CN103226815A - Low dose CT image filtering method - Google Patents

Abstract

The invention discloses a low dose CT (Computerized Tomography) image filtering method, and belongs to the technical field of medical image processing. The method comprises the steps that a dictionary learning method is used for filtering block noise in the original low dose CT image, and an obtained image is subjected to unsharp filtering for enhancing image border information inhibited by the dictionary learning method. The method can effectively inhibit star strip artifacts and noise in the low dose CT image, and improve the quality of the CT image.

Description

A kind of low dosage CT image filtering method

Technical field

The present invention relates to a kind of image filtering method, relate in particular to a kind of low dosage CT image filtering method, belong to technical field of medical image processing.

Background technology

As the effective clinical diagnose instrument of present a kind of routine, X ray computer fault imaging (X-ray Computerized Tomography, CT) can on mm-scale, obtain the dampening information of human body different tissues clearly, thereby provide abundant 3 D human body organ-tissue information for clinician's diagnosis and prevention for X ray.Today, as a kind of maturation and universally recognized inspection method clinically, CT has become one of indispensable main tool in the radiodiagnosis field.Yet, along with CT tomoscan popularizing in clinical diagnosis especially routine inspection, radiation dose problem in the CT scan has caused that people more and more pay close attention to, and vast amount of clinical shows that paranormal CT radiation dose easily brings out diseases such as human body pathobolism and even cancer.Yet, in present CT equipment, to increase the block distortion of reconstructed image and star strip artifact by adjustable pipe electric current and voltage to reduce CT scan dosage, and reduce the quality that CT rebuilds, thereby influence the diagnosis rate of clinician abnormal structure with directivity.

The algorithm of current raising low dosage CT picture quality mainly is divided into based on the projector space data and the two big classes image space data, algorithm based on the projector space data mainly comes to provide noise data for projection still less for reconstruction by the recovery denoising of CT data for projection, to improve the image reconstruction quality under the low dosage condition, for example current (the Filtered Backprojection of filtered back projection that in Clinical CT is rebuild, generally uses, FBP) algorithm, handle by built-in frequency domain filtering exactly and suppress pseudo-shadow and noise, also have some scholars to propose to set up data model and set up recovery algorithms based on this and suppress the noise in the CT data for projection under the low dosage condition at projector space.The research of these class methods is difficult in the reality be used widely owing to be subjected to the restriction that commercial CT data for projection can't obtain.Another kind of method is directly to improve the low dosage picture quality of having rebuild by the image space treatment technology, have and do not rely on original projection data and the fast advantage of processing speed, usually the denoising of using nonlinear disposal route to keep image edge information, method as anisotropic filter (nonlinear or anisotropic filter) or small echo (wavelet), yet this quasi-nonlinear side is mainly based on the local message of image, be difficult to obtain good treatment effect, for example, block distortion and star strip artifact in the low dosage CT image can't be effectively suppressed, also new pseudo-shadow can be in processing, produced.

Propose recently based on big neighborhood territory pixel weighted mean (Weighted Intensity Averaging over Large-scale Neighborhoods, WIA-LN) image denoising algorithm belongs to second class methods, the method can utilize the global information of CT image area to obtain better denoising effect, its principle is to think in the Clinical CT image, the pixel that belongs to different organ or decay tissue often in image distribution in a bigger yardstick, and the pixel that belongs to homolog or decay tissue often has similar neighborhood information, so each pixel is weighted the noise that on average can effectively suppress in the image at one according to its surrounding tissue similarity in than large scale.Definition f _iWith

Represent the gray-scale value of pixel i in former low dosage CT figure and the processing back low dosage CT figure respectively, the thinking of WIA-LN method is as follows:

${\hat{f}}_i=\underset{j\∈N_i}{\mathrm{\Σ}}{w}_{\mathrm{ij}}{f}_j/\underset{j\∈N_i}{\mathrm{\Σ}}{w}_{\mathrm{ij}}---\left(1\right)$

$w_{\mathrm{ij}}=\exp (-\frac{{\left|\right|n_i-n_j\left|\right|}_{2,\mathrm{\α}}^2}{\mathrm{\β}\left|n\right|})---\left(2\right)$

Here, f _jExpression search neighborhood N _iThe gray-scale value of interior pixel point i, the weight w in (2) _IjSimilarity between remarked pixel point i and the pixel j is estimated, by calculating neighborhood N _iInterior pixel point i and the pixel j institutional framework similarity around separately Obtain.Here n _iAnd n _jBe the similarity neighborhood of remarked pixel point i and the pixel j tissue around separately,

Be similarity piece n _iAnd n _jBetween Gauss's Weighted distance, wherein α represents two-dimentional gaussian kernel variance, | n| is as the pixel summation in each similarity piece of a homogenization parametric representation, for the image of different noise levels.Need control this, the smooth effect of WIA-LN method by regulating β.

The WIA-LN method is verified its treatment effect in the low dosage abdominal CT images, be reduced to abdominal CT images under original 1/5th the condition at tube current and still can obtain the better image recovery effects in handling, consider the linear relationship of dosage and tube current, the WIA-LN denoising method can make patient in abdomen scanning suffered dosage is reduced to original 1/5th, yet this kind method has certain limitation, easily the star strip artifact in the CT image under the low dosage condition as the information in the image, thereby can't effectively suppress the star strip artifact that under the low-dose scanning condition, is prone in the CT image, the generation of these star strip artifacts is because tissue has bigger decay to the X ray projection of some angle, generally appear in the CT scan image at the more position of high density (as bone), as shoulder, position such as chest and vertebra.Under the low dosage condition of tube current or voltage reduction, because the penetrability of X ray and the decline of energy, the star strip artifact will be more serious.

Summary of the invention

The technical problem to be solved in the present invention is to overcome the prior art deficiency, and a kind of low dosage CT image filtering method is provided, and can effectively suppress star strip artifact and noise in the low dosage CT image, improves CT picture quality.

The present invention is specifically by the following technical solutions:

A kind of low dosage CT image filtering method at first uses the method for dictionary study that original low dosage CT image is carried out filtering, then the image that obtains is carried out non-sharpening filtering again.

Preferably, the method for described use dictionary study is carried out filtering to original low dosage CT image, specifically in accordance with the following methods:

Step 1, use dct transform obtain DCT dictionary D ₀

Step 2, the original low dosage CT image I for the treatment of filtering are carried out bulk and are decomposed, and the bulk that obtains image I is decomposed and expressed matrix M;

Step 3, with matrix M as the input data, with dictionary D ₀As initial dictionary, use the K-SVD training algorithm to dictionary training renewal, the dictionary D after obtaining training ₁

Step 4, matrix M is carried out sparse expression, obtain sparse matrix of coefficients X; Use dictionary D then ₁Multiply each other with sparse matrix of coefficients X and to obtain matrix Y_Denoise;

Step 5, matrix Y_Denoise is recombinated, obtain filtering Y as a result according to block method of decomposing in the step 2; Y and original low dosage CT image I are carried out interpolation, obtain final filtered low dosage CT image Z.

Further, described non-sharpening filtering specifically in accordance with the following methods: each pixel among the image Z is carried out smoothing processing respectively, obtain the image Z ' after the smoothing processing; Image Z and image Z ' are carried out the difference computing, obtain both error images; Use the error image obtain that image graph is carried out interpolation as Z at last, obtain image after the final filtering.

The present invention learns two kinds of strategies of denoising and non-sharpening filtering with dictionary and combines, and at first utilizes the block distortion in the original low dosage CT of the dictionary learning method filtering image, utilizes non-sharpening filtering method that repressed marginal information is strengthened then.The inventive method can effectively suppress star strip artifact and the noise in the low dosage CT image, improves CT picture quality.

Description of drawings

Fig. 1 is a low dosage CT image;

Fig. 2 is a normal dose CT image;

Fig. 3 carries out image after the Filtering Processing for adopting the WIA-LN method to the low dosage image of Fig. 1;

Fig. 4 carries out image after the Filtering Processing for adopting the inventive method to the low dosage image of Fig. 1.

Embodiment

Below in conjunction with accompanying drawing technical scheme of the present invention is elaborated.

Thinking of the present invention is star strip artifact and the noise that is difficult to effectively removal with existing method in the low dosage CT image, dictionary is learnt denoising and two kinds of strategies of non-sharpening filtering organically combine, at first utilize the block distortion in the original low dosage CT of the dictionary learning method filtering image, utilize non-sharpening filtering method that repressed marginal information is strengthened then.

The bright method of this law is specific as follows:

Utilize the block distortion in the original low dosage CT of the dictionary learning method filtering image, the following method of concrete employing in this embodiment:

1) use discrete cosine transform (Discrete Cosine Transform is called for short dct transform), generate the DCT dictionary, method is as follows: given sequence x (n), and n=0,1 ... N ²-1, its discrete cosine transform:

$X_c\left(0\right)=\frac{1}{\sqrt{N}}\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}x\left(n\right)---\left(3\right)$

$X_c\left(k\right)=\sqrt{\frac{2}{N}}\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}x\left(n\right)\cos \frac{(2n+1)\mathrm{k\π}}{2N}$

Wherein the transformation range of k be [0, H ²-1], generating a size with this formula so is H ²* N ²Two-dimensional matrix, be initial dictionary D ₀

2) the low dosage CT image I for the treatment of filtering is carried out the bulk decomposition, and the bulk that obtains image I is decomposed the expression matrix M.

The block meaning of decomposing is as follows: suppose that image is carried out size to be decomposed for the piece of H*H, promptly in the image from top to bottom, from left to right, successively each is big or smallly become a H again for the fritter of H*H ²* 1 vector, these vectors are formed matrix M;

3) to dictionary training renewal, the dictionary D after obtaining training ₁Use the K-SVD algorithm to train in the present embodiment, the input data of algorithm are to treat the block matrix M that obtains afterwards of decomposing of low dosage CT image I of filtering, and initial dictionary is the dictionary D that obtains in the step 1) ₀

4) to step 2) matrix M that obtains carries out sparse expression, can obtain sparse matrix of coefficients X; Use the dictionary D after training then ₁Multiply each other with sparse matrix of coefficients X and to obtain matrix Y_Denoise, i.e. Y_Denoise=D*X;

5) with Y_Denoise according to the 2nd) block decomposition method of step recombinates and obtains filtering Y as a result; Y is carried out interpolation with treating the low dosage CT image I of filtering, obtain final filtering Z as a result.

Utilize non-sharpening filtering method that repressed marginal information is strengthened then, specific as follows:

Each pixel among the image Z is carried out smoothing processing respectively, obtain the image Z ' after the smoothing processing; Image Z and image Z ' are carried out the difference computing, obtain both error images; Use the error image obtain that image graph is carried out interpolation as Z at last, obtain image after the final filtering.

In order to verify the effect of the inventive method, carried out following contrast experiment:

At first, obtain the low dosage CT image (as shown in Figure 1) at same position, use WIA-LN method and the inventive method that low dosage CT image shown in Figure 1 is handled respectively.For relatively, gather the normal dose CT image (Fig. 2) at same position simultaneously.CT equipment is one 16 row CT (Somatom Sensation16), the condition of scanning is 120kVp and 5mm bed thickness, adopt the FBP method to rebuild, other parameters adopt machine default values, and the CT image of normal dose and the CT image of low dosage are respectively by being made as 270mA to the tube current parameter and 70mA obtains.

Using WIA-LN method and the inventive method that low dosage CT image shown in Figure 1 is handled the result who obtains distinguishes as shown in Figure 3, Figure 4.At first two kinds of resulting results of method are carried out visual assessment, compare low dosage CT image and normal dose CT image (Fig. 2) that WIA-LN method (Fig. 3) and the inventive method (Fig. 4) are handled, can see that the WIA-LN method can't effectively suppress the star strip artifact when suppressing noise, and using the picture quality after the inventive method is handled obviously to improve, star strip artifact and noise have all obtained effective inhibition.

For the validity of checking the inventive method of quantizing, we have calculated the variance in the selected homogeneous zone (zone shown in the broken circle frame among Fig. 1 to Fig. 4), and variance is defined as here:

$\mathrm{Var}=\frac{1}{\mathrm{np}-1}{\mathrm{\Σ}}_{j=1}^{\mathrm{np}}{(x_j-\stackrel{\‾}{x})}^2---\left(4\right)$

Wherein np represents the number of pixels in the selection area, x _jWith Represent single pixel CT value and mean CT-number (Housfield units in this selection area respectively, HU), can see that from following table 1 disposal route of the present invention can reduce the variance of original low dosage CT image significantly, obtain near the variance of normal dosage CT image in selected homogeneous zone.

Table 1

Claims (3)

1. a low dosage CT image filtering method is characterized in that, at first uses the method for dictionary study that original low dosage CT image is carried out filtering, then the image that obtains is carried out non-sharpening filtering again.

2. low dosage CT image filtering method according to claim 1 is characterized in that, the method for described use dictionary study is carried out filtering to original low dosage CT image, specifically in accordance with the following methods:

Step 1, use dct transform obtain DCT dictionary D ₀

Step 2, the original low dosage CT image I for the treatment of filtering are carried out bulk and are decomposed, and the bulk that obtains image I is decomposed and expressed matrix M;

Step 3, with matrix M as the input data, with dictionary D ₀As initial dictionary, use the K-SVD training algorithm to dictionary training renewal, the dictionary D after obtaining training ₁

Step 4, matrix M is carried out sparse expression, obtain sparse matrix of coefficients X; Use dictionary D then ₁Multiply each other with sparse matrix of coefficients X and to obtain matrix Y_Denoise;

Step 5, matrix Y_Denoise is recombinated, obtain filtering Y as a result according to block method of decomposing in the step 2; Y and original low dosage CT image I are carried out interpolation, obtain final filtered low dosage CT image Z.

3. as low dosage CT image filtering method as described in the claim 2, it is characterized in that, described non-sharpening filtering specifically in accordance with the following methods: each pixel among the image Z is carried out smoothing processing respectively, obtains the image Z ' after the smoothing processing; Image Z and image Z ' are carried out the difference computing, obtain both error images; Use the error image obtain that image graph is carried out interpolation as Z at last, obtain image after the final filtering.

Images