CN105719256A - PET image partial volume correction method based on guiding of structural image - Google Patents

Abstract

The invention relates to a PET image partial volume correction method based on guiding of a structural image. After that rigid registration is carried out on the structural image and a PET image, a median of the PET image is filtered, the structural image is defined as a voxel effective neighborhood, the PET image whose median is filtered and the voxel effective neighborhood are used to define a guiding capacity effect correction model, the structural image serves as a guiding image to convert the guiding capacity effect correction model, minimizing solution is carried out on an obtained equation with a constraint objective function in a linear regression method, and an output PET image after de-noising and partial volume effect correction is obtained. According to the invention, the high-definition structural image is used to guide de-noising and reduction of the partial volume effect, a non-iteration analytic algorithm is used so that the method of the invention is simpler, the effective neighborhood is selected to adjust the PET image adaptively, the resolution of the PET image is improved, formation of noises is inhibited, and PET detection capability and quantification precision of small structures are ensured.

Description

A kind of PET image partial volume correction method guided based on structural images

Technical field

The invention belongs to the technical field of general image real time transfer or generation, particularly to a kind of PET image partial volume correction method guided based on structural images.

Background technology

PET full name is positron e mission computed tomography (PositronEmissionTomography), it it is the clinical Imaging Techniques of the field of nuclear medicine advanced person, its substantially method be usually in biological life metabolism by certain material necessary material, as: glucose, protein, nucleic acid, fatty acid, on labelling, short-life radionuclide is (such as F18, carbon 11 etc.) inject after human body, by lesions position, focus is understood in the picked-up of tracer, and then disease is made the technology of diagnosis.

MRI full name is magnetic resonance (MagneticResonanceImaging), utilizes atomic nucleus at a kind of imaging technique of the magnetic field internal resonance reconstructed imaging of produced signal.The application of fast-changing gradient magnetic, is greatly accelerated the speed of NMR (Nuclear Magnetic Resonance)-imaging, makes this technology become a reality in the application of clinical diagnosis, scientific research, has greatly promoted developing rapidly of medical science, neuro physiology and cognitive neuroscience.

CT full name is ComputedTomography, i.e. CT scan, it utilizes the X-ray beam of Accurate collimation, the detector high with sensitivity together makes profile scanning one by one around a certain position of human body, there is sweep time fast, the features such as image is clear, can be used for the inspection etc. of multiple disease.

Although as PET imaging subject continuous advancement, such as introduce point spread function etc., but PET image yet suffers from the low signal-to-noise ratio brought due to the restriction of implantation dosage and instrumental sensitivity and the volume effect brought due to the spatial resolution of imaging system and the restriction of tissue fraction effect (tissuefractioneffect).

Rebuild bearing calibration after voxel level and need not assume that in region, activity is consistent, single voxel can be corrected, its for PET image itself be iterated deconvolution processing just can each voxel of correction chart picture, but trimming process can introduce high-level noise.In order to suppress the increase of noise, intermediate value priori and wavelet filtering are introduced in iterative deconvolution process, and unfortunately these deconvolution algorithms all can exist this artifact of gilbert.The PET partial volume correction method guided based on structural images priori has caused extensive concern, but, the partial volume correction that existing structural images priori guides is based on anatomic image area information more, this type of method first has to structural images is carried out Accurate Segmentation, and structural images segmentation there is no accurately and the method for robust.Additionally, this type of method needs to assume that in structural region, the distribution of PET activity is consistent, thus strongly limit the application of this type of method.

Chinese patent " a kind of voxel level PET image partial volume correction method based on MR information guidance ", the patent No. is 201410183320.5, disclose the voxel level PET image partial volume correction method based on MR information guidance, it also adopts MR information guidance to complete the partial volume correction of PET image, but it is a kind of iterative algorithm, algorithm is complicated, computationally intensive and is difficult to determine optimum iterations.

Summary of the invention

Present invention solves the technical problem that and be, low signal-to-noise ratio that PET image is brought due to the restriction of implantation dosage and scan sensitivity and the volume effect brought due to the spatial resolution of imaging system and the restriction of tissue fraction effect (tissuefractioneffect).Although most PET imaging systems have employed the iterative reconstruction algorithm based on maximal possibility estimation, but the volume effect that the PET image rebuild yet suffers from the spatial resolution of low signal-to-noise ratio problem and the imaging system brought due to the restriction of implantation dosage and system sensitivity and tissue fraction effect limits and brings.Although noise can be suppressed by early stage stopping iteration or by adopting wave filter to suppress after rebuilding, but early stage stops alternative manner and tends not to reach algorithm for reconstructing convergence thus tending not to the image that enough generations are detailed.Gaussian filtering denoising method is that after reconstruction the most frequently used in current clinic, image filtering method but gaussian filtering are likely to while eliminating noise and eliminate important picture structure, this will further decrease spatial resolution, thus the problem weakening the quantified precision of power of test and little structure.The invention provides a kind of PET image partial volume correction method guided based on structural images.

The technical solution adopted in the present invention is, a kind of PET image partial volume correction method guided based on structural images, said method comprising the steps of:

Step 1.1: be utilized respectively structural images imaging device and PET imaging device gathers the structural images of same object and the initial data of PET, obtain the structural images G and PET image P of object, obtain the resolution of structural images and PET imaging device simultaneously；

Step 1.2: the structural images G obtained in step 1.1 and PET image P is carried out Rigid Registration；

Step 1.3: the PET image P obtained in step 1.1 is carried out medium filtering, obtains the PET image P after medium filtering^median；

Step 1.4: the structural images G obtained in step 1.2 is defined the effective neighborhood ω of voxel_k；

Step 1.5: utilize the P obtained in step 1.3^medianNeighborhood ω effective in the voxel obtained in step 1.4_kDefinition guides volume effect calibration model；

Step 1.6: the guiding volume effect calibration model of step 1.5, as navigational figure, is converted, obtains the equation of belt restraining object function by structural images G step 1.2 obtained；

Step 1.7: the equation of belt restraining object function step 1.6 obtained adopts linear regression method to carry out minimizing solving, obtains the output PET image after denoising and partial volume effect correction.

Preferably, described structural images G is CT or MR image.

Preferably, in described step 1.3, medium filtering adopts formula (I),

P^median=Median (P, ω_m)(I)

Wherein, ω_mFor medium filtering window, P^medianPET image after filtering for medium filtering.

Preferably, in described step 1.3, ω_mMedium filtering window for (3 × 3 × 3).

Preferably, in described step 1.4, for pixel k and its neighborhood N_kEffective neighborhood ω according to following rule definition pixel k_k:

Neighborhood N for pixel k each in 3-D view_kIn each pixel i, if meetIt is classified as effective neighborhood ω_k, wherein μ is the self-defined edge decision threshold in structural images G.

Preferably, the span of described self-defined edge decision threshold μ is 0.5≤μ≤1.

Preferably, in step 1.5, guiding volume effect calibration model is formula (II),

$Q_i=a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median}---\left(II\right)$

Wherein, a_kAnd b_kFor at neighborhood ω_kIn the linear coefficient that remains unchanged, i is neighborhood ω_kIn any one pixel,I is each pixel value image equal to 1, and H is the point spread function of PET imaging device,Accord with for convolution operation.

Preferably, in described step 1.6, the equation of belt restraining object function is formula (III),

$E(a_k,b_k)=\underset{i\∈{\mathrm{\ω}}_k}{\Σ}({\left(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median}\right)}^2+{\mathrm{\ϵa}}_k^2)---\left(III\right)$

Wherein, ε is regularization parameter, is used for controlling a_kSpan；A_kAnd b_kFor at neighborhood ω_kIn the linear coefficient that remains unchanged, i is neighborhood ω_kIn any one pixel,I is each pixel value image equal to 1, and H is the point spread function of PET imaging device,Accord with for convolution operation.

Preferably, the span of described regularization parameter ε is 0≤ε≤1.

Preferably, in described step 1.7, according to formula (III), to a_kAnd b_kTake partial differential and be set as 0, obtaining formula (IV) and formula (V),

$\frac{\∂E}{\∂a_k}=2\underset{i\∈{\mathrm{\ω}}_k}{\Σ}\left(\right(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median}){\left(HG\right)}_i+{\mathrm{\ϵa}}_k)=0---\left(IV\right)$

$\frac{\∂E}{\∂b_k}=2\underset{i\∈{\mathrm{\ω}}_k}{\Σ}{\left(HI\right)}_i(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median})=0---\left(V\right)$

Making described output PET image is formula (IV),

Q_i=a_k(HG)_i+b_k(HI)_i(IV)

Wherein, $a_k=\frac{{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i^{median}\right){\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2-{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right){\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HI\right)}_i{P}_i^{median}\right)}{\left|\mathrm{\ω}\right|\mathrm{\ϵ}{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2+{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HG\right)}_i^2-{\left({\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HG\right)}_i{\left(HI\right)}_i\right)}^2},$

$b_k=\frac{{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i^{median}\right)-a_k{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right)}{{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2}.$

The invention provides a kind of PET image partial volume correction method guided based on structural images, by structural images G and PET image P is carried out Rigid Registration position corresponding after, PET image P is carried out medium filtering, and structural images G is defined the effective neighborhood ω of voxel_k, utilize the PET image P after medium filtering^medianNeighborhood ω effective in voxel_kDefinition guides volume effect calibration model, and utilize structural images G as navigational figure, convert guiding volume effect calibration model, adopt linear regression method to carry out minimizing solving the equation obtaining belt restraining object function, obtain the output PET image after denoising and partial volume effect correction；The present invention utilizes high-resolution structural images guide PET image denoising and reduce partial volume effect, adopt non-iterative Analytic Calculation Method, holistic approach is more easy, self-adaptative adjustment PET image is chosen by effective neighborhood, the formation of noise is suppressed, it is ensured that the power of test of PET and the quantified precision of little structure while improving the image resolution ratio of PET imaging.

Detailed description of the invention

Below in conjunction with embodiment, the present invention is described in further detail, but protection scope of the present invention is not limited to this.

The present invention relates to a kind of PET image partial volume correction method guided based on structural images, said method comprising the steps of:

Step 1.1: be utilized respectively structural images imaging device and PET imaging device gathers the structural images of same object and the initial data of PET, obtain the structural images G and PET image P of object, obtain the resolution of structural images and PET imaging device simultaneously；

Step 1.2: the structural images G obtained in step 1.1 and PET image P is carried out Rigid Registration；

Step 1.3: the PET image P obtained in step 1.1 is carried out medium filtering, obtains the PET image P after medium filtering^median；

Step 1.4: the structural images G obtained in step 1.2 is defined the effective neighborhood ω of voxel_k；

Step 1.5: utilize the P obtained in step 1.3^medianNeighborhood ω effective in the voxel obtained in step 1.4_kDefinition guides volume effect calibration model；

Step 1.6: the guiding volume effect calibration model of step 1.5, as navigational figure, is converted, obtains the equation of belt restraining object function by structural images G step 1.2 obtained；

Step 1.7: the equation of belt restraining object function step 1.6 obtained adopts linear regression method to carry out minimizing solving, obtains the output PET image after denoising and partial volume effect correction.

In the present invention, structural images G and PET image P is carried out Rigid Registration position corresponding after, PET image P is carried out medium filtering, and structural images G is defined the effective neighborhood ω of voxel_k, utilize the PET image P after medium filtering^medianNeighborhood ω effective in voxel_kDefinition guides volume effect calibration model, and utilize structural images G as navigational figure, convert guiding volume effect calibration model, adopt linear regression method to carry out minimizing solving the equation obtaining belt restraining object function, obtain the output PET image after denoising and partial volume effect correction.

In the present invention, adopt and guide volume effect calibration model, utilize structural images G as navigational figure to guiding volume effect calibration model to convert the equation obtaining belt restraining object function, carry out minimizing the output PET image after obtaining denoising and partial volume effect correction after solving with linear regression method, namely the information that non-iterative linear operation self adaptation introduce structural images is adopted to get involved effective neighborhood so that quality and the precision of PET image are higher.

The present invention utilizes high-resolution structural images guide PET image denoising and reduce partial volume effect, adopt non-iterative Analytic Calculation Method, holistic approach is more easy, adaptively selected by effective neighborhood, the formation of noise is suppressed, it is ensured that the power of test of PET and the quantified precision of little structure while improving the image resolution ratio of PET imaging.

Described structural images G is CT or MR image.

In the present invention, it is that generally there is high-resolution due to structural images using structural images as navigational figure, can denoising largely and reduce volume effect, it is ensured that the quality of the middle PET image of the present invention and integrity, it is ensured that the power of test of PET and the quantified precision of little structure.

In the present invention, generally, guide structure image G is CT or MR image.

In described step 1.3, medium filtering adopts formula (I),

P^median=Median (P, ω_m)(I)

Wherein, ω_mFor medium filtering window, P^medianPET image after filtering for medium filtering.

In described step 1.3, ω_mMedium filtering window for (3 × 3 × 3).

Present invention, it is desirable to PET image P carries out medium filtering, the size of medium filtering window needs to be made suitable adjustment by different situations.

In the present invention, medium filtering for the Mesophyticum of each point value in one neighborhood of this point of the value of any in digital picture or Serial No. is replaced, allows the actual value that the pixel value of surrounding is close, thus eliminating the noise spot isolated.

In described step 1.4, for pixel k and its neighborhood N_kEffective neighborhood ω according to following rule definition pixel k_k:

Neighborhood N for pixel k each in 3-D view_kIn each pixel i, if meetIt is classified as effective neighborhood ω_k, wherein μ is the self-defined edge decision threshold in structural images G.

The span of described self-defined edge decision threshold μ is 0.5≤μ≤1.

In the present invention, the neighborhood of each pixel of two dimensional image is 8, and the neighborhood of each pixel of 3-D view is 26.

In the present invention, μ is the self-defined edge decision threshold in structural images G, due toAnd 0.5≤μ≤1, namely represent the neighborhood N of this pixel k_kIn pixel i exist, be edge.

In the present invention, obtaining effective neighborhood ω_kAfter, it is possible to effective figure territory (in effective contiguous range) the adaptively selected guiding in structural images G so that the active position image in the PET image after Rigid Registration is relatively sharp.

In step 1.5, guiding volume effect calibration model is formula (II),

$Q_i=a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median}---\left(II\right)$

Wherein, a_kAnd b_kFor at neighborhood ω_kIn the linear coefficient that remains unchanged, i is neighborhood ω_kIn any one pixel,I is each pixel value image equal to 1, and H is the point spread function of PET imaging device,Accord with for convolution operation.

In the present invention, generally assume that the output image Q of model_iBeing centrally located at pixel k place, it is the guide structure image G window ω being positioned at pixel k after volume effect convolution_kLinear transformation, thus derive guiding volume effect straightening die pattern (II).

In described step 1.6, the equation of belt restraining object function is formula (III),

$E(a_k,b_k)=\underset{i\∈{\mathrm{\ω}}_k}{\Σ}({\left(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median}\right)}^2+{\mathrm{\ϵa}}_k^2)---\left(III\right)$

Wherein, ε is regularization parameter, is used for controlling a_kSpan；A_kAnd b_kFor at neighborhood ω_kIn the linear coefficient that remains unchanged, i is neighborhood ω_kIn any one pixel,I is each pixel value image equal to 1, and H is the point spread function of PET imaging device,Accord with for convolution operation.

The span of described regularization parameter ε is 0≤ε≤1.

In the present invention, due to needs denoising as far as possible, therefore minimize noise and meet constraint equation and can obtain the equation of needs.Specifically, filter the image obtained and by the equation being minimized in native window and strengthen a by ε_kWeight obtain.

In the present invention, compared with the resolution of PET, the imaging resolution of structural images such as CT or MR image is much higher, and it is often deemed to be impulse response, it is contemplated that the resolution between MRI and PET is not mated, therefore proposes the equation of above-mentioned belt restraining object function.

In the present invention, when identifying image, According with for convolution operation, H is the point spread function of the space invariance of PET imaging device, and it is by the approximate match being widely used as anisotropic Gaussian function, and in the problem guiding filtering, a_kAnd b_kLocal mean value Yong Lai not update each voxel.

In described step 1.7, according to formula (III), to a_kAnd b_kTake partial differential and be set as 0, obtaining formula (IV) and formula (V),

$\frac{\∂E}{\∂a_k}=2\underset{i\∈{\mathrm{\ω}}_k}{\Σ}\left(\right(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i){\left(HG\right)}_i+{\mathrm{\ϵa}}_k)=0---\left(IV\right)$

$\frac{\∂E}{\∂b_k}=2\underset{i\∈{\mathrm{\ω}}_k}{\Σ}{\left(HI\right)}_i(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i)=0---\left(V\right)$

Making described output PET image is formula (IV),

Q_i=a_k(HG)_i+b_k(HI)_i(IV)

Wherein, $a_k=\frac{{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i\right){\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2-{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right){\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HI\right)}_i{P}_i\right)}{\left|\mathrm{\ω}\right|\mathrm{\ϵ}{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2+{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HG\right)}_i^2-{\left({\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HG\right)}_i{\left(HI\right)}_i\right)}^2},$

$b_k=\frac{{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i\right)-a_k{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right)}{{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2}.$

In the present invention, formula (VII) can be obtained from formula (V),

$b_k=\frac{{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i\right)-a_k{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right)}{{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2}---\left(VII\right),$

Bring formula (VII) into formula (IV) and following formula can be obtained,

$a_k\underset{i\∈{\mathrm{\ω}}_k}{\Σ}({\left({\left(HG\right)}_i\right)}^2+\mathrm{\ϵ}\underset{i\∈{\mathrm{\ω}}_k}{\Σ}I)+\frac{1}{{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2}(\underset{i\∈{\mathrm{\ω}}_k}{\Σ}{\left(HG\right)}_i{P}_i-a_k\underset{i\∈{\mathrm{\ω}}_k}{\Σ}{\left(HI\right)}_i{\left(HG\right)}_i)\underset{i\∈{\mathrm{\ω}}_k}{\Σ}{\left(HI\right)}_i{\left(HG\right)}_i-\underset{i\∈{\mathrm{\ω}}_k}{\Σ}{P}_i{\left(HG\right)}_i=0$

, obtain $a_k=\frac{{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i\right){\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2-{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right){\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HI\right)}_i{P}_i\right)}{\left|\mathrm{\ω}\right|\mathrm{\ϵ}{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2+{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HG\right)}_i^2-{\left({\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HG\right)}_i{\left(HI\right)}_i\right)}^2},$

$b_k=\frac{{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i\right)-a_k{\Σ}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right)}{{\Σ}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2}.$

The inventive method is simple, and computing is convenient, and image definition has higher guarantee, and is different from prior art.

Present invention solves the technical problem that and be, low signal-to-noise ratio that PET image is brought due to the restriction of implantation dosage and scan sensitivity and the volume effect brought due to the spatial resolution of imaging system and the restriction of tissue fraction effect (tissuefractioneffect).Although most PET imaging systems have employed the iterative reconstruction algorithm based on maximal possibility estimation, but the volume effect that the PET image rebuild yet suffers from the spatial resolution of low signal-to-noise ratio problem and the imaging system brought due to the restriction of implantation dosage and system sensitivity and tissue fraction effect limits and brings.Although noise can be suppressed by early stage stopping iteration or by adopting wave filter to suppress after rebuilding, but early stage stops alternative manner and tends not to reach algorithm for reconstructing convergence thus tending not to the image that enough generations are detailed.Gaussian filtering denoising method is that after reconstruction the most frequently used in current clinic, image filtering method but gaussian filtering are likely to while eliminating noise and eliminate important picture structure, this will further decrease spatial resolution, thus the problem weakening the quantified precision of power of test and little structure.The invention provides a kind of PET image volume effect bearing calibration guided based on structural images., by structural images G and PET image P is carried out Rigid Registration position corresponding after, PET image P is carried out medium filtering, and structural images G is defined the effective neighborhood ω of voxel_k, utilize the PET image P after medium filtering^medianNeighborhood ω effective in voxel_kDefinition guides volume effect calibration model, and utilize structural images G as navigational figure, convert guiding volume effect calibration model, adopt linear regression method to carry out minimizing solving the equation obtaining belt restraining object function, obtain the output PET image after denoising and partial volume effect correction；The present invention utilizes high-resolution structural images guide PET image denoising and reduce partial volume effect, adopt non-iterative analytical Calculation mode, holistic approach is more easy, adjustment PET image is chosen by the self adaptation of effective neighborhood, the formation of noise is suppressed, it is ensured that the power of test of PET and the quantified precision of little structure while improving the image resolution ratio of PET imaging.

Claims (10)

1. the PET image partial volume correction method guided based on structural images, it is characterised in that: said method comprising the steps of:

Step 1.1: be utilized respectively structural images imaging device and PET imaging device gathers the structural images of same object and the initial data of PET, obtain the structural images G and PET image P of object, obtain the resolution of structural images and PET imaging device simultaneously；

Step 1.2: the structural images G obtained in step 1.1 and PET image P is carried out Rigid Registration；

Step 1.3: the PET image P obtained in step 1.1 is carried out medium filtering, obtains the PET image P after medium filtering^median；

Step 1.4: the structural images G obtained in step 1.2 is defined the effective neighborhood ω of voxel_k；

Step 1.5: utilize the P obtained in step 1.3^medianNeighborhood ω effective in the voxel obtained in step 1.4_kDefinition guides volume effect calibration model；

Step 1.6: the guiding volume effect calibration model of step 1.5, as navigational figure, is converted, obtains the equation of belt restraining object function by structural images G step 1.2 obtained；

Step 1.7: the equation of belt restraining object function step 1.6 obtained adopts linear regression method to carry out minimizing solving, obtains the output PET image after denoising and partial volume effect correction.

2. a kind of PET image partial volume correction method guided based on structural images according to claim 1, it is characterised in that: described structural images G is CT or MR image.

3. a kind of PET image partial volume correction method guided based on structural images according to claim 1, it is characterised in that: in described step 1.3, medium filtering adopts formula (I),

P^median=Median (P, ω_m)(I)

Wherein, ω_mFor medium filtering window, P^medianPET image after filtering for medium filtering.

4. a kind of PET image partial volume correction method guided based on structural images according to claim 3, it is characterised in that: in described step 1.3, ω_mMedium filtering window for (3 × 3 × 3).

5. a kind of PET image partial volume correction method guided based on structural images according to claim 1, it is characterised in that: in described step 1.4, for pixel k and its neighborhood N_kEffective neighborhood ω according to following rule definition pixel k_k:

Neighborhood N for pixel k each in 3-D view_kIn each pixel i, if meetIt is classified as effective neighborhood ω_k, wherein μ is the self-defined edge decision threshold in structural images G.

6. a kind of PET image partial volume correction method guided based on structural images according to claim 5, it is characterised in that: the span of described self-defined edge decision threshold μ is 0.5≤μ≤1.

7. a kind of PET image partial volume correction method guided based on structural images according to claim 1, it is characterised in that: in step 1.5, guiding volume effect calibration model is formula (II),

$Q_i=a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median}---\left(II\right)$

Wherein, a_kAnd b_kFor at neighborhood ω_kIn the linear coefficient that remains unchanged, i is neighborhood ω_kIn any one pixel,I is each pixel value image equal to 1, and H is the point spread function of PET imaging device,Accord with for convolution operation.

8. a kind of PET image partial volume correction method guided based on structural images according to claim 1, it is characterised in that: in described step 1.6, the equation of belt restraining object function is formula (III),

$E(a_k,b_k)=\underset{i\∈{\mathrm{\ω}}_k}{\Σ}({\left(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median}\right)}^2+{\mathrm{\ϵa}}_k^2)---\left(III\right)$

Wherein, ε is regularization parameter, is used for controlling a_kSpan；A_kAnd b_kFor at neighborhood ω_kIn the linear coefficient that remains unchanged, i is neighborhood ω_kIn any one pixel,I is each pixel value image equal to 1, and H is the point spread function of PET imaging device,Accord with for convolution operation.

9. a kind of PET image partial volume correction method guided based on structural images according to claim 8, it is characterised in that: the span of described regularization parameter ε is 0≤ε≤1.

10. a kind of PET image partial volume correction method guided based on structural images according to claim 6, it is characterised in that: in described step 1.7, according to formula (III), to a_kAnd b_kTake partial differential and be set as 0, obtaining formula (IV) and formula (V),

$\frac{\∂E}{\∂a_k}=2\underset{i\∈{\mathrm{\ω}}_k}{\Σ}\left(\right(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median}){\left(HG\right)}_i+{\mathrm{\ϵa}}_k)=0---\left(IV\right)$

$\frac{\∂E}{\∂b_k}=2\underset{i\∈{\mathrm{\ω}}_k}{\Σ}{\left(HI\right)}_i(a_k{\left(HG\right)}_i+b_k{\left(HI\right)}_i-P_i^{median})=0---\left(V\right)$

Making described output PET image is formula (IV),

Q_i=a_k(HG)_i+b_k(HI)_i(IV)

Wherein, $a_k=\frac{{\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i^{median}\right){\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2-{\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right){\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}\left({\left(HI\right)}_i{P}_i^{median}\right)}{\left|\mathrm{\ω}\right|{\mathrm{\ϵ\Σ}}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2+{\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2{\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}{\left(HG\right)}_i^2-{\left({\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}{\left(HG\right)}_i{\left(HI\right)}_i\right)}^2},$

$b_k=\frac{{\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{P}_i^{median}\right)-a_k{\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}\left({\left(HG\right)}_i{\left(HI\right)}_i\right)}{{\mathrm{\Σ}}_{i\∈{\mathrm{\ω}}_k}{\left(HI\right)}_i^2}.$