Vessel registration method, device and equipment based on nested search and medium

Abstract

The application discloses a blood vessel registration method and device based on nested search, equipment and medium, relates to the technical field of image data processing, and comprises the following steps: dividing the rotation angle search range of the 3D blood vessel center line of a target blood vessel into a plurality of sub-search bodies, and obtaining a 2D projection center line of the 3D center line by using the sub-search bodies; dividing the current translation search range corresponding to the sub-search body into a plurality of sub-search surfaces, and obtaining a translated center line by translating the projection center line by using the translation parameters of the sub-search surfaces; searching for a target optimal translation parameter of the current sub-search body based on the center line and a 2D blood vessel center line of the target blood vessel; determining a rotation parameter boundary to generate a theoretical optimal rotation search condition based on the target optimal translation parameter; traversing each sub-search body and dividing a new current rotation angle range until a sub-search body meeting the condition is found; and registering 3D images by using the target optimal translation parameter and the rotation parameter of the sub-search body. The efficiency of blood vessel registration is improved.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, and in particular to a method, apparatus, device, and medium for blood vessel registration based on nested search. Background Technology

[0002] Currently, most vascular registration technologies employ 3D-2D registration methods based on iterative optimization or exhaustive search. These methods generally suffer from high search space dimensionality, large computational load, and low search efficiency, making it difficult to accurately match the dual requirements of accuracy and efficiency in vascular registration for scenarios such as clinical diagnosis and surgical navigation. Consequently, it is challenging to achieve rapid and accurate alignment of 3D and 2D vascular images.

[0003] In summary, improving the efficiency of vascular registration is a problem that needs to be solved in this field. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for vessel registration based on nested search, thereby improving the efficiency of vessel registration. The specific solution is as follows: In a first aspect, this application discloses a blood vessel registration method based on nested search, including: Extract the 3D and 2D centerlines of the target blood vessel from the 3D and 2D images, respectively. An initial rotation angle range is constructed for the 3D blood vessel centerline, and the initial rotation angle range is determined as the current rotation angle range. The current rotation angle range is divided into multiple sub-search volumes. The 3D blood vessel centerline is rotated and 2D projected using the current sub-search volumes to obtain the 2D projected centerline. The preset optimal rotation parameter boundary is determined as the current optimal rotation parameter boundary of the current rotation angle range. The current translation search range corresponding to the current sub-search body is divided into multiple sub-search surfaces. The 2D projection center line is translated using the translation parameters of each sub-search surface to obtain the translated center line corresponding to each sub-search surface. Based on the translated center line and the 2D blood vessel center line, the target optimal translation parameters of the current sub-search body are searched from the translation parameters of each sub-search surface. The rotation parameter boundary of the current sub-search volume is determined based on the target optimal translation parameters, and the theoretically optimal rotation search conditions are generated based on the rotation parameter boundary and the current optimal rotation parameter boundary. If the current sub-search volume does not satisfy the theoretical optimal rotation search condition, then the current optimal rotation parameter boundary is updated, and the next sub-search volume is determined as the new current sub-search volume. Then, the process jumps back to the step of rotating and projecting the 3D blood vessel centerline using the current sub-search volume. After traversing all the sub-search bodies under the current rotation angle range, a new current rotation angle range is constructed based on each sub-search body, and the process jumps back to the step of dividing the current rotation angle range into multiple sub-search bodies; If the current sub-search volume satisfies the theoretically optimal rotation search condition, then the 3D image is registered and transformed using the target optimal translation parameter and the rotation parameter of the current sub-search volume.

[0005] Secondly, this application discloses a blood vessel registration device based on nested search, comprising: The centerline extraction module is used to extract the 3D and 2D centerlines of the target blood vessel from the 3D and 2D images, respectively. The rotation projection module is used to construct the initial rotation angle range of the 3D blood vessel centerline, and determine the initial rotation angle range as the current rotation angle range. The current rotation angle range is divided into multiple sub-search bodies. The 3D blood vessel centerline is rotated and 2D projected using the current sub-search bodies to obtain the 2D projected centerline. The preset optimal rotation parameter boundary is determined as the current optimal rotation parameter boundary of the current rotation angle range. The translation parameter determination module is used to divide the current translation search range corresponding to the current sub-search body into multiple sub-search surfaces, translate the 2D projection center line using the translation parameters of each sub-search surface to obtain the translated center line corresponding to each sub-search surface, and search for the target optimal translation parameter of the current sub-search body from the translation parameters of each sub-search surface based on the translated center line and the 2D blood vessel center line. The condition generation module is used to determine the rotation parameter boundary of the current sub-search volume based on the target optimal translation parameters, and to generate theoretically optimal rotation search conditions based on the rotation parameter boundary and the current optimal rotation parameter boundary; The rotation parameter determination module is used to update the current optimal rotation parameter boundary if the current sub-search volume does not meet the theoretical optimal rotation search conditions, and determine the next sub-search volume as the new current sub-search volume, and jump back to the step of rotating and 2D projecting the 3D blood vessel centerline using the current sub-search volume; The re-jump module is used to construct a new current rotation angle range based on each of the sub-search bodies after traversing all the sub-search bodies under the current rotation angle range, and then re-jump to the step of dividing the current rotation angle range into multiple sub-search bodies; The registration transformation module is used to perform registration transformation on the 3D image using the target optimal translation parameters and the rotation parameters of the current sub-search volume if the current sub-search volume satisfies the theoretical optimal rotation search conditions.

[0006] Thirdly, this application discloses an electronic device, comprising: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the aforementioned disclosed nested search-based vessel registration method.

[0007] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed nested search-based blood vessel registration method.

[0008] The beneficial effects of this application are as follows: As can be seen, this application avoids directly processing massive amounts of 3D and 2D images by first extracting the 3D and 2D vessel centerlines as registration benchmarks, significantly reducing the computational load for registration. Simultaneously, relying on the core morphological features of the vessel orientation and branches fully preserved by the centerline, the morphological accuracy of registration is improved. By dividing the current rotation angle range into multiple sub-search volumes and combining these sub-search volumes to rotate and 2D project the 3D vessel centerline, an ordered subdivision of the 3D rotation parameter search space is achieved. Combined with subsequent iterations, the optimal rotation parameter range can be gradually narrowed, avoiding the inefficient computation of traditional mesh exhaustive search methods while ensuring the rotation parameters are optimized. The global coverage of the search effectively solves the problem of existing registration methods easily getting trapped in local optima. By dividing the translation search range corresponding to the current sub-search volume into multiple sub-search surfaces, the translation parameters of the sub-search surfaces are used to translate the 2D projection centerline. Based on the matching relationship between the translated centerline and the 2D blood vessel centerline, the target optimal translation parameters of the current sub-search volume are searched. This achieves a dedicated fit between the translation parameters and the corresponding rotation parameters, ensuring that the translation parameters can accurately match the projection centerline under the current rotation state, improving the accuracy of the registration parameter solution. At the same time, by judging the subdivision of the sub-search surfaces, the invalid translation search space is further reduced, and the search load of the translation parameters is reduced. The rotation parameter boundary of the current sub-search volume is determined based on the target optimal translation parameters, and the theoretical optimal rotation search conditions are generated in combination with the current optimal rotation parameter boundary. This provides a clear criterion for the iterative optimization of rotation parameters, giving the iteration process a clear convergence direction and enabling the rapid selection of sub-search volumes with optimal potential, further improving the efficiency of registration parameter solution. Through an iterative update mechanism when the current sub-search volume does not meet the theoretical optimal rotation search conditions, namely updating the current optimal rotation parameter boundary and switching to the next sub-search volume to re-execute the rotation and projection process, and constructing a new rotation angle range based on the sub-search volume and re-dividing the sub-search volume, nested iteration of rotation and translation parameters is realized. The process involves continuous optimization to narrow down the range of optimal registration parameters, gradually approaching the global optimum and ensuring the global optimality of the registration parameters. When the current sub-search volume meets the theoretically optimal rotation search conditions, the corresponding target optimal translation and rotation parameters are used to perform registration transformation on the 3D image, achieving a precise mapping of the optimal registration parameters to the 3D image. This ensures that the final registered 3D image and 2D image achieve accurate matching in terms of vascular spatial location and morphology, improving the overall registration effect. At the same time, the entire process, through the subdivision and iteration of sub-search volumes and sub-search surfaces, significantly improves registration efficiency while ensuring registration accuracy and global optimum, adapting to the actual application needs of clinical medical image registration. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0010] Figure 1 This is a flowchart of a blood vessel registration method based on nested search disclosed in this application; Figure 2 This is a schematic diagram of a specific indeterminate sphere disclosed in this application; Figure 3 This is a schematic diagram of a specific rotational indeterminate circle disclosed in this application; Figure 4 This is a specific registration comparison diagram disclosed in this application; Figure 5 This is a schematic diagram of a blood vessel registration device based on nested search disclosed in this application; Figure 6 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0012] Currently, most vascular registration technologies employ 3D-2D registration methods based on iterative optimization or exhaustive search. These methods generally suffer from high search space dimensionality, large computational load, and low search efficiency, making it difficult to accurately match the dual requirements of accuracy and efficiency in vascular registration for scenarios such as clinical diagnosis and surgical navigation. Consequently, it is challenging to achieve rapid and accurate alignment of 3D and 2D vascular images.

[0013] Therefore, this application provides a vascular registration scheme based on nested search to improve the efficiency of vascular registration.

[0014] See Figure 1 As shown in the figure, this application discloses a blood vessel registration method based on nested search, including: Step S11: Extract the 3D vessel centerline and 2D vessel centerline from the 3D and 2D images of the target vessel, respectively.

[0015] The process involves acquiring 2D and 3D images of the target blood vessel, and then performing vessel segmentation on both images. Segmentation can be achieved using methods such as thresholding, region growing, level set-based methods, and tracking methods. Deep learning methods, primarily the UNet series, can also be used to obtain 2D and 3D vessel segmentation results. Skeleton thinning or distance transform methods are then employed to extract the 2D and 3D vessel centerlines from the segmentation results. The skeleton thinning method is an iterative algorithm that peels pixels layer by layer from the boundary of the segmentation result until only a single-pixel-wide centerline skeleton remains. Finally, the 3D and 2D vessel centerlines of the target blood vessel are obtained.

[0016] Step S12: Construct the initial rotation angle range of the 3D blood vessel centerline, and determine the initial rotation angle range as the current rotation angle range. Divide the current rotation angle range into multiple sub-search bodies, and use the current sub-search bodies to rotate and 2D project the 3D blood vessel centerline to obtain the 2D projected centerline. Then, determine the preset optimal rotation parameter boundary as the current optimal rotation parameter boundary of the current rotation angle range.

[0017] In this embodiment, dividing the current rotation angle range into multiple sub-search bodies includes: adding the initial search body corresponding to the current rotation angle range as the parent search body to the rotation search priority queue, and popping the parent search body from the rotation search priority queue in descending order of the upper boundary of the search parameter boundary and a second preset pop-out number, thereby dividing the parent search body into multiple sub-search bodies; the second preset pop-out number is 1.

[0018] When constructing the initial rotation angle range of the 3D blood vessel centerline, it is set as a cube [-pi,pi]³ with radius pi. This cube is the initial rotation angle range. At the same time, this initial rotation angle range is determined as the current rotation angle range. The initial search volume corresponding to the current rotation angle range is added to the rotation search priority queue as the parent search volume. Then, the parent search volume is popped from the rotation search priority queue. It should be noted that the number of parent search volumes popped each time is 1, and they need to be popped in descending order of the upper boundary of the search parameter boundary. The parent search volume is divided into multiple child search volumes. That is, the cube that is used as the current rotation angle range is taken out from the rotation search priority queue, and its xyz axis lengths are halved respectively. In this way, the cube of the current rotation angle range is divided into 8 equal child cubes. These 8 child cubes are the corresponding multiple child search volumes, which serve as the new 3D rotation angle search space.

[0019] The current sub-search volume is determined from multiple sub-search volumes. First, the center position of the current sub-search volume is calculated. The center position is used as the corresponding rotation angle to perform a rotation transformation on the 3D blood vessel centerline. Then, a projection matrix is ​​generated based on the camera parameters of the 2D-DSAdicom image. The 3D blood vessel centerline after rotation transformation is projected onto the 2D plane using the projection matrix. The rotation and 2D projection operations of the 3D blood vessel centerline are completed to obtain the projected centerline.

[0020] The preset optimal rotation parameter boundary is the initial rotation angle range, and the preset optimal rotation parameter boundary is determined as the current optimal rotation parameter boundary of the current rotation angle range.

[0021] Step S13: Divide the current translation search range corresponding to the current sub-search volume into multiple sub-search surfaces, and use the translation parameters of each sub-search surface to translate the 2D projection centerline to obtain the translated centerline corresponding to each sub-search surface. Based on the translated centerline and the 2D blood vessel centerline, search for the target optimal translation parameters of the current sub-search volume from the translation parameters of each sub-search surface.

[0022] In this embodiment, dividing the current translation search range corresponding to the current sub-search body into multiple sub-search surfaces includes: constructing an initial search surface corresponding to the current translation search range corresponding to the current sub-search body; adding the initial search surface as a parent search surface to the translation search priority queue; popping the parent search surface from the translation search priority queue according to the order of the upper boundary of the translation parameter boundary from largest to smallest and a first preset pop-up number, and dividing the parent search surface into multiple sub-search surfaces; the first preset pop-up number is 1.

[0023] The projection center line obtained by rotating and projecting the current sub-search volume is set as a square [-t, t]² with half side length t as the initial current translation search range. This square current translation search range is determined as the initial search surface and added to the translation search priority queue as the parent search surface. Then, the parent search surfaces in the translation search priority queue are popped. It should be noted that only one parent search surface is popped at a time, and the parent search surface popped each time is the parent search surface with the largest upper boundary of the translation parameter boundary in the translation search priority queue. That is, the parent search surfaces in the translation search priority queue are popped in descending order of the upper boundary of the translation parameter boundary each time, and only one parent search surface is popped at a time. Then, the popped parent search surface is divided into multiple sub-search surfaces, that is, its xy axis length is halved. In this way, the square of the current translation search range is divided into 4 equal sub-squares. These 4 sub-squares are the corresponding multiple sub-search surfaces, which serve as the new 2D translation search space.

[0024] In this embodiment, the step of searching for the target optimal translation parameters of the current sub-search volume from the translation parameters of each of the sub-search surfaces based on the translated centerline and the 2D blood vessel centerline includes: determining a preset optimal translation parameter boundary as the current optimal translation parameter boundary of the current translation search range; determining the current sub-search surface from multiple sub-search surfaces, and determining the translation parameter boundary of the current sub-search surface; generating theoretically optimal translation search conditions based on the translation parameter boundary and the current optimal translation parameter boundary; if the current sub-search surface does not satisfy the theoretically optimal translation search conditions, then updating the current... The previous optimal translation parameter boundary is determined, and the next sub-search surface is identified as the new current sub-search surface. The process then jumps back to the step of determining the translation parameter boundary of the current sub-search surface. After traversing all the sub-search surfaces under the current translation search range, a new current translation angle range is constructed based on each sub-search surface, and the process jumps back to the step of dividing the current translation search range corresponding to the current sub-search volume into multiple sub-search surfaces. If the current sub-search surface satisfies the theoretical optimal translation search condition, the translation parameter of the current sub-search surface is identified as the target optimal translation parameter of the current sub-search volume.

[0025] The preset optimal translation parameter boundary is determined as the current optimal translation parameter boundary of the current translation search range, providing an initial judgment criterion for the translation search. The translation parameter boundary of the current sub-search surface in the translation search priority queue is determined, i.e., the lower and upper registration bounds corresponding to the current sub-search surface are calculated. The theoretically optimal translation search conditions are generated based on the translation parameter boundaries and the current optimal translation parameter boundary. If the current sub-search surface does not meet the theoretically optimal translation search conditions, i.e., the upper registration bound of the current sub-search surface is less than the current optimal registration lower bound, the search of that search space is directly terminated. If the lower registration bound of the current sub-search surface is greater than the current optimal registration lower bound, the translation space is set as optimal and the optimal translation space is updated, thus completing the update of the current optimal translation parameter boundary. The next sub-search surface is then determined as the new current sub-search surface, and the search jumps back to determining the translation search priority queue. The steps for determining the translation parameter boundary of the current sub-search surface: After traversing all sub-search surfaces under the current translation search range, if no optimal translation parameter satisfying the termination condition is found, a new current translation angle range is constructed based on each sub-search surface. That is, the unpruned sub-search surfaces are added back to the translation search priority queue as a new search space, and the process jumps back to the step of dividing the current translation search range corresponding to the current sub-search volume into multiple sub-search surfaces, continuing to divide and search the new translation search range. If the current sub-search surface satisfies the theoretical optimal translation search condition, that is, the lower registration bound of the current sub-search surface is equal to the upper registration bound of the current optimal registration, or the side length of the sub-search surface is less than the specified side length threshold, then the translation parameter of the current sub-search surface is determined as the target optimal translation parameter of the current sub-search volume, that is, the center position of the sub-search surface is returned as the optimal translation parameter.

[0026] In this embodiment, determining the translation parameter boundary of the current sub-search surface in the translation search priority queue includes: determining the first distance between each feature point in the translated centerline corresponding to the current sub-search surface and the 2D blood vessel centerline, and determining the number of first feature points whose first distance is less than a first preset threshold; determining the radius of the translation uncertainty circle based on the side length of the current sub-search surface, and determining the sum of the radius of the translation uncertainty circle and the first preset threshold as a new first preset threshold, and determining the number of second feature points whose first distance is less than the new first preset threshold; and determining the number of first feature points and the number of second feature points as the first lower boundary and the first upper boundary in the translation parameter boundary of the current sub-search surface, respectively.

[0027] First, for the projected centerline after transformation by the translation parameters corresponding to the center position of the current sub-search surface, i.e. the translated centerline, calculate the Euclidean distance from each feature point to the nearest point of the 2D blood vessel centerline as the first distance. Then, determine whether each first distance is less than a preset distance threshold, i.e. the first preset threshold. Count the total number of feature points that meet the determination condition as the number of first feature points. Then, based on the attribute that the current sub-search surface is a square, calculate the diagonal length sqrt(2) corresponding to its side length. t (where t is half the side length of the initial square of the translation search range) is used as the radius of the translation uncertainty circle. The radius of the translation uncertainty circle is summed with the first preset threshold, and the resulting value is used as the new first preset threshold. Based on the first distance of each feature point calculated previously, it is determined again whether each distance is less than the new first preset threshold. The total number of feature points that meet the new determination condition is counted as the second feature point number. Finally, the first feature point number obtained by the statistics is directly determined as the first lower boundary, i.e., the lower registration boundary, in the translation parameter boundary of the current sub-search surface. The second feature point number obtained by the statistics is directly determined as the first upper boundary, i.e., the upper registration boundary, in the translation parameter boundary of the current sub-search surface. This constitutes the translation parameter boundary of the current sub-search surface.

[0028] In this embodiment, the theoretically optimal translation search condition is that the first lower boundary is equal to the upper boundary in the current optimal translation parameter boundary.

[0029] The theoretically optimal translation search condition is that the first lower boundary is equal to the upper boundary of the current optimal translation parameter boundary. Specifically, during the 2D translation search of the projection centerline corresponding to the current sub-search volume based on the branch-boundary method, the first lower boundary representing the lower registration boundary in the translation parameter boundary of the current sub-search surface is numerically compared with the upper boundary representing the overall optimal registration upper boundary in the current optimal translation parameter boundary. When the two values ​​are equal, the theoretically optimal translation search condition is satisfied. The fulfillment of this condition means that the translation parameters corresponding to the current sub-search surface can achieve optimal registration between the projection centerline and the 2D vessel centerline. At this point, there is no need to continue searching for other sub-search surfaces. The center position of the current sub-search surface can be directly determined as the target optimal translation parameter corresponding to the current sub-search volume, and the current 2D translation search process is terminated. This condition is also the core judgment basis for the branch-boundary method to achieve pruning and fast convergence to the optimal solution in the translation search, which can effectively reduce the invalid search space and improve the efficiency of 2D translation search.

[0030] In this embodiment, updating the current optimal translation parameter boundary includes: if the first lower boundary is greater than the lower boundary in the current optimal translation parameter boundary, then updating the lower boundary in the current optimal translation parameter boundary to the first lower boundary.

[0031] If the first lower boundary is greater than the lower boundary in the current optimal translation parameter boundary, then the lower boundary in the current optimal translation parameter boundary is updated to the first lower boundary. Specifically, after performing a 2D translation search on the projection centerline corresponding to the current sub-search volume based on the branch-and-bound method and calculating the first lower boundary in the translation parameter boundary of the current sub-search surface, the value of the first lower boundary is compared with the value of the lower boundary already set in the current optimal translation parameter boundary. If the value of the first lower boundary of the current sub-search surface is larger, it means that the translation parameters corresponding to the sub-search surface can achieve a better registration effect. At this time, the lower boundary in the current optimal translation parameter boundary is directly replaced with the larger first lower boundary, completing the update of the current optimal translation parameter boundary. This ensures that the lower boundary of the current optimal translation parameter boundary always represents the optimal registration lower bound among all searched sub-search surfaces up to the present, providing a more accurate quantitative basis for subsequent pruning operations and the determination of the optimal translation parameters, and ensuring that the translation search can converge to the sub-search space with a better registration effect.

[0032] In this embodiment, constructing a new current translation angle range based on each of the sub-search surfaces includes: if the first upper boundary is less than the lower boundary in the current optimal translation parameter boundary, then ending the search of the current sub-search surface; if the first upper boundary is not less than the lower boundary in the current optimal translation parameter boundary, then retaining the current sub-search surface in the translation search priority queue; and constructing a new current translation angle range using each of the sub-search surfaces in the translation search priority queue.

[0033] If the first upper boundary is less than the lower boundary of the current optimal translation parameter boundary, the search process for the current sub-search surface ends; if the first upper boundary is not less than the lower boundary of the current optimal translation parameter boundary, the current sub-search surface is kept in the translation search priority queue; a new current translation angle range is constructed using the sub-search surfaces in the translation search priority queue. Specifically, after performing a 2D translation search on the current sub-search volume based on the branch-and-boundary method and calculating the first upper boundary of the current sub-search surface (i.e., the registration upper boundary), the first upper boundary is numerically compared with the lower boundary of the current optimal translation parameter boundary (i.e., the global optimal registration lower boundary). If the first upper boundary of the current sub-search surface is less than the lower boundary of the current optimal translation parameter boundary, it means that there is no energy within the sub-search surface. If a translation parameter is better than the current optimal value, there is no value in continuing the search. The current sub-search surface is directly removed from the translation search priority queue and pruned. If the first upper boundary of the current sub-search surface is not less than the lower boundary of the current optimal translation parameter boundary, it means that there is still a possibility of finding a better translation parameter within the sub-search surface. The current sub-search surface should be kept in the translation search priority queue. After traversing all sub-search surfaces, all the remaining sub-search surfaces in the translation search priority queue are used as a new search space. The spatial ranges of these sub-search surfaces are integrated to form a new current translation angle range, providing a new search range for subsequent sub-search surface division and translation parameter evaluation operations, and realizing the gradual convergence of the translation search space towards the optimal solution.

[0034] In this embodiment, before updating the boundary of the current optimal translation parameter, the method further includes: if the side length of the current sub-search surface is less than the first side length threshold, then the translation parameter of the current sub-search surface is determined as the target optimal translation parameter of the current sub-search volume.

[0035] Before updating the current optimal translation parameter boundary, if the side length of the current sub-search surface is less than the first side length threshold, the translation parameters of the current sub-search surface are determined as the target optimal translation parameters of the current sub-search volume. Specifically, in the pre-step of performing a 2D translation search on the projection centerline corresponding to the current sub-search volume based on the branch-and-bound method, calculating the translation parameter boundary of the current sub-search surface, and preparing to update the current optimal translation parameter boundary, the spatial scale of the current sub-search surface is first determined, that is, checking whether the actual side length of the sub-search surface as a square is less than the pre-set first side length threshold. This threshold is a quantification standard for determining whether the translation search space has reached a sufficiently high level of refinement. If the side length of the current sub-search surface is less than the first side length threshold, it means that the translation search space has converged to a sufficiently small range, and the corresponding translation parameters can meet the accuracy requirements of 3D-2D blood vessel registration. There is no need to continue to divide, evaluate, and update the boundary of the current optimal translation parameters of the sub-search surface. The translation parameters corresponding to the center position of the current sub-search surface can be directly determined as the target optimal translation parameters of the current sub-search volume, and the 2D translation search process for the current sub-search volume is terminated. In this way, while ensuring the registration accuracy, the invalid calculation caused by over-search is avoided, and the efficiency of the overall translation search is improved.

[0036] Step S14: Determine the rotation parameter boundary of the current sub-search volume based on the target optimal translation parameters, and generate the theoretically optimal rotation search conditions based on the rotation parameter boundary and the current optimal rotation parameter boundary.

[0037] In this embodiment, determining the rotation parameter boundary of the current sub-search body based on the target optimal translation parameter includes: determining the number of registration points corresponding to the target optimal translation parameter as the second lower boundary in the rotation parameter boundary of the current sub-search body; determining the rotation uncertainty sphere of the 3D centerline point under the current sub-search body, and projecting the rotation uncertainty sphere to obtain a rotation uncertainty circle, and determining the sum of the radius of the rotation uncertainty circle and the second preset error as a new second preset error; translating the rotated 3D blood vessel centerline using the target optimal translation parameter to obtain a translated point set, determining the second distance between each feature point in the translated point set and the 2D blood vessel centerline; determining the number of third feature points in the translated point set whose second distance is less than the new second preset error, and determining the number of third feature points as the second upper boundary in the rotation parameter boundary of the current sub-search body.

[0038] After obtaining the optimal translation parameters of the target corresponding to the current sub-search volume using the branch-and-bound method, the number of registration points corresponding to these optimal translation parameters is directly used as the second lower boundary in the rotation parameter boundary of the current sub-search volume, i.e., the lower registration boundary of the rotation subspace. Then, based on the rotation space attribute of the current sub-search volume as a cube, the size of the rotation uncertainty sphere of the 3D centerline point in the rotation space of this sub-cube is calculated, for example... Figure 2 As shown, this uncertain sphere characterizes the range of deviation between any rotation within the sub-cube and the rotation at the center of the cube, as illustrated in the following formula: ; ; Among them, R c R represents the rotation represented by the center position of the cube. u Represents any rotation within the sub-cube, where x represents the center line point and x0 is the center of rotation; the formula represents the rotation R around x0 for any point x. c Subtract the point's rotation around x0 by R. u The result is less than the radius δ of the uncertain sphere.

[0039] The rotationally uncertain sphere is then projected onto a 2D plane using the camera parameters of the 2D-DSA dicom image, such as... Figure 3 As shown, the corresponding rotational uncertainty circle is obtained and its radius is calculated. The radius of the rotational uncertainty circle is summed with the pre-set second preset error, and the resulting value is used as the new second preset error. Then, using the obtained target optimal translation parameters, a translation transformation is performed on the 3D blood vessel centerline after rotation through the current sub-search volume center position to obtain the translated point set. The Euclidean distance from each feature point in the translated point set to the nearest point of the 2D blood vessel centerline is calculated as the second distance. Each second distance is determined to be less than the new second preset error. The total number of feature points that meet the determination condition is counted as the third feature point number. Finally, the third feature point number is determined as the second upper boundary in the current sub-search volume rotation parameter boundary, that is, the registration upper boundary of the rotation subspace. Thus, the second lower boundary and the second upper boundary together constitute the complete rotation parameter boundary of the current sub-search volume, providing a quantitative boundary basis for subsequent pruning of 3D rotation search and determination of optimal rotation parameters.

[0040] In this embodiment, the theoretically optimal rotation search condition is that the second lower boundary is equal to the upper boundary of the current optimal rotation parameter boundary.

[0041] The theoretically optimal rotation search condition is that the second lower boundary equals the upper boundary of the current optimal rotation parameter boundary. Specifically, in the nested 3D rotation search of the 3D vessel centerline based on the branch-and-bound method, after obtaining the target optimal translation parameters of the current sub-search volume through 2D translation search and calculating the second lower boundary of the rotation parameter boundary of the sub-search volume, the second lower boundary is numerically compared with the upper boundary of the current optimal rotation parameter boundary, which represents the global optimal registration effect. When the two values ​​are equal, the theoretically optimal rotation search condition is satisfied. The fulfillment of this condition indicates that the rotation parameters corresponding to the current sub-search volume, combined with the obtained optimal translation parameters, can achieve global alignment between the 3D vessel centerline projection and the 2D vessel centerline. Optimal registration occurs when there are no better rotation parameters within the current rotational subspace. There is no need for further segmentation, evaluation, or searching of other rotational sub-search bodies. The rotation parameters corresponding to the center position of the current sub-search body can be directly determined as the globally optimal rotation parameters, terminating the current 3D rotational search process. This condition is the core criterion for the branch-bound method to prune and quickly converge to the globally optimal rotational solution in the 3D rotational search stage. This condition effectively eliminates invalid rotational search spaces, significantly improving the efficiency of the 3D rotational search while ensuring the global optimality of the registration results. This aligns with the design intent of this nested search 3D-2D vessel registration method, which does not rely on initial values ​​and pursues global optimality.

[0042] Step S15: If the current sub-search volume does not satisfy the theoretical optimal rotation search condition, then update the current optimal rotation parameter boundary, determine the next sub-search volume as the new current sub-search volume, and jump back to the step of rotating and 2D projecting the 3D blood vessel centerline using the current sub-search volume.

[0043] In this embodiment, updating the current optimal rotation parameter boundary includes: if the second upper boundary is less than the lower boundary in the current optimal rotation parameter boundary, then ending the search of the current sub-search body; if the second upper boundary is not less than the lower boundary in the current optimal rotation parameter boundary, then retaining the current sub-search body in the rotation search priority queue; and constructing a new current rotation angle range using each of the sub-search bodies in the rotation search priority queue.

[0044] If the current sub-search volume does not meet the theoretical optimal rotation search conditions, then update the current optimal rotation parameter boundary, determine the next sub-search volume as the new current sub-search volume, and jump back to the step of rotating the 3D blood vessel centerline and projecting it into 2D using the current sub-search volume.

[0045] The process of updating the current optimal rotation parameter boundary is as follows: If the second upper boundary is less than the lower boundary of the current optimal rotation parameter boundary, the search of the current sub-search body ends; if the second upper boundary is not less than the lower boundary of the current optimal rotation parameter boundary, the current sub-search body is kept in the rotation search priority queue; a new current rotation angle range is constructed using the sub-search bodies in the rotation search priority queue. Specifically, when performing a 3D rotation search based on the branch-and-boundary method and determining that the current sub-search body does not meet the theoretical optimal rotation search condition that the second lower boundary is equal to the upper boundary of the current optimal rotation parameter boundary, the current optimal rotation parameter boundary is first updated. During the update, the second upper boundary of the current sub-search body's rotation parameter boundary is numerically compared with the lower boundary of the current optimal rotation parameter boundary. If the second upper boundary is less than the lower boundary of the current optimal rotation parameter boundary, it means that there is no registration result better than the current global optimum in the rotation space of the current sub-search body, and there is no value in continuing the search. The current sub-search body is directly pruned. If the second upper boundary is not less than the lower boundary of the current optimal rotation parameter boundary, it means that there is still a possibility of finding better rotation parameters in the rotation space of the current sub-search body. The current sub-search volume needs to be kept in the rotation search priority queue. After the determination operation is completed, all the remaining sub-search volumes in the rotation search priority queue are integrated, and the spatial range of these sub-search volumes is used to construct a new current rotation angle range. The boundary of the current optimal rotation parameter is updated. Then, the next sub-search volume in the rotation search priority queue is determined as the new current sub-search volume. The process jumps back to the step of rotating and projecting the 3D blood vessel centerline using the current sub-search volume. The complete set of operations of rotation, projection, 2D translation search, rotation parameter boundary calculation and optimality determination is performed on the new current sub-search volume. In this way, the 3D rotation search space is gradually pruned and converged until a sub-search volume that meets the theoretical optimal rotation search conditions is found or the preset search termination condition is reached.

[0046] In this embodiment, before updating the current optimal rotation parameter boundary, the method further includes: if the side length of the current sub-search volume is less than the first side length threshold, then the 3D image is registered and transformed using the lower boundary in the current optimal rotation parameter boundary and the target optimal translation parameter of the current sub-search volume.

[0047] Before updating the current optimal rotation parameter boundary, if the side length of the current sub-search volume is less than the first side length threshold, then the 3D image is registered using the lower boundary of the current optimal rotation parameter boundary and the target optimal translation parameter of the current sub-search volume. Specifically, after performing 3D rotation search based on the branch-and-boundary method, solving for the target optimal translation parameter and calculating the rotation parameter boundary corresponding to the current sub-search volume, and before performing the current optimal rotation parameter boundary update operation, a pre-judgment step is performed to first determine the scale of the actual side length of the current sub-search volume as a cube, checking whether it is less than the pre-set first side length threshold. This threshold is a quantitative standard for measuring whether the 3D rotation search space has converged to a sufficiently fine level to meet the accuracy requirements of 3D-2D vessel registration. If the side length of the current sub-search volume is less than the first threshold, then the 3D image is registered. The condition of a long threshold indicates that the rotation search space no longer needs to be further segmented and searched, and the accuracy of its corresponding rotation parameters is already suitable for the registration requirements. At this point, subsequent operations such as rotation search priority queue pruning and optimal rotation parameter boundary updates are no longer performed. Instead, the rotation parameters corresponding to the lower boundary representing the global optimal registration effect in the current optimal rotation parameter boundary are directly retrieved. Combined with the previously obtained target optimal translation parameters of the current sub-search volume, these two parameters are used as the final registration transformation parameters. A registration transformation operation combining rotation and translation transformations is performed on the original 3D vascular image to complete the registration of the 3D image and the 2D vascular image. This terminates excessive 3D rotation search while ensuring registration accuracy, avoids invalid calculations, and improves the overall efficiency of 3D-2D vascular registration.

[0048] Step S16: After traversing all the sub-search bodies under the current rotation angle range, construct a new current rotation angle range based on each sub-search body, and jump back to the step of dividing the current rotation angle range into multiple sub-search bodies.

[0049] After traversing all sub-search bodies within the current rotation angle range, a new current rotation angle range is constructed based on each sub-search body. The process then jumps back to the step of dividing the current rotation angle range into multiple sub-search bodies. Specifically, during the nested search of 3D rotation based on the branch-and-bound method, each sub-cube-shaped sub-search body within the current rotation angle range is processed sequentially. This involves performing 3D rotation transformation, 2D projection, 2D translation search to solve for the target optimal translation parameters, calculating the second upper and lower bounds of the rotation parameter boundary, comparing with the current optimal rotation parameter boundary, and pruning and retaining operations on each sub-search body. After traversing all sub-search bodies, a selection and integration process is performed. Sub-search bodies whose second upper bound is not less than the lower bound of the current optimal rotation parameter boundary and have continuing search value are retained in the rotation search priority queue as the core components. The spatial range of these retained sub-search bodies is integrated to form the new current rotation angle. The new rotation angle range is a precise convergence of the original search range, eliminating invalid spaces with no search value. Then, it jumps back to the step of dividing the current rotation angle range into multiple sub-search volumes. The branch operation is performed again on the newly constructed current rotation angle range, and each sub-search volume is halved along the xyz axis to be cut into 8 new sub-cubes. The subsequent nested search process of rotation transformation, projection, translation search, boundary calculation and pruning judgment continues. In this way, through the iterative operation of traversing, filtering, constructing new ranges and cutting sub-search volumes, the 3D rotation search space is continuously converged towards the direction of the globally optimal rotation parameters until a sub-search volume that meets the theoretically optimal rotation search conditions or the side length of the sub-search volume is less than a specified threshold is found. Finally, the globally optimal solution for 3D-2D blood vessel registration is obtained. At the same time, the exhaustive calculation of mesh search is avoided by relying on this iterative convergence method, which greatly reduces the computational amount of 3D rotation search.

[0050] Step S17: If the current sub-search volume satisfies the theoretical optimal rotation search condition, then the 3D image is registered and transformed using the target optimal translation parameter of the current sub-search volume and the rotation parameter of the current sub-search volume.

[0051] If the current sub-search volume satisfies the theoretically optimal rotation search condition, then the 3D image is registered using the target optimal translation parameter and the rotation parameter of the current sub-search volume. Specifically, in the nested 3D-2D vessel registration 3D rotation search process based on the branch-boundary method, when it is determined that the current sub-search volume satisfies the theoretically optimal rotation search condition that the second lower boundary is equal to the upper boundary of the current optimal rotation parameter boundary, it indicates that the rotation parameter corresponding to the current sub-search volume is a 3D rotation parameter that can achieve the globally optimal registration effect. At this time, the rotation parameter corresponding to the center position of the current sub-search volume is directly retrieved, and the target optimal translation parameter previously obtained for the current sub-search volume through 2D translation search is also retrieved. These two parameters are used as the final registration transformation parameters for this 3D-2D vessel registration. Based on these rotation parameters, a 3D rotation transformation is performed on the original 3D image of the target vessel. Then, combined with the target optimal translation parameters, a 2D translation transformation is performed on the 3D image after the rotation transformation. The combination of rotation and translation transformations completes the overall registration transformation of the 3D image. This ensures that the transformed 3D vessel centerline, when projected onto the 2D plane, can achieve precise spatial alignment and matching with the 2D vessel centerline, achieving the global optimal effect of 3D-2D vessel registration. At this point, there is no need to perform subsequent rotation search, projection, and translation search operations on other sub-search volumes. The entire nested search process of 3D rotation is terminated directly, and the registration is completed.

[0052] like Figure 4 The diagram shows a specific registration comparison, where blue represents 2D points and red represents the projection results of 3D points. Figure 4 (a) in the image represents the image before registration. Figure 4 (b) in the figure represents the result after registration.

[0053] The beneficial effects of this application are as follows: As can be seen, this application avoids directly processing massive amounts of 3D and 2D images by first extracting the 3D and 2D vessel centerlines as registration benchmarks, significantly reducing the computational load for registration. Simultaneously, relying on the core morphological features of the vessel orientation and branches fully preserved by the centerline, the morphological accuracy of registration is improved. By dividing the current rotation angle range into multiple sub-search volumes and combining these sub-search volumes to rotate and 2D project the 3D vessel centerline, an ordered subdivision of the 3D rotation parameter search space is achieved. Combined with subsequent iterations, the optimal rotation parameter range can be gradually narrowed, avoiding the inefficient computation of traditional mesh exhaustive search methods while ensuring the rotation parameters are optimized. The global coverage of the search effectively solves the problem of existing registration methods easily getting trapped in local optima. By dividing the translation search range corresponding to the current sub-search volume into multiple sub-search surfaces, the translation parameters of the sub-search surfaces are used to translate the 2D projection centerline. Based on the matching relationship between the translated centerline and the 2D blood vessel centerline, the target optimal translation parameters of the current sub-search volume are searched. This achieves a dedicated fit between the translation parameters and the corresponding rotation parameters, ensuring that the translation parameters can accurately match the projection centerline under the current rotation state, improving the accuracy of the registration parameter solution. At the same time, by judging the subdivision of the sub-search surfaces, the invalid translation search space is further reduced, and the search load of the translation parameters is reduced. The rotation parameter boundary of the current sub-search volume is determined based on the target optimal translation parameters, and the theoretical optimal rotation search conditions are generated in combination with the current optimal rotation parameter boundary. This provides a clear criterion for the iterative optimization of rotation parameters, giving the iteration process a clear convergence direction and enabling the rapid selection of sub-search volumes with optimal potential, further improving the efficiency of registration parameter solution. Through an iterative update mechanism when the current sub-search volume does not meet the theoretical optimal rotation search conditions, namely updating the current optimal rotation parameter boundary and switching to the next sub-search volume to re-execute the rotation and projection process, and constructing a new rotation angle range based on the sub-search volume and re-dividing the sub-search volume, nested iteration of rotation and translation parameters is realized. The process involves continuous optimization to narrow down the range of optimal registration parameters, gradually approaching the global optimum and ensuring the global optimality of the registration parameters. When the current sub-search volume meets the theoretically optimal rotation search conditions, the corresponding target optimal translation and rotation parameters are used to perform registration transformation on the 3D image, achieving a precise mapping of the optimal registration parameters to the 3D image. This ensures that the final registered 3D image and 2D image achieve accurate matching in terms of vascular spatial location and morphology, improving the overall registration effect. At the same time, the entire process, through the subdivision and iteration of sub-search volumes and sub-search surfaces, significantly improves registration efficiency while ensuring registration accuracy and global optimum, adapting to the actual application needs of clinical medical image registration.

[0054] See Figure 5 As shown in the figure, this application discloses a blood vessel registration device based on nested search, including: The centerline extraction module 11 is used to extract the 3D centerline and 2D centerline of the target blood vessel from the 3D image and 2D image, respectively. The rotation projection module 12 is used to construct the initial rotation angle range of the 3D blood vessel centerline, and determine the initial rotation angle range as the current rotation angle range. The current rotation angle range is divided into multiple sub-search bodies. The 3D blood vessel centerline is rotated and 2D projected using the current sub-search bodies to obtain the 2D projected centerline. The preset optimal rotation parameter boundary is determined as the current optimal rotation parameter boundary of the current rotation angle range. The translation parameter determination module 13 is used to divide the current translation search range corresponding to the current sub-search body into multiple sub-search surfaces, use the translation parameters of each sub-search surface to translate the 2D projection center line to obtain the translated center line corresponding to each sub-search surface, and search for the target optimal translation parameter of the current sub-search body from the translation parameters of each sub-search surface based on the translated center line and the 2D blood vessel center line; The condition generation module 14 is used to determine the rotation parameter boundary of the current sub-search body based on the target optimal translation parameters, and to generate theoretically optimal rotation search conditions based on the rotation parameter boundary and the current optimal rotation parameter boundary; The rotation parameter determination module 15 is used to update the current optimal rotation parameter boundary if the current sub-search volume does not meet the theoretical optimal rotation search conditions, and determine the next sub-search volume as the new current sub-search volume, and jump back to the step of rotating and 2D projecting the 3D blood vessel centerline using the current sub-search volume; The re-jump module 16 is used to construct a new current rotation angle range based on each of the sub-search bodies after traversing all the sub-search bodies under the current rotation angle range, and then re-jump to the step of dividing the current rotation angle range into multiple sub-search bodies. The registration transformation module 17 is used to perform registration transformation on the 3D image using the target optimal translation parameter and the rotation parameter of the current sub-search volume if the current sub-search volume satisfies the theoretical optimal rotation search condition.

[0055] Furthermore, embodiments of this application also provide an electronic device. Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0056] Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Specifically, it may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. Furthermore, the memory 22, as a carrier for resource storage, may be a read-only memory, random access memory, a disk, or an optical disk, etc., and the resources stored thereon include an operating system 221, computer programs 222, and data 223, etc., and the storage method may be temporary storage or permanent storage.

[0057] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed nested search-based vessel registration method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0058] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0059] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0060] The present invention provides a detailed description of a method, apparatus, device, and medium for vascular registration based on nested search. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only intended to help understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A blood vessel registration method based on nested search, characterized in that, include: Extract the 3D and 2D centerlines of the target blood vessel from the 3D and 2D images, respectively. An initial rotation angle range is constructed for the 3D blood vessel centerline, and the initial rotation angle range is determined as the current rotation angle range. The current rotation angle range is divided into multiple sub-search volumes. The 3D blood vessel centerline is rotated and 2D projected using the current sub-search volumes to obtain the 2D projected centerline. The preset optimal rotation parameter boundary is determined as the current optimal rotation parameter boundary of the current rotation angle range. The current translation search range corresponding to the current sub-search body is divided into multiple sub-search surfaces. The 2D projection center line is translated using the translation parameters of each sub-search surface to obtain the translated center line corresponding to each sub-search surface. Based on the translated center line and the 2D blood vessel center line, the target optimal translation parameters of the current sub-search body are searched from the translation parameters of each sub-search surface. The rotation parameter boundary of the current sub-search volume is determined based on the target optimal translation parameters, and the theoretically optimal rotation search conditions are generated based on the rotation parameter boundary and the current optimal rotation parameter boundary. If the current sub-search volume does not satisfy the theoretical optimal rotation search condition, then the current optimal rotation parameter boundary is updated, and the next sub-search volume is determined as the new current sub-search volume. Then, the process jumps back to the step of rotating and projecting the 3D blood vessel centerline using the current sub-search volume. After traversing all the sub-search bodies under the current rotation angle range, a new current rotation angle range is constructed based on each sub-search body, and the process jumps back to the step of dividing the current rotation angle range into multiple sub-search bodies; If the current sub-search volume satisfies the theoretical optimal rotation search condition, then the 3D image is registered and transformed using the target optimal translation parameter and the rotation parameter of the current sub-search volume; The step of dividing the current translation search range corresponding to the current sub-search volume into multiple sub-search surfaces includes: Construct an initial search surface corresponding to the current translation search range of the current sub-search body, and add the initial search surface as the parent search surface to the translation search priority queue; The parent search surfaces in the translation search priority queue are popped out according to the order of the upper boundary of the translation parameter boundary from largest to smallest and the first preset pop-out number, and the parent search surfaces are divided into multiple sub-search surfaces; the first preset pop-out number is 1. Accordingly, the step of searching for the target optimal translation parameters of the current sub-search volume from the translation parameters of each of the sub-search surfaces based on the translated centerline and the 2D vessel centerline includes: The preset optimal translation parameter boundary is determined as the current optimal translation parameter boundary of the current translation search range; The current sub-search surface is determined from the multiple sub-search surfaces, and the translation parameter boundary of the current sub-search surface is determined. The theoretically optimal translation search conditions are generated based on the translation parameter boundary and the current optimal translation parameter boundary. If the current sub-search surface does not satisfy the theoretical optimal translation search condition, then update the current optimal translation parameter boundary, determine the next sub-search surface as the new current sub-search surface, and jump back to the step of determining the translation parameter boundary of the current sub-search surface; After traversing all the sub-search surfaces under the current translation search range, a new current translation search range is constructed based on each sub-search surface, and the process jumps back to the step of dividing the current translation search range corresponding to the current sub-search body into multiple sub-search surfaces; If the current sub-search surface satisfies the theoretical optimal translation search condition, then the translation parameter of the current sub-search surface is determined as the target optimal translation parameter of the current sub-search volume; Determining the translation parameter boundary of the current sub-search surface includes: Determine the first distance between each feature point in the translated centerline corresponding to the current sub-search surface and the 2D blood vessel centerline, and determine the number of first feature points whose first distance is less than a first preset threshold; The radius of the translation uncertainty circle is determined based on the side length of the current sub-search surface, and the sum of the radius of the translation uncertainty circle and the first preset threshold is determined as a new first preset threshold. The number of second feature points whose first distance is less than the new first preset threshold is also determined. The number of the first feature points and the number of the second feature points are respectively determined as the first lower boundary and the first upper boundary in the translation parameter boundary of the current sub-search surface.

2. The vessel registration method based on nested search according to claim 1, characterized in that, The theoretically optimal translation search condition is that the first lower boundary is equal to the upper boundary of the current optimal translation parameter boundary; updating the current optimal translation parameter boundary includes: If the first lower boundary is greater than the lower boundary in the current optimal translation parameter boundary, then the lower boundary in the current optimal translation parameter boundary is updated to the first lower boundary; Accordingly, constructing a new current translation search range based on each of the sub-search surfaces includes: If the first upper boundary is smaller than the lower boundary in the current optimal translation parameter boundary, then the search for the current sub-search surface ends. If the first upper boundary is not less than the lower boundary in the current optimal translation parameter boundary, then the current sub-search surface is retained in the translation search priority queue; A new current translation search range is constructed using each of the sub-search surfaces in the translation search priority queue.

3. The vessel registration method based on nested search according to claim 2, characterized in that, The process of dividing the current rotation angle range into multiple sub-search volumes includes: The initial search body corresponding to the current rotation angle range is added to the rotation search priority queue as the parent search body. The parent search body is popped from the rotation search priority queue in descending order of the upper boundary of the search parameter boundary and the second preset popping number, and the parent search body is divided into multiple child search bodies. The second preset popping number is 1. Accordingly, determining the rotation parameter boundary of the current sub-search volume based on the target optimal translation parameters includes: The number of registration points corresponding to the optimal translation parameters of the target is determined as the second lower boundary in the rotation parameter boundary of the current sub-search volume; Determine the rotational uncertainty sphere under the current sub-search volume by the 3D centerline point, and project the rotational uncertainty sphere to obtain the rotational uncertainty circle. The sum of the radius of the rotational uncertainty circle and the second preset error is determined as the new second preset error. The 3D blood vessel centerline after rotation is translated using the target optimal translation parameters to obtain a translated point set, and the second distance between each feature point in the translated point set and the 2D blood vessel centerline is determined. Determine the number of third feature points in the translated point set whose second distance is less than the new second preset error, and determine the number of third feature points as the second upper boundary in the rotation parameter boundary of the current sub-search volume.

4. The vessel registration method based on nested search according to claim 3, characterized in that, The theoretically optimal rotation search condition is that the second lower boundary is equal to the upper boundary of the current optimal rotation parameter boundary; updating the current optimal rotation parameter boundary includes: If the second upper boundary is less than the lower boundary in the current optimal rotation parameter boundary, then the search for the current sub-search volume ends; If the second upper boundary is not less than the lower boundary in the current optimal rotation parameter boundary, then the current sub-search volume is retained in the rotation search priority queue; A new range of current rotation angles is constructed using each of the sub-search bodies in the rotation search priority queue.

5. The vessel registration method based on nested search according to claim 4, characterized in that, Before updating the current optimal translation parameter boundary, the method further includes: If the side length of the current sub-search surface is less than the first side length threshold, then the translation parameter of the current sub-search surface is determined as the target optimal translation parameter of the current sub-search volume. Accordingly, before updating the current optimal rotation parameter boundary, the process further includes: If the side length of the current sub-search volume is less than the first side length threshold, then the 3D image is registered and transformed using the lower boundary of the current optimal rotation parameter boundary and the target optimal translation parameter of the current sub-search volume.

6. A blood vessel registration device based on nested search, characterized in that, The steps for implementing the nested search-based vessel registration method as described in any one of claims 1 to 5 include: The centerline extraction module is used to extract the 3D and 2D centerlines of the target blood vessel from the 3D and 2D images, respectively. The rotation projection module is used to construct the initial rotation angle range of the 3D blood vessel centerline, and determine the initial rotation angle range as the current rotation angle range. The current rotation angle range is divided into multiple sub-search bodies. The 3D blood vessel centerline is rotated and 2D projected using the current sub-search bodies to obtain the 2D projected centerline. The preset optimal rotation parameter boundary is determined as the current optimal rotation parameter boundary of the current rotation angle range. The translation parameter determination module is used to divide the current translation search range corresponding to the current sub-search body into multiple sub-search surfaces, translate the 2D projection center line using the translation parameters of each sub-search surface to obtain the translated center line corresponding to each sub-search surface, and search for the target optimal translation parameter of the current sub-search body from the translation parameters of each sub-search surface based on the translated center line and the 2D blood vessel center line. The condition generation module is used to determine the rotation parameter boundary of the current sub-search volume based on the target optimal translation parameters, and to generate theoretically optimal rotation search conditions based on the rotation parameter boundary and the current optimal rotation parameter boundary; The rotation parameter determination module is used to update the current optimal rotation parameter boundary if the current sub-search volume does not meet the theoretical optimal rotation search conditions, and determine the next sub-search volume as the new current sub-search volume, and jump back to the step of rotating and 2D projecting the 3D blood vessel centerline using the current sub-search volume; The re-jump module is used to construct a new current rotation angle range based on each of the sub-search bodies after traversing all the sub-search bodies under the current rotation angle range, and then re-jump to the step of dividing the current rotation angle range into multiple sub-search bodies; The registration transformation module is used to perform registration transformation on the 3D image using the target optimal translation parameters and the rotation parameters of the current sub-search volume if the current sub-search volume satisfies the theoretical optimal rotation search conditions.

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the nested search-based vessel registration method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when executed by a processor, the computer program implements the steps of the nested search-based vessel registration method as described in any one of claims 1 to 5.

Images