KR101258814B1 - Nonrigid registration method and system with density correction of each tissue and rigidity constraint of tumor in dynamic contrast-enhanced breast mr images - Google Patents

Abstract

PURPOSE: A nonrigid registration method and system applying the tumor rigidity constraint and the density correction of each tissue in a dynamic contract enhanced breast MR image are provided to accurately classify a tissue by regularizing a contrast density. CONSTITUTION: A breast region is automatically divided by removing a breast skin and a pectoral(S110). A fuzzy C-means grouping method is used in the divided breast region. A breast tissue is classified into fat, the mammary gland, a tumor, and a blood vessel(S120). The tumor is divided by using a density and a morphological operation(S130). The density is corrected in the fat and the mammary gland(S140). [Reference numerals] (AA) Start; (BB) End; (S110) Breast division based on slope and direction information; (S120) Tissue classification using a fuzzy C-means grouping method; (S130) Tumor division based on brightness value and shape information; (S140) Rigid body matching based on normalized mutual information and brightness value correction; (S150) Non-rigid body matching applying tumor rigid constraints;

Description

Nonrigid Registration Method and System with Density Correction of Each Tissue and Rigidity Constraint of Tumor in Dynamic Contrast-Enhanced Breast MR Images}

The present invention relates to a non-rigid registration method and system, and more particularly, to a non-rigid registration method and system applying tissue-specific brightness correction and tumor stiffness constraint in dynamic contrast enhanced breast MR imaging.

Analyzing tumor perfusion on dynamic contrast-enhanced MR breast images (DCE-MR) taken four to five times at regular intervals before and after the contrast agent was diagnosed It is one of the non-invasive methods that enable treatment planning and observation of effects of malignant tumors.

Since dynamic contrast-enhanced MR breast imaging can be partially altered inside and outside of the breast due to respiratory differences and heart rate, non-rigid registration between pre and post-enhancement imaging is required.

Embodiments of the present invention have been made in accordance with the necessity as described above, and an object of the present invention is to provide a non-rigid registration method and system applying brightness correction and tumor stiffness constraint for each tissue in a dynamic contrast enhanced breast MR image.

Non-rigid registration method according to an embodiment of the present invention for achieving the above object, the step of automatically segmenting the breast region by removing the breast skin and pectoral muscle using the slope and direction information; Classifying breast tissue into fat, mammary tissue, tumor, and blood vessel using fuzzy C-means clustering technique in the breast region divided by the automatic segmentation step; Automatically extracting the location of the tumor by using brightness information according to the degree of contrast in the tumor information classified by the classification step or the target image and dividing the tumor site by using the brightness value and morphological calculation; Performing a rigid registration based on normalized mutual information between the target and the source image, and correcting brightness values of fat and mammary gland tissue having different brightness values under the influence of contrast medium; And performing daemon-based non-rigid registration applying constraints to maintain the rigid structure in the divided tumor areas and applying normalized weights to each breast tissue to reduce deformation due to the difference in brightness values of the tumor areas. It is characterized by.

Herein, the enhancement image before the contrast agent injection is defined as a target image, and the enhancement image after the contrast agent injection is a source image, and the source image may be configured as a data set continuously photographed at predetermined intervals.

In the automatic segmentation step, after anisotropic diffusion filtering is applied to remove noise while maintaining important information such as edges, the automatic segmentation step is performed to correct the degree of contrast of breast tissue that is different for each data. By performing the tissue classification after normalizing the brightness value, the accuracy of performing breast segmentation and tissue classification can be improved.

The tumor site segmentation step may determine the location of the tumor site from the tumor information classified as fuzzy C-mean clustering, and if the location of the tumor site is difficult to locate using the classified tumor information due to low contrast, the target and The location of the tumor can be automatically extracted by using the contrast image of the tumor between the source images.

In the rigid body matching and brightness value correcting step, the global motion can be corrected by obtaining an optimal rigid body transform without being affected by the difference between the brightness values of the target and source images. Value correction can be used to correct the effects of contrast agents on fat and mammary glands in the source image.

The non-rigid registration step may correct local movement while reducing volume and morphological deformation of the tumor region between the breast region in the target and source images due to respiration difference and heartbeat.

Non-rigid registration system according to an embodiment of the present invention for achieving the above object, the breast segmentation unit for automatically segmenting the breast area by removing the breast skin and pectoral muscle using the slope and direction information; A tissue classification unit classifying breast tissue into fat, mammary tissue, tumor, and blood vessel using a fuzzy C-means clustering technique in the breast region divided by the breast division; Tumor segmentation unit for automatically extracting the location of the tumor using the difference of the brightness value according to the degree of contrast in the tumor information or target and source image classified by the tissue classification unit, and segmenting the tumor site using the brightness value and morphological operations ; A matching and correction unit configured to perform rigid registration based on normalized mutual information between the target and the source image, and to correct brightness values of fat and mammary gland tissue having different brightness values under the influence of contrast medium; And a non-rigid registration unit that performs a daemon-based non-rigid registration applying constraints to maintain rigid structures and reducing normalized weights for each breast tissue in order to reduce deformation due to differences in brightness values of the tumor site. do.

Herein, the enhancement image before the contrast agent injection is defined as a target image, and the enhancement image after the contrast agent injection is a source image, and the source image may be configured as a data set continuously photographed at predetermined intervals.

After applying anisotropic diffusion filtering to remove noise while maintaining important information such as the breast segmentation part and the edge, the breast region is divided and brightness is corrected to correct the degree of contrast of the breast tissue that is different for each data. By performing tissue classification after normalizing the values, the accuracy of performing breast division and tissue classification can be improved.

The tumor segmentation unit may determine the location of the tumor site from the tumor information classified by fuzzy C-mean clustering, and when the location of the tumor site is difficult to locate using the classified tumor information due to low contrast, the target and source images The difference in the contrast of the liver tumor can be used to automatically extract the location of the tumor.

The matching and correction unit may correct the global motion by obtaining an optimal rigid body transformation without being affected by the difference between the brightness values of the target and the source image. In this case, the matching amount may be reduced by sampling, and the brightness value correction may be performed. In the source image, the effect of contrast agent on fat and mammary tissue can be corrected.

The non-rigid registration unit may correct local movement while reducing volume and shape deformation of the tumor region between the breast region in the target and source images due to respiration difference and heartbeat.

In the MR image according to an embodiment of the present invention, the breast and tumor segmentation and matching method and system can be effectively used for diagnosing the tumor malignancy and observing the therapeutic effect of the malignant tumor, and photographing with other modalities such as X-ray and ultrasound. It can be applied to breast and tumor analysis through registration with the breast image.

1 is a flow chart showing a non-rigid registration method applying the brightness value correction and tumor stiffness constraint for each breast tissue according to an embodiment of the present invention.
2 is a diagram comparing anisotropic diffusion filtering results according to k values.
3 is a diagram illustrating a breast region division process.
4 is a diagram showing breast tissue classification.
5 is a view showing a tumor location extraction and segmentation process.
6 is a diagram comparing histograms before and after correcting brightness values.
7 is a view comparing the results of non-rigid registration according to whether the tumor site stiffness constraint applied.
8 is a view showing a non-rigid registration result according to whether brightness value correction and tumor stiffness constraint are applied.
9 is a view showing the results of evaluating the accuracy of non-rigid registration applying the brightness correction and tumor stiffness constraint.
10 is a view showing a rigid and non-rigid matching execution time.
FIG. 11 is a diagram schematically illustrating a non-rigid registration system applying brightness correction and tumor rigidity constraint for each breast tissue according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, a detailed description of known techniques well known to those skilled in the art may be omitted.

In describing the constituent elements of the present invention, the same reference numerals may be given to constituent elements having the same name, and the same reference numerals may be given thereto even though they are different from each other. However, even in such a case, it does not mean that the corresponding component has different functions according to the embodiment, or does not mean that the different components have the same function. It should be judged based on the description of each component in the example.

In the following description of the embodiments of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

In addition, in describing the component of this invention, terms, such as 1st, 2nd, A, B, (a), (b), can be used. These terms are intended to distinguish the constituent elements from other constituent elements, and the terms do not limit the nature, order or order of the constituent elements. When a component is described as being "connected", "coupled", or "connected" to another component, the component may be directly connected or connected to the other component, Quot; may be "connected," "coupled," or "connected. &Quot;

As a previous study to correct regional motion of breast in DCE-MR breast imaging, Rueckert et al. Proposed non-rigid registration using b-spline-based free-form deformation. In this method, optimal deformation vectors are obtained by minimizing the cost function consisting of the normalized mutual information and the smoothing function using the second derivative to evaluate the similarity between images. In this method, there was no difference in the volume and shape of tumors before and after contrast imaging, but it was transformed by the change of brightness due to contrast agent.

Various non-rigid registration studies have been proposed to maintain the rigid structure of the tumor to correct this. Tanner et al., When performing b-spline-based non-rigid registration, coaxial the control points corresponding to the tumor site, obtain an average vector of the surrounding deformation vectors, and apply the same to the tumor site. Rigid body transformation in the and directions. Rohlfing et al. Also defined the cost function for optimization in b-spline-based non-rigid registration as the sum of normalized mutual information and Jacobian determinant, and minimized the cost Non-rigid registration was performed while reducing volume and shape deformation. In addition, Li et al. Proposed a brightness value-based non-rigid matching algorithm using a radial basis function.

The radial-based function is an approximation model that defines the corresponding reference point in the pre- and post-enhancement image and interpolates the local deformation according to the distance from the reference point for pixels other than the reference point. Jacobian determinants for tumor sites were added to the cost function to reduce tumor volume and morphology. In addition to methods of restricting the stiffness of the tumor site, studies have been proposed to reduce the deformation of the tumor and increase the accuracy of non-rigid registration by reducing the difference in brightness due to the contrast agent between before and after contrast imaging. Sun et al. Normalized brightness values using average and variance of brightness values based on conditional probability density function in pre- and post-enhanced augmented images in brightness-based non-rigid registration.

The cost function was calculated using the difference of brightness values in the pre and post contrast enhancement images with normalized brightness values, and the area with the difference in brightness values above a certain threshold such as tumor was not calculated. Reduced. Hayton et al. Performed a non-rigid registration after compensating for the problem that the same tissue in the breast had different brightness values due to the influence of artifacts such as breast coils through MR field imaging through bias field correction. . In addition, Zheng et al. Classified the breast into 15 regions using graph cuts, generated contrast enhancement maps, and then performed non-rigid registration after reducing the effects of contrast using contrast enhancement maps in augmented images.

In the present invention, we propose a demo-based non-rigid registration technique for correcting the brightness of each breast tissue in pre- and post-enhanced augmented images and correcting the movement of the breast region while reducing tumor volume and shape deformation.

Since the degree of contrast is different for each breast tissue before and after contrast imaging, the breast region is segmented based on the slope and direction information and the tissue is classified using fuzzy C-means clustering. . In addition, in order to give rigidity to the tumor site, the tumor site is segmented using brightness and shape information. Normalized mutual information-based rigid matching, brightness correction, and non-rigid registration with tumor stiffness constraints reduce the deformation of the breast area that may be caused by the effects of contrast agents, and the tumor area is rigidly transformed while the breast area is deformed, thereby preforming the breast. · Compensate for global and regional movements between augmented images.

1 is a flow chart of non-rigid registration applying the brightness correction and tumor stiffness constraint for each breast tissue proposed in the present invention. The augmented image before the contrast agent injection is defined as a target image, and the augmentation image after the contrast agent injection is defined as the source image, and the source image is composed of four data sets taken continuously at intervals of 84 seconds. The non-rigid registration method proposed in the present invention is largely composed of four steps.

First, the breast region and the pectoral muscle are removed using the slope and direction information, and the breast region is automatically segmented, and the breast tissue is classified into fat, mammary tissue, tumor, and blood vessel using fuzzy C-means clustering technique. (S110). Second, the location of the tumor is automatically extracted by using the difference in brightness value according to the degree of contrast in the classified tumor information or the target image and the source image, and the tumor area is segmented using the brightness value and the morphological operation (S120). Third, normalized mutual information-based rigid body registration between target and source images is performed, and brightness values are corrected for fat and mammary gland tissues having different brightness values under the influence of contrast medium (S130). Fourth, in order to reduce the deformation caused by the difference in the brightness value of the tumor site, a constraint based on maintaining the rigid structure at the tumor site is performed, and daemon-based non-rigid registration is applied by applying normalized weights to each breast tissue (S140).

First, the process of step S110 and step S120 will be described.

Due to the influence of the contrast agent, the target and source images have different distributions of brightness values, which may be a factor that can cause deformation in areas that should not be deformed when performing daemon-based non-rigid registration. In order to reduce the occurrence of such a false deformation, it is necessary to correct the brightness between the same breast tissue in the two images, and to do this, segmentation and tissue classification of the breast region must be performed first.

Segmentation and tissue classification of the breast region are performed on a source image showing the degree of contrast between different breast tissues in a dynamic contrast enhanced MR image. Anisotropic diffusion filtering is first applied to MR images to improve the accuracy of breast segmentation and tissue classification. An anisotropic diffusion filter is a filtering technique that removes noise while reducing loss of important information such as edges having large differences in brightness values in MR images, and is performed using Equations 1 and 2 below.

[Equation 1]

&Quot; (2) "

In this case, Δ is a Laplacian, ∇ is a slope, div (·) is a divergence operator. c (x, y, t) is a divergence coefficient of the (x, y) coordinate at time t. In general, the gradient function of the image is used. Equation 2 is a slope function defined for c (x, y, t). k is a constant that controls the sensitivity of the edge and is defined as a value between 0 and 1.

2 shows the results of anisotropic diffusion filtering according to k values. (A) of FIG. 2 is an original source image, and (b), (c) and (d) of FIG. 2 show results of k values of 0.2, 0.5, and 0.8 at 50 identical repetitions. As the value of k increases, the noise is removed relatively, but the image tends to smooth. In the present invention, the k value is experimentally defined as 0.5 to remove the noise and to prevent the image from being smoothed much.

3 shows a process of dividing a breast region in each slice after applying an anisotropic diffusion filter.

Figure 3 (a) shows a process of extracting the boundary points of the air-breast skin and breast skin-breast. The images were taken in the sagittal direction, and the air region showed an average signal intensity of 0 to 50 and the breast skin region showed a signal intensity of 150 to 450. The air-breast skin boundary point is extracted by using the difference of brightness values for each tissue, and the point of inclination in the normal direction within 5 mm from the extracted air-breast skin boundary points to the inside of the breast is defined as the breast skin-breast boundary point. Extract.

Figure 3 (b) is the result of removing the breast skin after interpolation of the extracted breast skin-breasts using b-splines.

In addition, (c) of Figure 3 shows the process of extracting the boundary points of the breast-thoracic muscle. We define the region of interest in V-shape and extract the vertical lines with high slope through linear filtering. Extract the point with the largest slope (i, j) from the y-axis centerline indicated by the dotted line and track the point with the largest slope in both the up and down directions. At this time, the breast-thoracic boundary point is extracted by tracing the point where the slope is the largest among five points existing within 2 pixels to the left and 2 pixels to the right from the x coordinate of the point (i, j).

FIG. 3 (d) shows the final breast region segmentation result of removing the pectoral muscle after interpolating breast-thoracic boundary points extracted from the result of FIG. 3 b using b-splines.

Within the segmented breast region, the breast tissues of fat, mammary tissue, tumor and blood vessels are present and the degree of contrast is different for each data, so the maximum and minimum brightness values in the segmented breast region are normalized to a value between 0 and 1. Let's do it. At this time, the pixel corresponding to the lower 5% of the total number of pixels is 0, and the pixel corresponding to the upper 5% is 1.

FIG. 4A is a histogram of normalized brightness values in a divided breast region with respect to one data. In the source image, fat appears darker than mammary tissue and tumor, while mammary tissue and tumor are contrasted, resulting in relatively bright brightness. To classify breast tissue into fat, mammary gland, tumor and blood vessel, it is divided into six layers using fuzzy C-means clustering technique. In this case, the clustering technique is performed on a well-concealed source image so that the tissue can be classified well. The data with malignant tumors are performed on the first source image and the fourth source image. Each pixel in the breast has a probability of belonging to six layers and the sum of belonging probabilities must be one. The initial belonging probability of each pixel for clustering is automatically defined using the histogram of FIG. Divide the normalized brightness histogram into six ranges: 0.0, 0.0 to 0.25, 0.26 to 0.50, 0.51 to 0.75, 0.76 to 0.99, and 1.0. Give a probability of belonging to 1.0, and pixels corresponding to 0.0 to 0.25 are 0.25, 0.5, 0.25, and 0.26 to 0.50 for layers 1, 2, and 3, respectively, 0.25, 0.25, and 0.5 for layers 1, 2, and 3, respectively. Gave the probability of belonging. Pixels with normalization values of 0.51 to 0.75, which are likely to correspond to the mammary gland, give the probability of belonging to 0.7 and 0.3 in layers 4 and 5, and pixels corresponding to 0.76 to 0.99 in layers 4 and 5, respectively. Gave a chance to belong. A pixel corresponding to a normalization value of 1.0, which is likely to correspond to a tumor, gave Tier 6 a probability of membership of 1.0. The average brightness value is obtained for pixels having a high probability of belonging to each layer, and the probability of belonging to each layer is repeatedly calculated according to the difference between the brightness value of each pixel and the average brightness value of each layer. In this case, if the difference between the previously calculated membership probability and the currently calculated membership probability hardly changes below a certain value (0.01), the convergence is converged, and each pixel is classified into the layer having the largest membership probability among the six layers. Figure 4 (b) is the result classified into six layers (three breast tissues) through the fuzzy C-means clustering.

Next, the process of step S130 will be described.

Tumor segmentation should be performed because the tumor site needs to be known in order to give rigid constraints to the tumor when performing non-rigid registration. Since the tumor site classified in step S120 tends to be smallly divided and contains blood vessels, it is necessary to accurately segment the tumor site.

If the location of the tumor site is low, such as a benign tumor, and the location of the tumor site is difficult to determine using the tumor information classified in step S120, the location of the tumor is automatically extracted using the difference in the contrast between the target and source images. .

5 shows the process of automatic extraction and segmentation of tumor location.

In order to automatically extract the tumor location, a difference image is generated by subtracting the brightness value of the target image from the source image as shown in FIG. 5 (a), and average filtering as shown in FIG. 5 (b) to remove the noise. filtering). In order to remove small components, a gray-scale opening operation is performed as shown in (c) of FIG. 5. Histogram analysis automatically defines the threshold with the peak brightness value × 0.5, which occupies the largest number of pixels, creates a binary image using the defined threshold, and then generates the highest volume through 3D connected component labeling. Select a large component. At this time, in order to prevent the internal tissues such as the lungs and the heart from being selected as the largest component, the components are selected in the breast region divided in step S110. 5 (d) shows a result of showing the selected component as a binary image, and this region becomes the location of the tumor. As shown in (e) of FIG. 5, the region of interest of the tumor is defined as 1.5 times larger in the optimal volume of the selected component using the tumor region or the difference image classified as fuzzy C-means clustering. The initial tumor site is segmented by performing the 3D region growth method within the region of interest of the tumor with the seed site as a seed point extracted from the source image. Inside the tumor, areas that are not contrasted due to necrosis may appear, and hole filling is performed to include these areas as tumor areas. Hole filling results in a result as shown in FIG. 5 (f) by inverting the result after performing the 3D region growth method with the seed point against the background of the initial tumor segment in the tumor region of interest. The tumor is divided as shown in (g) of FIG. 5 by performing a close operation, which is a morphological operation, to remove blood vessels connected to the tumor.

Next, the process of step S140 will be described.

Since the target and source images are photographed at different time points at predetermined time intervals (which can be executed at 84 second intervals), positional and rotational differences may occur due to respiratory differences and heartbeats. Rigid registration is performed to compensate for this global motion.

In the present invention, the normalized mutual information-based rigid body matching using the distribution of the brightness values is performed to obtain an optimal rigid body transform vector without being affected by the difference between the brightness values of the target and source images. At this time, the normalized mutual information is calculated by sampling at intervals of 4 pixels on the x and y axes and 2 slices on the z axis in the original image to reduce the amount of computation.

Since the mammary gland is more contrasted than the fat, the difference in brightness between the target and the source image is large, whereas the fat is not much larger than the mammary gland. As described above, since the degree of contrast for each breast tissue is different, the brightness value correction for each breast tissue (fat, mammary tissue) is performed using Equation 3 after performing rigid registration.

&Quot; (3) "

x means pixels corresponding to adipose or mammary gland except tumor area, and V _s (x) and V ′ _s (x) refer to brightness values of the original source image and brightness values of the source image after correction, respectively. P _s is the peak brightness value that occupies the largest number of pixels in the histogram of fat or mammary tissue in the source image, and P _t is the highest brightness value in the histogram of the corresponding tissue in the target image.

By performing histogram matching using the difference between the highest brightness values of the source and target images, the brightness value of the source image is corrected, and after brightness correction, pixels having a brightness value of 0 or less have a brightness value of zero. In addition, brightness value smoothing using a Gaussian mask is performed to continuously create discrete brightness values that appear at the border of breast tissue.

FIG. 6 analyzes histograms of fat and mammary gland corresponding to target and source images before and after brightness correction. FIG. 6A shows a result of analyzing histograms of a target image and four source images photographed after imaging. 6 (b) shows that the brightness value distribution of the source images is similar to the target image as a histogram of the target and the source image after the brightness value correction.

Finally, the process of step S150 will be described.

Non-rigid registration is required to correct regional movement between breast regions in target and source images due to respiratory differences and heartbeats. The difference in brightness due to the influence of contrast agent on the tumor site alters the shape and volume of the tumor when daemon-based non-rigid registration is performed. In order to solve this problem, the present invention performs daemon-based non-rigid registration applying rigid constraints to divided tumor sites.

The breast region of the source image is modified similarly to the breast region of the target image, and the deformation vector u (x) of each pixel except for the tumor region in the breast is calculated as in Equation 4.

&Quot; (4) "

In this case, V _t (x) and V _s (T (x)) denote brightness values of the pixel x in the breast region of the target image and the breast region of the rigid-converted source image, respectively, and ∇V _t and ∇V _s The slope in the target image and the source image, ∥ · ∥ means the magnitude of the slope. α and β were defined as 0.5 for each weight. In daemon-based non-rigid registration, the slope determines the direction of the transform vector and the difference in brightness between the two images determines the size of the transform vector. Since the larger the difference between the brightness values between the two images, the larger the size of the modified vector is, the difference in the brightness value of the tumor caused by the influence of the contrast agent on the target and source images is a factor that changes the shape and volume of the tumor due to the modified vector.

7 (a) shows that the tumor volume is reduced and the shape is modified as a result of the daemon-based non-rigid registration. In order to constrain the stiffness of the tumor region, as shown in FIG. Obtain

[Equation 5]

In this case, Ω means the area around the tumor and N means the number of pixels in the area. Since all pixels corresponding to the tumor site have the modified vector u _tumor having the same direction and size, the rigid structure of the tumor site is maintained.

Since the breast is a soft tissue, the strain vector at one pixel must appear continuously without overlapping the surrounding strain vectors. Vector smoothing using a Gaussian mask was performed for regularization of the modified vector and normalization of 1.0, 0.7, 0.5, 0.3, and 0.0 according to the tumor, the area around the tumor, the mammary tissue, the adipose, and other regions. Weighted. The tumor site is given the highest weight and the lower the distance from the tumor, the lower the weight.The higher the weight, the greater the degree of vector smoothing to maintain the rigid structure, and the lower the weight, the smaller the degree of vector smoothing. To make it happen.

Non-rigid matching is repeated until the cost function is minimized and the cost function values obtained at the previous and present time are almost unchanged below the threshold (0.05), and the cost function is defined as in Equation 6.

&Quot; (6) "

Г is the breast region, and V _t and V _s are the brightness values of pixel x in the target and source images. Φ is the tumor site and N is the total number of pixels in the tumor site. α is defined as 1.0 in the present invention as a weight. J is Jacoby's determinant, which means that if the absolute value of J is greater than 1 for pixel x, the periphery of x is expanded, if it is less than 1, it is reduced, and if it is 1, there is no change. Through this, the degree of deformation of the tumor volume can be measured, and as the value increases, the tumor area is deformed much. Therefore, the deformation of the tumor area is reduced by performing non-rigid registration so that CF is minimized.

The experiments consisted of a total of six patient data with one malignant or benign tumor taken with Siemens Sonata 1.5T set with a sequence of TR = 4.9 sec, TE = 1.83 sec, flip angle = 12 °, FOV = 170 × 170 mm. Was performed against. Each experimental data consisted of five DCE-MR volume images taken continuously before and after the injection of contrast medium, with 448 × 448 pixels and 0.38 × 0.38 mm ² pixel size. The other two data images were 896 × 896 pixels, the pixel size was 0.19 × 0.19mm ^2, and the data was reduced to 448 × 448 pixels with a pixel size of 0.38 × 0.38mm ² . The slice interval is 1.0 ~ 1.4mm and the number of slices is 96.

In order to evaluate the non-rigid registration method applying the brightness correction and tumor stiffness constraint proposed in the present invention, the results of visual evaluation, accuracy evaluation and execution time were analyzed.

FIG. 8 shows the results of non-rigid registration before and after non-rigid registration according to whether brightness correction and tumor rigidity constraints are applied to patient data with malignant and benign tumors.

8 (a) and 8 (e) show a source image and a target image, respectively. Tumors and mammary tissues in the source image appear brighter than the corresponding tissues in the target image due to the influence of contrast medium. 8 (b) shows the result similar to the original source image because the global motion difference is corrected with the movement and rotation conversion to the transformed source image after rigid registration, but the degree is not large.

In addition, Figure 8 (c) is a modified source image after performing the conventional daemon-based non-rigid registration, the difference in brightness and volume between the source and the target image due to the contrast medium to generate a modified vector, thereby reducing the volume and shape of the tumor Transformed. However, after the non-rigid registration with proposed brightness correction and tumor stiffness constraint, the tumor volume and morphology were hardly changed as shown in Fig. 8 (d).

(F)-(i) of FIG. 8 is a difference image between the source image of (a)-(d) and the target image (e) of FIG. 8 and shows a matching error between the two images. When comparing (h) and (i) of FIG. 8, the proposed method more accurately corrected the contrast and motion difference in the breast region excluding the tumor, compared to the conventional daemon-based non-rigid registration. It can be seen that the difference in brightness due to the contrast agent is more clearly shown.

The sum of squared difference (SSD) of the entire breast area was measured in order to compare the results of non-rigid registration according to the adjustment of the brightness value and the application of the constraint of tumor stiffness. The SSD between the source and the target image is calculated using Equation 7.

[Equation 7]

In this case, u and v mean a source and a target image, respectively. x is the 3D coordinate of voxels in the entire breast region Γ in the target image, N is the total number of voxels in the breast region, and T (x) is the coordinate of voxel x after non-rigid registration. The brightness difference is lower as the accuracy of matching between the two images is higher.

9 shows the results of SSD measurement. The measured values per data were calculated as averages after performing non-rigid registration between the target image and the four source images. In FIG. 9, the initial error of SSD for 6 data was 28.53 ± 6.20 on average and decreased slightly to 27.40 ± 5.88 after rigid registration. After the daemon-based non-rigid registration, the SSD value decreased to 24.56 ± 5.50, but the volume and shape of the tumor were severely reduced as shown in FIG. After non-rigid registration using the proposed brightness correction and tumor stiffness constraint, the number of erroneous deformation vectors caused by the influence of contrast medium was reduced and the breast area was deformed, resulting in the reduction of SSD to 10.55 ± 4.29. This results in a 13.9% reduction in the conventional daemon-based non-rigid registration method and a 63.0% reduction in the proposed method, compared to the initial SSD. By correcting this, we can increase the matching accuracy.

Since the tumors in the source image show bright brightness values and the non-contrast tumors in the target image appear dark brightness values, the tumor volume decreases when the tumors with bright brightness values are non-rigid matched with dark brightness values. To assess the effect of tumor stiffness constraints on maintaining tumor volume, tumor volume reduction rates before and after rigid and non-rigid registration were measured using Equation 8.

[Equation 8]

In this case, V _after and V _before refer to the number of voxels of the tumor volume automatically divided in the source image before and after registration, respectively.

Table 1 shows the tumor volume reduction rate for the six data.

[Table 1]

There was no change in tumor volume after rigid registration, and tumor volume was reduced by an average of 36.71% after conventional daemon-based non-rigid registration. However, after performing the non-rigid registration with the proposed brightness correction and tumor stiffness constraint, the tumor volume reduction rate was 9.38%, which prevented the reduction of the tumor volume by 27.33%. The reduction in tumor volume after tumor stiffness constraints was due to the normalization of the deformation vector in the peri-tumor region, thus affecting the border of the tumor.

10 shows the performance time of rigid and non-rigid registration for six data. Rigid matching time ranged from 31.88 to 47.16 seconds with mean and standard deviation of 33.29 ± 1.93 seconds. In the case of rigid registration, the optimization was performed by applying sampling at intervals of 4 pixels on the x and y axes and 2 slices on the z axis, thereby reducing the calculation amount by 32 times compared to before applying the sampling. The performance time of the proposed non-rigid registration ranged from 64.63 to 107.14 seconds and the mean and standard deviation were 84.81 ± 11.05 seconds. In this case, since the deformation vector is calculated only for the voxels corresponding to the breast region, the execution time of the proposed non-rigid registration is affected by the breast size of the patient.

The present invention proposes a global and regional motion correction method between before and after contrast enhancement images for perfusion analysis of tumors in dynamic contrast enhanced MR breast images. Normalized mutual information-based rigid-body matching using sampling was able to robustly compensate for global movements even with differences in brightness due to contrast agents. The breast region and breast muscles were removed to correct the brightness of each breast tissue. The breast region was divided and the tissues in the divided breast region were classified into adipose, mammary tissue, tumor and blood vessel using fuzzy C-means clustering technique. In addition, the tumor region was segmented by using brightness and shape information to constrain the tumor stiffness. Non-rigid registration using brightness correction and tumor stiffness constraints for each breast tissue improves the accuracy of registration by reducing the deformation that may occur due to the influence of contrast agent in the breast area. Tumor deformity was reduced during registration. As a result, the difference in initial brightness between pre and post-enhancement images decreased from 28.53 ± 6.2 to 10.55 ± 4.3 after non-rigid registration, and prevented reduction of tumor volume by 27.33% after application of tumor stiffness constraint. there was. The average execution time of the rigid registration was 33.29 ± 1.93 seconds and the average execution time of the non-rigid registration was 84.81 ± 11.05 seconds.

In the MR image proposed in the present invention, the breast and tumor segmentation and matching method can be effectively used for diagnosing the tumor malignancy and observing the therapeutic effect of the malignant tumor, and comparing the breast image with other modalities such as X-ray and ultrasound. It can be applied to the analysis of breast and tumor through registration.

Meanwhile, the non-rigid registration method applying the brightness value correction and the tumor stiffness constraint for each breast tissue of FIG. 1 may be implemented by the non-rigid registration system as shown in FIG. 11. In this case, the non-rigid registration system may include a breast divider 210, a tissue sorter 220, a tumor divider 230, a match and corrector 240, and a non-rigid matcher 250.

Here, the breast dividing unit 210, the tissue sorting unit 220, the tumor dividing unit 230, the matching and correcting unit 240, and the non-rigid registration unit 250 are the step S110, the step S120, the step S130, and the step, respectively. Functions and operations may be performed at step S140 and step S150, and a detailed description thereof will be omitted.

The present invention is not necessarily limited to these embodiments, as all the constituent elements constituting the embodiment of the present invention are described as being combined or operated in one operation. In other words, within the scope of the present invention, all of the components may be selectively operated in combination with one or more. In addition, although all of the components may be implemented as one independent hardware, some or all of the components may be selectively combined to perform a part or all of the functions in one or a plurality of hardware. As shown in FIG. In addition, such a computer program may be stored in a computer-readable medium such as a USB memory, a CD disk, a flash memory, etc., and read and executed by a computer, thereby implementing embodiments of the present invention. As the storage medium of the computer program, a magnetic recording medium, an optical recording medium, a carrier wave medium, or the like may be included.

Furthermore, all terms including technical or scientific terms have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, unless otherwise defined in the Detailed Description. Terms used generally, such as terms defined in a dictionary, should be interpreted to coincide with the contextual meaning of the related art, and shall not be interpreted in an ideal or excessively formal sense unless explicitly defined in the present invention.

The above description is merely illustrative of the technical idea of the present invention, and those skilled in the art to which the present invention pertains may make various modifications and variations without departing from the essential characteristics of the present invention. In addition, the embodiments disclosed in the present invention are not intended to limit the scope of the present invention but to limit the scope of the technical idea of the present invention. Accordingly, the scope of protection of the present invention should be construed according to the claims, and all technical ideas within the scope of equivalents should be interpreted as being included in the scope of the present invention.

Claims (12)

Automatically segmenting the breast region by removing breast skin and pectoral muscle using tilt and direction information;
Classifying breast tissue into fat, mammary tissue, tumor, and blood vessel using fuzzy C-means clustering technique in the breast region divided by the automatic segmentation step;
Automatically extracting the location of the tumor by using brightness information according to the degree of contrast in the tumor information classified by the classification step or the target image and dividing the tumor site by using the brightness value and morphological calculation;
Performing a rigid registration based on normalized mutual information between the target and the source image, and correcting brightness values of fat and mammary gland tissue having different brightness values under the influence of contrast medium; And
Performing daemon-based non-rigid registration applying constraints to maintain the rigid structure in the divided tumor areas and applying normalized weights to each breast tissue to reduce deformation due to the difference in brightness values of the tumor areas;
Non-rigid registration method comprising a.
The method of claim 1,
Non-rigid registration method characterized in that the enhancement image before the contrast medium injection (target), the contrast image after the contrast medium injection is defined as the source (source), the source image is composed of a data set continuously photographed at a predetermined interval .
The method of claim 1, wherein the automatic splitting step,
After applying anisotropic diffusion filtering to remove noise while maintaining important information such as edges, segment the breast area, and normalize the brightness values to correct the degree of contrast of the breast tissue, which is different for each data. A non-rigid registration method characterized by increasing the accuracy of performing breast division and tissue classification by performing tissue classification.
The method of claim 1, wherein the tumor segmentation step,
Tumor difference between the target and source images when the location of the tumor site can be determined from the tumor information classified by fuzzy C-mean clustering, and the location of the tumor site is difficult to locate using the classified tumor information due to low contrast. Non-rigid registration method, characterized in that for automatically extracting the location of the tumor using.
The method of claim 1, wherein the rigid matching and brightness value correction step,
It is possible to compensate for global movement by obtaining the optimal rigid body transformation without being influenced by the difference in brightness value between target and source images.In this case, the calculation amount can be reduced by sampling, and the brightness value correction can reduce fat and wire in the source image. A non-rigid registration method comprising correcting the influence of contrast agent on a tissue.
The method of claim 1, wherein performing the non-rigid registration step,
A non-rigid registration method comprising correcting local movement while reducing volume and morphological changes of tumor regions between breast regions in the target and source images due to respiratory differences and heartbeats.
A breast dividing unit for automatically dividing a breast region by removing breast skin and pectoral muscle using tilt and direction information;
A tissue classification unit classifying breast tissue into fat, mammary tissue, tumor, and blood vessel using a fuzzy C-means clustering technique in the breast region divided by the breast division;
Tumor segmentation unit for automatically extracting the location of the tumor using the difference of the brightness value according to the degree of contrast in the tumor information or target and source image classified by the tissue classification unit, and segmenting the tumor site using the brightness value and morphological operations ;
A matching and correction unit configured to perform rigid registration based on normalized mutual information between the target and the source image, and to correct brightness values of fat and mammary gland tissue having different brightness values under the influence of contrast medium; And
A non-rigid registration unit that performs daemon-based non-rigid registration applying constraints to maintain rigid structures and applying normalized weights for each breast tissue to reduce deformation due to differences in brightness values of the tumor site;
Non-rigid matching system comprising a.
8. The method of claim 7,
A non-rigid registration system, characterized in that the augmented image before the contrast agent injection is defined as a target image, and the augmented image after the contrast medium is injected as the source image, and the source image is composed of data sets continuously photographed at predetermined intervals. .
According to claim 7, wherein the breast portion,
After applying anisotropic diffusion filtering to remove noise while maintaining important information such as edges, segment the breast area, and normalize the brightness values to correct the degree of contrast of the breast tissue, which is different for each data. A non-rigid registration system characterized by increasing the accuracy of performing breast division and tissue classification by performing tissue classification.
The method of claim 7, wherein the tumor divider,
Tumor difference between the target and source images when the location of the tumor site can be determined from the tumor information classified by fuzzy C-mean clustering, and the location of the tumor site is difficult to locate using the classified tumor information due to low contrast. Non-rigid registration system, characterized in that for automatically extracting the location of the tumor using.
The method of claim 7, wherein the matching and correcting unit,
It is possible to compensate for global movement by obtaining the optimal rigid body transformation without being influenced by the difference in brightness value between target and source images.In this case, the calculation amount can be reduced by sampling, and the brightness value correction can reduce fat and wire in the source image. A non-rigid registration system characterized by correcting the influence of contrast agent on a tissue.
The method of claim 7, wherein the non-rigid matching portion,
A non-rigid registration system, characterized in that it compensates for regional movement while reducing volume and morphological changes of the tumor region between the breast region in the target and source images due to respiratory differences and heartbeats.

Images