Trauma localization and registration method applied to orthopedic robot-assisted surgery

Abstract

This invention relates to the field of skeletal feature analysis technology, specifically to a trauma localization and registration method applied to orthopedic robot-assisted surgery. The method divides the area to be localized by dividing the point cloud data into skeletal region clusters; high-density blocks are selected based on the high-density distribution of the point cloud data; curvature analysis of the local surface is performed based on the high-density blocks to obtain local surface feature degree; combining the distribution density of the point cloud data and the local surface feature degree, the corner point allocation importance of different skeletal region clusters is determined, and the optimal corner point detection quantity is determined and adjusted based on the relationship between the expected corner point allocation importance and the detection quantity, resulting in optimized point cloud data for registration. This invention improves the robustness of corner point feature position matching during registration by analyzing the curvature and density characteristics of skeletal features in different regions, thereby enhancing the accuracy of registration during trauma localization.

Description

Technical Field

[0001] This invention relates to the field of skeletal feature analysis technology, and more specifically to a trauma localization and registration method applied to orthopedic robot-assisted surgery. Background Technology

[0002] In orthopedic robot-assisted surgery, wound localization and registration is a crucial component. It utilizes image recognition to assist surgeons in accurately locating the wound site for precise surgical manipulation. Commonly used algorithms and methods involve image recognition, data registration, and accuracy assurance. Among these, shape-based registration is a classic and effective method. Since bones are typically rigid objects, iterative closest point (ICP) registration algorithms can be employed to accurately match preoperative models with real-time intraoperative image data. The ICP algorithm performs registration by minimizing the distance between point clouds, and this algorithm greatly aids in 3D bone localization during orthopedic surgery.

[0003] The registration algorithm relies heavily on the effectiveness of feature extraction, specifically the corner marking, during the registration process. However, inconsistencies in environmental factors such as pre- and intra-operative shooting angles and lighting conditions often lead to deviations in the number and location of corner points. In other words, different shooting environments result in varying degrees of salience in different locations. Inaccuracies in corner marking during the pre-registration preparation stage can lead to suboptimal registration results due to environmental shooting issues. Summary of the Invention

[0004] To address the technical problems of inaccurate corner point marking during the pre-registration preparation stage in existing technologies, which can lead to suboptimal registration results due to environmental imaging issues, this invention aims to provide a trauma localization and registration method for orthopedic robot-assisted surgery. The specific technical solution adopted is as follows:

[0005] This invention provides a trauma localization and registration method for orthopedic robot-assisted surgery, the method comprising:

[0006] Intraoperative 3D image point cloud data is acquired in the area to be located; clustering is performed based on the number of point cloud data per unit volume block in the 3D image and the distance between unit volume blocks to obtain skeletal region clusters;

[0007] Analyze the high-density distribution of point cloud data in a unit volume block to identify high-density blocks; obtain the local surface feature degree of each high-density block based on the curvature of the space composed of unit volume blocks within a local area.

[0008] Based on the distribution of local surface features of high-density blocks in each skeletal region cluster and the distribution of point cloud data in unit volume blocks, the corner point assignment importance of each skeletal region cluster is determined; based on the correspondence between the expected corner point assignment importance and the number of corner point detections, combined with the corner point assignment importance of skeletal region clusters, the optimized number of corner point detections for each skeletal region cluster is obtained.

[0009] The number of point cloud data in each skeletal region cluster is adjusted based on the number of corner detections to obtain optimized point cloud data for localization and registration.

[0010] Furthermore, the method for obtaining the skeletal region clusters includes:

[0011] The density value of each unit volume block is determined based on the amount of point cloud data in each unit volume block;

[0012] The difference in density values ​​between any two unit volume blocks is used as the density deviation between any two unit volume blocks; the product of the density deviation between any two unit volume blocks and the distance is used as the clustering metric between any two unit volume blocks.

[0013] Clustering of unit volume blocks is performed based on clustering metrics between unit volume blocks to obtain skeletal region clusters.

[0014] Furthermore, the method for obtaining the high-density block includes:

[0015] The number of point cloud data in each unit volume block is normalized to obtain the quantity index of each unit volume block.

[0016] Unit volume blocks with a quantity index greater than the preset high distribution threshold are designated as high-density blocks.

[0017] Furthermore, the method for obtaining the local surface feature degree includes:

[0018] For any high-density block, the surface of all point cloud data within a preset local area of ​​the high-density block is reconstructed to obtain the local surface of the high-density block.

[0019] The curvature value of the local surface is calculated as the local surface characteristic of the high-density block.

[0020] Furthermore, the method for obtaining the importance of corner point assignment includes:

[0021] For any skeletal region cluster, calculate the average number of point cloud data contained in all unit volume blocks in the skeletal region cluster to obtain the skeletal distribution significance value of the skeletal region cluster;

[0022] The mean of the local surface feature degree of all high-density blocks in the skeletal region cluster is normalized to obtain the skeletal structure significance value of the skeletal region cluster.

[0023] The product of the skeletal distribution significance value and the skeletal structure significance value of the skeletal region cluster is used as the corner point assignment importance of the skeletal region cluster.

[0024] Furthermore, the method for obtaining the optimized number of corner detections includes:

[0025] For any skeletal region cluster, the ratio of the expected corner point assignment importance to the corner point assignment importance of that skeletal region cluster is used as the importance percentage of that skeletal region cluster.

[0026] Multiply the importance percentage of the skeletal region cluster by the product of the preset correction coefficient to obtain the corner detection percentage of the skeletal region cluster.

[0027] The ratio of the expected number of corner detections to the proportion of corner detections in the skeletal region cluster is used as the optimized number of corner detections for the skeletal region cluster.

[0028] Furthermore, the method for acquiring the optimized point cloud data includes:

[0029] For any skeletal region cluster, when the number of optimized corner detections is less than the total number of point cloud data in the skeletal region cluster, the unit volume blocks in the skeletal region cluster are arranged in ascending order of the number of point cloud data in the unit volume blocks to obtain a removal sequence; the point cloud data is removed sequentially according to the arrangement order of the unit volume blocks in the removal sequence until the total number of point cloud data in the skeletal region cluster is the same as the number of optimized corner detections, thus obtaining optimized point cloud data;

[0030] When the number of optimized corner detections is greater than the total number of point cloud data in the skeletal region cluster, the unit volume blocks in the skeletal region cluster are arranged in descending order of the number of point cloud data in the unit volume blocks to obtain an addition sequence; the point cloud data is interpolated and added sequentially according to the arrangement order of the unit volume blocks in the addition sequence until the total number of point cloud data in the skeletal region cluster is the same as the number of optimized corner detections, thus obtaining optimized point cloud data;

[0031] When the number of optimized corner detections equals the total number of point cloud data in the skeletal region cluster, the point cloud data in the skeletal region cluster is used as the optimized point cloud data.

[0032] Further, the point cloud data is removed sequentially according to the order of unit volume blocks in the removal sequence until the total number of point cloud data in the skeletal region cluster is the same as the number of optimized corner detections, thus obtaining optimized point cloud data, including:

[0033] The unit volume blocks in the elimination sequence are traversed sequentially, and the mean distance between each point cloud data in the unit volume block and other point cloud data is calculated as the dispersion of each point cloud data.

[0034] When the dispersion of point cloud data exceeds the preset dispersion threshold, the corresponding point cloud data is taken as data to be removed, and the data to be removed is removed in descending order of dispersion.

[0035] When the dispersion of the non-existent point cloud data is greater than the preset dispersion threshold, the point cloud data corresponding to the maximum dispersion is removed; the process continues until the total number of point cloud data in the removal sequence is the same as the number of optimized corner detections, and then the optimized point cloud data is obtained.

[0036] Further, determining the density value of each unit volume block based on the number of point cloud data in each unit volume block includes:

[0037] When point cloud data exists in the unit volume block, the quantity of point cloud data in the unit volume block is used as the density value of the unit volume block; otherwise, the preset density value is used as the density value of the unit volume block; the preset density value is a non-zero positive integer.

[0038] Furthermore, the unit volume block is a 3D image with a volume of 1. A cube of a certain size.

[0039] The present invention has the following beneficial effects:

[0040] This invention performs preliminary clustering based on the density distribution of point cloud data in the region to be located, obtaining the classification results of point cloud data in representing different skeletal regions, i.e., skeletal region clusters. This facilitates subsequent analysis of the different skeletal structure representation capabilities of different skeletal region clusters. High-density blocks are selected through high-density distribution within unit volume blocks for significant analysis of local arrangement curvature. The high-density distribution analysis limits the volume blocks that may be skeletal parts for surface curvature analysis, obtaining local surface feature degree, reflecting the degree to which the point cloud data in the local space can reflect the curved structural features of the skeleton. Finally, combining the distribution density and local surface feature degree in the divided skeletal region clusters, the optimal corner point allocation importance for feature corner point matching analysis is determined in different skeletal region clusters, in order to reduce the influence of corner points with weak representation due to factors such as ambient lighting. By determining the relationship between the expected corner point allocation importance and the detection quantity, the optimal number of corner point detections for each cluster is determined. Based on the optimal detection quantity, the amount of point cloud data is adjusted to improve registration accuracy. This invention analyzes the curvature and density characteristics of skeletal features in different regions to determine the importance of corner point allocation within those regions, thereby improving the robustness of corner point feature position matching during registration and enhancing the accuracy of registration during trauma localization. Attached Figure Description

[0041] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of a trauma localization and registration method for orthopedic robot-assisted surgery, provided in one embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram of local point cloud data of a knee region provided in an embodiment of the present invention. Detailed Implementation

[0044] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a trauma localization and registration method for orthopedic robot-assisted surgery proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0046] The following description, in conjunction with the accompanying drawings, details a specific scheme for a trauma localization and registration method for orthopedic robot-assisted surgery provided by the present invention.

[0047] The main steps of the ICP registration algorithm used to match preoperative models with intraoperative real-time image data include:

[0048] 1) Feature Extraction: Key features in the skeletal structure are extracted by processing preoperative images (such as CT or MRI images). Commonly used feature extraction methods include SIFT (Scale-Invariant Feature Transform) and SURF (Speeded UpRobust Features).

[0049] 2) Preliminary Alignment (Coarse Registration): This step typically involves coarse registration (based on manual markers, rough registration methods, etc.) to roughly align the two images. This is achieved through translation and rotation between the images. For example, in knee joint registration, preoperative and intraoperative images are first roughly aligned using known skeletal landmarks (such as the center of the patella).

[0050] 3) Fine Registration (Precise Alignment): Fine alignment is achieved by calculating the geometric transformation between two sets of feature points. This typically uses a method that minimizes the error function, aiming to minimize the distance between the two sets of feature points. The most commonly used fine registration algorithm is the Iterative Closest Point (ICP) algorithm, the process of which is as follows:

[0051] Closest point matching: The ICP algorithm finds the closest point pair between two sets of data points (e.g., the preoperative skeletal model and the skeletal point cloud in the real-time image during surgery).

[0052] Transformation optimization: Based on the current registration results, translation and rotation matrices are calculated to minimize the distance between point clouds. This transformation is then used to update the registration results.

[0053] Iterative process: This process is repeated until the registration error (the distance between point clouds) is less than a set threshold. Ultimately, the preoperative 3D model and the point cloud in the intraoperative image will be precisely aligned.

[0054] Preoperative 3D model acquisition mainly includes: Image acquisition: CT (computed tomography) is used to acquire images of the knee region. Scanning procedure: The patient remains still, and positioning is performed according to the requirements of the image acquisition equipment. The scanner acquires tomographic image data of the knee region from multiple angles. Point cloud data generation: Using multi-layer tomographic image data (CT or MRI images), reconstruction algorithms, such as voxel reconstruction, are used to convert the sliced ​​tomographic image data into point cloud data in the 3D image, forming a 3D model.

[0055] In orthopedic robot-assisted surgery, registration algorithms are typically used for localization during the registration phase. However, because these algorithms rely on corner point positions obtained from preprocessing, errors can occur when corner point positions deviate under different environments. This invention improves the robustness of corner point position analysis and acquisition by quantifying the features of the skeletal region, thereby enhancing the accuracy of the localization results. Please refer to [link to relevant documentation]. Figure 1 The diagram illustrates a flowchart of a trauma localization and registration method for orthopedic robot-assisted surgery according to an embodiment of the present invention. The method includes the following steps:

[0056] S1: Acquire point cloud data of intraoperative 3D images in the area to be located; perform clustering based on the number of point cloud data per unit volume block in the 3D images and the distance between unit volume blocks to obtain skeletal region clusters.

[0057] In this embodiment of the invention, for bone trauma sites, such as the knee region, a C-arm machine is typically used during surgery to acquire real-time images, obtain point cloud data of three-dimensional images of the knee region, and use filters, such as median filtering, to remove noise in the images, improve image quality, and align the point cloud map of the three-dimensional image during surgery with the preoperative three-dimensional model so that the two can correspond in region.

[0058] After preprocessing, corner point comparison and registration can be performed between images to achieve intraoperative identification and positioning. However, considering the inconsistencies in environmental factors such as shooting angles and lighting conditions before and during surgery, the selection of feature corner points is optimized and adjusted before corner point matching to improve the reliability of the registration results.

[0059] Point cloud data, composed of various skeletal structures such as edges, reflects the arrangement and degree of arrangement of different bones in the knee region. Please refer to [link / reference]. Figure 2 This illustration shows a schematic diagram of local point cloud data of a knee region provided by an embodiment of the present invention. The knee region has relatively obvious skeletal features, mainly composed of the femur, tibia, and patella. Such regions typically possess relatively distinct imaging features, specifically:

[0060] The femoral condyle has a clear outline and a curved surface, especially on the articular surface, where it has a more obvious edge feature; the tibia is relatively straight, and the top of the bone (tibial plateau) has a complex shape, often showing a more obvious curved or step-like outline; the patella is usually hemispherical, located in front of the knee joint, with clear edges and a relatively smooth surface structure.

[0061] Furthermore, due to the different skeletal features in the knee region, the degree of point cloud aggregation is inconsistent. The most significant manifestation is that the density value per unit volume of point cloud data is inconsistent in different skeletal feature regions. Therefore, different features can be divided by the density of skeletal point cloud, which can serve as a preliminary label for the clustering of different skeletal regions.

[0062] For example, the more complex the shape or the clearer the skeletal region, the higher its corresponding point cloud density value, such as the patella and femur. Conversely, the more blurred the shape or the smaller the volume occupied by the bone itself, the lower its corresponding point cloud density value, such as articular cartilage or ligaments.

[0063] First, the point cloud results are clustered based on the density values ​​per unit volume to obtain the region division results. At this point, each region is roughly a skeleton type of division item. When performing corner point optimization later, different importance assignments can be calculated for different corner point types to improve optimization efficiency.

[0064] In this embodiment of the invention, a unit volume block is defined as a 3D image with a volume of 1. For cubic blocks of varying sizes, a higher distribution of point cloud data within a unit volume block indicates higher density. Skeletal region clusters can be obtained based on the density and distance between unit volume blocks.

[0065] Preferably, in this embodiment of the invention, the method for obtaining skeletal region clusters includes:

[0066] The density value of each unit volume block is determined based on the number of point cloud data points within it. In this embodiment of the invention, to ensure the clusterability of all volume blocks, when point cloud data is present in a unit volume block, the number of point cloud data points in the unit volume block is used as the density value of the unit volume block; otherwise, a preset density value is used. It is understood that the preset density value is set to ensure that the unit block is meaningful during clustering; therefore, the preset density value is a non-zero positive integer, set to 1 in this embodiment. The specific value can be adjusted by the implementer according to the specific implementation scenario.

[0067] Furthermore, the density difference between any two unit volume blocks is used as the density deviation between them, reflecting the degree of deviation in density representation. The smaller the density deviation, the more similar the density distribution characteristics. Then, combined with distance, the product of the density deviation and the distance between any two unit volume blocks is used as the clustering metric between them. The clustering metric serves as a dimension for the clustering process.

[0068] Finally, the unit volume blocks are clustered based on the clustering metric between them to obtain skeletal region clusters. In this embodiment of the invention, the K-means clustering algorithm is used. The smaller the clustering metric, that is, the closer the density values ​​and the closer the distance between unit volume blocks, the higher the probability that they will be clustered into one class. This is because they have similar or equal characteristics, and the skeletal region components they reflect are more similar. It should be noted that the clustering process is a technique well known to those skilled in the art, and will not be described or limited here.

[0069] At this point, the initial division of several skeletal regions has been completed, and each skeletal region cluster represents a type of skeletal characteristic.

[0070] S2: Analyze the high-density distribution of point cloud data in a unit volume block to identify high-density blocks; based on the curvature of the space composed of unit volume blocks within a local area, obtain the local surface characteristic of the high-density block.

[0071] The edges of knee bones, such as the femur, tibia, and patella, typically exhibit significant contrast variations, making them prime locations for the SIFT algorithm to detect strong corner points. These corner points often appear at the bends of the bone contours, particularly at the femoral condyle, tibial plateau, and patellar edges. Additionally, the rounded structure of the patella itself, and the contact area between the patella and femur, create a distinct boundary. The SIFT algorithm can extract corner points at these boundaries or abrupt changes in surface curvature, especially at the superior and inferior edges of the patella or the glenoid fossa.

[0072] Therefore, different bone regions will exhibit different detection results due to their own structural characteristics. When the corner detection operator performs detection on different bone regions, the higher the edge curvature value composed of point cloud data in the bone image structure, the higher the probability of the corresponding corner point appearing, and the better it reflects the structural characteristics of the bone, thus providing more important value in subsequent registration.

[0073] Therefore, based on the skeletal structure information that can be represented by a unit volume block, and by analyzing the curvature value through local geometric morphology to characterize local surface features, it can be understood that skeletal structures mostly appear in clearly defined skeletal regions. Thus, curvature analysis is only performed on high-density distributed unit blocks.

[0074] Therefore, high-density volume blocks are first determined by high quantity distribution. In this embodiment of the invention, the quantity of point cloud data in each unit volume block is normalized to obtain a quantity index for each unit volume block, reflecting the degree of density distribution. Unit volume blocks with a quantity index greater than a preset high distribution threshold are considered high-density blocks. The higher the quantity index, the higher the proportion of point cloud data in the unit block. In this embodiment of the invention, the preset high distribution threshold is set to 0.8 to filter high-density distributed unit volume blocks. The specific value can be adjusted by the implementer. It should be noted that normalization is a well-known technique in the art, and linear normalization or standard normalization methods can be used, which are not limited here.

[0075] The surface curvature of the high-density block can be further analyzed to reflect the amount of skeletal structural information. Preferably, in this embodiment of the invention, the method for obtaining the local surface characteristics of the high-density block includes:

[0076] For any high-density block, surface reconstruction is performed on all point cloud data within a preset local area of ​​the high-density block to obtain the local surface of the high-density block. Surface reconstruction extracts a continuous surface mesh, such as a triangular mesh, from discrete point cloud data. In this embodiment of the invention, the preset local area is set as a sphere with a radius of 10 unit volume blocks centered on the high-density block. The specific size of the area can be adjusted by the implementer. The Poisson Surface Reconstruction algorithm is used to convert the discrete point cloud data into a continuous three-dimensional surface mesh as the local surface. Poisson Surface Reconstruction is a reconstruction method based on the Poisson equation. It generates a smooth, continuous three-dimensional surface mesh by modeling the normal direction and the position of the points in the point cloud data.

[0077] Therefore, the curvature value of the local surface can be further calculated as the local surface characteristic of the high-density block. In this embodiment of the invention, the principal curvature method (Curvature Computation for Triangular Meshes) can be used. First, the normal vector of each point is calculated, and then the curvature is estimated based on the change in the normal vectors of adjacent triangles. It should be noted that the application of the surface reconstruction algorithm and the calculation of the surface curvature value are both public technical means well known to those skilled in the art, and will not be described in detail here.

[0078] This completes the further analysis of the skeletal structure information represented by the unit volume block.

[0079] S3: Determine the corner point assignment importance of each skeletal region cluster based on the distribution of local surface features of high-density blocks in each skeletal region cluster and the distribution of point cloud data in unit volume blocks; Based on the correspondence between the expected corner point assignment importance and the number of corner point detections, and combined with the corner point assignment importance of the skeletal region cluster, obtain the optimized number of corner point detections for each skeletal region cluster.

[0080] Within each cluster region, the curvature also reflects the likelihood of corner points appearing, which is the effectiveness of corner point appearance. Effectiveness refers to the structural characteristics that the corner point can reflect, rather than noise corner points caused by environmental issues such as angles and gloss, which affect the final registration effect.

[0081] Therefore, by combining the local surface feature degree reflecting corner characteristics in the skeletal region clusters, and the distribution of point cloud data reflecting the prominence of skeletal regions, a comprehensive importance analysis of corner assignment is performed on each skeletal region cluster. Preferably, in this embodiment of the invention, the method for obtaining the importance of corner assignment includes:

[0082] First, for any skeletal region cluster, calculate the average number of point cloud data contained in all unit volume blocks in the skeletal region cluster to obtain the skeletal distribution significance value of the skeletal region cluster. The denser the distribution of unit point cloud data in the cluster, the higher the probability that the skeletal region cluster is a skeletal distribution.

[0083] Furthermore, the mean of the local surface feature degree of all high-density blocks in the skeletal region cluster is normalized to obtain the skeletal structure significance value of the skeletal region cluster. The higher the feature degree exhibited by the overall high-density blocks, the more significant the skeletal structure features in the skeletal region cluster.

[0084] Finally, the product of the skeletal distribution significance value and the skeletal structure significance value of the skeletal region cluster is used as the corner point assignment importance of that skeletal region cluster. A higher corner point assignment importance means a sufficient number of corner points are needed in the region cluster to avoid corner point registration problems caused by insufficient angle or lighting. Conversely, a lower corner point assignment importance corresponds to a lower skeletal region density and fewer point clouds affecting the skeletal structure. Theoretically, this results in fewer corner points in the region cluster, avoiding false corners caused by factors such as concentrated lighting.

[0085] As an example, the expression for assigning importance to corner points is: In the formula, Represented as the first The importance of corner points in each skeletal region cluster. Represented as the first The total number of unit volume blocks in each skeletal region cluster. Represented as the first The number of point cloud data in a unit volume block. Represented as the first The total number of high-density blocks in each skeletal region cluster. Represented as the first Local surface characteristic degree of a high-density block, This is represented as a normalization function.

[0086] This completes the analysis of the importance of corner point allocation in different skeletal region clusters. Furthermore, by analyzing the ideal expected number of corner points and the distribution relationship of corner point importance, we can obtain the number of corner points required for each skeletal region cluster under the expected relationship, that is, optimize the number of corner point detections.

[0087] In this embodiment of the invention, the expected relationship between the importance of corner point assignment and the number of intersection point detections is as follows: In the formula, represents the number of corner points detected in the expected skeletal region, represents the importance of corner point allocation in the expected skeletal region, and represents the correction coefficient. The specific value is adjusted by the implementer according to the specific implementation scenario and is not limited here.

[0088] By combining the importance of corner point distribution in skeletal region clusters, the ideal number of corner points corresponding to each skeletal region cluster can be obtained. This is then analyzed by substituting the proportions into a relational expression. Preferably, in this embodiment of the invention, the method for optimizing the acquisition of the number of detected corner points includes:

[0089] For any skeletal region cluster, the ratio of the expected corner point importance to the actual corner point importance of that cluster is used as the importance percentage of that cluster. Ideal corner points are then obtained through a proportional relationship. The corner point detection percentage of the cluster is obtained by multiplying this percentage by a preset correction coefficient. The ratio of the expected number of corner point detections to this percentage is used as the optimized number of corner point detections for that cluster. The preferred number of corner point detections for the cluster can be directly obtained by substituting this relationship into the formula. As an example, the expression for optimizing the number of corner points is: In the formula, Represented as the first The importance of corner points in each skeletal region cluster. Represented as the first Optimize the number of corner detections for each skeletal region cluster.

[0090] Thus, by analyzing the importance of the required corner points in different regions after division, the analysis of the ideal corner point distribution for each skeletal region cluster is completed.

[0091] S4: Adjust the number of point cloud data in each skeletal region cluster based on the optimized number of corner detections to obtain optimized point cloud data for localization and registration.

[0092] The distribution of the current point cloud data can be adjusted by optimizing the number of corner detections. When there are more point cloud data points in a region than the optimal number of corner detections, the best corner points can be obtained by removing them. Conversely, when there are fewer point cloud data points than the optimal number of corner detections, the ideal number of detections can be increased by adding corner points to obtain the optimal point cloud data.

[0093] Therefore, preferably, in this embodiment of the invention, the method for optimizing point cloud data acquisition includes:

[0094] For any skeletal region cluster, if the number of optimized corner detections is less than the total number of point cloud data in the cluster, it indicates that there is a large amount of point cloud data in the cluster, which needs to be removed. The unit volume blocks in the skeletal region cluster are arranged in ascending order of the number of point cloud data in the unit volume block to obtain the removal sequence. Removal is carried out according to the density distribution in the unit volume block. The lower the density in the unit volume block, the weaker the information representation ability, and the greater the possibility of removal.

[0095] Therefore, point cloud data is removed sequentially according to the arrangement order of unit volume blocks in the removal sequence until the total number of point cloud data in the skeletal region cluster is the same as the number of optimized corner detections, thus obtaining optimized point cloud data. In this embodiment of the invention, the unit volume blocks in the removal sequence are traversed sequentially, and the average distance between each point cloud data in the unit volume block and other point cloud data is calculated as the dispersion of each point cloud data, reflecting the redundancy of the point cloud data. When the dispersion of point cloud data is greater than a preset dispersion threshold, it indicates that the dispersion of the point cloud data is relatively large, and the degree of deletion is better. The corresponding point cloud data is taken as the data to be removed, and the data to be removed is removed in descending order of dispersion. During this process, if the total number of point cloud data in the removal sequence is the same as the number of optimized corner detections, the process stops; otherwise, the unit volume blocks are traversed again after removal.

[0096] When no point cloud data has a dispersion greater than a preset dispersion threshold, it indicates that the point cloud data distribution is relatively optimal. In this case, only the point cloud data corresponding to the maximum dispersion is removed, and the process is repeated sequentially. The removal sequence is iteratively traversed until the total number of point cloud data in the removal sequence is the same as the number of optimized corner detections. The remaining point cloud data is the optimized point cloud data. In this embodiment of the invention, the preset dispersion threshold is set to 0.8. The specific value can be adjusted by the implementer according to the specific implementation situation, and is not limited here.

[0097] When the number of optimized corner detections is greater than the total number of point cloud data in the skeletal region cluster, it indicates that the overall point cloud data in the cluster is insufficient and needs to be added. The unit volume blocks in the skeletal region cluster are arranged in descending order of the number of point cloud data in the unit volume block to obtain the addition sequence. For unit volume blocks with more distributed point cloud data, the higher the probability of priority addition.

[0098] Point cloud data is added sequentially by interpolation according to the order of unit volume blocks in the addition sequence, until the total number of point cloud data in the skeletal region cluster is the same as the number of optimized corner detections, thus obtaining optimized point cloud data. In this embodiment of the invention, one point cloud data is added sequentially to each unit volume block using interpolation, following the order of unit volume blocks in the addition sequence. The addition sequence can be iteratively added, stopping only when the total number of point cloud data in the addition sequence is the same as the number of optimized corner detections. The remaining point cloud data is the optimized point cloud data. It should be noted that the interpolation method is a well-known technique to those skilled in the art and will not be described in detail here.

[0099] When the number of optimized corner detections equals the total number of point cloud data in the skeletal region cluster, it indicates that the data distribution in the skeletal region cluster is relatively ideal, and the point cloud data in the skeletal region cluster is used as the optimized point cloud data.

[0100] Furthermore, the optimized point cloud data can be used to register the intraoperative data of the area to be located with the preoperative model's ICP point cloud, reducing the registration group error caused by environmental factors and improving the registration effect.

[0101] This invention performs preliminary clustering based on the density distribution of point cloud data in the region to be located, obtaining the classification results of point cloud data in representing different skeletal regions, i.e., skeletal region clusters. This facilitates subsequent analysis of the different skeletal structure representation capabilities of different skeletal region clusters. High-density blocks are selected through high-density distribution within unit volume blocks for significant analysis of local arrangement curvature. The high-density distribution analysis limits the volume blocks that may be skeletal parts for surface curvature analysis, obtaining local surface feature degree, reflecting the degree to which the point cloud data in the local space can reflect the curved structural features of the skeleton. Finally, combining the distribution density and local surface feature degree in the divided skeletal region clusters, the optimal corner point allocation importance for feature corner point matching analysis is determined in different skeletal region clusters, in order to reduce the influence of corner points with weak representation due to factors such as ambient lighting. By determining the relationship between the expected corner point allocation importance and the detection quantity, the optimal number of corner point detections for each cluster is determined. Based on the optimal detection quantity, the amount of point cloud data is adjusted to improve registration accuracy. This invention analyzes the curvature and density characteristics of skeletal features in different regions to determine the importance of corner point allocation within those regions, thereby improving the robustness of corner point feature position matching during registration and enhancing the accuracy of registration during trauma localization.

[0102] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0103] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for trauma localization and registration applied to orthopedic robot-assisted surgery, characterized in that, The methods include: Acquire point cloud data of intraoperative 3D images of the area to be located; Clustering is performed based on the number of point cloud data distributions per unit volume block in the 3D image and the distance between unit volume blocks to obtain skeletal region clusters; The method for obtaining skeletal region clusters includes: determining the density value of each unit volume block based on the number of point cloud data in each unit volume block; using the difference in density values ​​between two unit volume blocks as the density deviation between two unit volume blocks; using the product of the density deviation between two unit volume blocks and the distance as the clustering metric between two unit volume blocks; and clustering the unit volume blocks based on the clustering metric between unit volume blocks to obtain skeletal region clusters. Analyze the high-density distribution of point cloud data in a unit volume block to identify high-density blocks; obtain the local surface feature degree of each high-density block based on the curvature of the space composed of unit volume blocks within a local area. Based on the distribution of local surface features of high-density blocks in each skeletal region cluster and the distribution of point cloud data in unit volume blocks, the corner point assignment importance of each skeletal region cluster is determined; based on the correspondence between the expected corner point assignment importance and the number of corner point detections, combined with the corner point assignment importance of skeletal region clusters, the optimized number of corner point detections for each skeletal region cluster is obtained. The number of point cloud data in each skeletal region cluster is adjusted based on the number of corner detections to obtain optimized point cloud data for localization and registration. The method for obtaining the importance of corner assignment includes: for any skeletal region cluster, calculating the average number of point cloud data contained in all unit volume blocks in the skeletal region cluster to obtain the skeletal distribution significance value of the skeletal region cluster; normalizing the average local surface feature value of all high-density blocks in the skeletal region cluster to obtain the skeletal structure significance value of the skeletal region cluster; and using the product of the skeletal distribution significance value and the skeletal structure significance value of the skeletal region cluster as the corner assignment importance of the skeletal region cluster. The method for acquiring optimized point cloud data includes: for any skeletal region cluster, when the number of optimized corner detections is less than the total number of point cloud data in the skeletal region cluster, the unit volume blocks in the skeletal region cluster are arranged in ascending order of the number of point cloud data in the unit volume blocks to obtain a removal sequence; point cloud data is removed sequentially according to the arrangement order of the unit volume blocks in the removal sequence until the total number of point cloud data in the skeletal region cluster is the same as the number of optimized corner detections, thus obtaining optimized point cloud data; when the number of optimized corner detections is greater than the number of optimized corner detections in the skeletal region cluster, the method for acquiring optimized point cloud data is: for any skeletal region cluster, when the number of optimized corner detections is less than the total number of point cloud data in the skeletal region cluster, the unit volume blocks in the skeletal region cluster are arranged in ascending order of the number of point cloud data in the unit volume blocks to obtain a removal sequence; when the number of optimized corner detections is greater than the total number of point cloud data in the skeletal region cluster, the method for acquiring optimized point cloud data is: for any ... When determining the total number of point cloud data in a domain cluster, the unit volume blocks in the skeletal region cluster are arranged in descending order of the number of point cloud data in each unit volume block to obtain an addition sequence. Point cloud data is then added sequentially according to the order of the unit volume blocks in the addition sequence until the total number of point cloud data in the skeletal region cluster is the same as the number of optimized corner detections, thus obtaining optimized point cloud data. When the number of optimized corner detections equals the total number of point cloud data in the skeletal region cluster, the point cloud data in the skeletal region cluster is used as optimized point cloud data.

2. The trauma localization and registration method for orthopedic robot-assisted surgery according to claim 1, characterized in that, Methods for obtaining high-density blocks include: The number of point cloud data in each unit volume block is normalized to obtain the quantity index of each unit volume block. Unit volume blocks with a quantity index greater than the preset high distribution threshold are designated as high-density blocks.

3. The trauma localization and registration method for orthopedic robot-assisted surgery according to claim 1, characterized in that, Methods for obtaining local surface feature degree include: For any high-density block, the surface of all point cloud data within a preset local area of ​​the high-density block is reconstructed to obtain the local surface of the high-density block. The curvature value of the local surface is calculated as the local surface characteristic of the high-density block.

4. The trauma localization and registration method for orthopedic robot-assisted surgery according to claim 1, characterized in that, Methods for optimizing the acquisition of corner detection count include: For any skeletal region cluster, the ratio of the expected corner point assignment importance to the corner point assignment importance of that skeletal region cluster is used as the importance percentage of that skeletal region cluster. Multiply the importance percentage of the skeletal region cluster by the product of the preset correction coefficient to obtain the corner detection percentage of the skeletal region cluster. The ratio of the expected number of corner detections to the proportion of corner detections in the skeletal region cluster is used as the optimized number of corner detections for the skeletal region cluster.

5. The trauma localization and registration method for orthopedic robot-assisted surgery according to claim 1, characterized in that, Point cloud data is removed sequentially according to the order of unit volume blocks in the removal sequence until the total number of point cloud data in the cluster of the skeletal region is the same as the number of optimized corner detections, resulting in optimized point cloud data, including: The unit volume blocks in the elimination sequence are traversed sequentially, and the mean distance between each point cloud data in the unit volume block and other point cloud data is calculated as the dispersion of each point cloud data. When the dispersion of point cloud data exceeds the preset dispersion threshold, the corresponding point cloud data is taken as data to be removed, and the data to be removed is removed in descending order of dispersion. When the dispersion of the non-existent point cloud data is greater than the preset dispersion threshold, the point cloud data corresponding to the maximum dispersion is removed; the process continues until the total number of point cloud data in the removal sequence is the same as the number of optimized corner detections, and then the optimized point cloud data is obtained.

6. The trauma localization and registration method for orthopedic robot-assisted surgery according to claim 1, characterized in that, The density value of each unit volume block is determined based on the amount of point cloud data in each unit volume block, including: When point cloud data exists in the unit volume block, the quantity of point cloud data in the unit volume block is used as the density value of the unit volume block; otherwise, the preset density value is used as the density value of the unit volume block; the preset density value is a non-zero positive integer.

7. The trauma localization and registration method for orthopedic robot-assisted surgery according to claim 1, characterized in that, A unit volume block is a 3D image with a volume of 1. A cube of a certain size.

Images