CN103034989A - Low-dosage CBCT (Cone Beam Computed Tomography) image denoising method based on high-quality priori image - Google Patents

Abstract

The invention relates to a low-dosage CBCT (Cone Beam Computed Tomography) image denoising method based on a high-quality priori image. The method comprises the following steps: needing to obtain a high-quality high signal to noise ratio three-dimensional volume data with the same scanning part, and using the high-quality image as the denoising priori information; after each CBCT scanning, processing voxels one by one in a reconstructed image; aiming at some voxel to be denoised, selecting peripheral zones centered on the voxel and forming an image block, and calculating an SSIM (structural similarity) value to find image blocks similar with the image block from the high-quality priori image; weighted averaging the center voxels of all the similar image blocks, and therefore obtaining the result of the denoised voxel; calculating all the voxels one by one, and finally obtaining the denoised CBCT three-dimensional image. As the method uses the high-quality high signal to noise ratio image as the priori image, the denoising effect can be greatly increased, and the registration is not needed in the denoising process. Each scanning can use the same one priori image to process, and the feasibility of the method is enhanced.

Description

A kind of low dosage CBCT image de-noising method based on the high-quality prior image

Technical field

The present invention relates to radiant image and technical field of image processing, relate to CBCT image denoising disposal route, particularly a kind of low dosage CBCT image de-noising method based on the high-quality prior image.

Background technology

Noise is to affect CBCT(Cone-beam CT, conical beam CT) key factor of picture quality, especially (low dosage is herein without clear and definite especially official definition at low dosage, it is generally acknowledged, when being lower than the employed normal dose of clinical diagnosis CT, namely can be called low dosage) scanning situation under, it is more serious that noise becomes, this moment, it may bury in oblivion focus, had influence on diagnosis, and it is very important therefore to study suitable Image Denoising Technology.At present, the denoising thinking for low dosage CT mainly comprises the projection domain denoising, choose reasonable filter function or iterative constrained item in process of reconstruction, the denoising of reconstructed image territory.Specific to denoising method, at present the comparatively popular minimized denoising method of the denoising method that comprises wavelet multiresolution, TV, based on the denoising method of partial differential equation etc.These methods can reduce the noise level of image effectively, improve the SNR value of image, but also bring inevitably image blurring and loss details simultaneously, as shown in Figure 1.Therefore, the new image processing algorithm of research keeps detailed information very important for promoting CBCT picture quality simultaneously.

In recent years, along with the development of mathematical theory and image analysis technology, obtained rapidly people's concern based on the denoising method of redundant data and image similarity.Most representative is exactly that the non-local mean method that proposes of Buades A is (referring to A.Buades, B.Coll, J.M.Morel.A review of image denoising algorithms, with a new one.Multiscale Model.Simul, 2005,4 (2): 490-530.), it has adopted the image natural redundancies, utilize the similar area in every width of cloth image to carry out denoising, can obtain good reservation to details and marginal information.The method may be summarized as follows: for a given Noise image v={v (i) | i ∈ I}, I presentation video coordinate system.Estimated value after using weighted mean that the non-local mean algorithm estimates all pixels of full figure as this denoising for pixel i so:

$\mathrm{NL}\left[v\right]\left(i\right)=\frac{1}{C_i}\underset{j\∈I}{\mathrm{\Σ}}w(i,j)v\left(j\right)$

Wherein, weighted value w (i, j) depends on the similarity between pixel i and the pixel j.C _iIt is normalization coefficient.How similarity and the choose reasonable weight between decision block and the piece is the key of this algorithm, and general similarity can be determined by the Gauss's weighted euclidean distance between neighborhood gray-scale value vector:

Show between two pixels more similarly if distance is nearer, then weighting weight is larger.Weighting function can be set is: w (i, j)=exp (d (i, j)/h ²).Here h is filtering parameter, the speed of control characteristic function decay.This algorithm can reduce the noise level in the image effectively, and keeps detailed information and marginal information.But for the strong data of noise amplitude, its denoising effect is still limited.

The introducing of prior imformation has very important meaning for image denoising.If can obtain the similar image of a width of cloth high-quality, just can effectively recover Noise information.Someone has proposed a kind of ODCT(Oracle-based DCT filter) hard-threshold filtering noise reduction algorithm, use first one-level dct transform threshold denoising to obtain clean image, carry out again secondary filter one time, thereby obtain clean image (referring to Onur G.Guleryuz, Weighted Averaging for Denoising WithOvercomplete Dictionaries.[J] .IEEE Trans.Image Process., 2007.16 (12): 3020-3034).In addition, also have a kind of method based on dictionary learning, train dictionary (a kind of sparse expression of image) by clean image first, more noisy image is carried out denoising.But in Medical Image Processing, high-quality image often is difficult to obtain.In addition, even obtained high-quality image by again scanning, also need to carry out the processing such as registration, assurance structure and edge mate just can be used for the denoising operation fully, and difficulty is very large, lacks practicality.

At present, utilize the patented claim of the image de-noising method of non-local mean to have a lot, such as " the non-local mean image de-noising method of integrated structure information " (application number: 201110091450.2), " non-local mean de-noising method for natural image " (application number: 201010271499.1), and " based on the non-local mean denoising method of associating similarity " (application number: 201110282126.9) etc.But these methods are all carried out denoising based on the redundancy of image self, do not introduce suitable prior image.When the priori image is introduced, can effectively promote the effect of denoising.In the denoising method for low dosage CT, invention " data for projection recovers the non local average low dosage CT method for reconstructing of guiding " (application number: 201010517537.7) proposed a kind of disposal route, obtain a low noise reconstructed image with low dosage CT projection sinogram by BM3D denoising and reconstruction first, be weighted the average filter denoising with this width of cloth image for the low dosage strong noise image of direct reconstruction, thereby realize the effect that low dosage CT rebuilds.Invention " based on the low dosage CT image processing method of wavelet space directivity filtering " (application number: 201010595896.4) proposed a kind of image de-noising method based on wavelet space that carries out in projection domain.Similarly comprise " based on the low dosage CT image rebuilding method of standard dose image redundancy information " (application number: 201010555893.8) with the present invention.The method provides the image of a secondary normal dose by previous scan-data, carries out low-dose CT scanning one time again.Image and the low dosage CT of normal dose are carried out registration, and then carry out denoising by the similarity between the image block.It is twice different scanning of same object that the method requires prior image and low dosage image, and having relatively high expectations for registration.And the present invention requires prior image to get final product to the similar position of low dosage image scanning, can be from different patients, and do not need accurate registration process.

Summary of the invention

The technology of the present invention is dealt with problems: overcome the deficiencies in the prior art, a kind of low dosage CBCT image de-noising method based on the high-quality prior image is provided, adopted the high s/n ratio qualitative picture as priori, utilize in the priori data and the block similarity matching for the treatment of denoising low dosage CBCT, so that denoising effect is improved significantly, solve the excessive problem of low dosage CBCT picture noise, improve picture quality, improve the signal to noise ratio (S/N ratio) of image, do not lose simultaneously detailed information and marginal texture information as far as possible.

The technology of the present invention solution: a kind of principle of the low dosage CBCT image de-noising method based on the high-quality prior image is, but in the CBCT image for the scanning of Different Individual same area, because the difference of human body, scanning position, scanning mode, scanning angle etc. are all distinct, although two width of cloth images are similar positions, its one-piece construction difference is still very large.But little image block is processed and when mating, wherein have the similar part of many height when two width of cloth images are divided into.The present invention has then utilized the similarity of these image blocks, for treating that denoising low dosage CBCT image carries out similar block search by tissue points in prior image, and the similar image piece that searches is weighted average treatment, thereby reaches the purpose of denoising.Because prior image has good picture quality, so the introducing of its information can effectively improve picture quality.Simultaneously, the present invention has avoided the step of image registration, and the exact matching that does not need to carry out prior image He treat the denoising image has reduced the requirement for prior image, the error of having avoided the registration out of true to bring.

The inventive method specific implementation step is as follows:

The first step: the obtaining of high-quality prior image

Use the image of high s/n ratio as the prior imformation of denoising calculating.Requirement to this image comprises the following aspects: (1) has similar structural information and similar intensity profile to treating the denoising image.Therefore, require and treat that denoising image scanning position is similar, thereby guarantee that CBCT rebuilds the volume data that obtains and treat that the denoising image has higher similarity.In general, for the CBCT equipment of same use, such as oral cavity CBCT, the oromaxillo-facial region data that all gather zone similarity get final product.(2) this image need to have preferably picture quality, comprises high s/n ratio, more rich detailed structure information.

For these two needs, can get access to prior image and as the denoising database by following several modes.

(1) select higher X-ray tube electric current to scan;

(2) each projection angle Multiple-Scan stack;

(3) slow down rotational speed, gather the more projected angle number of degrees between certain Rotary District;

(4) other can improve the means of CBCT picture quality.

Second step: image denoising

(1) for prior image P with treat that the denoising data carry out pre-service, carries out simple location matches.This matching process can manually be finished, and also can use the automatic calculating method of image centroid and main shaft coupling, so that corresponding organ site and the angle of two width of cloth images approaches.

(2) in treating denoising CBCT data reconstruction, choose and treat denoising tissue points x.The size of choosing centered by x is the image block v(x of n * n).

(3) in prior image, search for.Block-by-block calculate in the prior image centered by tissue points y image block v(y) and structural similarity sex index SSIM (x, y) v(x).

$\mathrm{SSIM}(x,y)=\frac{\left({2\mathrm{\μ}}_x{\mathrm{\μ}}_y\right)({2\mathrm{\σ}}_{\mathrm{xy}}+c_2)}{({\mathrm{\μ}}_x^2+{\mathrm{\μ}}_y^2+c_1)({\mathrm{\σ}}_x^2+{\mathrm{\σ}}_y^2+c_2)}$

μ wherein _x, μ _yBe respectively the voxel mean value of two image blocks, σ _x, σ _yBe respectively the standard deviation of two image block voxel values, σ _XyBe the covariance of two image block voxel values, c ₁=(k ₁L) ², c ₂=(k ₂L) ², L is the dynamic range of two image block gray-scale values to be calculated;

(4) setting threshold ε is as the standard of similarity judgement.As SSIM (x, y)〉during ε, think that then two image block centrosome vegetarian refreshments are similar, then this tissue points can be used for pending tissue points x is weighted average denoising.Otherwise, and think two tissue points dissmilarities, in the weighted mean process, do not use.

(5) choose all similitude data, for treating that denoising tissue points x is weighted on average, the data after the acquisition denoising:

$v^{\′}\left(x\right)=\frac{1}{C\left(x\right)}\underset{y\∈P}{\mathrm{\Σ}}w(x,y)v\left(y\right)$

Y is the tissue points in the noisy image in all similar areas that search, and w (x, y) is weighting function, and C (x) is the weight normalized factor, and its computing method are:

$C\left(x\right)=\underset{y\∈P}{\mathrm{\Σ}}w(x,y).$

The computing method of weighting function w (x, y) are also according to SSIM(x, y) carry out conversion and obtain.Because SSIM(x, y) value be [1,1], and when its value absolute value near 1 the time, two image blocks are more similar.And traditional non-local mean adds temporary, and distance is that 0 o'clock two image block is more similar, and weighted value is maximum.Therefore, definition Similarity Structure parameter: S (x, y)=1-|SSIM (x, y) |.Weighting function can be written as:

w(i,j)=exp(S(x,y)/h ²)

Wherein h is filtering parameter, generally becomes certain linear relationship with the poor σ of prior image noise criteria.

The method can directly be carried out denoising to three-dimensional data.In addition, the similarity of judgement image block also can be utilized other similar approach in the second step, is not confined to computation structure index of similarity SSIM, and the present invention is just as example.

In addition, use the method, can obtain a kind of CBCT image de-noising method of based on database.Namely by the clinical scanning data, set up a series of patients database, the image in the database all uses normal dose to obtain, and has preferably signal to noise ratio (S/N ratio) and picture quality.After this Database, when carrying out low dosage CBCT image denoising, can choose the most similar image in database, flow process is processed according to the method described above.This process can be finished online by network.

The present invention's advantage compared with prior art is:

(1) the present invention has adopted the high s/n ratio qualitative picture as priori, while combining image denoising process, so that denoising effect is improved significantly, solve the excessive problem of CBCT picture noise, improve picture quality, improve the signal to noise ratio (S/N ratio) of image, do not lose simultaneously detailed information and marginal texture information as far as possible.

(2) the present invention requires lower for prior image, Structure of need information accuracy registration not, only need scanning the position be close to get final product, do not need to carry out registration in the denoising process, and each scanning all can use same width of cloth prior image to process, strengthen the feasibility of method, reduced the difficulty of implementing.In actual applications, can set up first the prior data bank of a cover high-quality.In the clinical practice, can share a cover prior image information for different patients' scanning and carry out denoising.

Description of drawings

The confirmatory experiment of Fig. 1 for the inventive method is carried out, upper figure is employed high-quality prior image, figure below is the low dosage CT image for the treatment of denoising;

Fig. 2 is for using the inventive method to carry out the as a result comparison diagram of denoising front and back;

Fig. 3 amplifies comparison diagram for the part of using the inventive method to carry out the denoising front and back;

Fig. 4 is realization flow figure of the present invention;

The confirmatory experiment of Fig. 5 for using oral cavity CBCT clinical data to carry out, left figure is employed high-quality prior image, right figure is the low dosage CBCT image for the treatment of denoising;

Fig. 6 is for using the inventive method oral cavity CBCT low dose imaging data to be carried out the as a result comparison diagram of denoising front and back;

Fig. 7 amplifies comparison diagram for the part of using the inventive method that oral cavity CBCT low dose imaging data are carried out the denoising front and back.

Embodiment

As shown in Figure 3, the inventive method was divided into for two steps, and is specific as follows:

The first step is obtaining of high-quality prior image, the image denoising of second step for carrying out based on this prior image.For different patients' CBCT scanning, if scanning area and organ-tissue structure are similar, generally can adopt same width of cloth prior image to process.Because CBCT scanning obtains is three-dimensional data, therefore for be that three-dimensional data is processed.But clear and easy for what narrate, the below uses two-dimentional tomography that the method is described.The method can directly extend to three-dimensional situation.

Be specially:

The first step: the obtaining of high-quality prior image

The prior imformation that this step need to use the image of high s/n ratio to calculate as denoising.Requirement to this image comprises the following aspects: (1) has similar structural information and similar intensity profile to treating the denoising image.Therefore, require and treat that denoising image scanning position is similar, thereby guarantee that CBCT rebuilds the volume data that obtains and treat that the denoising image has higher similarity.In general, scan same position and obtain prior image and get final product, oral cavity CBCT for example, two width of cloth images all gather the oral jaw face data and get final product.(2) this image need to have preferably picture quality, comprises high s/n ratio, more rich detailed structure information.

For these two needs, can get access to prior image and as the denoising database by following several modes.

(1) select higher X-ray tube electric current to scan;

(2) each projection angle Multiple-Scan stack;

(3) slow down rotational speed, gather the more projected angle number of degrees between certain Rotary District;

(4) use multi-detector diagnosis CT generating three-dimensional volume data, and it is proofreaied and correct, CT value and space voxel size are corresponded in the corresponding CBCT system, as prior image.

Second step: image denoising

(1) for prior image P with treat that the denoising data carry out pre-service, carries out simple position and slightly mate.This matching process can manually be finished, and also can use the automatic calculating method of image center of gravity and main shaft coupling, so that organ site corresponding to two width of cloth images and angle approach, thereby can reduce the size of region of search.

(2) in treating denoising CBCT data reconstruction, choose and treat denoising tissue points x.The size of choosing centered by x is the image block v(x of n * n).

(3) in prior image, search for.Block-by-block calculate in the prior image centered by voxel y image block v(y) and structural similarity sex index SSIM (x, y) v(x).

$\mathrm{SSIM}(x,y)=\frac{\left({2\mathrm{\μ}}_x{\mathrm{\μ}}_y\right)({2\mathrm{\σ}}_{\mathrm{xy}}+c_2)}{({\mathrm{\μ}}_x^2+{\mathrm{\μ}}_y^2+c_1)({\mathrm{\σ}}_x^2+{\mathrm{\σ}}_y^2+c_2)}$

μ wherein _x, μ _yBe respectively the voxel mean value of two image blocks, σ _x, σ _yBe respectively the standard deviation of two image block voxel values, σ _XyBe the covariance of two image block voxel values, c ₁=(k ₁L) ², c ₂=(k ₂L) ², L is the dynamic range of two image block gray-scale values to be calculated;

(4) setting threshold ε is as the standard of similarity judgement.As SSIM (x, y)〉during ε, think that then two image blocks are similar data, the voxel piece in the prior image can be used for tissue points x is carried out denoising.

(5) choose all similitude data, for treating that denoising tissue points x is weighted on average, the data after the acquisition denoising:

$v^{\′}\left(x\right)=\frac{1}{C\left(x\right)}\underset{y\∈P}{\mathrm{\Σ}}w(x,y)v\left(y\right)$

Y is the tissue points in the noisy image in all similar areas that search, and w (x, y) is weighting function, and C (x) is the weight normalized factor.Wherein, the computing method of weighting function are also according to SSIM(x, y) carry out conversion and obtain.Because SSIM(x, y) value be [1,1], and when its value absolute value near 1 the time, two image blocks are more similar.And traditional non-local mean adds temporary, and distance is that 0 o'clock two image block is more similar, and weighted value is maximum.Therefore, definition Similarity Structure parameter: S (x, y)=1-|SSIM (x, y) |.Weighting function can be written as:

w(i,j)=exp(S(x,y)/h ²)

Wherein h is filtering parameter, and the poor σ of noise criteria general and in the prior image becomes certain linear relationship.

The similarity of judgement image block also can be utilized other similar approach in the second step, is not confined to computation structure index of similarity SSIM, as using the modes such as Euclidean distance of Gauss's weighting.The present invention is just as example.

The applicant uses the CT database to carry out confirmatory experiment.Fig. 1 is the prior image that uses and the low dosage CT experiment for the treatment of denoising, and upper figure is the high-quality prior image, and figure below is the low dosage CT image for the treatment of denoising.Its scanning be similar position, but come from different patients.Fig. 2 is image after the denoising and the contrast of low dosage original image, and left figure is original image, and right figure is the image after the denoising.Fig. 3 is the contrast of partial enlarged drawing picture, and left figure is original low dosage image, and right figure is the image after the denoising.Can find out by Detail contrast, use the inventive method, can effectively suppress noise, improve picture quality.

In addition, the applicant also makes the oral cavity CBCT of LifeView instrument development carry out the denoising experiment of clinical data.As prior image, simultaneously, this image uses traditional Non Local Means method to carry out denoising, has guaranteed signal to noise ratio (S/N ratio) and the quality of prior image with one group of X-ray tube voltage 100Kev, tube current oromaxillo-facial region scan rebuilding data that are 4mA.Treat that the denoising image is the low dosage CBCT image that arrives that adopts 100Kev tube voltage, 2mA tube current to scan.The left figure of Fig. 5 is the prior image that uses, and right figure is the low dosage CBCT image for the treatment of denoising.Fig. 6 can find to use this kind method to guarantee detailed information in denoising for the image that uses the denoising of this kind method and obtain and the contrast of former low dosage CBCT image.Fig. 7 is topography's contrast of amplifying before and after the denoising.Can find that use picture quality and the signal to noise ratio (S/N ratio) of the present invention under can Effective Raise CBCT low-dose scanning, more traditional method can not brought image blurring, has preserved image detail and edge.

The present invention is with respect to former various denoising methods, and owing to the prior image of having introduced high s/n ratio, so its denoising ability is stronger, and has utilized the similarity of image-region, and is better for the preservation of detailed information.Prior image of the present invention is selected more flexible, all can use identical prior image for each scanning of different patients, and practicality is stronger.

The non-elaborated part of the present invention belongs to techniques well known.

The above; only be part embodiment of the present invention, but protection scope of the present invention is not limited to this, any those skilled in the art are in the technical scope that the present invention discloses; the variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.

Claims (4)

1. low dosage CBCT image de-noising method based on the high-quality prior image is characterized in that performing step is as follows:

The first step: the prior image P of high-quality obtains

Adopt the image of high s/n ratio to calculate prior imformation as denoising; The image request of described high s/n ratio comprises following: (1) has similar structural information and similar intensity profile to treating the denoising image, to treat that denoising image scanning position is similar, thereby guarantee that CBCT rebuilds the volume data that obtains and treat that the denoising image has high similarity; (2) this image need to have high s/n ratio and abundant detailed structure information; Described high s/n ratio is through experiment, should be than the low dosage CBCT figure image height for the treatment of denoising more than 50%;

Second step: image denoising

(21) for prior image P with treat that the denoising data carry out pre-service, carrying out the position slightly mates, this matching process can manually be finished, and also can use the automatic calculating method of image centroid and main shaft coupling, so that corresponding organ site and the angle of two width of cloth images approaches;

(22) in treating denoising CBCT data reconstruction, choose and treat denoising voxel x, the size of choosing again centered by x is the image block v(x of n * n * n);

(23) search in prior image P, block-by-block calculates the image block v(y centered by voxel y among the prior image P) and structural similarity sex index SSIM (x, y) v(x):

$\mathrm{SSIM}(x,y)=\frac{\left({2\mathrm{\μ}}_x{\mathrm{\μ}}_y\right)({2\mathrm{\σ}}_{\mathrm{xy}}+c_2)}{({\mathrm{\μ}}_x^2+{\mathrm{\μ}}_y^2+c_1)({\mathrm{\σ}}_x^2+{\mathrm{\σ}}_y^2+c_2)}$

μ wherein _x, μ _yBe respectively the voxel mean value of two image blocks, σ _x, σ _yBe respectively the standard deviation of two image block voxel values, σ _XyBe the covariance of two image block voxel values, c ₁=(k ₁L) ², c ₂=(k ₂L) ², L is the dynamic range of two image block gray-scale values to be calculated;

(24) standard judged as similarity of setting threshold ε is as SSIM (x, y)〉during ε, think that then two image blocks are similar, its center voxel can come tissue points x is carried out denoising by weighted mean; Otherwise, then think and when weighted average calculation, do not adopt two image block dissmilarities;

(25) the center voxel that uses all similar image pieces that search in the previous step is for treating that denoising tissue points x is weighted on average, thus the voxel value after the acquisition denoising, and the formula of weighted average calculation is as follows:

$v^{\′}\left(x\right)=\frac{1}{C\left(x\right)}\underset{y\∈P}{\mathrm{\Σ}}w(x,y)v\left(y\right)$

Wherein, w (x, y) is weighting function, the computing method of weighting function are according to SSIM(x, y) carry out conversion and obtain, because SSIM(x, y) value is [1,1], and when its value absolute value near 1 the time, two image blocks are more similar, definition Similarity Structure parameter: S (x, y)=1-|SSIM (x, y) |, weighting function is written as:

w(i,j)=-exp(S(x,y)/h ²)

Wherein h is filtering parameter, and poor σ is linear with noise criteria, and C (x) is the weight normalized factor, and its computing method are: $C\left(x\right)=\underset{y\∈P}{\mathrm{\Σ}}w(x,y).$

2. a kind of low dosage CBCT image de-noising method based on the high-quality prior image according to claim 1, it is characterized in that: described method can directly be carried out denoising to the volume data of three-dimensional.

3. a kind of low dosage CBCT image de-noising method based on the high-quality prior image according to claim 1 is characterized in that: the similarity of judging image block in the described second step also can be utilized the Euclidean distance computing method.

4. a kind of low dosage CBCT image de-noising method based on the high-quality prior image according to claim 1 is characterized in that: get access to prior image by one of following several modes in the described first step:

(1) select the X-ray tube electric current to scan;

(2) each projection angle Multiple-Scan stack;

(3) slow down rotational speed, gather the more projected angle number of degrees between certain Rotary District;

(4) the use spiral CT scans and obtains the prior image of high s/n ratio, and resulting CT value and spatial discrimination are proofreaied and correct, and corresponds in the CBCT system that treats denoising.

Images