CN102903117A - 3D (three-dimensional) image registration method and device based on conformal geometric algebra - Google Patents

Abstract

The invention discloses a three-dimensional image registration method and device based on conformal geometric algebra, uses conformal geometric algebra to reconstruct the positional relationship constraints of 3D medical images, analyzes the conformal geometric transformation of medical images, and constructs a A new similarity measure for 3D medical image registration, based on which a 3D medical image registration algorithm is proposed for 3D registration of CT and MR_T1 images. The direct alignment of 3D data is realized, which can better locate the 3D positions of tissues and organs, making the registration results more intuitive, the registered images clearer, and the registration accuracy higher.

Description

A three-dimensional image registration method and device based on conformal geometric algebra

technical field

The invention relates to the field of medical image processing, in particular to a three-dimensional image registration method and device based on conformal geometric algebra.

Background technique

From the data dimension, the registration of medical images can be divided into three types: 2D/2D, 2D/3D and 3D/3D. For 3D/3D registration, the registration data is three-dimensional data, and the types of geometric transformations are more and more complex. It is more difficult to find parameters for optimization, and it is easier to fall into local optimum, and the space complexity and time complexity of the whole registration process are higher than Much higher than 2D/2D registration. In medical applications such as neurosurgery navigation, medical image processing technology is very critical, but it also faces many problems, especially in 3D/3D registration. From the relevant literature, we can find that many methods to solve the problem are worth learning from. For example, Hsu and Loew first proposed a fully automatic multi-modal medical image 3D registration method based on hierarchical feature extraction; Li Wenlong et al. used non-uniform B-spline deformable bodies instead of general cubic B-spline deformable bodies to describe imaging Based on the nonlinear movement of tissues, a 3D nonlinear medical image registration based on free deformation was proposed; Harmouche et al. established a hinge model for 3D registration of spine MR and X-ray by calculating intervertebral deformation.

In the registration of medical images, the existing 3D registration schemes mostly assume that the corresponding relationship between the registration points is known to analyze how to deform, or already know how to deform, and only need to obtain the corresponding relationship between the registration points, which is difficult The geometric position of the registration is described, so that the three-dimensional display of the registered medical image is not clear, and the registration accuracy is not high.

Therefore, the prior art still needs to be improved and developed.

Contents of the invention

The technical problem to be solved by the present invention is to provide a three-dimensional image registration method and device based on conformal geometric algebra for the above-mentioned defects of the prior art, so as to make the registered image clearer and the registration accuracy higher.

The technical solution adopted by the present invention to solve technical problems is as follows:

A three-dimensional image registration method based on conformal geometric algebra, which includes the following steps:

A. Select the reference image and the floating image for registration, perform edge detection on the reference image and the floating image, obtain the corresponding edge contour, and use the image segmentation algorithm to extract the features of the reference image and the floating image point;

B. According to the feature points of the reference image and the floating image, under the framework of conformal geometry algebra, generate the feature vector of the feature points;

C. Perform several rotations and translations to the feature vectors of the feature points of the floating image, and calculate the similarity measure of the feature vectors of the feature points of the reference image and the floating image after each rotation and translation;

D. When the similarity measure is less than a predetermined threshold or the number of rotations and translations is greater than or equal to a predetermined number of times, output the feature vectors of the feature points of the floating image at this time, generate a registered floating image, and register The aligned floating image is fused with the reference image to obtain a final registered image.

In the three-dimensional image registration method based on conformal geometric algebra, the step A further includes: pre-processing the distribution ratio and size range of the reference image and the floating image to be consistent.

In the three-dimensional image registration method based on conformal geometric algebra, in the step A, edge detection is performed on the reference image and the floating image using a canny operator.

The three-dimensional image registration method based on conformal geometric algebra, wherein, the step C also includes:

C1. Obtain the closest point on the floating image corresponding to the feature point of the reference image, and calculate a rotation operator and a translation operator according to the feature vector of the closest point;

C2. Perform corresponding rotation and translation on the feature vectors of the feature points of the floating image according to the rotation operator and the translation operator.

The three-dimensional image registration method based on conformal geometric algebra, wherein the similarity measure in the step C is calculated by the following formula:

Wherein, S _MN is the similarity measure, Xi _and Y _j are the feature vectors of the feature points of the floating image and the reference image respectively, Si is the minimum inner product of Xi and Yj, and both i and j are natural numbers.

The three-dimensional image registration method based on conformal geometric algebra, wherein, the step D also includes:

When the similarity measure is greater than or equal to the predetermined threshold and the number of rotations and translations is less than a predetermined number of times, continue to rotate and translate the feature vectors of the feature points of the floating image until the similarity measure is less than a predetermined threshold The value or the number of rotations and translations is greater than or equal to the predetermined number of times.

In the three-dimensional image registration method based on conformal geometric algebra, the predetermined threshold value is 1.0×10 ^-3 , and the predetermined number of times is 1000.

A three-dimensional image registration device based on conformal geometric algebra, wherein the device includes:

The feature point extraction unit is used to perform edge detection on the selected reference image and floating image for registration, obtain corresponding edge contours, and use an image segmentation algorithm to extract feature points of the reference image and floating image;

A feature vector conversion unit, configured to generate feature vectors of the feature points under the framework of conformal geometry algebra according to the feature points of the reference image and the floating image extracted by the feature point extraction unit;

A rotation and translation unit is used to perform rotation and translation operations on the feature vectors of the feature points;

A similarity measure calculation unit, configured to calculate the similarity measure of the feature vectors of the feature points of the reference image and the floating image after each rotation and translation operation performed by the rotation and translation unit;

A registration unit, configured to output the feature vectors of the feature points of the floating image at this time when the similarity measure is less than a predetermined threshold or the number of rotations and translations is greater than or equal to a predetermined number, and generate a registered floating image , and fuse the registered floating image with the reference image to obtain a final registered image.

The three-dimensional image registration device based on conformal geometric algebra, wherein the device also includes:

The image preprocessing unit is used to process the selected reference image and the floating image in advance, and process the resolution and size range of the selected reference image and the floating image to be consistent.

The three-dimensional image registration method and device based on conformal geometric algebra provided by the present invention realizes the direct alignment of three-dimensional data, can better locate the three-dimensional positions of tissues and organs, makes the registered images clearer, and has higher registration accuracy. Higher, the image display is more accurate.

Description of drawings

Fig. 1 is a flow chart of the three-dimensional image registration method based on conformal geometric algebra provided by the present invention.

Fig. 2 is a diagram of the first 8 slices of CT in an embodiment of the three-dimensional image registration method based on conformal geometric algebra provided by the present invention.

Fig. 3 is a diagram of the first 8 layers of MR_T1 in an embodiment of the three-dimensional image registration method based on conformal geometric algebra provided by the present invention.

Fig. 4 is a three-dimensional brain model reconstructed from different angles according to the CT image shown in Fig. 2 .

Fig. 5 is a three-dimensional brain model reconstructed from different angles according to the MR_T1 map shown in Fig. 3 .

Fig. 6 is a three-dimensional effect diagram of different angles after fusion of registered CT and MR_T1.

FIG. 7 is a partial slice diagram of the three-dimensional rendering shown in FIG. 6 .

Fig. 8 is a structural block diagram of a three-dimensional image registration device based on conformal geometric algebra provided by the present invention.

Fig. 9 is a structural block diagram of a preferred embodiment of a three-dimensional image registration device based on conformal geometric algebra provided by the present invention.

Detailed ways

The present invention uses conformal geometric algebra to reconstruct the positional relationship constraints of 3D medical images, analyzes the conformal geometric transformation of medical images, and constructs a new 3D medical image registration similarity measure, based on which a 3D medical image registration similarity measure is proposed. The quasi-algorithm is used for 3D registration of CT and MR_T1 images, so as to realize the direct alignment of 3D data, better locate the 3D positions of tissues and organs, and intuitively reflect the registration results.

In order to make the object, technical solution and advantages of the present invention more clear and definite, the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

Referring to Fig. 1, Fig. 1 is a flow chart of the three-dimensional image registration method based on conformal geometric algebra provided by the present invention, including the following steps:

Step S100, select the reference image and floating image for registration, perform edge detection on the reference image and floating image, obtain the corresponding edge contour, and use image segmentation algorithm to extract the reference image and floating image Feature points;

Step S200, according to the feature points of the reference image and the floating image, generate feature vectors of the feature points under the framework of conformal geometry algebra;

Step S300, perform several rotations and translations on the feature vectors of the feature points of the floating image, and calculate the similarity measure of the feature vectors of the feature points of the reference image and the floating image after each rotation and translation;

Step S400, when the similarity measure is less than a predetermined threshold or the number of rotations and translations is greater than or equal to a predetermined number, output the feature vectors of the feature points of the floating image at this time, generate a registered floating image, and The registered floating image is fused with the reference image to obtain a final registered image.

The above steps are illustrated and described in detail below in conjunction with specific embodiments.

First, select the reference image to be registered and the floating image , and the reference image and the floating image are preprocessed to make the resolution and size range of the two in the effective area consistent, so that the registration is more accurate.

和浮动图像

采用canny算子进行边缘检测，得到相应的边缘轮廓，根据得到的边缘轮廓，利用图像分割算法，提取参考图像

和浮动图像中的特征点的集合，分别为

和

。 In step S100, the test image

and the floating image

Using the canny operator for edge detection to obtain the corresponding edge contour, according to the obtained edge contour, use the image segmentation algorithm to extract the reference image

and the floating image The set of feature points in is respectively

and

.

和

转换成形式，因此，转换后

，

。之后对浮动图像的特性点的特征矢量

进行K次迭代的旋转和平移运算，首先计算旋转算子

和平移算子

，其中，K_min=0，K_max=1000，

和分别表示第K次迭代的旋转算子和平移算子。在计算时，和

通过公式

平方和最小为约束条件来求出，其中，S_i为X_i与Y_j的内积最小值。也即是获取浮动图像上与所述参照图像的特征点相对应的最近点，然后根据所述最近点的特征矢量，计算旋转算子和平移算子

，在计算出旋转算子

和平移算子

后，对浮动图像的特征点的特征矢量进行相应的旋转和平移运算，X_i变换后为

，

。然后计算每次迭代旋转和平移后X_i与Y_j的相似测度，计算公式为：

。 In order to achieve better registration, the present invention further processes the above feature points, specifically, under the framework of Conformal Geometric Algebra (CGA), preferably space, will

and

converted to of the form, therefore, after conversion

,

. After that, the feature vectors of the feature points of the floating image

Carry out rotation and translation operations for K iterations, first calculate the rotation operator

and translation operator

, where K _min =0, K _max =1000,

and Denote the rotation operator and translation operator of the K-th iteration, respectively. When calculating, and

by formula

The minimum sum of squares is used as a constraint condition to obtain, among them, S _i is the minimum value of the inner product of Xi _and Y _j . That is to obtain the closest point on the floating image corresponding to the feature point of the reference image, and then calculate the rotation operator according to the feature vector of the closest point and translation operator

, after calculating the rotation operator

and translation operator

After that, the corresponding rotation and translation operations are performed on the feature vectors of the feature points of the floating image, and _Xi is transformed into

,

. Then calculate the similarity measure between X _i and Y _j after each iteration of rotation and translation, the calculation formula is:

.

判断，当所述相似测度小于一预定阀值

（

）或者旋转和平移的次数k大于或者预定次数K_max=1000时，输出所述浮动图像此时的特征点的特征矢量，并生成配准后的浮动图像，完成初步配准，并将配准后的浮动图像与参照图像N融合，得到最终的配准图像。否则将迭代次数加1，继续进行迭代旋转和平移运算，直至所述相似测度

小于预定阀值

或者旋转和平移的次数K大于预定次数K_max。 Calculate the similarity measure by the above formula After that, the similarity measure

judgment, when the similarity measure less than a predetermined threshold

(

) or when the number k of rotation and translation is greater than or predetermined times K _max =1000, output the feature vector of the feature point of the floating image at this time, and generate the registered floating image, complete the preliminary registration, and register The final floating image is fused with the reference image N to obtain the final registration image. Otherwise, add 1 to the number of iterations, and continue to perform iterative rotation and translation operations until the similarity measure

less than the predetermined threshold

Or the number K of rotation and translation is greater than the predetermined number K _max .

In order to illustrate the above-mentioned registration process more concretely and intuitively, the present invention uses experimental image data as an example for illustration.

The present invention selects the CT and MR_T1 sequence of the head scan in "The Retrospective Image Registration Evaluation Project" of Vanderbit University in the United States as an example for description. CT is a reference image, and MR_T1 is a floating image.

Specifically, the CT map has 28 slices with a pixel size of 0.65mm in the x and y directions and 4.0mm in the z direction. Figure 2 is the image of the first 8 layers of CT. It can be seen that due to the different positions in the Z direction, the human brain bone information displayed on each layer is different. Comparing the same layers of each CT image, it can be seen that they have similar outer contours, but the internal tissues displayed are more or less different due to different imaging focuses. It can be seen from Fig. 2 that imaging at different Z-axis positions reflects different details of patient tissue.

The MR_T1 map has 26 layers with a pixel size of 1.25mm in the x and y directions and 4.0mm in the z direction. Figure 3 is the first 8 layers of MR_T1. It can be seen that due to the different positions in the Z direction, the soft tissue information of the patient's brain displayed on each layer is different.

Three-dimensional images can be reconstructed using the sequence diagrams shown in Figures 2 and 3. Figure 4 shows three-dimensional brain models reconstructed from CT images at different angles. Different parts of the brain skeleton can be seen from different angles. Figure 5 shows the three-dimensional brain models reconstructed from different angles based on the MR_T1 image. Since the MR_T1 image reflects the main soft tissue information of the brain, the three-dimensional model image reconstructed based on the MR_T1 image has fewer see-through parts, only between internal tissues and bones. There is a small amount of gap perspective in between.

和一个旋转算

，也就是说，MR_T1在平移算子

和一个旋转算子

的作用下将和CT配准。在实际算法中，由于采用的寻优策略是基于ICP算法的，实际上MR_T1是在平移算子

和一个旋转算

的连续的作用下取得配准。配准后，MR_T1和CT图骨骼部分合并，而MR_T1中的软组织部分将填入CT图的骨骼空隙中，图6是经过配准后的CT和MR_T1融合的不同角度的三维效果图。 Before registration, from the perspective of five-dimensional conformal geometric algebra, the difference between the CT image and the MR_T1 image in the world coordinate system is a translation operator

and a rotation

, that is, MR_T1 in the translation operator

and a rotation operator

It will be registered with CT under the action of . In the actual algorithm, since the optimization strategy adopted is based on the ICP algorithm, in fact MR_T1 is shifting the operator

and a rotation

The registration is obtained under the continuous action of . After registration, the bone parts of MR_T1 and CT images are merged, and the soft tissue part in MR_T1 will be filled in the bone space of the CT image. Figure 6 is a three-dimensional rendering of the fusion of CT and MR_T1 at different angles after registration.

Figure 7 is a part of the sliced image after the fusion of the CT image and the MR_T1 image after registration. The dark part is the tissue belonging to CT, and the light part is the tissue belonging to MR_T1. It can be seen from Figure 7 that the registered slices are different from the pre-registered CT and MR_T1 slices. The same tissues in the CT and MR_T1 slices are aligned and fused, while different tissues are displayed complementary to each other in position. This enables the final registered images to better locate the three-dimensional positions of tissues and organs, making the registration results more intuitive.

It should be pointed out that the above-mentioned FIG. 2 to FIG. 7 are only effect diagrams used for explaining the registration process, and are not used to limit this embodiment.

Based on the above three-dimensional image registration method, the present invention also provides a three-dimensional image registration device based on conformal geometric algebra, as shown in FIG. 8, wherein the shown device includes:

The feature point extraction unit 10 is used to perform edge detection on the selected reference image and floating image for registration, obtain corresponding edge contours, and use an image segmentation algorithm to extract feature points of the reference image and floating image ;

A feature vector conversion unit 20, configured to generate feature vectors of the feature points according to the feature points of the reference image and the floating image extracted by the feature point extraction unit 10 under the framework of conformal geometric algebra;

A rotation and translation unit 30, configured to perform rotation and translation operations on the feature vectors of the feature points;

A similarity measure calculation unit 40, configured to calculate the similarity measure of the feature vectors of the feature points of the reference image and the floating image after each rotation and translation operation performed by the rotation and translation unit 30;

The registration unit 50 is configured to output the feature vectors of the feature points of the floating image at this time when the similarity measure is less than a predetermined threshold or the number of rotations and translations is greater than or equal to a predetermined number, and generate a registered floating image. image, and fuse the registered floating image with the reference image to obtain a final registered image.

Further, in order to perform image registration more accurately, as shown in FIG. 9 , the device further includes: an image preprocessing unit 60, which is used to process the selected reference image and floating image in advance, and the selected reference image Resolution and size ranges are handled consistently for images and floats.

In summary, through the 3D image registration method and device based on conformal geometric algebra provided by the present invention, the direct alignment of 3D data is realized, the 3D position of tissues and organs can be better located, and the registration result is more intuitive. The registered images are clearer and the registration accuracy is higher.

It should be understood that the application of the present invention is not limited to the above examples, and those skilled in the art can make improvements or transformations according to the above descriptions, and all these improvements and transformations should belong to the protection scope of the appended claims of the present invention.

Claims (9)

1. A three-dimensional image registration method based on conformal geometric algebra, is characterized in that, comprises the following steps: A. Select the reference image and the floating image for registration, perform edge detection on the reference image and the floating image, obtain the corresponding edge contour, and use the image segmentation algorithm to extract the features of the reference image and the floating image point; B. According to the feature points of the reference image and the floating image, under the framework of conformal geometry algebra, generate the feature vector of the feature points; C. Perform several rotations and translations to the feature vectors of the feature points of the floating image, and calculate the similarity measure of the feature vectors of the feature points of the reference image and the floating image after each rotation and translation; D. When the similarity measure is less than a predetermined threshold, or the number of rotations and translations is greater than or equal to a predetermined number of times, output the feature vector of the feature point of the floating image at this time, generate a registered floating image, and The registered floating image is fused with the reference image to obtain a final registered image. 2. The three-dimensional image registration method based on conformal geometric algebra according to claim 1, wherein the step A further comprises: pre-processing the distribution ratio and size range of the reference image and the floating image to be consistent . 3. The three-dimensional image registration method based on conformal geometric algebra according to claim 1, characterized in that, in the step A, the canny operator is used to perform edge detection on the reference image and the floating image. 4. the three-dimensional image registration method based on conformal geometric algebra according to claim 1, is characterized in that, described step C also comprises: C1. Obtain the closest point on the floating image corresponding to the feature point of the reference image, and calculate a rotation operator and a translation operator according to the feature vector of the closest point; C2. Perform corresponding rotation and translation on the feature vectors of the feature points of the floating image according to the rotation operator and the translation operator. 5. The three-dimensional image registration method based on conformal geometric algebra according to claim 1, wherein the similarity measure in the step C is calculated by the following formula:

Wherein, S _MN is the similarity measure, Xi _and Y _j are the feature vectors of the feature points of the floating image and the reference image respectively, Si is the minimum inner product of Xi and Yj, and both i and j are natural numbers. 6. the three-dimensional image registration method based on conformal geometric algebra according to claim 1, is characterized in that, described step D also comprises: When the similarity measure is greater than or equal to the predetermined threshold and the number of rotations and translations is less than a predetermined number of times, continue to rotate and translate the feature vectors of the feature points of the floating image until the similarity measure is less than a predetermined value The threshold or the number of rotations and translations is greater than or equal to the predetermined number of times. 7 . The three-dimensional image registration method based on conformal geometric algebra according to claim 1 , wherein the predetermined threshold value is 1.0×10 ⁻³ , and the predetermined number of times is 1000. 8. A three-dimensional image registration device based on conformal geometric algebra, characterized in that the device comprises: The feature point extraction unit is used to perform edge detection on the selected reference image and floating image for registration, obtain corresponding edge contours, and use an image segmentation algorithm to extract feature points of the reference image and floating image; A feature vector conversion unit, configured to generate feature vectors of the feature points under the framework of conformal geometry algebra according to the feature points of the reference image and the floating image extracted by the feature point extraction unit; A rotation and translation unit is used to perform rotation and translation operations on the feature vectors of the feature points; A similarity measure calculation unit, configured to calculate the similarity measure of the feature vectors of the feature points of the reference image and the floating image after each rotation and translation operation performed by the rotation and translation unit; A registration unit, configured to output the feature vectors of the feature points of the floating image at this time when the similarity measure is less than a predetermined threshold or the number of rotations and translations is greater than or equal to a predetermined number, and generate a registered floating image , and fuse the registered floating image with the reference image to obtain a final registered image. 9. The three-dimensional image registration device based on conformal geometric algebra according to claim 8, wherein the device further comprises: The image preprocessing unit is used to pre-process the selected reference image and the floating image, and process the resolution and size range of the selected reference image and the floating image to be consistent.

Images