KR101144579B1 - Apparatus and method for nonrigid lung registration between end-exhale and end-inhale computed tomography scan - Google Patents

Abstract

The present invention provides a method for registration of a computed tomography (CT) image of a chest taken respectively on an exhalation and an inspiration. According to the image matching method of the present invention, a rigid body transformation or an affine transformation is firstly performed on an exhalation image and an inspiration image, and an end-exhale and an intake image step registration process in which non-rigid body registration is performed according to the slope of each cell of the cell.

In the matching method of the present invention, in the non-rigid matching process of the second step, an active cell and an inactive cell are separated from each other in order to accelerate the matching process and reduce the calculation amount, Can be performed.

Computed tomography, Rigid body matching, Non - rigid body matching, Active cell, Active - cell

Description

[0001] APPARATUS AND METHOD FOR NONRIGID LUNG REGISTRATION BETWEEN END-EXHALE AND END-INHALE COMPUTED TOMOGRAPHY SCAN [0002]

The present invention relates to a method and apparatus for performing lung image matching between thoracic CT images obtained by computed tomography (CT) at the expiration and inspiration, respectively.

Lung cancer is the most common cancer in the world and is a fatal disease that kills more than 1.3 million people annually. The number of deaths from lung cancer in Korea surpassed 12,000 in 2002, and mortality from lung cancer was the highest. Pulmonary nodules are commonly found on chest radiography and have clinical forms such as inflammatory granuloma, benign tumors, and malignant tumors (lung cancer). Therefore, it is very important to determine whether the pulmonary nodule is benign / malignant as well as the simple detection of pulmonary nodule in the early diagnosis and rapid treatment of lung cancer.

It is known that discrimination of pulmonary nodule positivity / malignancy is difficult for a skilled expert, and the accuracy of discrimination is known to be about 2/3 even when computed tomography (CT) images are used. In the actual medical field, for this reason, it relies heavily on direct verification of the living body for the positive / malignant final discrimination of pulmonary nodules.

On the other hand, as the patient continues to breathe alive, the volume and shape of the lung changes according to the patient's breathing state. Pulmonary maturation of the lungs around the pulmonary tissues in the expiratory or inspiratory, ie, different respiratory tracts, is essential for clinical applications such as treatment planning, quantitative assessment of obstructive pulmonary disease, and pulmonary motion analysis.

In particular, lung deformation due to respiration is difficult to correct with conventional rigid body registration or affine registration, and calibration requires non-rigid body matching. Rigid or affine matching is used to correct for translational mismatch or global expansion of the lung tissue, but enough degrees of freedom to compensate for the local deformation of the lung due to breathing. Because they can not have.

As an example of non-rigid body matching, a method of calculating a dense displacement vector from all the pixels of a CT image and matching each pixel has been proposed. However, this method has a problem that the amount of data to be processed is too large when the resolution of the CT image is high, so that the amount of calculation required for the matching process is too large and the execution time is long.

[M. M. Coselmon, J. M. Balter, D. L. McShan, and M. L. Kessler, "Mutual information based CT registration of the lung at exhalation and inhale breathing states using thin-plate splines," Med. Phys. Vol. 31, pp. 2942-2948, 2004.], Coselmon et al. Proposed a method for non-rigid body matching of lung tissue between CT images obtained from different respiratory conditions. According to this, thin-plate spline warping is performed so that mutual information between two images is maximized by using 30 pairs of manually detected control points.

E. Rietzel and G. T. Chen, "Deformable registration of 4D computed tomography data," Med. Phys. Vol. 33, pp. 4423-4430, 2006.], Rietzel et al. Proposed a B-spline-based matching method to match four-dimensional CT images of patients with non-small cell lung cancer. According to this, a matching technique is applied only to the lung and mediastinum to reduce misregistration due to discontinuity between the large lung tissue and the relatively small motion rib.

D. A. Torigian, W. B. Gefter, and J. D. Affuso, K. Emami and L. Dougherty, "Application of an optical flow method to an inspiratory and expiratory lung MDCT to assess regional air trapping: a feasibility study, Am. J. Roentgenol. Vol. 188, pp. 276-280, 2007, and L. Dougherty, J. C. Asmuth and W. B. Gefter, "Alignment of CT lung volumes with an optical flow method," Acad. Radiol. Vol. 10, pp. 249-254, 2003.], a method based on optical flow has been proposed to visualize the difference in brightness value between the lungs and the expiratory-inspiratory CT images.

However, the above-mentioned methods are specialized in the matching process for the wide-area movement of the lung tissue or in the matching process for the local deformation in the state in which the wide-area matching is completed. Therefore, if the wide- There is a problem that the calculation amount is very large and thus the time required for the optimum convergence process becomes long. Therefore, it is necessary to develop a non-rigid body matching method that can stably and optimally converge the large deformation of the lung, as well as accelerate the matching process.

SUMMARY OF THE INVENTION The present invention has been made in order to solve the above problems of the prior art, and it is an object of the present invention to provide an End-Exhale CT Scan and an End-Inhale CT Scan obtained by Computed Tomography (CT) And an image matching method and apparatus for shortening the time required for the matching process. It is another object of the present invention to provide an image matching method and apparatus capable of increasing the accuracy of matching between an exhaled CT image and an inspiratory CT image.

In order to achieve the above object, first, a global registration of an exhalation CT image to an inspiratory CT image is performed using rigid body matching, and then a non-rigid body registration of the globally matched exhaled CT image is performed . By performing global matching through rigid body matching, base data for non-rigid body matching is prepared.

The present invention aims to provide a non-rigid body matching method capable of accelerating the matching process by reducing unnecessary calculation processes in the non-rigid body matching process. In each step of the non-rigid body matching which is repeatedly performed, designation of an area having high matching degree as an inactive cell can reduce unnecessary calculation and reduce deformation folding due to deformation.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a matching method capable of quickly searching for an optimal convergence solution according to a corresponding aspect of an expiration CT image and an inspiration CT image. The non-rigid body matching method of the present invention can search for an optimal convergent solution that rapidly converges even in a region where the gradient of the inspiration CT image is weak by using the inclination information of the exhalation CT image and the inspiration CT image together.

An image matching method according to an embodiment of the present invention matches a lung lung image and an inspiration lung image generated by a computed tomography (CT) image. The image matching method includes dividing the lung, bronchus, and pulmonary vascular region from the lung lung image, generating a floating image, dividing the lung, bronchus, and pulmonary vascular region from the lung lung image, and generating a reference image do. The image matching method includes a first matching step of matching the floating image with respect to the reference image using an affine transformation based on narrowband distance propagation, a step of matching the first matching floated image and the active image of the reference image separating an active cell and an inactive cell from each other, performing non-rigid-body matching of the first-matched floating image with respect to the reference image using the tilting information of the active cell of the first-matched floating image; And subtracting the brightness value of the reference image from the brightness value of the non-rigid-body-matched floating image, and image-coding the brightness value difference.

At this time, the non-rigid-body matching step of the image matching method calculates a demon force based on the slope information of the boundary voxel of the reference image and the slope information of the boundary voxel of the first- And locally modifying a boundary surface of the primary-matched floating image according to the daemon force to a boundary surface direction of the reference image.

Also, the step of calculating the daemon force of the image matching method may calculate the daemon force by multiplying the slope information of the interface voxel of the reference image and the slope information of the interface voxel of the primary-matched floating image by a weight, .

Also, the non-rigid body matching step of the image matching method may include calculating the daemon force and locally modifying the boundary surface of the primary matched floating image to the boundary of the reference image to obtain an optimal convergence value . ≪ / RTI >

The step of separating the active and inactive cells of the primary-matched floating image and the reference image of the image matching method may include the following steps.

Determining whether a difference in brightness value between the corresponding first-match floating image and a corresponding cell of the reference image is less than a first threshold, calculating a daemon force for the corresponding cell, Is less than the second threshold value, and if the brightness value difference between the corresponding cells is less than the first threshold value and the norm of the daemon force is less than the second threshold value, To an inactive cell.

The image matching apparatus according to an embodiment of the present invention performs matching of an expiration lung image and an inspiration lung image obtained by a CT scan. The image matching apparatus divides the pulmonary, bronchial and pulmonary vascular regions from the expiratory lung image, generates a floating image, divides the pulmonary, bronchial and pulmonary vascular regions from the inspiratory lung image, and generates a reference image A wide matching unit for matching the floating image with the reference image using affine transformation based on narrowband distance propagation, separating the active and inactive cells of the primary matching floating image and the reference image, A non-rigid body matching unit for performing non-rigid body matching of the primary matched floating image to the reference image using the slope information of the active cell of the primary matched floating image, and a non-rigid body matching unit for matching the brightness value of the reference image with the non-rigid matched floating image And an image coding unit for subtracting the brightness value from the brightness value and image-coding the brightness value difference.

Wherein the non-rigid body matching unit of the image matching apparatus calculates the daemon force based on the slope information of the interface voxel of the reference image and the slope information of the interface voxel of the primary matched floating image, It is possible to locally transform the boundary surface of the floating image to the boundary surface direction of the reference image.

Wherein the non-rigid body matching unit of the image matching apparatus separates active and inactive cells among corresponding cells of the reference image and the primary-matched floating image based on the daemon force, It is possible to locally transform the boundary surface of the primary-matched floating image to the boundary surface direction of the reference image.

Wherein the non-rigid body matching unit of the image matching apparatus recomputes the demon force for the locally deformed floating image and the reference image, re-separates the active and inactive cells according to the re-computed daemon force, The boundary surface of the locally deformed floating image can be re-transformed.

According to the image matching method and apparatus of the present invention, the time required for the matching process between the end-Exhale CT scan and the end-intake CT scan obtained by computed tomography (CT) Can be shortened. Also, the image matching method and apparatus of the present invention can improve the accuracy of matching between an expiration CT image and an inspiration CT image.

The present invention firstly performs global registration of an exhalation CT image to an inspiratory CT image using rigid body matching, and then non-rigidly matches the globally matched exhaled CT image. By performing global matching through rigid body matching, underlying data for non-rigid matching can be provided.

In the present invention, unnecessary calculations can be reduced and deformation folding due to deformation can be reduced by designating regions having high matching degrees as inactive cells at each step of non-rigid body matching which is repeatedly performed. The present invention can accelerate the matching process by reducing unnecessary calculation processes in the non-rigid matching process.

The image matching method of the present invention can quickly search for an optimal convergence solution by using inclination information of an expiration CT image and an inspiration CT image together. There has been a case where a long time is required for the convergence of the final solution according to the correspondence relationship between the expiration CT image and the inspiration CT image. The present invention uses the inclination information of the exhalation CT image and the inspiration CT image together in a complementary manner, thereby accelerating the convergence process even when it is difficult to accelerate in the prior art.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to or limited by the embodiments. Like reference symbols in the drawings denote like elements.

FIG. 1 is a flowchart illustrating an image matching method according to an exemplary embodiment of the present invention. Referring to FIG.

Referring to FIG. 1, the lung image matching method divides regions corresponding to pulmonary, bronchial, and pulmonary blood vessels in an exhaled CT image and an inspired CT image (S110). The lung image matching method divides lung area, bronchial area, and pulmonary vein area from each of expiratory CT and inspiration CT images in order to limit the calculation range of the matching to the lung area, Only the region can be specified as the calculation range for matching.

In this specification, in order to simplify the following description, an excitation CT image divided into only a closed region is defined as a reference image, and an expired CT image separated from a closed region is defined as a floating image. The reason why the inspiratory CT image is the reference image is that the closed area of the inspiratory CT image is generally larger than the closed area of the expiratory CT image and the scarce image matching method has the same or similar shape as the reference image by expanding the floating image .

In the waste image matching method, affinity image matching is performed on the reference image (the inspected CT image of the closed region) separated from the floating image (closed CT image of the closed region) (S120). At this time, affine image matching is performed using narrowband distance propagation.

Affine matching is a matching process using affine transformation, which means that the floating image is affine transformed to match the reference image. The affine transformation is a linear transformation assuming a rigid body, and takes the form of T (x) = Ax + b, where T (x) is the affine transformation of a particular position vector x in the floating image. Where A is a matrix and b is a position vector with the same dimensions as x.

Affinity matching treats the lung area as a rigid body, and therefore, it can not solve the mismatch of shapes due to the lung tissue deformed by the exhalation and the inspiration. Therefore, the reference image and the floating image are globally registered by affine matching, and then locally matched by non-rigid matching.

Non-rigid body matching is performed by steps S130 to S160. In the closed-image matching method, a voxel having a predetermined size is set on the lung interface of each of the reference image and the floating image, and a gradient of the CT image is obtained in the voxel. In the waste image matching method, a corresponding voxel of the reference image and the floating image is searched and a demon force is calculated according to the corresponding voxel inclination information (S130).

The closed image matching method divides each of the reference image and the floating image into predetermined cells and searches corresponding cells in the reference image and the floating image. The waste image matching method determines whether the corresponding cell is an active cell or an inactive cell (S140).

It is determined whether or not there is an active cell (S150), and if no active cell remains, the process is terminated. With respect to the active cell, if the active cell remains, the voxel at the interface of the floating image is deformed according to the daemon force and matched to the corresponding voxel of the reference image (S160).

In this case, in the waste image matching method, the brightness value of the reference image may be subtracted from the brightness value of the floating image, and the difference of brightness values may be color-coded to visualize a difference between brightness values between the reference image and the floating image .

Some known structures among the detailed structures used in the present invention have been partially described in the following documents. See, e.g., P. J. Burt and E. H. Adelson, "The laplacian pyramid as a compact image code," IEEE Trans. on Comm. Vol. 31, pp. 532-540, 1983.], a process of generating a hierarchical multi-resolution image from each CT image using a Laplacian pyramid was introduced, and a method of calculating the displacement between corresponding two CT images was proposed.

D. Sarrut, V. Boldea, S. Miguet and C. Ginestet, "Simulation of four-dimensional CT images from deformable registration between inhale and exhale breath-hold CT scans," Med. Phys. Vol. 33, pp. 605-617, 2006], JP Thirion, "Image matching as a diffusion process: An analogy with Maxwell ' s demons, Image Anal. Vol. 2, pp. 243-260, 1998, and X. Pennec, C. Cachier, N. Ayache, "Understanding the demon's algorithm: 3D non-rigid registration by gradient descent," in Medical Image Computing and Computer-Assisted Intervention, MICCAI'99 , Vol. 1679 , pp. 597-605] proposed a lung matching method based on the so-called daemon algorithm. The idea of correcting the brightness value of the exhaled CT image by subtracting the average brightness value between the corresponding regions of the two images along the z-axis direction of the volume was introduced in order to consider the change in the brightness value of the lung during the patient's breathing. In general, the lung tissue on the expiratory CT image is smaller than the lung tissue on the inspiratory CT image and the density of the lung tissue is high. Differences in density of pulmonary tissue caused differences in brightness values in CT images, and a matching method was introduced to correct for differences in brightness values.

H. Wang, L. Dong, J. O'Daniel, R. Mohan, AS Garden, KK Ang, DA Kuban, M. Bonnen, JY Chang and R. Cheung, radiation therapy, "Phys. Med. Biol. Vol. 50, pp. 2887-2905, 2005.), a method of accelerating the above-mentioned demon algorithm using gradient information of an expiration CT image has been introduced. In this study, we propose a procedure to modify the boundary of pulmonary tumor manually drawn on the expiratory CT image to correspond to the inspiratory CT image.

[W. G. Lu, M. L. Chen, G. H. Olivera, K. J. Ruchala and T. R. Mackie, "Fast free-form deformable registration via calculus of variations," Phys. Med. Biol. Vol. 49, pp. 3067-3087, 2004.] introduced a technique of expressing the optimization problem of the matching process as a set of partial differential equations using a calculus of variation. According to this, a Gauss-Seidel method has been introduced to recompute the exhalation image to the inspiratory image by repeatedly calculating the partial differential equations.

In step S110, a slope and a brightness value distribution are used to divide the lung region from each of the exhalation CT image and the inspiration CT image. First, the initial lung area and boundary are divided by the resolution based on the brightness value of the CT scan, and the bronchial and pulmonary blood vessels are excluded from the segment.

The brightness values of the CT image are represented by the Hounsfield Unit (HU) scale and have different values depending on the material. Air is defined as -1000 HU, fat has a value of -120 HU, and water has a value of 0 HU. Generally, muscle has about 40 HU, contrast is about 130 HU, and bone has more than 400 HU.

The image matching method divides the lung and bronchi using a 3-dimensional region growing method with a threshold value of -400 HU from the CT image. The three-dimensional region growth method is performed as follows. First, we set up a seed voxel located in the organ. The region is grown by including neighboring voxels having a brightness value of -950 HU or less from the seed voxel in the region. The coordinates of the seed voxel are determined by averaging the coordinates of the voxels having brightness values smaller than the threshold value among the voxels located at the center of the upper sectional image. Pulmonary blood vessels detect and divide areas with brightness values greater than -400 HU in the segmented lung area. Then, a gradient profile having an adaptive length is generated and analyzed to set a range in which the boundary line can move. The setting of this range has the effect of preventing the boundary line from leaking out of this area and enhancing the efficiency of boundary movement. The boundary line is propagated according to the velocity function within the set range. The velocity function prevents the boundary line from converging to the local maximum or minimum value of the local slope based on the slope and brightness distribution.

The image matching method of the present invention performs global rigid matching in step S120 and performs local non-rigid matching in steps S130 to S160. During CT acquisition, the position, volume, and shape of the lungs between the expiratory and inspiratory images are different due to patient movement and respiration. The affine registration described above can be applied to compensate for the movement variations and global expansion of the lungs. However, for local deformation such as lung distortion, matching is not completed with affine matching. Therefore, a non-rigid body matching process is necessary to fully reflect the movement of the lung during breathing.

In step S120, the lungs of the floating image are matched to the reference image by a surface-based registration method using narrow-band distance propagation. First, an optimal cube including the lungs extracted from the reference image and the floating image is generated, and an initial matching between the optimal boundary volume of the reference image and the optimal boundary volume of the floating image is performed to correct the movement variation. After generating the 3D distance map using the narrowband propagation from the lung boundary, the image is matched to the position where the distance difference between the two boundaries is minimum through the selective distance measurement. The narrow-band-distance propagation is a process of locally setting a plurality of imaginary boundaries extended from the closed boundary and matching the images to positions where the distance between the boundaries of the reference image and the boundary among the plurality of boundaries is minimum. That is, a first boundary extended from the boundary surface of the floating image, a second boundary extended from the first boundary, and a third boundary extended from the second boundary are set, and then the boundary between the first and the third boundary Select the boundary with the smallest distance difference.

A more detailed description of a boundary-based registration method using narrow-band distance propagation is given in [H. Hong, J. Lee and Y. Yim, "Automatic lung nodule matching on sequential CT images," Comput. Biol. Med. Vol. 38, pp. 623-634, 2008.) and Lee Jung-jin, Hong Helen, Shin Young-gil, Automatic Lung Registration Using Local Distance Propagation, Software and Applications, Vol. 32, No. 1, January 2005. .

In step S130, the lungs of the affine-matched floating image are transformed using the non-rigid matching method based on the daemon algorithm. The daemon algorithm locates the daemon in specific voxels in the reference image and uses the daemon force to spread the floating image to the reference image. In the conventional daemon algorithm mentioned in the cited document, the daemon force u (x) of the voxel x is calculated using the following equation (1).

[Equation 1]

(X) represents the brightness value of the voxel x of the reference image, and Ex (T (x)) represents the brightness value of the voxel T (x) of the primary-matched floating image. Ins is the gradient of the partial derivatives of the voxel x of the reference image. α is a parameter that defines the magnitude of the daemon force, and the daemon force is limited to ½ α.

The power of the daemon, u (x), is the slope of the sum of squared difference (SSD) used to evaluate the similarity measure between two images. The SSD of the two images is minimized by extending the floating image in the slope direction of the reference image. . This force is defined as the reference gradient force because it is determined by the slope of the reference image. The SSD is defined as shown in Equation (2) below.

&Quot; (2) "

However, since this force is set as in the conventional daemon algorithm (Equation (1) above), it acts mainly at the lung boundary of the reference image, and therefore, the distance from the lung boundary of the reference image and indirectly influenced by the reference image Deformation is slow at the lung boundary of the image. In the conventional daemon algorithm, the deadweight boundary of the floating image is matched against the reference image, but the daemon force is calculated using only the scale obtained from the boundary surface of the reference image. As a result, the optimal convergence process is delayed.

Therefore, the waste image matching method of the present invention can accelerate the optimal convergence process by suggesting a daemon algorithm that uses the slope information of the reference image and the slope information of the floating image together. Even if the slope information of the boundary surface of the reference image is small, the slope information of the floating image is large in a region where the deformation of the boundary surface of the floating image is large. On the contrary, even if the slope information of the boundary surface of the floating image is small, when the deformation of the boundary surface of the reference image is large, fast optimum convergence can be achieved according to the daemon algorithm newly proposed by the present invention.

In the present invention, the daemon force is calculated by summing the weights of the floating gradient force based on the slope information of the reference image and the slope information of the floating image based on the slope information of the reference image as shown in Equation (3).

&Quot; (3) "

At this time,? Is a weight for the floating slope force. Ex is the slope information in the corresponding voxel T (x) of the floating image.

The floating tilt force helps to pull the reference boundary of the reference image mainly by acting on the lung boundary of the floating image with the reference tilt force. In the present invention, each of the floating slope force and the reference slope force is added with the weight multiplied. At this time, the weight? Is fixed during the optimization process.

Depending on the embodiment, the parameter beta is initially set to a value greater than 0.5 and may be set to decrease as the number of iterations increases. Initially, the difference in pulmonary distance between the two images is so large that the lung of the floating image is pulled as close to the reference image as possible by applying a relatively large weight to the floating gradient force. The accuracy of registration can be improved in the lower part of the lung with regional deformation by emphasizing the reference tilt force by reducing β after moving the lung reference image close to the lung reference image. Depending on the embodiment,? May initially be set to 0.6 and may be reduced by 0.1 as the registration process progresses.

The daemon algorithm using the slope information of the floating image performs matching by repeatedly applying the following optimization process. The number of iterations in the inner loop of the optimization process is denoted by n, and the number of iterations in the outer loop is denoted by N. The inner loop calculates the displacement field by repeatedly applying the following four steps. First, the force is calculated at each daemon position using Equation 2 so that the similarity between the two images is high (S130). Second, the field of force is smoothed using a Gaussian filter so that a force field is generated continuously and smoothly between neighboring force vectors. Third, accumulate forces to calculate displacements. Fourth, smooth the displacement field. The field and displacement field of the force are smoothed using the following equation (4).

&Quot; (4) "

Where V is the set of neighboring voxels of x in the Gaussian filter and σ is the standard deviation of the Gaussian distribution.

The inner loop of the optimization process is repeated until the following two stop conditions are satisfied. The normalized sum of absolute difference (NSAD) of Equation (5) and the displacement field norm || U (x) || The optimization process stops when one of the two values diverges. The outer loop of the optimization process repeats the above process of calculating the displacement field a predetermined number of times. The final displacement field is calculated by accumulating the calculated displacement field in each step N.

&Quot; (5) "

&Quot; (6) "

In this case, M represents the number of voxels in each image, and U (x) represents the displacement in each voxel x.

To achieve fast and efficient matching for high-distortion areas, a multi-resolution volume is created. A low-resolution volume Ex _k and Ins _k (k = 1, 2, 3, 4) are generated from the reference image Ex ₀ and the floating image Ins ₀ using a Gaussian pyramid [15]. As the parameter k increases, the size of the low resolution volume decreases and the number of voxels contained in the unit cell increases. As the original image has a resolution of 512 × 512 × 512, the volume is reduced to 256 × 256 × 256, 128 × 128 × 128, 64 × 64 × 64, and 32 × 32 × 32 as k increases from 1 to 4 , And the cell size increases by 2 × 2 × 2, 4 × 4 × 4, 8 × 8 × 8, and 16 × 16 × 16. The displacement field between the corresponding low resolution volumes is sequentially calculated from the lowest resolution step (k = 4) to the highest step (k = 1). After the optimization process is completed in one step, the calculated displacement is used as the initial value for optimization of the next resolution. A trilinear interpolation is used to transform the low-resolution displacement into the next-stage displacement with high resolution.

In the matching method of the present invention, the weighted sum of the reference tilt force and the floating tilt force is used to deform the floating image corresponding to the reference image. This weighting sum enables fast convergence even at the lung boundary of the floating image with weak reference slope force. The present invention rapidly converges the convergence process by adaptively selecting β during the optimization process and repeats the convergence process, thereby increasing the accuracy of matching at the lower part of the lung with local deformation.

When the lung is matched between the reference image and the floating image, the displacement of each cell is calculated by accumulating the force. If the change in displacement is negligibly small, the cell is no longer deformed. In this specification, an active-cell-based daemon algorithm is provided that selectively optimizes according to the degree of matching to avoid unnecessary computation for such well-matched cells.

The activity of each cell in the low resolution volume is determined by evaluating the corresponding intercell similarity of the reference image and the floating image. A cell satisfying both of the following two conditions is regarded as sufficiently matched and designated as an inactive cell (S140). If any one of the following equations (7) and (8) is not satisfied, it is designated as an active cell.

&Quot; (7) "

&Quot; (8) "

(세번째 조건식)

(Third conditional expression)

는 기울기 벡터

와

간의 각도이다.In this case, Ex _k (T (x)) and Ins _k (x) are brightness values in the cell x of the k-th low-resolution volume of the floating image and the reference image, respectively. d _k (x) is the absolute value of the difference in brightness values of Ex _k (T (x)) and Ins _k (x).

Is a slope vector

Wow

.

Equation (7) indicates that the brightness difference between the reference image and the floating image for the k-th cell should be smaller than the first threshold value T ₁ . The first threshold value T ₁ is determined according to the following equation (9) in consideration of the distribution of the brightness difference.

&Quot; (9) "

In this case, μ _dk represents the average of d _k , δ _dk represents the standard deviation of d _k , and ω _N is a weight for adjusting the ratio of active-cells, and ranges from 0 to 1.

Since the histogram of d _k tends to decrease monotonically from 0 HU to the right, 0 is assigned to the mean μ _dk . The parameter ω _N is initially set to 0.1 to set the threshold by 10% of the standard deviation. This value can be conservatively set so as not to degrade the accuracy of the match while reducing the run time. As the number of iterations increases, ω _N may increase by 30% of the standard deviation to reduce the active-cell ratio.

The following equation (8) indicates that the norm of the daemon force should be negligibly small. Depending on the embodiment, the force having a magnitude of less than 0.25 has little effect on the displacement, so the threshold T ₂ can be set to 0.25.

The equation (third conditional expression) indicates that the angle between the slope vector of the reference image and the floating image should be smaller than the threshold value T2. T2 was experimentally set at 30 degrees.

In this case, the cell and the voxel may be different names of the same segment or may represent differently segmented regions. The cell is set to a lower resolution, and the voxel may be set more precisely.

FIG. 2 is a diagram illustrating a process in which the ratio of active-cells gradually decreases while the matching process is repeatedly performed. Green highlighted active cells. Most of the cells not included in the closed area and the cells in the upper part of the deformed lung were set as inactive cells after the first matching process was performed. On the other hand, cells with large deformation at the lower right side of the lung remain active - cell even after the third matching process is performed. The ratio measured from FIG. 2 decreased to 11.03%, 7.1%, and 4.81% as the number of repetitions N increased.

After the inactive cell is designated in step S140, it is determined in step S150 whether there is any active cell remaining. If no active cell remains, the process terminates, and if active cell remains, step S160 is performed on the active cell. Three - dimensional Gaussian smoothing of force field and displacement field is performed on the active - cell and active - cell neighboring cells to prevent discontinuity between active - cell and inactive - cell.

In the image matching method of the present invention, the reference image and the floating image are matched through non-rigid body matching, and then the accuracy of registration is confirmed, and a subtracted image is obtained by subtracting the two images to visualize the difference in brightness value between images. In order to easily identify the difference in brightness value, the brightness value of the subtraction image is mapped into the color range and expressed as a color map image.

The active-cell-based daemon algorithm determines cell activity by evaluating the degree of matching between the corresponding cells of the two images to be matched. By performing optimization only for active-cells, it is possible to prevent unnecessary accumulation of displacement. As a result, the registration process can be accelerated and the deformation folding phenomenon can be reduced.

3 is a view showing a lung non-rigid body matching process in a coronal view. A coronal view (coronal plane or frontal plane) is an arbitrary profile that divides the human body forward and backward. 3 (a) shows an expiration CT image in which a closed region is separated. (b) shows an aerial-converted expiration CT image, that is, a floating image. (c) shows the non-rigid-matched floating image with respect to the reference image.

(D) represents an inspiratory CT image, that is, a reference image, and (e) represents a displacement field set for the reference image. (f) shows a color-coded color map image.

Hereinafter, an example in which the image matching method of the present invention is actually applied will be described. The present invention has been applied to both chest CT images obtained at maximal expiration and maximal inspiration. Five patients were subjected to thin-section CT scanning at 0.7 mm intervals using a Sensation 16 scanner (Siemens, Erlangen, Germany). All images have a resolution of 512 x 512 pixels, the size of each pixel is 0.55 mm to 0.63 mm, and the number of images per scan is 421 to 480. Table 1 shows the resolution and voxel size of the volume data used in the experiment.

[Table 1]

data resolution Voxel size (mm) Ex-Ins1 512 × 512 × 421 0.57 x 0.57 x 0.7 Ex-Ins2 512 × 512 × 480 0.63 x 0.63 x 0.7 Ex-Ins3 512 × 512 × 441 0.55 x 0.55 x 0.7 Ex-Ins4 512 × 512 × 435 0.62 x 0.62 x 0.7 Ex-Ins5 512 × 512 × 448 0.60 x 0.60 x 0.7

The mean volume (standard deviation) of the left and right lungs was 2185.94 (688.53) cc on the expiratory CT image and 3867.86 (d) on the inspiratory CT image, 649.93) cc. The mean difference in volume between the two images was 954.78 (347.13) cc in the right and 793.63 (203.4) cc in the left. It can be seen that the increase in volume during inspiration was greater in the blindness than in the left arm.

Visual evaluation, accuracy evaluation, and execution time were selected as items for evaluating the results of the lung non-rigid body matching method of the present invention. First, the effect obtained by using the slope information of the floating image is compared with the conventional method (method A) using the reference tilt force and the image matching method (method B) of the present invention using the reference tilt force and the floating tilt force together, The effect of active-cell use was evaluated by comparing the image matching method (method B) of the present invention using the active-cell and the conventional method (method C) not distinguishing the active-cell / inactive-cell. The same parameters were used for Gaussian smoothing of the above three methods. The standard deviation σ for Gaussian smoothing of the field of force and the displacement field was set to 2. The size of the Gaussian filter was set to 5 × 5 × 5 for k 1 and 2, and 3 × 3 × 3 for k 3 and 4.

To compare Method A and Method B, the accuracy of the matching was evaluated by measuring the normalized cross correlation (NCC). NCC was measured for the lower lung 30% of the total lung and deformation.

4 is a comparative graph showing the NCC when the methods A and B are applied to the data Ex-Ins1 of Table 1, respectively. The NCC for the brightness value between the reference image and the deformed floating image is calculated using Equation (10).

&Quot; (10) "

는 부유영상의 평균 밝기 값,

는 기준영상의 평균 밝기값이며, σ_Ex는 부유영상의 밝기 값의 표준편차, σ_Ins는 기준영상의 밝기 값의 표준편차를 나타낸다.Here, x is an index of a voxel, M is a number of voxels, Ex (x) is a brightness value of a voxel x of a floating image, and Ins (x) is a brightness value of a voxel x of a reference image.

The average brightness value of the floating image,

Is the average brightness value of the reference image,? _Ex is the standard deviation of the brightness value of the floating image, and? _Ins is the standard deviation of the brightness value of the reference image.

The higher the NCC, the higher the degree of matching between the reference image and the floating image. Referring to FIG. 4, it can be seen that the NCC when method B is applied is higher in all the execution steps (the number of iterations) than the NCC when method A is applied. The maximum difference between the NCC values of the two methods was (a) 1.67% in the total lung area and (b) 8.75% in the lower 30% area. It can be seen that Method B (the present invention) is more effective in the region of the lower 30% of the lung where the deformation is large.

5 is a box plot showing the correlation distribution when applying the conventional method and the matching method of the present invention, respectively. Referring to FIG. 5, a box plot of the NCC between two measured images for five patient data is shown. (a) When Method B was applied to the entire lung area, the mean NCC was 0.954 and the standard deviation was 0.018. When Method A was applied, the mean NCC was 0.947 and the standard deviation was 0.023. The degree of matching is superior. It can be seen that NCC is higher than Method A when the method B is applied in the lower 30% region where the deformation due to the respiration state is severe, which is more effective matching means than the conventional art.

에서 변위 벡터를 U(

)로 표시한다. 변형 접힘 현상의 확률은 음수 야코비 행렬식 (negative Jacobian determinant)에 의하여 계산되며, 하기 수학식 11과 같이 나타내어질 수 있다.One of the items that can evaluate the effect of the matching method of the present invention is the probability and frequency of the deformed folding phenomenon. The lower the probability of the deformed folding phenomenon, the better the match degree. The position vector of the voxel with coordinates (x, y, z)

The displacement vector is denoted by U (

). The probability of the deformation-folding phenomenon is calculated by a negative Jacobian determinant and can be expressed by Equation (11).

&Quot; (11) "

=(x, y, z), U(

)=(Ux, Uy, Uz)로 나타내어진다.here

= (x, y, z), U (

) = (Ux, Uy, Uz).

The Jacobian determinant of the transform is a measure of the invertibility of the transform. Since the voxel with negative Jacobian determinant can not be reversible transformed, the transformed folding phenomenon occurs locally in such a voxel.

FIG. 6 is a box plot showing the probability of the negative Jacobian determinant when applying the prior art (method C) and the inventive matching method (method B), respectively. Method B follows the configuration of the present invention in which the active-cell and the inactive-cell are separated and the non-rigid-body matching process is no longer performed for the inactive-cell. As previously described, it is possible to use the newly proposed daemon force of a cell or cell (Equation 3) and norms as a criterion for distinguishing an active-cell from an inactive-cell. On the other hand, Method C is a conventional technique for performing a non-rigid matching process on all cell regions without dividing the active-cell and the inactive-cell.

The box plots in Fig. 6 show upper, lower, and average values. Referring to FIG. 6, when the method B is applied, the probability of the average 2.713% (standard deviation 1.261) and the probability of the average of 3.177% when the method C is applied (standard deviation 1.545) Has a lower deformation folding probability than Method C.

The matching method of the present invention can prevent unnecessary deformation from accumulating by calculating and smoothing the displacement only for the active-cell. Accordingly, the matching method of the present invention can lower the deformation folding probability and prevent image distortion during registration.

The matching method of the present invention has the effect of shortening the execution time of the overall matching process by using the active-cell based matching method. Such an effect can also be evaluated by comparing the total execution time by Method B and Method C.

Table 2 below is a chart comparing the execution times in each of the matching process steps of Method B and Method C.

[Table 2]

Steps to Perform Method B Method C Create a multi-resolution volume 76.67 76.96 Calculate length of force 8.76 14.85 Smoothing of power 127.83 363.37 Smoothing displacement field 116.77 417.18 Conversion of low resolution volumes 21.26 33.72 Determine cell activity 2.63 0.00 Conversion of original image 59.46 62.81 Total Run Time 413.38 968.89

[Unit: second]

Table 2 shows the execution time of each step of the matching process performed in a system equipped with 2.67 GHz Intel Pentium Core i7 and 3.23 GB memory. Method B performed a total of 6 minutes and 53 seconds and Method C performed 16 minutes and 9 seconds. Method B shows a 57% reduction in execution time for Method C. In particular, the execution time reduction effect is noticeable in the smoothing step, which shows a 72% reduction in execution time.

The smoothing process is performed to set the spatial scale of each cell to an equivalent relation before calculating the daemon force for each cell of the floating image and the reference image. The method B (present invention) However, since the method C performs a smoothing process for all cells, it is expected that the step is repeated and the shorter the execution time of the method B becomes, the more the matching result accumulates.

As a result, by using both the slope value obtained from the reference image and the slope value obtained from the floating image, it is possible to accelerate the process of searching for the optimum convergence value even if the deformation pattern of the lung tissue changes, and the active- It is possible to shorten the execution time and prevent the accumulation of unnecessary deformation, thereby reducing the occurrence probability or frequency of the deformation folding phenomenon.

The matching method according to an exemplary embodiment of the present invention may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The image matching method of the present invention may be implemented by a hardware module. At this time, the separation module can perform the function of dividing each part of the lung and separating the lung area from the initial CT image. On the other hand, the function of rigid body matching of the separately generated floating image to the reference image can be performed by the rigid body matching module. The function of non - rigid matching based on the rigid - body - matched floating image on the basis of the improved daemon force and the active cell search algorithm can be performed by a non - rigid matching module. At this time, the improved daemon force can be obtained by using both the slope of the boundary surface of the reference image and the slope of the boundary surface of the floating image.

While the invention has been shown and described with reference to certain preferred embodiments thereof, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the equivalents of the claims, as well as the claims.

FIG. 1 is a flowchart illustrating an image matching method according to an exemplary embodiment of the present invention. Referring to FIG.

FIG. 2 is a diagram illustrating a process in which the ratio of active-cells gradually decreases while the matching process is repeatedly performed.

Fig. 3 is a view showing a lung non-rigid body matching process on the coronal plane.

Fig. 4 is a comparative graph showing the correlation when the conventional method and the matching method of the present invention are applied to the data Ex-Ins1.

5 is a box plot showing the correlation distribution when applying the conventional method and the matching method of the present invention, respectively.

6 is a box plot showing the probability of the negative Jacobian determinant when applying the conventional method and the matching method of the present invention, respectively.

Claims (9)

A method for matching a lung lung image and an inspiratory lung image generated by a computed tomography image, Dividing lung, bronchus, and pulmonary vascular regions from the expiration lung image and generating a floating image; Dividing lung, bronchus and pulmonary vascular regions from the inspiratory lung images and generating a reference image; Performing a first-order matching of the floating image with respect to the reference image using affine transformation based on narrow-band-distance propagation; Separating active and inactive cells of the primary matched floating image and the reference image; Performing non-rigid-body matching of the first-matched floating image with respect to the reference image using slope information of active cells of a first-matched floating image; And Subtracting the brightness value of the reference image from the brightness value of the non-rigid-body-matched floating image, and image-coding the brightness value difference Wherein the method comprises the steps of: The method according to claim 1, The non-rigid body matching step Calculating a daemon force based on slope information of a boundary surface voxel of the reference image and slope information of a boundary surface voxel of the primary matched floating image; And And locally transforming a boundary surface of the primary-matched floating image according to the daemon force to a boundary surface direction of the reference image Wherein the method comprises the steps of: 3. The method of claim 2, The step of calculating the daemon force And calculating the daemon force by multiplying the slope information of the boundary voxel of the reference image and the slope information of the boundary voxel of the primary matched floating image by the weights. 3. The method of claim 2, The non-rigid body matching step Calculating the daemon force, and locally modifying the boundary surface of the primary matched floating image to the boundary plane of the reference image to search for an optimum convergence value. The method according to claim 1, The step of separating the active and inactive cells of the primary matched floating image and the reference image comprises: Determining whether a difference in brightness value between the first-matched floating image and a corresponding cell of the reference image is smaller than a first threshold value; Calculating a daemon force for the corresponding cell; Determining whether the norm of the daemon force is less than a second threshold; And Designating the corresponding cell as an inactive cell if the brightness difference between the corresponding cells is smaller than the first threshold value and the norm of the daemon force is smaller than the second threshold value Wherein the method comprises the steps of: An apparatus for matching a lung lung image and an inspiratory lung image generated by a computed tomography image, An image segmentation unit for segmenting lung, bronchus and pulmonary vascular regions from the lung lung image, generating a floating image, dividing lung, bronchial and pulmonary vascular regions from the lung lung image, and generating a reference image; A wide matching unit for performing a first-order matching of the floating image with respect to the reference image using affine transformation based on the narrow-band distance propagation; And a non-rigid-body matching unit for performing a non-rigid matching of the first-matched floating image with respect to the reference image by using the first-matched floating image and the active and inactive cells of the reference image and using the slope information of the active- A rigid body matching portion; And An image coding unit for subtracting the brightness value of the reference image from the brightness value of the non-rigid-body-matched floating image, and image-coding the brightness value difference, And an image pickup device for picking up an image. The method according to claim 6, The non-rigid body matching portion Calculating a daemon force based on the slope information of the interface voxel of the reference image and the slope information of the interface voxel of the primary matched floating image, Wherein the first matching unit transforms the boundary surface of the first matched floating image locally in the direction of the boundary surface of the reference image according to the daemon force. 8. The method of claim 7, The non-rigid body matching portion Separating an active cell and an inactive cell among corresponding cells of the reference image and the primary-matched floating image based on the daemon force, Wherein the active matching unit locally transforms the boundaries of the primary matched floating image to the boundary of the reference image according to the daemon force. 8. The method of claim 7, The non-rigid body matching portion Recalculating the demon force for the locally deformed floating image and the reference image, re-separating the active and inactive cells according to the re-computed daemon force, and re-transforming the locally deformed floating image boundary To-waste image matching device.

Images