CN107133975B

CN107133975B - Heart CT-TEE method for registering based on valve alignment and probability graph

Info

Publication number: CN107133975B
Application number: CN201710242632.2A
Authority: CN
Inventors: 缑水平; 陈琳琳; 庄建; 黄美萍; 杨淑媛; 焦李成; 黄力宇; 李军
Original assignee: GUANGDONG PROV CARDIOVASCULAR DISEASE INST; Xidian University
Current assignee: GUANGDONG PROV CARDIOVASCULAR DISEASE INST; Xidian University
Priority date: 2017-04-14
Filing date: 2017-04-14
Publication date: 2019-07-23
Anticipated expiration: 2037-04-14
Also published as: CN107133975A

Abstract

The invention discloses it is a kind of based on valve alignment and probability graph heart CT-TEE method for registering, mainly solve the problems, such as that prior art cardiac CT is difficult to be registrated with TEE image since mode difference is huge.Its realization process is: interacting formula to CT and TEE image respectively and divides and introduce heart valve endpoint location, segmented image and heart valve endpoint location is obtained, using the spatial position of valve as prior information；Basic registration is carried out to CT and TEE image based on prior information；Region enhancing is carried out to CT and TEE image, and grey level enhancement is carried out to its segmentation figure, generates the probability graph of CT and TEE image；Normalization based on probability graph and the initial parameter using the transformation matrix of basis registration as optimizing algorithm in final registration, are finally registrated CT and TEE image.The present invention can more accurately realize being registrated to CT and TEE cardiac image, can be used for the recognition and tracking to cardiac anatomy.

Description

Heart CT-TEE method for registering based on valve alignment and probability graph

Technical field

The invention belongs to technical field of image processing, in particular to the method for registering of a kind of couple heart CT and TEE image can For the recognition and tracking to cardiac anatomy.

Background technique

With the fast development of medical imaging technology and computer processing technology, the mode of medical imaging is more and more richer Richness, such as CT, MR, PET, SPECT, ultrasound image.There is very big otherness, multi-modal medicine between different image modes A variety of images are combined by the image-forming information of fusion different modalities, show several on same piece image by image registration The image-forming information of image, the purpose is to which diversified information to be accurately fused in same piece image, so as to more smart Really lesion and anatomical structure from all angles.

According to different weighing criterias, image registration has different mode classifications, for example, according to the classification of space dimensionality, According to the classification of mapping mode, the classification for the feature being based on according to the classification of optimization algorithm, according to algorithm, according to used by The classification etc. of similarity measure；According to Spatial Dimension, image registration can be divided into the registration of 2D-2D, 2D-3D, 3D-3D.According to sky Between mapping mode difference, can be divided into rigid body translation and non-rigid transformation.The selection of optimization algorithm is also in image registration Multiplicity, usually there are gradient descent method, Newton method, Powell method, genetic algorithm etc..The feature that medical image can extract is very It is abundant, generally include characteristic point, surface texture, image pixel intensities and surface etc..Registration based on characteristic point, which refers to, to be passed through It is chosen at the feature point set that can be positioned for geometrically having special meaning, such as discontinuity point, the turning point of figure, line intersects Point, medically point with anatomically significant etc., and carry out coordinate matching；Registration based on surface refers to by way of segmentation Extract feature space of the profile of interesting image regions as registration；Registration based on pixel value, which refers to, utilizes entire image Pixel value or voxel value constitutive characteristic space；For medical image, the registration based on surface refers to by subject Internal fixation mark object or the mark point that determination on the image is obtained to internal injection developing materials.According to used phase Like the difference that property is estimated, image registration can be divided into registration, registration based on mutual information based on cross-correlation etc. again.

Mutual information method by propositions such as Collignon is one of the hot spot studied in recent years, especially in multi-modal medicine In image registration field.At abroad, carrying out medical figure registration using mutual information method at first is Viola etc..Have again later Person proposes many innovatory algorithms on the basis of mutual information, and Maes etc. proposes normalized mutual information, reduces traditional mutual Information is as measure function to the susceptibility of two width registration image overlapping region.Josien proposes a kind of mutual information combination gradient Similarity measurement criterion GMI, be successfully applied on the registration of the heterologous image such as MR, CT, PET.Fan et al. is by wavelet transformation In conjunction with mutual information, it is successfully applied on the registration of visible images and infrared light image.But in heart CT and TEE image On registration, since two kinds of image mode othernesses on pixel value are significant, and the edge and texture representative model of heart TEE image Paste, these medical image registration methods are difficult to carry out it accurate effective registration.

Summary of the invention

It is an object of the invention to be directed to the greatest differences of CT and TEE image image mode, propose a kind of based on valve The heart CT-TEE method for registering of alignment and probability graph, it is real by merging structure and texture information under both image modes Now to the accuracy registration of heart CT-TEE image.

To achieve the above object, the present invention includes the following steps:

(1) respectively to cardiac CT image I_R(x) and TEE image I_F(x) formula segmentation is interacted, CT image I is obtained_R(x) Segmentation figure, i.e. area-of-interest G_R(x) and TEE image I_F(x) segmentation figure, i.e. area-of-interest G_F(x)；

(2) CT image I is introduced while interacting formula segmentation_R(x) 3 heart valve endpoints or midpoint are sat Mark x_1r=(i_1r,j_1r),x_2r=(i_2r,j_2r),x_3r=(i_3r,j_3r), as 3 characteristic points of CT image, while introducing TEE figure As I_F(x) in CT image I_R(x) the corresponding heart valve endpoint of 3 characteristic points or midpoint coordinate x_1f=(i_1f,j_1f), x_2f=(i_2f,j_2f),x_3f=(i_3f,j_3f), as 3 characteristic points of TEE image, by 3 characteristic points and TEE image of CT image 3 characteristic points constitute 3 pairs of characteristic points pair, the prior information that will be registrated based on this 3 pairs of characteristic points pair；

(3) by CT image I_R(x) it is used as reference picture, by TEE image I_F(x) it is used as floating image, is believed based on valve priori Breath carries out basic registration to reference picture and floating image, obtains the transformation matrix T of basis registration₁, and obtain basis registration knot Fruit S₁(x)；

(4) CT image I is set_R(x) enhancing matrix V_R(x), respectively to CT image I_R(x) and area-of-interest G_R(x) into The enhancing of row region, obtains enhanced CT image I_Rh(x) and enhanced area-of-interest G_Rh(x), TEE image I is set_F(x) Enhancing matrix V_F(x), respectively to TEE image I_F(x) and area-of-interest G_F(x) region enhancing is carried out, is obtained enhanced TEE image I_Fh(x) and enhanced area-of-interest G_Fh(x)；

(5) it is based on the enhanced CT image I in region_Rh(x) and enhanced area-of-interest G_Rh(x), CT image I is generated_R (x) probability graph P_R(x), it is based on the enhanced TEE image I in region_Fh(x) and enhanced area-of-interest G_Fh(x), it generates TEE image I_F(x) probability graph P_F(x)；

(6) to the two probability graph P generated in (5)_R(x) and P_F(x) similarity measurement is carried out, and by base obtained in (3) The transformation matrix T of plinth registration₁As the initial parameter of optimizing algorithm in final registration, it is based on probability graph P_R(x) and P_F(x) phase Like property to CT image I_R(x) and TEE image I_F(x) it is finally registrated, acquires transformation matrix T₂, and obtain final registration result S₂(x)。

Compared with the prior art, the present invention has the following advantages:

1, the present invention is based in practical application to position of valve information demand and TEE image in valve imaging clearly And the visible Realistic Analysis in position of valve in CT image, the spatial positional information of valve is introduced in the registration of basis, is made with this Feature Points Matching is carried out for prior information, greatly improves the validity of CT Yu TEE image registration；

2, the transformation matrix that the present invention obtains basis registration is as the initial ginseng of optimizing algorithm Powell in final registration Number, avoid Powell algorithm because initial parameter selection it is improper and enter local optimum the problem of, improve local optimal searching The efficiency and accuracy of algorithm；

3, the present invention is based on region enhancing generating probability figure is carried out to region of interest, the life of probability graph is greatly simplified It is combined at process, and by the probability graph generated in the present invention with normalized mutual information, is greatly improved similarity measurement Accuracy and reliability；

Detailed description of the invention

Fig. 1 is realization general flow chart of the invention；

Fig. 2 is cardiac CT image used in the present invention, that is, the reference picture being registrated；

Fig. 3 is heart TEE image used in the present invention, that is, the floating image being registrated；

Fig. 4 is the probabilistic image generated in the present invention based on Fig. 2；

Fig. 5 is the probabilistic image generated in the present invention based on Fig. 3；

Fig. 6 is that the present invention carries out the registration image after basis is registrated with Fig. 3 to Fig. 2；

Fig. 7 is that the present invention carries out fused blending image to Fig. 2 and Fig. 6；

Fig. 8 is the registration image after the present invention is finally registrated Fig. 2 with Fig. 3；

Fig. 9 is that the present invention carries out fused blending image to Fig. 2 and Fig. 8.

Specific embodiment

Below in conjunction with attached drawing, specific embodiments of the present invention and effect are made further explanation and description:

Referring to Fig.1, the present invention is based on valve alignment and the heart CT-TEE method for registering images of probability graph mutual information

Implementation step is as follows:

Step 1: to the cardiac CT image I of input_R(x) and TEE image I_F(x) formula segmentation is interacted.

1a) input cardiac CT image I_R(x), as shown in Fig. 2, artificially choosing CT image I_R(x) atrium dextrum and active in Region where arteries and veins, and a small amount of several target pixel points and several background dots in the region are chosen, iteration for several times is carried out, is obtained The area-of-interest G of CT image_R(x)；

1b) input TEE image I_F(x), as shown in figure 3, artificially choosing TEE image I_F(x) atrium dextrum and aorta in The region at place, and a small amount of several target pixel points and several background dots in the region are chosen, iteration for several times is carried out, is obtained The area-of-interest G of TEE image_F(x)。

Step 2: obtaining heart valve prior information.

2a) while Interactive Segmentation, CT image I is introduced_R(x) 3 heart valve endpoints or midpoint coordinate x_1r=(i_1r,j_1r),x_2r=(i_2r,j_2r),x_3r=(i_3r,j_3r), as 3 characteristic points of CT image, while introducing TEE image I_F(x) in CT image I_R(x) the corresponding heart valve endpoint of 3 characteristic points or midpoint coordinate x_1f=(i_1f,j_1f), x_2f=(i_2f,j_2f),x_3f=(i_3f,j_3f), 3 characteristic points as TEE image；

It can 2b) be divided into the characteristic of bicuspid valve, tricuspid valve, aorta petal or pulmonary valve according to heart valve, be schemed with CT 3 characteristic points of picture and 3 characteristic points of TEE image constitute 3 pairs of characteristic points pair, and by this 3 pairs of characteristic points to as valve elder generation Test information.

Step 3: to CT image I_R(x) and TEE image I_F(x) basic registration is carried out.

3a) by CT image I_R(x) it is used as reference picture, by TEE image I_F(x) it is used as floating image；

3b) it is based on valve prior information, i.e. the position coordinates x of 3 based on reference picture characteristic point_1r=(i_1r,j_1r), x_2r=(i_2r,j_2r),x_3r=(i_3r,j_3r) and floating image 3 characteristic points position coordinates x_1f=(i_1f,j_1f),x_2f= (i_2f,j_2f),x_3f=(i_3f,j_3f), the transformation matrix of coordinates T of 3 pairs of characteristic points pair is acquired as follows₁:

3c) it is based on transformation matrix T₁, affine transformation is carried out to floating image, and basis registration knot is obtained by cubic interpolation Fruit S₁(x): S₁(x)=T₁×I_F(x), as shown in fig. 6, Fig. 2 is merged with Fig. 6, fusion figure is obtained, as shown in fig. 7, can from Fig. 7 To find out, after the registration of basis, reference picture Fig. 2 and floating image Fig. 3 have geographically obtained rough matching.

Step 4: to CT image I_R(x) and its area-of-interest G_R(x) region enhancing is carried out.

4a) according to CT image I_R(x) pixel distribution setting enhancing matrix V_R(x):

Set V_R(x) size and CT image I_R(x) in the same size, and by V_R(x) consistent with area-of-interest coordinate in Point be set as one be greater than 0 fixed value, set is 80 in the present invention, and the point of remaining position is set as 0；

4b) to CT image I_R(x) and area-of-interest G_R(x) grey level enhancement is carried out, enhanced CT image I is obtained_Rh(x) With enhanced area-of-interest G_Rh(x):

I_Rh(x)=I_R(x)+V_R(x)

G_Rh(x)=G_R(x)+V_R(x)。

Step 5: to TEE image I_F(x) and its area-of-interest G_F(x) region enhancing is carried out.

5a) according to TEE image I_F(x) pixel distribution setting enhancing matrix V_F(x):

Set V_F(x) size and TEE image I_F(x) in the same size, and by V_F(x) in area-of-interest coordinate one The point of cause is set as a fixed value greater than 0, is set as 80 in the present invention, the point of remaining position is set as 0；

5b) to TEE image I_F(x) and its area-of-interest G_F(x) grey level enhancement is carried out, enhanced TEE image is obtained I_Fh(x) and enhanced area-of-interest G_Fh(x):

I_Fh(x)=I_F(x)+V_F(x)

G_Fh(x)=G_F(x)+V_F(x)。

Step 6: generating CT image I_R(x) probability graph P_R(x)。

6a) the enhanced CT image I in region obtained based on step 4_Rh(x) and enhanced area-of-interest G_Rh(x), It generates by the enhanced CT image I in region_Rh(x) about area-of-interest G after enhancing_Rh(x) probability density function:

The probability density function is equivalent to a probability retrieval table, and when inputting some pixel, output is the pixel The probability value that point is likely located in area-of-interest；

6b) by the enhanced CT image I in region_Rh(x) it is used as probability density function f_R(i) input obtains CT image I_R (x) probability graph P_R(x): P_R(x)=f_R(x),x∈I_Rh(x), as shown in Figure 4.

Step 7: generating TEE image I_F(x) probability graph P_F(x)。

7a) the enhanced TEE image I in region obtained based on step 5_Fh(x) and enhanced area-of-interest G_Fh(x), It generates by the enhanced TEE image I in region_Fh(x) about area-of-interest G after enhancing_Fh(x) probability density function:

7b) by the enhanced TEE image I in region_Fh(x) it is used as probability density function f_F(i) input obtains TEE image I_F(x) probability graph P_F(x): P_F(x)=f_F(x),x∈I_Fh(x), as shown in Figure 5.

Step 8: to CT image I_R(x) and TEE image I_F(x) it is finally registrated.

8a) to probability graph P_R(x) and P_F(x) similarity measurement is carried out

Similarity measurement criterion can be mutual information, normalized mutual information, cross-correlation, gradient mutual information etc., and the present invention adopts It is normalized mutual information, solution procedure is as follows:

8a1) find out image I_R(x) probability graph P_R(x) entropy H_R(x) and image I_F(x) probability graph P_F(x) entropy

H_F(x):

Wherein, P (P_RIt (x)) is probability graph P_R(x) probability density function, P (P_FIt (x)) is probability graph P_F(x) probability is close Spend function；

8a2) find out probability graph P_R(x) and P_F(x) combination entropy H_R,F(x):

Wherein, P (P_R(x),P_FIt (x)) is probability graph P_R(x) and P_F(x) joint probability density function；

8a3) obtain probability graph P_R(x) and P_F(x) normalized mutual information NMPI:

8b) to probability graph P_R(x) and P_F(x) normalized mutual information NMPI carries out optimizing

There are many type of optimizing algorithm, as particle swarm algorithm, simulated annealing, ant group algorithm, gradient descent method, Powell method, the present invention in using Powell method, the specific implementation process is as follows:

8b1) by the transformation matrix T of basis registration₁Initial parameter as Powell method；

8b2) obtained by Powell method optimizing as probability graph P_R(x) and P_FWhen normalized mutual information NMPI maximum (x) Transformation matrix of coordinates T₂, by TEE image I_F(x) it is coordinately transformed to obtain final registration result S₂(x): S₂(x)=T₂× I_F(x), as shown in figure 8, Fig. 2 is merged with Fig. 8, fusion figure is obtained, as shown in Figure 9.

Come as can be seen from Figure 9, after being finally registrated, reference picture Fig. 2 has obtained accurately matching with floating image Fig. 3 Quasi- effect.

Above description is only example of the present invention, does not constitute any limitation of the invention, it is clear that for this It, all may be without departing substantially from the principle of the invention, structure after having understood the content of present invention and principle for the professional in field In the case of, various modifications and variations in form and details are carried out, but these modifications and variations based on inventive concept are still Within the scope of the claims of the present invention.

Claims

1. a kind of heart CT-TEE method for registering images based on valve alignment and probability graph, comprising:

(1) respectively to cardiac CT image I_R(x) and TEE image I_F(x) formula segmentation is interacted, CT image I is obtained_R(x) sense is emerging Interesting region G_R(x) and TEE image I_F(x) area-of-interest G_F(x)；

(2) CT image I is introduced while interacting formula segmentation_R(x) 3 heart valve endpoints or midpoint coordinate x_1r =(i_1r,j_1r),x_2r=(i_2r,j_2r),x_3r=(i_3r,j_3r), as 3 characteristic points of CT image, while introducing TEE image I_F (x) in CT image I_R(x) the corresponding heart valve endpoint of 3 characteristic points or midpoint coordinate x_1f=(i_1f,j_1f),x_2f =(i_2f,j_2f),x_3f=(i_3f,j_3f), as 3 characteristic points of TEE image, by 3 characteristic points of CT image and TEE image 3 characteristic points constitute 3 pairs of characteristic points pair, the prior information that will be registrated based on this 3 pairs of characteristic points pair；

(3) by CT image I_R(x) it is used as reference picture, by TEE image I_F(x) it is used as floating image, is based on valve prior information, Basic registration is carried out to reference picture and floating image, obtains the transformation matrix T of basis registration₁, and obtain basic registration result S₁ (x)；

(4) CT image I is set_R(x) enhancing matrix V_R(x), respectively to CT image I_R(x) and area-of-interest G_R(x) area is carried out Domain enhancing, obtains enhanced CT image I_Rh(x) and enhanced area-of-interest G_Rh(x), TEE image I is set_F(x) increasing Strong matrix V_F(x), respectively to TEE image I_F(x) and area-of-interest G_F(x) region enhancing is carried out, enhanced TEE figure is obtained As I_Fh(x) and enhanced area-of-interest G_Fh(x)；

(5) it is based on the enhanced CT image I in region_Rh(x) and enhanced area-of-interest G_Rh(x), CT image I is generated_R(x) Probability graph P_R(x), it is based on the enhanced TEE image I in region_Fh(x) and enhanced area-of-interest G_Fh(x), TEE image is generated I_F(x) probability graph P_F(x)；

(6) to the two probability graph P generated in (5)_R(x) and P_F(x) similarity measurement is carried out, and basis obtained in (3) is matched Quasi- transformation matrix T₁As the initial parameter of optimizing algorithm in final registration, it is based on probability graph P_R(x) and P_F(x) similitude To CT image I_R(x) and TEE image I_F(x) it is finally registrated, acquires transformation matrix T₂, and obtain final registration result S₂ (x)。

2. matching according to the method described in claim 1, wherein carrying out basis to reference picture and floating image described in step (3) Standard carries out as follows:

(3.1) based on the 3 CT image I introduced in step (2)_R(x) characteristic point coordinate x_1r=(i_1r,j_1r),x_2r=(i_2r, j_2r),x_3r=(i_3r,j_3r) and 3 TEE image I_F(x) characteristic point coordinate x_1f=(i_1f,j_1f),x_2f=(i_2f,j_2f),x_3f= (i_3f,j_3f) coordinate correspondence relationship, acquire the transformation matrix T of coordinate transform as follows₁:

(3.2) it is based on transformation matrix T₁Affine transformation is carried out to floating image, acquires the result S of basis registration₁(x):

S₁(x)=T₁×I_F(x)。

3. according to the method described in claim 1, wherein setting CT image I in step (4)_R(x) enhancing matrix V_R(x), be by CT image I_R(x) size sets V_R(x) size is consistent with its, and by V_R(x) equal with the consistent point of area-of-interest coordinate in It is set as a fixed value greater than 0, the point of remaining position is set as 0.

4. according to the method described in claim 1, wherein respectively to CT image I in step (4)_R(x) and area-of-interest G_R(x) Region enhancing is carried out, is carried out by following formula:

I_Rh(x)=I_R(x)+V_R(x)

G_Rh(x)=G_R(x)+V_R(x)

Wherein I_RhIt (x) is enhanced CT image, G_RhIt (x) is enhanced area-of-interest.

5. according to the method described in claim 1, wherein setting TEE image I in step (4)_F(x) enhancing matrix V_F(x), it is By TEE image I_F(x) size sets V_F(x) size is consistent with its, and by V_F(x) consistent with area-of-interest coordinate in Point is set as a fixed value greater than 0, and the point of remaining position is set as 0.

6. according to the method described in claim 1, wherein respectively to TEE image I in step (4)_F(x) and area-of-interest G_F(x) Region enhancing is carried out, is carried out by following formula:

I_Fh(x)=I_F(x)+V_F(x)

G_Fh(x)=G_F(x)+V_F(x)

Wherein I_FhIt (x) is enhanced TEE image, G_FhIt (x) is enhanced area-of-interest.

7. according to the method described in claim 1, being wherein based on the enhanced CT image I in region in step (5)_Rh(x) and enhance Area-of-interest G afterwards_Rh(x), CT image I is generated_R(x) probability graph P_R(x), it carries out as follows:

(5.1) it generates by the enhanced CT image I in region_Rh(x) about area-of-interest G after enhancing_Rh(x) probability density letter Number f_R(i):

(5.2) by the enhanced CT image I in region_Rh(x) it is used as probability density function f_R(i) input obtains CT image I_R(x) Probability graph P_R(x):

P_R(x)=f_R(x),x∈I_Rh(x)。

8. according to the method described in claim 1, being wherein based on the enhanced TEE image I in region in step (5)_Fh(x) and enhance Area-of-interest G afterwards_Fh(x), TEE image I is generated_F(x) probability graph P_F(x), it carries out as follows:

(5.3) it generates by the enhanced TEE image I in region_Fh(x) about area-of-interest G after enhancing_Fh(x) probability density Function f_F(i):

(5.4) by the enhanced TEE image I in region_Fh(x) it is used as probability density function f_F(i) input obtains TEE image I_F (x) probability graph P_F(x):

P_F(x)=f_F(x),x∈I_Fh(x)。

9. according to the method described in claim 1, wherein to the two probability graph P generated in (5) in step (6)_R(x) and P_F(x) Similarity measurement is carried out, is carried out as follows:

(6.1) using normalized mutual information as probability graph P_R(x) and P_F(x) similarity measurement criterion；

(6.2) image I is found out_R(x) probability graph P_R(x) entropy H_R(x) and image I_F(x) probability graph P_F(x) entropy H_F(x):

Wherein, P (P_RIt (x)) is probability graph P_R(x) probability density function, P (P_FIt (x)) is probability graph P_F(x) probability density letter Number；

(6.3) probability graph P is found out_R(x) and P_F(x) combination entropy H_R,F(x):

(6.4) according to probability graph P_R(x) entropy H_R(x), probability graph P_F(x) entropy H_F(x), probability graph P_R(x) and probability graph P_F(x) Combination entropy H_R,F(x), two probability graph P are obtained_R(x) and P_F(x) normalized mutual information NMPI are as follows:

10. according to the method described in claim 1, probability graph P will be wherein based in step (6) in (3)_R(x) and P_F(x) phase Like property to CT image I_R(x) and TEE image I_F(x) it is finally registrated, is carried out as follows:

(6.5) by the transformation matrix T of basis registration₁Initial parameter as Powell method；

(6.6) it is obtained by Powell method optimizing as probability graph P_R(x) and P_F(x) seat when normalized mutual information NMPI maximum Mark transformation matrix T₂, by TEE image I_F(x) it is coordinately transformed to obtain final registration result S₂(x):

S₂(x)=T₂×I_F(x)。