to each pixel.

Described a kind of PET cutting based on figure and the lung neoplasm dividing method of CT image, is characterized in that: wherein, and PET energy function E _pET, utilize the distribution characteristics of PET image and the pixel of tumour and position feature, comprise that the regional cost function of four part: figure in cutting, boundary cost function, shape are apart from constraint function and monotonically decreasing function;

1) regional cost function:

$D_u(f_u=1)=\left\{\begin{array}{cc}0& ifS\left(u\right)<S_L\\ \frac{T_{\max }}{1+\exp (-(\frac{S\left(u\right)-S_L}{S_M-S_L}+a)/b)},& \mathrm{if}{S}_L<S\left(u\right)<S_M\\ T_{\max },& \mathrm{ifS}\left(u\right)>S_M\end{array}\right.$

D _u(f _u＝0)＝T _max-D _u(f _u＝1)

Wherein: S _m, S _lbe respectively that SUV peaked 50% and 15%, S (u) are the SUV value of pixel u, a, b are two coefficients, T _maxfor the maximal value of cost function allowing;

2) boundary cost function:

$W_{\mathrm{uv}}=-\log (1-\exp \left(\frac{-{|\&dtri;G|}^2(u,v)}{{2\mathrm{\&sigma;}}_g^2}\right))$

| ▽ G| ²(u, v) is the gradient information of pixel u, v, σ _gfor gaussian coefficient;

3) shape is apart from fettering function:

$S_u\left(f_u\right)=T_{\max }(1-\exp (-\frac{d(u,x_o)}{r_o}))$

R _ofor the radius of target area, d (u, x _o) for pixel u be x to target area coordinate _oeuclidean distance; T _maxfor the maximal value of cost function allowing;

4) monotonically decreasing function:

$A_u=\left\{u_i\right|\mathrm{if}\left|\right|x_{u_i}-x_{u_{\max }}\left|\right|\&GreaterEqual;\left|\right|x_{u_j}-x_{u_{\max }}\left|\right|\mathrm{andSUV}\left(u_i\right)\&le;\mathrm{SUV}\left(u_j\right)\}$

A _ufor the set of the point on dull descent direction, M _ufor the cost function that dullness declines, T _maxfor the maximal value of cost function allowing, x is coordinate figure, p is two coefficients;

Therefore, PET energy function formula is as follows:

$E_{\mathrm{PET}}\left(f\right)=\mathrm{\&alpha;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{D}_u\left(f_u\right)+\mathrm{\&beta;}\underset{(u,v)\&Element;N_{\mathrm{PET}}}{\mathrm{\&Sigma;}}{V}_{\mathrm{uv}}(f_u,f_v)+\mathrm{\&gamma;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{S}_u\left(f_u\right)+\mathrm{\&lambda;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{M}_u\left(f_u\right)$

G is the PET figure of setting up, and u, v are the pixels in PET figure, N _pETbe the set of the neighborhood point of PET figure, f is the mark of distributing to each pixel, and α, β, γ, λ are respectively the coefficients that each energy function is corresponding.

Described a kind of PET cutting based on figure and the lung neoplasm dividing method of CT image, is characterized in that: wherein, and CT energy function E _cT, comprise that the regional cost function of figure in cutting, boundary cost function, shape are apart from constraint function;

1) regional cost function:

$D_{u^{,}}(f_{u^{,}}=0)=-\log (1-\mathrm{Pr}(g_{u^{,}}=1))\&Proportional;-(1-\exp \left(\frac{{(g_{u^{,}}-\stackrel{\&OverBar;}{g})}^2}{{\mathrm{\&sigma;}}^2}\right))$

$D_{u^{,}}(f_{u^{,}}=1)=-\log \mathrm{Pr}\left(g_{u^{,}}\right|f_{u^{,}}=1)\&Proportional;\frac{{(g_{u^{,}}-\stackrel{\&OverBar;}{g})}^2}{{\mathrm{\&sigma;}}^2}$

σ is gaussian coefficient, g _{u '}for the brightness value of pixel, for the mean value of all pixel brightness;

2) boundary cost function:

$W_{u^{\&prime;}{v}^{\&prime;}}=-\log (1-\exp \left(\frac{-{\left|{\&dtri;G}^{\&prime;}\right|}^2(u^{\&prime;},v^{\&prime;})}{{{2\mathrm{\&sigma;}}_{g^{\&prime;}}}^2}\right))$

| ▽ G| ²(u', v') is pixel u ', the Grad of v ', σ _git is Gaussian parameter;

3) shape is apart from constraint:

$S_{u^{\&prime;}}\left(f_{u^{\&prime;}}\right)=1-\exp (-\frac{d(u^{\&prime;},x_o)}{r_o})$

R _ofor the radius of target area, and d (u ', x _o) be that pixel u ' is x to target area coordinate _oeuclidean distance;

Therefore, CT energy function formula is as follows:

$E_{\mathrm{CT}}\left(f\right)={\mathrm{\&alpha;}}^{,}\underset{u\&Element;G^{,}}{\mathrm{\&Sigma;}}{D}_{u^{,}}\left(f_{u^{,}}\right)+{\mathrm{\&beta;}}^{,}\underset{(u^{,},v^{,})\&Element;N_{\mathrm{CT}}}{\mathrm{\&Sigma;}}{V}_{u^{,}{v}^{,}}(f_{u^{,}},f_{v^{,}})+{\mathrm{\&gamma;}}^{,}\underset{u\&Element;G^{,}}{\mathrm{\&Sigma;}}{S}_{u^{,}}\left(f_{u^{,}}\right)$

The CT figure of G ' for setting up, u ', v ' are the pixels in PET figure, N _cTbe the set of the neighborhood point of CT figure, f is the mark of distributing to each pixel, and α ', β ', γ ' are the coefficients that each energy function is corresponding.

Described a kind of PET cutting based on figure and the lung neoplasm dividing method of CT image, is characterized in that: wherein, and context cost function E _context: the information of utilizing PET and CT to provide, adopts the difference of a cost function as punishment PET, CT cost;

$E_{\mathrm{context}}\left(f\right)=\underset{u\&Element;G,u^{\&prime;}\&Element;G^{\&prime;}}{\mathrm{\&Sigma;}}\mathrm{\&psi;exp}(\mathrm{\&delta;}(1-|N_u-N_{u^{\&prime;}}\left|\right)+H)$

N _u, N _{u '}be respectively the normalized value of PET, CT cost function, δ, ψ, H are coefficient.

In described step (9), " result of cutting apart and goldstandard are compared; use weighing criteria to quantize testing result " and specifically refer to: adopt DSC coefficient and Hausdorff distance to weigh the result of lesion segmentation, wherein, DSC coefficient is used for the tumour result cut apart of reflection and the registration in goldstandard region, and Hausdorff is apart from the compatible degree that reflects goldstandard and segmentation result border;

DSC coefficient:

$\mathrm{DSC}(U_1,U_2)=2\&CenterDot;\frac{|U_1\&cap;U_2|}{|U_1+U_2|}$

U1, U2 are respectively result and the goldstandard cut apart;

Hausdorff distance:

$\mathrm{HD}(X,Y)=H({\&PartialD;U}_1,{\&PartialD;U}_2)=\max \{\underset{x\&Element;X}{\sup }\underset{y\&Element;Y}{\inf }d(x,y),\underset{y\&Element;Y}{\sup }\underset{x\&Element;X}{\inf }d(x,y)\}$

represent boundary information; for cutting apart the set of boundary information of tumour and goldstandard, inf, sup represents respectively the upper bound and lower bound, d (x, y) is the Euclidean distance of x, y.

Beneficial effect: the lung neoplasm dividing method of a kind of PET cutting based on figure provided by the invention and CT image, utilize figure to cut (graph cut) algorithm, extract PET, the information of CT image and feature, provide a kind of tumour degree of accuracy higher, performance is more excellent, the better method of robustness, in conjunction with the metabolic information of PET image and the anatomic information of CT image, utilize the dullness decline feature of tumour brightness on PET image, positional information, the organization of human body information that the distributed intelligence of the SUV of the gradient information on border and PET image and CT provide and the positional information of tumour on CT, cut algorithm by figure, auto Segmentation tumour, make the tumour cut apart more accurate, for the treatment of tumour from now on lays the first stone.

Brief description of the drawings

Fig. 1 is the schematic flow sheet of the inventive method;

Fig. 2 is PET and CT image in the present invention: (a) PET image (b) CT image;

Fig. 3 is dsc analysis figure in the present invention;

Fig. 4 is the analysis chart of HD distance in the present invention.

Embodiment

Below in conjunction with accompanying drawing, the present invention is further described.

As depicted in figs. 1 and 2, first lung neoplasm dividing method of the present invention carries out the collection of PET, CT view data, by PET image is carried out to up-sampling, and PET and CT image is carried out to affine registration, makes the pixel on PET and CT image corresponding one by one; Tumor locus to image and non-tumor locus carry out the demarcation of Seed Points; Under clinical tumor scholar's help and supervision, obtain the goldstandard of tumour; By the information extraction to PET, CT, utilize Graph cut Algorithms Integration analysis extract PET and CT image on information, lung neoplasm is cut apart, test, draw last testing result.

The method of the invention is under the patronage of First Affiliated Hospital of Soochow University,Suzhou, obtain the patient data who suffers from non-small cell lung cancer, extraction and the fusion of cutting apart principle and mainly utilized information on PET, CT image of lung neoplasm, utilize this algorithm of Graph cut to calculate, analyze.Below in conjunction with Fig. 1 to Fig. 4, the embodiment of process in detail.

A lung neoplasm dividing method for the PET cutting based on figure and CT image, is characterized in that, comprises the following steps:

(1) the PET image getting is carried out to linear up-sampling, and PET and CT image are carried out to affine registration, make the point on PET and CT image corresponding one by one;

(2) cut algorithm needs according to figure, the Seed Points of manual marked tumor region and non-tumor region on the PET merging and CT image;

(3), to the medical image extracting, determine the goldstandard of tumour: under clinical tumor expert's guidance, utilize ITK-SNAP software, tumor region is demarcated;

(4), to PET and CT image filter filtering, smoothed image retains the information on border simultaneously;

(5) maximal value and the minimum value of the focus standard uptake value SUV on calculating PET image;

(6) cut according to figure the energy function proposing in algorithm, calculate the cost of the upper each pixel of PET and CT; (7) cut algorithm according to figure, for PET and CT image are set up two figures, and two width figures are connected with a special connection (d-link); (8) computational context cost function; This method adopts the algorithm of Graph Cut, extracts the image information of PET and CT, and tumour is cut apart.The essence of Graph Cut algorithm is the problem that the problem of cutting apart is changed into energy minimization, and this method is by PET and CT image co-registration, and allows the information of PET and CT image different, thus energy function formed by three parts, PET energy function E _pET, CT energy function E _cTand the punishment PET context cost function E different with CT information _context, total energy function is as follows:

E (f)=E _pET(f)+E _cT(f)+E _context(f); F is the mark of distributing to each pixel;

1, wherein: PET energy function E _pET, utilize the distribution characteristics of PET image and the pixel of tumour and position feature, comprise that the regional cost function of four part: figure in cutting, boundary cost function, shape are apart from constraint function and monotonically decreasing function;

1) regional cost function (region cost function):

$D_u(f_u=1)=\left\{\begin{array}{cc}0& ifS\left(u\right)<S_L\\ \frac{T_{\max }}{1+\exp (-(\frac{S\left(u\right)-S_L}{S_M-S_L}+a)/b)},& \mathrm{if}{S}_L<S\left(u\right)<S_M\\ T_{\max },& \mathrm{ifS}\left(u\right)>S_M\end{array}\right.$

D _u(f _u＝0)＝T _max-D _u(f _u＝1)

Wherein: S _m, S _lbe respectively that SUV peaked 50% and 15%, S (u) are the SUV value of pixel u, a, b are two coefficients, T _maxfor the maximal value of cost function allowing;

2) boundary cost function (boundary cost function):

$W_{\mathrm{uv}}=-\log (1-\exp \left(\frac{-{|\&dtri;G|}^2(u,v)}{{2\mathrm{\&sigma;}}_g^2}\right))$

| ▽ G| ²(u, v) is the gradient information of pixel u, v, σ _gfor gaussian coefficient;

3) shape is apart from fettering function (shape distance constraint):

$S_u\left(f_u\right)=T_{\max }(1-\exp (-\frac{d(u,x_o)}{r_o}))$

R _ofor the radius of target area (onject), d (u, x _o) for pixel u be x to target area coordinate _oeuclidean distance;

4) monotonically decreasing function (Monotonic Downhill cost function)

$A_u=\left\{u_i\right|\mathrm{if}\left|\right|x_{u_i}-x_{u_{\max }}\left|\right|\&GreaterEqual;\left|\right|x_{u_j}-x_{u_{\max }}\left|\right|\mathrm{andSUV}\left(u_i\right)\&le;\mathrm{SUV}\left(u_j\right)\}$

T _maxfor the maximal value of cost, x is coordinate figure, p is two coefficients;

Therefore, PET energy function formula is as follows:

$E_{\mathrm{PET}}\left(f\right)=\mathrm{\&alpha;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{D}_u\left(f_u\right)+\mathrm{\&beta;}\underset{(u,v)\&Element;N_{\mathrm{PET}}}{\mathrm{\&Sigma;}}{V}_{\mathrm{uv}}(f_u,f_v)+\mathrm{\&gamma;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{S}_u\left(f_u\right)+\mathrm{\&lambda;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{M}_u\left(f_u\right)$

G is the PET figure of setting up, and u, v are the pixels in PET figure, N _pETbe the set of the neighborhood point of PET figure, f is the mark of distributing to each pixel, and α, β, γ, λ are respectively the coefficients that each energy function is corresponding;

2, CT energy function E _cT, comprise that the regional cost function of figure in cutting, boundary cost function, shape are apart from constraint function;

1) regional cost function D _{u '}(f _{u '}):

$D_{u^{,}}(f_{u^{,}}=0)=-\log (1-\mathrm{Pr}(g_{u^{,}}=1))\&Proportional;-(1-\exp \left(\frac{{(g_{u^{,}}-\stackrel{\&OverBar;}{g})}^2}{{\mathrm{\&sigma;}}^2}\right))$

$D_{u^{,}}(f_{u^{,}}=1)=-\log \mathrm{Pr}\left(g_{u^{,}}\right|f_{u^{,}}=1)\&Proportional;\frac{{(g_{u^{,}}-\stackrel{\&OverBar;}{g})}^2}{{\mathrm{\&sigma;}}^2}$

σ is gaussian coefficient, g _{u '}for the brightness value of pixel, for the mean value of all pixel brightness;

2) boundary cost function:

$W_{u^{\&prime;}{v}^{\&prime;}}=-\log (1-\exp \left(\frac{-{\left|{\&dtri;G}^{\&prime;}\right|}^2(u^{\&prime;},v^{\&prime;})}{{{2\mathrm{\&sigma;}}_{g^{\&prime;}}}^2}\right))$

| ▽ G| ²(u', v') is pixel u ', the Grad of v ', σ _git is Gaussian parameter;

3) shape is apart from constraint (shape distance constraint)

$S_{u^{\&prime;}}\left(f_{u^{\&prime;}}\right)=1-\exp (-\frac{d(u^{\&prime;},x_o)}{r_o})$

R _ofor the radius of target area (onject), and d (u ', x _o) be that pixel u ' is x to target area coordinate _oeuclidean distance; T _maxfor the maximal value of cost function allowing;

Therefore, CT energy function formula is as follows:

$E_{\mathrm{CT}}\left(f\right)={\mathrm{\&alpha;}}^{,}\underset{u\&Element;G^{,}}{\mathrm{\&Sigma;}}{D}_{u^{,}}\left(f_{u^{,}}\right)+{\mathrm{\&beta;}}^{,}\underset{(u^{,},v^{,})\&Element;N_{\mathrm{CT}}}{\mathrm{\&Sigma;}}{V}_{u^{,}{v}^{,}}(f_{u^{,}},f_{v^{,}})+{\mathrm{\&gamma;}}^{,}\underset{u\&Element;G^{,}}{\mathrm{\&Sigma;}}{S}_{u^{,}}\left(f_{u^{,}}\right)$

The CT figure of G ' for setting up, u ', v ' are the pixels in PET figure, N _cTbe the set of the neighborhood point of CT figure, f is the mark of distributing to each pixel, and α ', β ', γ ' are the coefficients that each energy function is corresponding;

3, context cost function E _context: the information of utilizing PET and CT to provide, adopts the difference of a cost function as punishment PET, CT cost;

$E_{\mathrm{context}}\left(f\right)=\underset{u\&Element;G,u^{\&prime;}\&Element;G^{\&prime;}}{\mathrm{\&Sigma;}}\mathrm{\&psi;exp}(\mathrm{\&delta;}(1-|N_u-N_{u^{\&prime;}}\left|\right)+H)$

N _u, N _{u '}be respectively the normalized value of PET, CT cost function, δ, ψ, H are coefficient.

(9) algorithm cutting with figure is cut apart tumour on PET and CT image, and the result of cutting apart and goldstandard are compared, and uses weighing criteria to quantize testing result.

The tumour data that adopt First Affiliated Hospital of Soochow University,Suzhou to provide are provided in the present invention, have chosen 18 available patient datas, and adopt DSC coefficient and Hausdorff distance (Hausdorff distance, HD distance) to weigh the result of lesion segmentation; DSC coefficient is used for the tumour result cut apart of reflection and the registration in goldstandard region, and Hausdorff is apart from the compatible degree that reflects goldstandard and segmentation result border.The quantitative analysis results of these 18 data as shown in Figure 3 and Figure 4;

DSC coefficient:

$\mathrm{DSC}(U_1,U_2)=2\&CenterDot;\frac{|U_1\&cap;U_2|}{|U_1+U_2|}$

U1, U2 are respectively result and the goldstandard cut apart;

Hausdorff distance:

$\mathrm{HD}(X,Y)=H({\&PartialD;U}_1,{\&PartialD;U}_2)=\max \{\underset{x\&Element;X}{\sup }\underset{y\&Element;Y}{\inf }d(x,y),\underset{y\&Element;Y}{\sup }\underset{x\&Element;X}{\inf }d(x,y)\}$

U1, U2 are respectively result and the goldstandard cut apart, represent boundary information; for cutting apart the set of boundary information of tumour and goldstandard, inf, sup represents respectively the upper bound and lower bound, d (x, y) is the Euclidean distance of x, y.

Prove by experiment, the lung neoplasm dividing method of a kind of PET cutting based on figure provided by the invention and CT image, utilize figure to cut (graph cut) algorithm, extract PET, the information of CT image and feature, provide a kind of tumour degree of accuracy higher, performance is more excellent, the better method of robustness, in conjunction with the metabolic information of PET image and the anatomic information of CT image, utilize the dullness decline feature of tumour brightness on PET image, positional information, the organization of human body information that the distributed intelligence of the SUV of the gradient information on border and PET image and CT provide and the positional information of tumour on CT, cut algorithm by figure, auto Segmentation tumour, make the tumour cut apart more accurate, for the treatment of tumour from now on lays the first stone.The above is only the preferred embodiment of the present invention; be noted that for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (7)

1. a lung neoplasm dividing method for the PET cutting based on figure and CT image, is characterized in that, comprises the following steps:

(1) PET image is carried out to up-sampling, and PET and CT image are carried out to affine registration, make the point on PET and CT image corresponding one by one;

(2) cut algorithm needs according to figure, the Seed Points of manual marked tumor region and non-tumor region on the PET merging and CT image;

(3) to the medical image extracting, determine the goldstandard of tumour, tumor region is demarcated;

(4), to PET and CT image filter filtering, smoothed image retains the information on border simultaneously;

(5) maximal value and the minimum value of the focus standard uptake value SUV on calculating PET image;

(6) cut according to figure the energy function proposing in algorithm, calculate the energy function value of the upper each pixel of PET and CT;

(7) cut algorithm according to figure, for PET and CT image are set up two figures, and two width figures are connected to d-link with one connect;

(8) computational context cost function;

(9) algorithm cutting with figure is cut apart tumour on PET and CT image, and the result of cutting apart and goldstandard are compared, and uses weighing criteria to quantize testing result.

2. the lung neoplasm dividing method of a kind of PET cutting based on figure according to claim 1 and CT image, is characterized in that: in described step (1), PET image is carried out to up-sampling for PET image is carried out to linear up-sampling.

3. the lung neoplasm dividing method of a kind of PET cutting based on figure according to claim 1 and CT image, is characterized in that: in described step (6), energy function comprises: PET energy function E _pET, CT energy function E _cT, and the punishment PET context cost function E different with CT information _context, total energy function is as follows:

E (f)=E _pET(f)+E _cT(f)+E _context(f); F is the mark of distributing to each pixel.

4. the lung neoplasm dividing method of a kind of PET cutting based on figure according to claim 3 and CT image, is characterized in that: wherein, and PET energy function E _pET, utilize the distribution characteristics of PET image and the pixel of tumour and position feature, comprise that the regional cost function of four part: figure in cutting, boundary cost function, shape are apart from constraint function and monotonically decreasing function;

1) regional cost function:

$D_u(f_u=1)=\left\{\begin{array}{cc}0& ifS\left(u\right)<S_L\\ \frac{T_{\max }}{1+\exp (-(\frac{S\left(u\right)-S_L}{S_M-S_L}+a)/b)},& \mathrm{if}{S}_L<S\left(u\right)<S_M\\ T_{\max },& \mathrm{ifS}\left(u\right)>S_M\end{array}\right.$

D _u(f _u＝0)＝T _max-D _u(f _u＝1)

Wherein: S _m, S _lbe respectively that SUV peaked 50% and 15%, S (u) are the SUV value of pixel u, a, b are two coefficients, T _maxfor the maximal value of cost function allowing;

2) boundary cost function:

$W_{\mathrm{uv}}=-\log (1-\exp \left(\frac{-{|\&dtri;G|}^2(u,v)}{{2\mathrm{\&sigma;}}_g^2}\right))$

| ▽ G| ²(u, v) is the gradient information of pixel u, v, σ _gfor gaussian coefficient;

3) shape is apart from fettering function:

$S_u\left(f_u\right)=T_{\max }(1-\exp (-\frac{d(u,x_o)}{r_o}))$

R _ofor the radius of target area, d (u, x _o) for pixel u be x to target area coordinate _oeuclidean distance; T _maxfor the maximal value of cost function allowing;

4) monotonically decreasing function:

$A_u=\left\{u_i\right|\mathrm{if}\left|\right|x_{u_i}-x_{u_{\max }}\left|\right|\&GreaterEqual;\left|\right|x_{u_j}-x_{u_{\max }}\left|\right|\mathrm{andSUV}\left(u_i\right)\&le;\mathrm{SUV}\left(u_j\right)\}$

A _ufor the set of the point on dull descent direction, M _ufor the cost function that dullness declines, T _maxfor the maximal value of cost function allowing, x is coordinate figure, p is two coefficients;

Therefore, PET energy function formula is as follows:

$E_{\mathrm{PET}}\left(f\right)=\mathrm{\&alpha;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{D}_u\left(f_u\right)+\mathrm{\&beta;}\underset{(u,v)\&Element;N_{\mathrm{PET}}}{\mathrm{\&Sigma;}}{V}_{\mathrm{uv}}(f_u,f_v)+\mathrm{\&gamma;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{S}_u\left(f_u\right)+\mathrm{\&lambda;}\underset{u\&Element;G}{\mathrm{\&Sigma;}}{M}_u\left(f_u\right)$

G is the PET figure of setting up, and u, v are the pixels in PET figure, N _pETbe the set of the neighborhood point of PET figure, f is the mark of distributing to each pixel, and α, β, γ, λ are respectively the coefficients that each energy function is corresponding.

5. the lung neoplasm dividing method of a kind of PET cutting based on figure according to claim 3 and CT image, is characterized in that: wherein, and CT energy function E _cT, comprise that the regional cost function of figure in cutting, boundary cost function, shape are apart from constraint function;

1) regional cost function:

$D_{u^{,}}(f_{u^{,}}=0)=-\log (1-\mathrm{Pr}(g_{u^{,}}=1))\&Proportional;-(1-\exp \left(\frac{{(g_{u^{,}}-\stackrel{\&OverBar;}{g})}^2}{{\mathrm{\&sigma;}}^2}\right))$

$D_{u^{,}}(f_{u^{,}}=1)=-\log \mathrm{Pr}\left(g_{u^{,}}\right|f_{u^{,}}=1)\&Proportional;\frac{{(g_{u^{,}}-\stackrel{\&OverBar;}{g})}^2}{{\mathrm{\&sigma;}}^2}$

σ is gaussian coefficient, g _{u '}for the brightness value of pixel, for the mean value of all pixel brightness;

2) boundary cost function:

$W_{u^{\&prime;}{v}^{\&prime;}}=-\log (1-\exp \left(\frac{-{\left|{\&dtri;G}^{\&prime;}\right|}^2(u^{\&prime;},v^{\&prime;})}{{{2\mathrm{\&sigma;}}_{g^{\&prime;}}}^2}\right))$

| ▽ G| ²(u', v') is pixel u ', the Grad of v ', σ _git is Gaussian parameter;

3) shape is apart from constraint:

$S_{u^{\&prime;}}\left(f_{u^{\&prime;}}\right)=1-\exp (-\frac{d(u^{\&prime;},x_o)}{r_o})$

R _ofor the radius of target area, and d (u ', x _o) be that pixel u ' is x to target area coordinate _oeuclidean distance;

Therefore, CT energy function formula is as follows:

$E_{\mathrm{CT}}\left(f\right)={\mathrm{\&alpha;}}^{,}\underset{u\&Element;G^{,}}{\mathrm{\&Sigma;}}{D}_{u^{,}}\left(f_{u^{,}}\right)+{\mathrm{\&beta;}}^{,}\underset{(u^{,},v^{,})\&Element;N_{\mathrm{CT}}}{\mathrm{\&Sigma;}}{V}_{u^{,}{v}^{,}}(f_{u^{,}},f_{v^{,}})+{\mathrm{\&gamma;}}^{,}\underset{u\&Element;G^{,}}{\mathrm{\&Sigma;}}{S}_{u^{,}}\left(f_{u^{,}}\right)$

The CT figure of G ' for setting up, u ', v ' are the pixels in PET figure, N _cTbe the set of the neighborhood point of CT figure, f is the mark of distributing to each pixel, and α ', β ', γ ' are the coefficients that each energy function is corresponding.

6. the lung neoplasm dividing method of a kind of PET cutting based on figure according to claim 3 and CT image, is characterized in that: wherein, and context cost function E _context: the information of utilizing PET and CT to provide, adopts the difference of a cost function as punishment PET, CT cost;

$E_{\mathrm{context}}\left(f\right)=\underset{u\&Element;G,u^{\&prime;}\&Element;G^{\&prime;}}{\mathrm{\&Sigma;}}\mathrm{\&psi;exp}(\mathrm{\&delta;}(1-|N_u-N_{u^{\&prime;}}\left|\right)+H)$

N _u, N _{u '}be respectively the normalized value of PET, CT cost function, δ, ψ, H are coefficient.

7. the lung neoplasm dividing method of a kind of PET cutting based on figure according to claim 1 and CT image, it is characterized in that: in described step (9), " result of cutting apart and goldstandard are compared; use weighing criteria to quantize testing result " and specifically refer to: adopt DSC coefficient and Hausdorff distance to weigh the result of lesion segmentation, wherein, DSC coefficient is used for the tumour result cut apart of reflection and the registration in goldstandard region, and Hausdorff is apart from the compatible degree that reflects goldstandard and segmentation result border;

DSC coefficient:

$\mathrm{DSC}(U_1,U_2)=2\&CenterDot;\frac{|U_1\&cap;U_2|}{|U_1+U_2|}$

U1, U2 are respectively result and the goldstandard cut apart;

Hausdorff distance:

$\mathrm{HD}(X,Y)=H({\&PartialD;U}_1,{\&PartialD;U}_2)=\max \{\underset{x\&Element;X}{\sup }\underset{y\&Element;Y}{\inf }d(x,y),\underset{y\&Element;Y}{\sup }\underset{x\&Element;X}{\inf }d(x,y)\}$

U1, U2 are respectively result and the goldstandard cut apart; represent boundary information; for cutting apart the set of boundary information of tumour and goldstandard, inf, sup represents respectively the upper bound and lower bound, d (x, y) is the Euclidean distance of x, y.