WO2014094376A1

WO2014094376A1 - Automatic segmentation method for cervical cancer image based on t2-mri and dw-mri

Info

Publication number: WO2014094376A1
Application number: PCT/CN2013/070942
Authority: WO
Inventors: 李悟; 考月英; 田捷
Original assignee: 中国科学院自动化研究所
Priority date: 2012-12-19
Filing date: 2013-01-24
Publication date: 2014-06-26
Also published as: CN102999917A; CN102999917B

Abstract

An automatic segmentation method for a cervical cancer image based on T2-weighted magnetic resonance imaging (T2-MRI) and diffusion-weighted magnetic resonance imaging (DW-MRI), including: registering a DW-MR image to a T2-MR image using a nonlinear registration method, and classifying the registered DW-MR image; filtering the T2-MR image using a nonlinear anisotropic diffusion filtering technique, segmenting a bladder and a rectum, and segmenting a region of interest using the segmentation results of the bladder and the rectum; and performing accurate segmentation of a tumour on the region of interest of the T2-MR image and the DW-MR image using a combined maximum a posteriori probability (CMAP) method. The present invention makes full use of the effective information about a T2-MR image and a DW-MR image, can effectively overcome the influences of noise, partial volume effect and intensity overlap in the T2-MR image, is an accurate and effective segmentation method for cervical cancer, and has important clinical significance and application value for the prevention, diagnosis and treatment of cervical cancer.

Description

Automated segmentation method for cervical cancer images based on T2-MRI and DW-MRI This application claims the application number 201210554664. 3, the invention name is "T2-MRI and DW-MRI based cervical cancer" Priority of the Chinese Patent Application for the Automatic Image Segmentation Method, the entire contents of which are incorporated herein by reference. Technical field

The invention belongs to the field of image processing, and particularly relates to a method for automatically segmenting cervical cancer images based on T2-weighted nuclear magnetic resonance imaging T2-MRI and diffusion weighted magnetic resonance imaging DW-MRI. Background technique

Cervical cancer is one of the common malignant tumors that seriously threaten women's life and health. The accurate segmentation of cervical cancer has important clinical significance and application value for the prevention, diagnosis and treatment of cervical cancer.

With the development of imaging technology, medical image segmentation has become a key and challenging problem in the field of medical image analysis. The segmentation of cervical cancer is particularly complicated by the complex structure of the human abdomen. The single imaging mode T2-MRI does not fully display the effective information of cervical cancer. Figure 1 (a) and (b) show the T2-MR image and DW-MR image of the abdomen of cervical cancer patients. The cervix is located between the bladder and the rectum. It can be seen from Fig. 1 (a) that the T2-MR image has a higher spatial resolution and the tumor boundary is clearer, but the normal tissue of the cervix, the tumor, the bladder wall and the rectum have more severe intensity overlap with each other; Figure 1 (b) shows that the tumor has a significantly higher gray value in the DW-MR image, but its resolution is low and the tumor boundary is blurred. Therefore, the automatic realization of cervical cancer segmentation in a single imaging mode is quite difficult, and some conventional methods such as region growth and threshold cannot segment the tumor well. Summary of the invention

It is an object of the present invention to provide a cervical cancer image based on T2-MRI and DW-MRI. The dynamic segmentation framework and the method of accurately segmenting the cervical cancer tumor region by using the combined maximum posterior probability (CMAP) to accurately segment the cervical cancer.

To achieve the above objectives, a T2-weighted magnetic resonance imaging T2-MRI and diffusion-weighted magnetic resonance imaging DW-MRI automatic segmentation method for cervical cancer images, including:

Step 1: Register the DW-MR image into the T2-MR image by nonlinear registration method, and classify the registered DW-MR image;

Step 2: Filter the T2-MR image using nonlinear anisotropic diffusion filtering technique, segment the bladder and rectum, and segment the region of interest using the segmentation results of the bladder and rectum; Step 3: For T2-MR images The region of interest and the DW-MR image were combined with the maximum posterior probability CMAP for precise segmentation of the tumor.

The invention fully utilizes the effective information of the T2-MR image and the DW-MR image, and can effectively overcome the influence of noise, local volume effect and intensity overlap in the T2-MR image, and is an accurate and effective segmentation method for cervical cancer. It has important clinical significance and application value for the prevention, diagnosis and treatment of cervical cancer. DRAWINGS

Figure 1 is an anatomical view of a patient with cervical cancer, (a) is a T2-M image; (b) is a DW-MR image;

Figure 2 is an automatic segmentation frame diagram based on T2-MRI and DW-MRI;

Figure 3 is a flow chart of the Maximum Posterior Probability Method (CMAP);

Figure 4 is a graphical representation of the auto-segmentation framework of cervical cancer images based on T2-weighted magnetic resonance imaging (T2-MRI) and diffusion-weighted magnetic resonance imaging (DW-MRI), (a) is the original image; (b) red outline Intra-line is the region of interest containing tumor and normal tissue; (c) DW-MR image registered to T2-M image; (d) Cervical cancer segmentation result using MAP method only on T2-MR image (ie ^ (0) (e) Cervical cancer segmentation results using the CMAP method on T2-MR images and DW-M images (ie, ^ = 1); (0 Experts manually segmentation results. detailed description

In order to make the objects, the technical solutions and the advantages of the present invention more comprehensible, the present invention will be further described in detail below with reference to the accompanying drawings.

The core idea of the present invention is a T2-weighted magnetic resonance imaging (T2-MRI) and diffusion-weighted magnetic resonance imaging (DW-MRI) automatic segmentation framework for cervical cancer images and the use of joint maximum posterior probability (CMAP). The method for segmenting the cervical cancer tumor region includes the following steps: First, the DW-MR image is registered to the T2-MR image by using a nonlinear registration method (here, the mutual information registration method is taken as an example), and the registration is performed. DW-MR images are classified; then T2-MR images are filtered using nonlinear anisotropic diffusion filtering techniques (here, PM nonlinear anisotropic diffusion filtering is used as an example), then the bladder and rectum are segmented, and the bladder and The segmentation result of the rectum divides the region of interest (including normal tissues and tumors). Finally, the region of interest and the DW-MR image of the T2-MR image are combined with the maximum posterior probability (CMAP) for accurate segmentation of the tumor.

The automatic segmentation framework of cervical cancer images based on T2-weighted magnetic resonance imaging (T2-MRI) and diffusion-weighted magnetic resonance imaging (DW-MRI) provided according to the present invention and the use of joint maximization are described below in connection with specific embodiments. Probability of Probability (CMAP) A method for accurately segmenting the tumor area of a cervical cancer is described in detail. FIG. 2 is a flowchart of a T2 weighted magnetic resonance imaging (T2-MRI) and diffusion weighted magnetic resonance imaging (DW-MRI) cervical segmentation image automatic segmentation framework provided by the present invention, which comprises the following steps. :

Step 1: Use a nonlinear registration method (such as mutual information registration method, Demons algorithm, etc.) to register the DW-MR image to the T2-MR image, and sub-step the registered DW-MR image. Step 2: Adopt The nonlinear anisotropic diffusion filtering technique filters the T2-MR image, then segments the bladder and rectum, and uses the segmentation results of the bladder and rectum to segment the region of interest (including normal tissues and tumors);

Step 3: Accurate segmentation of the tumor is performed using the combined maximum posterior probability (CMAP) method for the region of interest and the DW-MR image of the T2-M image. Step 1 above includes the following two small steps: 1) Registering the DW-MR image to the T2-MR image using a nonlinear registration method, where the mutual information registration method is used as an example; 2) The registered DW-MR The image classification is an automatic threshold classification method to achieve the initial segmentation and localization of the tumor.

The above step 2 includes the following four steps: 1) filtering the T2-M image by using a nonlinear anisotropic diffusion filtering technique to preserve the edge information while removing noise, where PM nonlinear anisotropic diffusion filtering is taken as an example; Bladder segmentation; 3) rectal segmentation; 4) segmentation of the region of interest.

The formula of the PM nonlinear diffusion filter in the above first small step is as follows:

among them/. (0 is the pixel intensity value of the image at z, ζ· indicates that the position information of the image is a gradient operator, representing time, ·) is a diffusion coefficient, and the two forms are:

II V/ II ²

c(V/) = exp (-- ~ ^-)

Among them is the gradient threshold.

The method used for bladder segmentation in the second small step above is the active contour model. The model was built based on our prior knowledge that the bladder is the more uniform monolithic region with the highest gray value in the T2-M image of the abdominal cavity.

In the third small step above, the rectal segmentation uses the a priori knowledge that the rectum is located below the cervix and the preliminary segmentation results of the tumor in step 1. The fuzzy C-means algorithm is used on the T2-MR image of the bladder to obtain the rectal segmentation. result.

In the fourth small step above, the region of interest is segmented. On the T2-MR image of the bladder and rectum, the preliminary segmentation result of the tumor in step 1 is used, and the fuzzy C-means algorithm is used to segment the inclusion. The region of interest of the tumor and normal tissue, the results are shown in Figure 4 (b).

The above step 3 is to accurately segment the tumor by using the combined maximum posterior probability (CMAP) method for the region of interest and the DW-MR image of the T2-MR image, and the result is shown in Fig. 4(e).

The specific implementation process of the above joint maximum a posteriori probability (CMAP) method is specifically described below. Let F; denote the image, y = , "N, denote the gray value at the image _; W denotes the total number of pixels of the image. Suppose the image is to be classified into ^, to

= 1,2,... represents a pixel _; is classified as a class. The flow chart of the Joint Maximum A Posteriori Probability (CMAP) method is shown in Figure 3. The specific steps are as follows: 1) Calculate the energy function of the T2-M image^ ₂ ; 2) Calculate the energy function of the DW-M image t/ (x 3) Calculate the joint energy function of the T2-MR image and the DW-MR image ^ ₂ ( + ^/ ( ; 4) to determine whether the termination condition is satisfied, and if so, determine the type of tumor and normal tissue according to the principle of minimum energy, thereby Output the accurately segmented tumor area; return to step 1 if the termination condition is not met. The corresponding calculation formula in each small step is shown in the following description.

The traditional MAP segmentation algorithm is to obtain the segmentation result L^, which makes the posterior probability ^x| _y) the largest. which is:

Λ

X = arg max (x | y) According to the Bayesian formula, P(x| y;i can be expressed as - P(x\y)= ^P{y [ ^X ^ ^{x) ∞P(y\x)P (x) As can be seen from the above equation, the object of image segmentation is transformed into and the segmentation result X is obtained, so that Ρ(_ν| Ρ( ) is the largest, namely:

Λ

X = arg max ( | x)P(x)} where <y|x) and P (the probability density and prior probability respectively under a given partition 。 condition. ty|x) and ^) are calculated below.

Assuming the noise in the image is Gaussian white noise, the image model is: Where μ _Λ is the gray mean of the first type of tissue at the pixel, n _ik is the Gaussian white noise corresponding to the tissue at the pixel Z, and its distribution obeys the normal distribution W(0, σΐ, P(y|x) It can be expressed as:

P{y I =

Expressed by the Gibbs distribution as:

Where a constant, f / (x) is the energy function:

Wherein, is the neighborhood of the first position, t^x^ is a potential function, and δ is a constant.

Finally, the posterior probability formula is transformed into:

P(x I y) cxp -U(x)} The segmentation target is further transformed into the segmentation result X, which minimizes the energy function, ie

Arg min U (x)

The energy function f/ ₂ (x) of the T2-M image in the first small step described above can be calculated from the calculation formula of f/(x).

The energy function ί/ (χ) of the DW-MR image in the second small step above can be calculated according to the calculation formula of [/(X).

Since the posterior probability P ₇₂ (x|y _T2 ) of the T2-MR image and the posterior probability of the DW-MR image ^ Wy^) are independent, the CMAP segmentation method is to obtain the segmentation result L^, so that = arg max (P _T1 (x | y _T1 )P _DW (x \ y _DW )) converts the above expression into a formula based on the energy function in the MAP algorithm.

P _T1 I y _T 2 ) ^P DW ( ^X y _DW ) ^{∞ eX} p{" i ^U T2 ( ^X ) + β ^ϋ DW ( ^χ ))}

Where ? is the weight coefficient, which is used to balance the influence of the T2-MR image and the DW-MR image on the segmentation result. Finally, the problem of the CMAP segmentation algorithm is transformed into the minimum problem of the energy function:

Arg min(?7 _r2 (x) + βϋ _DW (χ)) The joint energy function f^W + ^^ x) of the T2-MR image and the DW-MR image of the third small step above can be based on the above [/ _f2 The formulas for (x) and [/^ (χ) are calculated.

When ^ = 0, the CMAP method becomes the MAP method. The experimental segmentation result at this time is shown in Fig. 4(d). When different values are taken, the influence of the T2-MR image and the DW-MR image on the segmentation result is also It is different.

To validate the method of the invention, we used manual segmentation of cervical cancer as a standard reference.

Figure 4 shows the effect of the T2-weighted magnetic resonance imaging (T2-MRI) and diffusion-weighted magnetic resonance imaging (DW-MRI) cervical cancer image segmentation framework experiments. (a) original image; (b) the red outline is the region of interest containing the tumor and normal tissue; (c) the DW-M image registered to the T2-M image; (d) only on the T2-M image Cervical cancer segmentation results using the MAP method (ie, when 0 = 0); (e) Cervical cancer segmentation results using the CMAP method on T2-MR images and DW-MR images (ie, ^ = 1); (f) Expert manual Segment the result.

Experiments have shown that the method of the present invention is based on T2-weighted magnetic resonance imaging (T2-MRI) and diffusion-weighted magnetic resonance imaging (DW-MRI) automatic segmentation framework for cervical cancer images and the use of joint maximum posterior probability (CMAP). The method of accurately segmenting the tumor area of cervical cancer accurately segmented the tumor area of the cervical cancer image and achieved the purpose of segmenting the tumor.

The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not In view of the above, it is to be understood that those skilled in the art are within the scope of the present invention within the scope of the present invention. Therefore, the scope of protection of the present invention should be The scope of protection of the request shall prevail.

Claims

Rights request

1. A T2-weighted magnetic resonance imaging T2-MRI and diffusion-weighted nuclear magnetic resonance imaging DW-MRI automatic segmentation method for cervical cancer images, including:

2. Method according to claim 1, characterized in that the non-linear registration method comprises a mutual information registration method or a Demons algorithm.

3. The method according to claim 1, wherein the classifying the registered DW-MR images adopts an automatic threshold classification method to achieve preliminary segmentation and localization of the tumor.

4. The method of claim 1 wherein the nonlinear anisotropic diffusion filtering technique comprises P-M nonlinear anisotropic diffusion filtering.

5. The method according to claim 4, characterized in that PM nonlinear anisotropic diffusion filtering is realized by the following formula -

among them/. "For the pixel intensity value of the image at z, V is the gradient operator, is the scatter operator, ί is the time, ·) is the diffusion coefficient, and the two forms are -

II V/ II ²

c(V/) = exp (-- ~ ^-)

Among them is the gradient threshold.

6. The method of claim 1 wherein said bladder segmentation employs an active contour model, wherein said bladder is a relatively uniform monolithic region having the highest gray value in the abdominal T2-MR image.

7. The method according to claim 1, wherein the rectal segmentation uses a priori knowledge that the rectum is located below the cervix and the preliminary segmentation result of the tumor in step 1, and is used on the T2-MR image of the bladder removal. The algorithm of fuzzy C-means is obtained, and the result of rectal segmentation is obtained.

8. The method according to claim 1, wherein the segmentation of the region of interest comprises using a fuzzy C-means algorithm on the T2-MR image of the removal of the bladder and the rectum, using the preliminary segmentation result of the tumor in step 1, The region of interest containing the tumor and normal tissue is segmented.

9. The method of claim 1 wherein said step 3 comprises:

1) Calculate the energy function t/ _T2 (x) of the T2-MR image _;

2) Calculate the energy function of the DW-MR image t/ (x);

3) Calculate the joint energy function of the T2-MR image and the DW-MR image ^ ₂ + ;

4) Determine whether the termination condition is satisfied. If it is satisfied, determine the type of tumor and normal tissue according to the principle of minimum energy, thereby outputting the accurately segmented tumor area, and return to step 1) if the termination condition is not satisfied.

10. Method according to claim 9, characterized in that the energy functions f / ₂ (x) and t / (x) are calculated by:

, - 1⁄2

+ ^1η ( ^σ )

)

Which represents the _image; gray value at, W represents the total number of pixels of the image. Suppose the image is to be classified into classes, with χ, .ϋ = ιυ representing the pixel skin as class /1.

Is the gray mean of the class I organization at the pixel ^, ^ the variance of the Gaussian white noise corresponding to the class 1 tissue at the pixel _Z , which is the neighborhood of the zth position, ί^χ, χ^ is the potential function, A constant, 3⁄4 is a collection of all pixels belonging to a class of organization.

The method according to claim 9, characterized in that the classification according to the principle that the joint energy function U _T2 (x) + U _DW (x) is the smallest is calculated by: XCMAP = + βυ _οπ (χ)) where ? is the weight coefficient

12. The method according to claim 11, characterized by comparing different ones? The segmentation result under the value gives accurate tumor segmentation results.