Method, device and equipment for registration of point cloud data for orthopedic surgery

Abstract

The disclosure provides a registration method, device and equipment for orthopedic surgery point cloud data, relating to the technical field of computer-aided orthopedic surgery and three-dimensional point cloud processing. The method comprises: obtaining a first point cloud set and a second point cloud set, the first point cloud set being obtained by modeling based on image data of a target bone region before surgery, and the second point cloud set being obtained by collecting point cloud data of the target bone region during surgery; for any point cloud set in the first point cloud set and the second point cloud set, determining a weight parameter of each point cloud in the point cloud set based on the curvature of each point cloud in the point cloud set; performing principal component feature extraction on the point cloud set based on the weight parameter of each point cloud in the point cloud set to determine a weighted principal component feature; and performing point cloud set registration on the first point cloud set and the second point cloud set based on the weighted principal component features of the first point cloud set and the second point cloud set respectively to obtain a registration result.

Description

Technical Field

[0001] This disclosure relates to the field of computer-aided orthopedic surgery and three-dimensional point cloud processing technology, specifically to a registration method, apparatus, and device for orthopedic surgical point cloud data, which can be applied to orthopedic surgical navigation systems. Background Technology

[0002] High tibial osteotomy (HTO) is a core knee-preserving surgery for the treatment of early-stage knee osteoarthritis. Registering the structured light point cloud of the exposed bone surface during surgery with the point cloud of the complete tibial model from preoperative CT is the key step in achieving precise transformation from preoperative planning to intraoperative operation.

[0003] In related technologies, principal component analysis can be used to extract principal component features from the point cloud to be registered for registration. However, traditional principal component analysis assigns equal weights to the point cloud, which is easily affected by flat areas and edge noise, causing the extracted principal component features to deviate from the long axis of the actual skeleton and reducing the accuracy of registration. Summary of the Invention

[0004] In view of this, the present disclosure provides a registration method, apparatus and device for orthopedic surgical point cloud data.

[0005] One aspect of this disclosure provides a registration method for point cloud data in orthopedic surgery. The method includes: acquiring a first point cloud set and a second point cloud set, wherein the first point cloud set is obtained by modeling based on image data of the target bone region before surgery, and the second point cloud set is obtained by collecting point cloud data of the target bone region during surgery; for any point cloud set in the first and second point cloud sets, determining the weight parameters of each point cloud in the point cloud set based on the curvature of each point cloud in the point cloud set; extracting principal component features from the point cloud set based on the weight parameters of each point cloud in the point cloud set to determine the weighted principal component features; and registering the first and second point cloud sets based on the respective weighted principal component features of the first and second point cloud sets to obtain a registration result.

[0006] According to embodiments of this disclosure, based on the weight parameters of each point cloud in the point cloud set, principal component features are extracted from the point cloud set to determine the weighted principal component features, including: based on the weight parameters of each point cloud in the point cloud set, performing weighted averaging on multiple point clouds in the point cloud set to obtain the weighted centroid of the point cloud set; and determining the principal component features of the point cloud set based on the weighted centroid.

[0007] According to embodiments of this disclosure, determining the principal component features of a point cloud set based on weighted centroids includes: constructing a weighted covariance matrix of the point cloud set based on the weighted centroids and the coordinates of each point cloud in the point cloud set; solving for the eigenvalues ​​of the weighted covariance matrix to obtain multiple weighted eigenvalues; determining a target weighted principal axis based on a target weighted eigenvalue among the multiple weighted eigenvalues, wherein the target weighted eigenvalue is greater than or equal to any other weighted eigenvalue among the multiple weighted eigenvalues; and obtaining the principal component features of the point cloud set based on the weighted centroids and the target weighted principal axis.

[0008] According to embodiments of this disclosure, determining the weight parameters of each point cloud in a point cloud set based on the curvature of each point cloud in the point cloud set includes: determining the geometric centroid of the point cloud set; determining the weight parameters of each point cloud in the point cloud set based on the curvature of each point cloud in the point cloud set and the relative distance between each point cloud in the point cloud set and the geometric centroid; wherein the weight parameters increase with increasing curvature and decrease with increasing relative distance.

[0009] According to embodiments of this disclosure, the curvature of each point cloud in a point cloud set is determined as follows: for any point cloud in the point cloud set, a local point cloud set and a local geometric centroid of the local point cloud set are determined based on the point cloud and multiple neighboring point clouds whose distances to the point cloud satisfy a preset distance; a local covariance matrix is ​​constructed based on each point cloud in the local point cloud set and the local geometric centroid; eigenvalues ​​are solved for the local covariance matrix to obtain multiple eigenvalues; the curvature of the point cloud is determined based on the ratio of a target eigenvalue among the multiple eigenvalues ​​to the sum of the multiple eigenvalues, wherein the target eigenvalue is less than or equal to any other eigenvalue among the multiple eigenvalues.

[0010] According to embodiments of this disclosure, the weighted principal component features include weighted centroids and target weighted principal axes; based on the respective weighted principal component features of the first point cloud set and the second point cloud set, the point cloud sets are registered to obtain a registration result, including: determining a translation vector to align the two weighted centroids based on the respective weighted centroids of the first and second point cloud sets; determining a rotation matrix to align the two target weighted principal axes based on the respective target weighted principal axes of the first and second point cloud sets; performing transformation processing on each point cloud in the second point cloud set based on the translation vector and the rotation matrix to obtain an initially registered second point cloud set; and fine-tuning the initially registered second point cloud set to obtain a target second point cloud set registered with the first point cloud set.

[0011] According to embodiments of this disclosure, fine-tuning the initially registered second point cloud set to obtain a target second point cloud set registered with the first point cloud set includes: performing multiple transformation processes on each point cloud in the initially registered second point cloud set based on preset transformation step sizes of multiple transformation dimensions to obtain multiple updated second point cloud sets; wherein, the multiple transformation dimensions include a spin dimension rotating about the target weighted principal axis, a tilt dimension rotating about an axis perpendicular to the target weighted principal axis, and a sliding dimension sliding along the direction of the target weighted principal axis; and determining the target second point cloud set from the multiple updated second point cloud sets based on the distance between each updated point cloud in each updated second point cloud set and the matching point cloud in the first point cloud set corresponding to each updated point cloud.

[0012] According to embodiments of this disclosure, determining a target second point cloud set from multiple updated second point cloud sets based on the distances between each updated point cloud in each updated second point cloud set and the matching point cloud corresponding to each updated point cloud in the first point cloud set includes: determining a distance set based on the distances between each updated point cloud in each updated second point cloud set and the matching point cloud corresponding to each updated point cloud in the first point cloud set; sorting the distances in the distance set in ascending order to determine the target distance in the distance set that is at the target sorting position; and determining the target second point cloud set from multiple updated second point cloud sets based on the target distance.

[0013] Another aspect of this disclosure provides a registration device for point cloud data in orthopedic surgery, comprising: an acquisition module for acquiring a first point cloud set and a second point cloud set, wherein the first point cloud set is obtained by modeling based on image data of a target bone region before surgery, and the second point cloud set is obtained by collecting point cloud data of the target bone region during surgery; a determination module for determining weight parameters of each point cloud in the first and second point cloud sets based on the curvature of each point cloud in the point cloud set; a feature extraction module for extracting principal component features from the point cloud sets based on the weight parameters of each point cloud in the point cloud sets to determine weighted principal component features; and a registration module for registering the first and second point cloud sets based on the respective weighted principal component features of the first and second point cloud sets to obtain a registration result.

[0014] Another aspect of this disclosure provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the methods described above.

[0015] According to embodiments of this disclosure, by assigning differentiated weight parameters to each point cloud based on its curvature, and then performing weighted principal component feature extraction, the contribution of point clouds in flat areas and point clouds in areas with prominent skeletal features to feature calculation can be weakened. This filters out interference from noise and redundant points, improves the distinguishability and matching reliability of preoperative modeling point clouds and intraoperative real-time acquisition point cloud features, and completes registration based on accurate weighted principal component features of the two sets of point clouds. This can effectively reduce registration offset problems caused by smooth areas of the bone surface and acquisition noise, and improve the accuracy and robustness of bone point cloud registration in orthopedic surgical scenarios. Attached Figure Description

[0016] The above and other objects, features and advantages of this disclosure will become clearer from the following description of embodiments of this disclosure with reference to the accompanying drawings, in which:

[0017] Figure 1 A flowchart is shown below illustrating a registration method for orthopedic surgical point cloud data according to an embodiment of this disclosure;

[0018] Figure 2 A flowchart is shown for a registration method for orthopedic surgical point cloud data according to another embodiment of the present disclosure;

[0019] Figure 3 A schematic diagram illustrating the principle of registration according to an embodiment of the present disclosure is shown;

[0020] Figure 4 A block diagram of a registration device for orthopedic surgical point cloud data is shown. Detailed Implementation

[0021] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0022] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0023] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0024] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0025] Figure 1 A flowchart of a registration method for orthopedic surgical point cloud data according to an embodiment of the present disclosure is shown.

[0026] like Figure 1 As shown, the method includes operations S110 to S140.

[0027] In operation S110, a first point cloud set and a second point cloud set are acquired. The first point cloud set is obtained by modeling based on image data of the target bone region before surgery, and the second point cloud set is obtained by collecting point cloud data of the target bone region during surgery.

[0028] In operation S120, for any point cloud set in the first point cloud set and the second point cloud set, the weight parameters of each point cloud in the point cloud set are determined based on the curvature of each point cloud in the point cloud set.

[0029] In operation S130, based on the weight parameters of each point cloud in the point cloud set, principal component features are extracted from the point cloud set to determine the weighted principal component features.

[0030] In operation S140, based on the weighted principal component features of the first point cloud set and the second point cloud set respectively, the point cloud sets are registered to obtain the registration result.

[0031] In embodiments of this disclosure, the target bone region may be the surgical area for a high tibial osteotomy, such as the proximal tibial region. This disclosure is not limited thereto; the target bone region may also be other bone regions.

[0032] The patient's bones can be scanned using medical imaging equipment such as CT / MRI, and the tomographic two-dimensional image can be reconstructed into a three-dimensional model. Then, the surface of the three-dimensional model is uniformly or discretely sampled to obtain multiple spatial points represented by three-dimensional coordinates (X, Y, Z), which is the first point cloud set.

[0033] For example, a CT scan of the target bone region can be performed before surgery to obtain CT image data. Based on the grayscale threshold characteristics of the bone, a threshold segmentation algorithm is used to remove non-bone tissues such as muscles, skin, and blood vessels, eliminating background noise and interference from irrelevant tissues. A three-dimensional surface mesh model of the target bone region can be reconstructed using the Marching Cubes (MC) algorithm. The three-dimensional surface mesh model is discretely sampled based on the local curvature of the mesh. For example, high-curvature key anatomical regions such as bony ridges, joint edges, and bony protrusions are sampled more densely, while flat areas are sampled uniformly and moderately to obtain an initial first point cloud set. A single-view point cloud visibility segmentation method based on spherical projection can be used to occlude and remove occlusions from the initial first point cloud set, thus retaining only the visible point clouds of the bony surface that can be exposed and effectively acquired during the surgical field, reducing the computational cost of subsequent point cloud registration.

[0034] During surgery, devices such as optical scanners and depth cameras can be used to directly expose and scan the patient's intraoperative skeletal area to obtain a second point cloud set.

[0035] For example, during surgery, after incising the skin and soft tissue to expose the proximal bone surface of the target bone, a binocular structured light scanning device projects complementary Gray and phase-shift structured light patterns onto the exposed bone surface, acquiring distorted fringe images modulated by the bone surface, and then reconstructing them to obtain an initial second point cloud set for the target bone region during surgery. The initial second point cloud set can also be preprocessed to remove irrelevant point clouds such as soft tissue, surgical instruments, and background from the original point cloud, retaining only the effective point cloud of the exposed bone surface. Outlier noise points can be removed using statistical filtering (SOR), and the point cloud data volume can be compressed through curvature adaptive downsampling, while fully preserving the high-curvature point cloud data such as the target bone edges and intercondylar spines, to obtain the second point cloud set.

[0036] Binocular structured light scanning equipment can complete the acquisition of three-dimensional information of bone surface in a short time, which significantly shortens the total time of HTO surgery, reduces the exposure time of soft tissue during surgery, and reduces the risk of infection; at the same time, it eliminates the dependence of registration accuracy on doctors' experience and improves the homogeneity of surgery.

[0037] The curvature of a point cloud is a geometric parameter used to characterize the degree of local surface curvature at a spatial point on the bone surface. Flat areas of the bone have small curvature values, while convex / concave structures have large curvature values. The greater the curvature, the higher the weighting parameter; the smaller the curvature, the lower the weighting parameter.

[0038] Principal component analysis (PCA) can extract three sets of orthogonal principal axes representing the spatial distribution trend of a point cloud set, thus describing the overall position, orientation, and size of the point cloud set. Weighted PCA extracts these three sets of orthogonal principal axes by weighting each point cloud with weight parameters. Point clouds with high curvature contribute more to the principal axis directions, while flat, low-curvature point clouds contribute less.

[0039] The weighted principal component features can include the weighted centroid representing the coordinates of the center of the weighted point cloud set and three sets of orthogonal weighted principal axes representing the overall three-dimensional orientation of the skeletal point cloud.

[0040] Based on the principal component features of the first and second point cloud sets, a set of spatial transformation matrices can be solved. After performing a spatial transformation on the second point cloud set using the spatial transformation matrices, the point clouds in the first and second point cloud sets can be aligned and fitted, thus achieving registration.

[0041] According to embodiments of this disclosure, by assigning differentiated weight parameters to each point cloud based on the curvature of the point cloud, and then performing weighted principal component feature extraction, the contribution of point clouds in flat areas and point clouds in areas with prominent skeletal features to feature calculation can be weakened, interference from noise and redundant points can be filtered, and the distinguishability and matching reliability of point cloud features between preoperative modeling and intraoperative real-time acquisition can be improved.

[0042] According to embodiments of this disclosure, the curvature of each point cloud in a point cloud set is determined as follows: for any point cloud in the point cloud set, a local point cloud set and a local geometric centroid of the local point cloud set are determined based on the point cloud and multiple neighboring point clouds whose distances to the point cloud satisfy a preset distance; a local covariance matrix is ​​constructed based on each point cloud in the local point cloud set and the local geometric centroid; eigenvalues ​​are solved for the local covariance matrix to obtain multiple eigenvalues; the curvature of the point cloud is determined based on the ratio of a target eigenvalue among the multiple eigenvalues ​​to the sum of the multiple eigenvalues, wherein the target eigenvalue is less than or equal to any other eigenvalue among the multiple eigenvalues.

[0043] It can calculate the Euclidean distance between the target point cloud and other point clouds in the same point cloud set, filter out k neighboring point clouds whose distance values ​​are less than or equal to a preset distance threshold, and combine the target point cloud and the selected k neighboring point clouds into a local point cloud set.

[0044] The three-dimensional (X, Y, Z) coordinates of each point cloud in the local point cloud set can be extracted. The sum of the X coordinates, Y coordinates, and Z coordinates of all point clouds can be accumulated and then divided by the total number of point clouds contained in the local point cloud set to obtain the local geometric centroid composed of the three-dimensional coordinates.

[0045] The offset vector of each point cloud relative to the centroid can be obtained based on the difference between the 3D coordinates of each point cloud in the local point cloud set and the 3D coordinates of the local geometric centroid. A 3x3 local covariance matrix is ​​constructed using all offset vectors. The local covariance matrix can completely characterize the discrete point distribution features of the local skeletal surface in 3D space.

[0046] Eigenvalue decomposition can be performed on the constructed 3rd-order local covariance matrix to obtain three eigenvalues. The curvature of the target point cloud is obtained based on the ratio of the minimum of the three eigenvalues ​​to the sum of the three eigenvalues, thus completing the solution for the point cloud curvature. By iterating through the point cloud set, the curvature of the entire point cloud can be obtained. For any single point cloud... Its curvature The calculation formula is:

[0047] ;

[0048] In the formula, , where are the three eigenvalues ​​of the local covariance matrix; Quantified a bit The degree of surface curvature at a point; the larger the value, the more significant the geometric features at that point.

[0049] By quantifying curvature using the ratio of eigenvalues ​​of the covariance matrix, relying solely on the original 3D coordinate data, we can accurately identify the geometrically significant parts of the target skeletal region, providing a reliable quantitative basis for subsequent point cloud weight allocation and weighted principal component feature extraction based on curvature.

[0050] According to embodiments of this disclosure, determining the weight parameters of each point cloud in a point cloud set based on the curvature of each point cloud in the point cloud set may include: determining the geometric centroid of the point cloud set; determining the weight parameters of each point cloud in the point cloud set based on the curvature of each point cloud in the point cloud set and the relative distance between each point cloud in the point cloud set and the geometric centroid; wherein the weight parameters increase with increasing curvature and decrease with increasing relative distance.

[0051] The relative distance between the point cloud and the geometric centroid can be calculated using Euclidean distance, based on the 3D coordinates of the point cloud and the 3D coordinates of the geometric centroid. The relative distance reflects the degree of spatial deviation of the point cloud; the larger the distance, the more likely the point is on the edge of the point cloud set, and the smaller the distance, the closer the point is to the geometric centroid, thus suppressing the interference of outliers on the principal axis features.

[0052] For example, weight parameters The calculation formula is as follows:

[0053] ;

[0054] in, For the curvature of the point cloud, The geometric center of the point cloud, For the i-th point cloud, To control the scale parameter of distance decay, exp() represents the natural exponential function. This represents the Euclidean distance between the point cloud and the geometric centroid. When Near the geometric centroid, the exponential term approaches 1, barely weakening the weighting parameter of the fundamental curvature. As the distance from the geometric centroid increases, the exponential term approaches 0, resulting in a decay suppression of the weighting parameters.

[0055] According to embodiments of this disclosure, weighting parameters are determined by simultaneously utilizing the curvature of the point cloud and the relative distance between the point cloud and the geometric centroid. Since curvature can distinguish anatomical feature points from flat redundant points, and relative distance can filter local noise and outliers, the weighted principal component features can focus more on the effective anatomical structure of the skeleton, reduce the influence of noise on the preoperative and intraoperative point cloud registration offset, and improve registration accuracy and algorithm robustness.

[0056] According to embodiments of this disclosure, principal component feature extraction of the point cloud set based on the weight parameters of each point cloud in the point cloud set, and determination of weighted principal component features may include: performing weighted averaging on multiple point clouds in the point cloud set based on the weight parameters of each point cloud in the point cloud set to obtain the weighted centroid of the point cloud set; and determining the principal component features of the point cloud set based on the weighted centroid.

[0057] The weighted average process uses the weight parameters corresponding to each point cloud as weighting coefficients, performs a weighted summation of the 3D coordinates of all point clouds, and then divides the sum by the mean of all weights.

[0058] For example, suppose the point cloud set contains N point clouds, and the i-th point cloud is... The weight parameters are Then the weighted centroid The calculation formula is as follows:

[0059] ;

[0060] Traditional equal-weighted geometric centroids are affected by outliers and random noise points acquired during surgery, causing the aggregate reference center to deviate from the actual main structure of the skeleton. The embodiments disclosed in this paper reduce the contribution of noise points by weighting, and the weighted centroid is dominated only by high curvature anatomical feature points, which fits the actual shape center of the skeleton and eliminates the reference offset error caused by noise.

[0061] According to embodiments of this disclosure, determining the principal component features of a point cloud set based on the weighted centroid may include: constructing a weighted covariance matrix of the point cloud set based on the weighted centroid and the coordinates of each point cloud in the point cloud set; solving for the eigenvalues ​​of the weighted covariance matrix to obtain multiple weighted eigenvalues; determining the target weighted principal axis based on the maximum value among the multiple weighted eigenvalues; and obtaining the principal component features of the point cloud set based on the weighted centroid and the target weighted principal axis.

[0062] For point clouds Calculate point clouds offset vector relative to the weighted centroid .

[0063] Weight of each point Construct a 3×3 weighted covariance matrix using the coefficients. The calculation formula is as follows;

[0064] ;

[0065] Through the Singular Value Decomposition (SVD) or eigenvalue decomposition yields three sorted eigenvalues. Its corresponding feature vector eigenvectors The three weighted principal axes, representing the global distribution of a point cloud set, are the principal component directions, where the largest eigenvalue is the largest eigenvalue. corresponding vector This is the target weighted principal axis, corresponding to the anatomical long axis of the target skeleton.

[0066] Using the methods described above, the first point cloud set can be obtained respectively. weighted center of mass Target-weighted principal axis and the second point cloud collection weighted center of mass Target-weighted principal axis .

[0067] By constructing a weighted covariance matrix based on weighted centroids, high-weight skeletal feature points can dominate the matrix calculation. The final output principal component features can highlight the distribution patterns of highly recognizable anatomical regions such as joints and bone spines, weaken the interference caused by flat bone surfaces and invalid noise points, and significantly improve the distinguishability and matching similarity between the preoperative modeling point cloud and the intraoperative collected point cloud features.

[0068] Figure 2 A flowchart of a registration method for orthopedic surgical point cloud data according to another embodiment of the present disclosure is shown.

[0069] Registration methods for orthopedic surgical point cloud data can include, for example, Figure 1 In addition to operations S110 to S140 shown, it may also include, for example: Figure 2 Operations S210 to S240 are shown.

[0070] In operation S210, based on the weighted centroids of the first and second point cloud sets, a translation vector is determined to align the two weighted centroids.

[0071] In operation S220, based on the target weighted principal axes of the first point cloud set and the second point cloud set, a rotation matrix is ​​determined to rotate and align the two target weighted principal axes.

[0072] In operation S230, based on the translation vector and rotation matrix, the point clouds in the second point cloud set are transformed to obtain the initially registered second point cloud set.

[0073] In operation S240, the second point cloud set initially registered is fine-tuned to obtain the target second point cloud set registered with the first point cloud set.

[0074] For example, the weighted centroid of the second point cloud set can be... Translate to the weighted centroid of the set of the first point cloud. Coincident, translation vector for:

[0075] ;

[0076] The target weighted principal axis for the second point cloud set The existence of 180° directional ambiguity can be addressed by setting symbolic variables. Traverse the target weighted principal axis The two states, positive and negative, are the target weighted principal axes to be aligned. It can be represented as:

[0077] ;

[0078] For example, the Rodriguez rotation formula can be used to construct... Rotate to Parallel reference alignment rotation matrix The calculation method is as follows:

[0079] First, calculate the unit vector k, the axis of rotation, and the angle of rotation θ between the two vectors:

[0080] ;

[0081] Then solve for the rotation matrix using Rodriguez's formula:

[0082] ;

[0083] Where I is the identity matrix. Let be the antisymmetric matrix of vector k.

[0084] We can start by translating the vector. The point cloud in the first point cloud set is translated and its centroid position is corrected; then a rotation matrix is ​​used. The point clouds in the first point cloud set are rotated and transformed, and the skeleton pose is corrected, so that the first point cloud set and the second point cloud set are initially registered.

[0085] By employing a weighted centroid and a target weighted principal axis, the calculation process amplifies the contribution of high-curvature anatomical feature points such as joints and bone spines, significantly reducing the impact of acquisition noise and invalid edge points on the solution of translation vectors and rotation matrices. This significantly reduces the initial alignment error of coarse registration, greatly reduces the number of iterations for further registration, and improves the final bone registration accuracy.

[0086] According to embodiments of this disclosure, transforming each point cloud in the second point cloud set based on translation vectors and rotation matrices to obtain a target second point cloud set registered with the first point cloud set may include: performing multiple transformations on each point cloud in the initially registered second point cloud set based on preset transformation step sizes of multiple transformation dimensions to obtain multiple updated second point cloud sets. The multiple transformation dimensions include a spin dimension rotating about the target weighted principal axis, a tilt dimension rotating about an axis perpendicular to the target weighted principal axis, and a sliding dimension sliding along the direction of the target weighted principal axis; and determining the target second point cloud set from the multiple updated second point cloud sets based on the distance between each updated point cloud in each updated second point cloud set and the matching point cloud corresponding to each updated point cloud in the first point cloud set.

[0087] The preset transformation step size is a small adjustment increment for each of the three dimensions: spin, tilt, and slide. The spin and tilt dimensions correspond to small angle steps, and the slide dimension corresponds to small spatial distance steps.

[0088] For example, a target weighted principal axis perpendicular to the first point cloud set can be constructed. tilting rotation axis The search range for the tilt angle α is set to [-π / 12, π / 12], and the preset transformation step size for the tilt dimension is Δα. The tilt rotation matrix is ​​constructed using the Rodriguez formula. Using a tilt rotation matrix The second point cloud set of the initial registration is transformed to control the overall pitch and yaw tilt angle of the target skeleton and correct the small skew error of the long axis of the target skeleton.

[0089] For example, with the target-weighted principal axis Let β be the rotation axis, set the search range for the spin angle β, and set the preset transformation step size for the spin dimension to Δβ. Construct the spin rotation matrix. Using spin rotation matrix The second point cloud set of the initial registration is transformed so that the second point cloud set rotates around the long axis of the target bone itself, correcting the small torsional deviation of the bone around its own long axis after coarse registration.

[0090] For example, it can be along the target weighted principal axis The search range is set for the sliding distance d, the preset transformation step size of the sliding dimension is Δd, and the corresponding translation correction vector is... Using translation correction vectors The second point cloud set of the initial registration is transformed to control the skeleton to slide back and forth along its own long axis and correct the small positional offset in the long axis direction.

[0091] Figure 3 A schematic diagram illustrating the principle of registration according to an embodiment of the present disclosure is shown.

[0092] like Figure 3 As shown, firstly, by translating the vector... The second point cloud set is translated so that its weighted centroid coincides with the weighted centroid of the second point cloud set (see Figure a). Secondly, a rotation matrix is ​​used... The second point cloud set is rotated so that its target weighted principal axis is aligned with that of the first point cloud set (see Figure b). Third, a tilt rotation matrix is ​​used. The pitch and tilt of the first point cloud set are finely adjusted to correct the pitch deviation between the first and second point cloud sets (see Figure c). Fourth, the spin rotation matrix is ​​used... A rotation transformation is performed on the second point cloud set to align the first and second point cloud sets circumferentially (see Figure d). Fifth, the translation correction vector is... The first point cloud set is finely adjusted along the weighted principal axis to obtain the updated second point cloud set (see Figure e).

[0093] For each updated second point cloud set obtained after each transformation, traverse each updated point cloud in the set, retrieve the point with the closest Euclidean distance in the first point cloud set as the matching point cloud, and calculate the three-dimensional Euclidean distance between each set of updated point clouds and matching point clouds.

[0094] In some examples, the total or average matching distance of the entire updated point cloud can be calculated as the registration error, and the updated second point cloud set with the smallest error can be selected as the target second point cloud set based on the registration error.

[0095] After initial registration based on the weighted centroid and the target weighted axis, the three dimensions of spin, tilt, and sliding are further decomposed based on the target weighted principal axis extending from the target skeleton body. This covers all the degrees of freedom of the target skeleton transformation. The optimal set is selected by updating the point cloud and matching the point cloud distance, and only the unique correct pose with perfect anatomical feature fit is retained. This effectively reduces the multiple solutions and pose flip ambiguity caused by the strong cylindrical symmetry of the target skeleton.

[0096] According to embodiments of this disclosure, determining a target second point cloud set from multiple updated second point cloud sets based on the distances between each updated point cloud in each updated second point cloud set and the matching point cloud corresponding to each updated point cloud in the first point cloud set may include: determining a distance set based on the distances between each updated point cloud in each updated second point cloud set and the matching point cloud corresponding to each updated point cloud in the first point cloud set; sorting the distances in the distance set in ascending order to determine the target distance in the distance set that is at the target sorting position; and determining the target second point cloud set from multiple updated second point cloud sets based on the target distance.

[0097] For example, the 90th percentile trimmed Hausdorff distance (THD) can be used. As a scoring criterion for registration quality, the smaller the score, the better the registration fit.

[0098] For each updated point cloud in the second point cloud set, calculate the nearest neighbor Euclidean distance from each updated point cloud to the matching point cloud in the first point cloud set, obtaining a distance set D. Sort the distance set D in ascending order; the target position is the 90th percentile. The position of ) will The distance value is used as the target distance. This can filter out spurious errors caused by 10% of edge overhangs and noise points, accurately evaluate the fit of the subject point cloud, and avoid misjudgment.

[0099] The target second point cloud set can be selected from the multiple updated second point cloud sets whose target distances are minimized.

[0100] For example, a total rigid body transformation matrix can also be constructed for the second point cloud set, and the second point cloud set can be transformed to obtain the target second point cloud set. The transformed coordinates P′ of any P in the second point cloud set are:

[0101] ;

[0102] ;

[0103] in, For the weighted centroid level of the second point cloud set, R is the weighted centroid of the first point cloud set; total Represents the rotation matrix for target weighted principal axis alignment. Rotation matrix of skew dimension Spin-dimensional rotation matrix The total transformation matrix; This represents the translation correction vector.

[0104] In some embodiments, a two-stage search strategy can also be employed to efficiently solve for the globally optimal total rigid body transformation matrix. The first stage is a coarse search: based on a preset step size, it traverses different parameter combinations of the total rigid body transformation matrix across multiple dimensions, calculating the corresponding parameters for each set. The first stage involves selecting the top 5 candidate total rigid body transformation matrices with the smallest target distance. The second stage is a refined search: for each of the 5 candidate total rigid body transformation matrices, a local refined search is performed within the neighborhood of each parameter with a preset step size of 1 / 5, ultimately outputting the result. The smallest rigid body transformation matrix is ​​used as the target total rigid body transformation matrix. The target total rigid body transformation matrix is ​​used to transform the second point cloud set to obtain the target second point cloud set.

[0105] In some embodiments, the target total rigid body transformation matrix can be used as the initial value, and the Generalized Iterative Closest Point (GICP) algorithm can be used for fine registration optimization. The GICP algorithm unifies point-to-point and point-to-plane ICP under a probabilistic framework, constructs plane-to-plane matching constraints based on the covariance matrix of the local surface of the point cloud, has stronger robustness to sampling differences of non-homogeneous point clouds, can achieve sub-millimeter-level surface fitting, and obtains the final total rigid body transformation matrix.

[0106] The final transformation matrix obtained from the precise registration can be output to the surgical navigation system to achieve spatial mapping between the preoperative CT coordinate system and the intraoperative world coordinate system. The navigation information such as the preoperative osteotomy line and osteotomy plane can be accurately converted to the anatomical position of the patient during the operation, providing a navigation reference for HTO surgery.

[0107] Experimental results based on prosthetic models show that, in a typical clinical scenario with a 30% overlap rate, the method of this disclosure achieves an average target registration error of only 0.708 mm in point cloud registration. All experimental results are below the 1 mm clinical safety threshold for orthopedic surgery, fully meeting the millimeter-level accuracy requirements for osteotomy positioning and force line correction in HTO surgery. This achieves precise spatial mapping from preoperative planning to the intraoperative scene, ensuring the accuracy of osteotomy plane positioning and force line correction, reducing the risk of under- or over-correction due to registration errors, and improving the clinical efficacy and safety of HTO surgery.

[0108] Furthermore, the method of this disclosure does not require any disruptive modifications to the existing HTO surgical procedure. Registration can be completed simply by adding a structured light scan after exposing the bone surface, without changing the surgeon's surgical habits. The output rigid body transformation matrix can be directly interfaced with existing orthopedic surgical navigation systems, demonstrating strong adaptability.

[0109] Figure 4 A block diagram of a registration device for orthopedic surgical point cloud data is shown.

[0110] like Figure 4 The modified device shown includes an acquisition module 410, a determination module 420, a feature extraction module 430, and a registration module 440.

[0111] The acquisition module 410 is used to acquire a first point cloud set and a second point cloud set. The first point cloud set is obtained by modeling based on the image data of the target bone region before surgery, and the second point cloud set is obtained by collecting the point cloud data of the target bone region during surgery.

[0112] The determination module 420 is used to determine the weight parameters of each point cloud in the point cloud set based on the curvature of each point cloud in the point cloud set for any point cloud set in the first point cloud set and the second point cloud set.

[0113] The feature extraction module 430 is used to extract principal component features from the point cloud set based on the weight parameters of each point cloud in the point cloud set, and to determine the weighted principal component features.

[0114] The registration module 440 is used to register the first point cloud set and the second point cloud set based on the weighted principal component features of their respective sets, and obtain the registration result.

[0115] It should be noted that the registration device part for orthopedic surgical point cloud data in the embodiments of this disclosure corresponds to the registration method part for orthopedic surgical point cloud data in the embodiments of this disclosure. The description of the data processing device part is specifically referred to in the data processing method part, and will not be repeated here.

[0116] Another aspect of this disclosure provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the methods described above.

[0117] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0118] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A registration method for point cloud data in orthopedic surgery, characterized in that, The method includes: A first point cloud set and a second point cloud set are obtained. The first point cloud set is obtained by modeling based on image data of the target bone region before surgery, and the second point cloud set is obtained by collecting point cloud data of the target bone region during surgery. For any point cloud set in the first point cloud set and the second point cloud set, the weight parameters of each point cloud in the point cloud set are determined based on the curvature of each point cloud in the point cloud set. Based on the weight parameters of each point cloud in the point cloud set, weighted principal component features are extracted from the point cloud set to determine the weighted principal component features; Based on the weighted principal component features of the first and second point cloud sets respectively, the point cloud sets are registered to obtain the registration result.

2. The method according to claim 1, characterized in that, The step of extracting principal component features from the point cloud set based on the weight parameters of each point cloud in the point cloud set, and determining the weighted principal component features, includes: Based on the weight parameters of each point cloud in the point cloud set, a weighted average is calculated on multiple point clouds in the point cloud set to obtain the weighted centroid of the point cloud set. Based on the weighted centroids, the principal component features of the point cloud set are determined.

3. The method according to claim 2, characterized in that, Based on the weighted centroids, the principal component features of the point cloud set are determined, including: Based on the weighted centroid and the coordinates of each point cloud in the point cloud set, construct the weighted covariance matrix of the point cloud set; The eigenvalues ​​of the weighted covariance matrix are obtained by solving for the eigenvalues; Based on the target weighted feature value among the multiple weighted feature values, a target weighted principal axis is determined, wherein the target weighted feature value is greater than or equal to any other weighted feature value among the multiple weighted feature values; Based on the weighted centroid and the target weighted principal axis, the principal component features of the point cloud set are obtained.

4. The method according to claim 1, characterized in that, Based on the curvature of each point cloud in the point cloud set, the weight parameters of each point cloud in the point cloud set are determined, including: Determine the geometric centroid of the point cloud set; Based on the curvature of each point cloud in the point cloud set and the relative distance between each point cloud in the point cloud set and the geometric centroid, the weight parameters of each point cloud in the point cloud set are determined; wherein, the weight parameters increase with the increase of the curvature and decrease with the increase of the relative distance.

5. The method according to claim 1, characterized in that, The curvature of each point cloud in the point cloud set is determined in the following way: For any point cloud in the point cloud set, based on the point cloud and multiple neighboring point clouds whose distance from the point cloud satisfies a preset distance, a local point cloud set and a local geometric centroid of the local point cloud set are determined. Based on each point cloud in the local point cloud set and the local geometric centroid, a local covariance matrix is ​​constructed. The eigenvalues ​​of the local covariance matrix are obtained by solving for the eigenvalues. The curvature of the point cloud is determined based on the ratio of the target feature value among the plurality of feature values ​​to the sum of the plurality of feature values, wherein the target feature value is less than or equal to any other feature value among the plurality of feature values.

6. The method according to any one of claims 1 to 5, characterized in that, The weighted principal component features include weighted centroids and target weighted principal axes; Based on the weighted principal component features of the first and second point cloud sets respectively, point cloud set registration is performed on the first and second point cloud sets to obtain the registration results, including: Based on the weighted centroids of the first and second point cloud sets respectively, a translation vector is determined to align the two weighted centroids. Based on the target weighted principal axes of the first point cloud set and the second point cloud set respectively, determine the rotation matrix for rotating and aligning the two target weighted principal axes; Based on the translation vector and the rotation matrix, each point cloud in the second point cloud set is transformed to obtain the initially registered second point cloud set. Fine-tune the initially registered second point cloud set to obtain the target second point cloud set registered with the first point cloud set.

7. The method according to claim 6, characterized in that, The step of fine-tuning the initially registered second point cloud set to obtain a target second point cloud set registered with the first point cloud set includes: Based on the preset transformation step size of each of the multiple transformation dimensions, the point clouds in the initially registered second point cloud set are transformed multiple times to obtain multiple updated second point cloud sets; wherein, the multiple transformation dimensions include a spin dimension that rotates about the target weighted principal axis, a tilt dimension that rotates about an axis perpendicular to the target weighted principal axis, and a sliding dimension that slides along the direction of the target weighted principal axis; The target second point cloud set is determined from multiple updated second point cloud sets based on the distance between each updated point cloud in each updated second point cloud set and the matching point cloud in the first point cloud set corresponding to each updated point cloud.

8. The method according to claim 7, characterized in that, Based on the distance between each updated point cloud in each updated second point cloud set and the matching point cloud corresponding to each updated point cloud in the first point cloud set, the target second point cloud set is determined from multiple updated second point cloud sets, including: Based on the distance between each updated point cloud in each updated second point cloud set and the matching point cloud in the first point cloud set corresponding to each updated point cloud, a distance set is determined; Sort the distances in the distance set in ascending order, and determine the target distance in the distance set that is at the target sorting position; Based on the target distance, the target second point cloud set is determined from multiple updated second point cloud sets.

9. A registration device for point cloud data in orthopedic surgery, characterized in that, include: The acquisition module is used to acquire a first point cloud set and a second point cloud set. The first point cloud set is obtained by modeling based on image data of the target bone region before surgery, and the second point cloud set is obtained by collecting point cloud data of the target bone region during surgery. The determination module is used to determine the weight parameters of each point cloud in the point cloud set based on the curvature of each point cloud in the point cloud set for any point cloud set in the first point cloud set and the second point cloud set. The feature extraction module is used to extract principal component features from the point cloud set based on the weight parameters of each point cloud in the point cloud set, and determine the weighted principal component features. The registration module is used to register the first point cloud set and the second point cloud set based on the weighted principal component features of the first point cloud set and the second point cloud set respectively, and obtain the registration result.

10. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

Images