Construction system for orthopedic surgery analysis model

Abstract

The invention relates to the technical field of medical informatization, and discloses a construction system for an orthopedic surgery analysis model, which comprises the following steps: acquiring original tomography data, and carrying out resampling and isotropic filtering processing on the original tomography data to obtain standard forearm three-dimensional image data; performing semantic recognition processing on the standard forearm three-dimensional image data to obtain a Boolean skeleton matrix; the positioning module is used for performing fracture surface positioning processing based on a Boolean skeleton matrix to extract fracture surface parameters; the skeleton processing module is used for performing skeletonization processing on the Boolean skeleton matrix to extract a radius center line and an ulna center line, and determining a functional rotation axis based on the radius center line and the ulna center line; and projecting the radius center line to the functional rotation axis to obtain a projection curve, and performing curvature analysis processing on the projection curve to obtain a radius arch top coordinate.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and more specifically, to a system for constructing orthopedic surgical analysis models. Background Technology

[0002] In orthopedic surgery, the precision of screw implantation directly affects fracture healing and functional recovery. Currently, after reconstructing a bone model using CT data, the screw entry point, direction, and length can be designed in a virtual environment, and the principle of safe zone avoidance can be introduced. The screw implantation path can be optimized by identifying the distribution of blood vessels and nerves.

[0003] For the fixation of forearm fractures, current techniques typically employ plate and screw internal fixation, exposing the bone through a palmar or dorsal approach and adjusting screw position using intraoperative fluoroscopy. However, current techniques still rely on the surgeon's experience and lack a systematic analysis of the coupling of anatomy and function, making it difficult to simultaneously protect forearm rotational function and optimize screw mechanical properties.

[0004] Existing technologies still have the following problems with the double-bone structure of the forearm:

[0005] 1. The pronation and supination functions of the forearm depend on the coaxial rotation of the radius and ulna around the forearm's radial artery (FRA), but traditional path planning does not take the FRA as a core constraint.

[0006] 2. The cortex is thickest and strongest at the apex of the physiological curvature of the mid-shaft radius (the arch apex). However, current techniques often randomly select the entry point based on the fracture location, without using curvature analysis to lock the arch apex as the optimal entry point. This results in insufficient screw holding force, especially in patients with osteoporosis, where loosening is likely to occur.

[0007] 3. The palmar side of the radius and the dorsal side of the ulna have sparse blood vessels and nerves, forming a natural safe passage. However, existing safe zone analyses mostly target single bones (such as the sacrum) and do not combine the anatomy of the interosseous membrane with the spatial relationship between the two bones to establish cross-bone constraints.

[0008] In summary, existing technologies for screw implantation in forearm fractures have issues regarding functional compatibility, mechanical stability, and safety. Summary of the Invention

[0009] This invention provides a system for constructing orthopedic surgical analysis models, solving the technical problems mentioned in the background art.

[0010] This invention provides a system for constructing orthopedic surgical analysis models, comprising:

[0011] The data acquisition module 101 is used to acquire raw tomographic scan data and resample and isotropically filter the raw tomographic scan data to obtain standard forearm three-dimensional image data.

[0012] Data processing module 102 is used to perform semantic recognition processing on standard forearm three-dimensional image data to obtain a Boolean skeleton matrix; wherein, element 1 in the Boolean skeleton matrix represents skeleton and element 0 represents non-skeleton.

[0013] The first analysis module 103 is used to perform fracture surface localization processing based on the Boolean skeleton matrix to extract fracture surface parameters, including: fracture surface normal vector and fracture surface passing point;

[0014] The second analysis module 104 is used to perform skeletonization processing on the Boolean skeleton matrix to extract the radial and ulnar central lines and determine the functional rotation axis based on the radial and ulnar central lines.

[0015] The third analysis module 105 is used to project the radial centerline onto the functional rotation axis to obtain the projection curve, and to perform curvature analysis on the projection curve to obtain the coordinates of the radial arch top.

[0016] The screw prediction module 106 is used to perform adaptive optimization based on the radius arch top coordinates, fracture surface parameters and Boolean skeleton matrix to determine the screw entry coordinates and screw entry direction; and to perform multi-constraint search processing based on the screw entry coordinates, screw entry direction, fracture surface parameters and Boolean skeleton matrix to obtain the screw exit coordinates and screw length.

[0017] Furthermore, semantic recognition processing is performed on the standard forearm 3D image data to obtain a Boolean skeleton matrix, including:

[0018] Three-dimensional orthogonal slices of standard forearm 3D image data are extracted and used as three-channel inputs to the pre-trained UNet neural network; wherein, the three-dimensional orthogonal slices include: XY plane slices, YZ plane slices and XZ plane slices;

[0019] The pre-trained UNet neural network outputs a three-dimensional skeleton probability matrix; where the elements in the three-dimensional skeleton probability matrix take values ​​ranging from 0 to 1, representing the probability value of the corresponding pixel in the standard forearm three-dimensional image data belonging to the skeleton.

[0020] The 3D skeleton probability matrix is ​​binarized based on a probability threshold to obtain an initial Boolean skeleton matrix; the initial Boolean skeleton matrix is ​​then subjected to a morphological closing operation to obtain a smoothed Boolean skeleton matrix.

[0021] Connectivity analysis is performed on smoothed Boolean matrices to extract clusters of elements within them.

[0022] The element clusters are sorted from largest to smallest based on the number of elements in the clusters, forming a feature sort;

[0023] Extract the element clusters with the first and second positions in the feature sorting and use them as the radius skeleton mask and ulna skeleton mask, respectively.

[0024] A Boolean skeleton matrix is ​​formed by combining the radius skeleton mask and the ulna skeleton mask.

[0025] Furthermore, fracture surface localization is performed based on a Boolean skeleton matrix to extract fracture surface parameters, including:

[0026] Edge detection is performed on the Boolean skeleton matrix to obtain a set of edge points;

[0027] Determine the gradient value of each edge point in the edge point set, and classify the edge points based on the gradient threshold, including: if the gradient value of an edge point is greater than or equal to the gradient threshold, then classify the edge point as a transition point to obtain a transition point set;

[0028] Connectivity analysis is performed on the jump points in the jump point set to extract the element clusters in the jump point set, and the element cluster with the largest number of element clusters is taken as the fracture surface point set.

[0029] Principal component analysis was used to perform planar fitting on the set of fracture surface points to obtain the fracture surface parameters NP, where NP = {n}. fr ,p fr}, n fr p represents the normal vector of the fracture surface. fr This indicates that the fracture surface has passed through a certain point.

[0030] Furthermore, the Boolean skeleton matrix is ​​subjected to skeletalization to extract the radial and ulnar central lines, and the functional rotation axis is determined based on the radial and ulnar central lines, including:

[0031] Step 41: Perform three-dimensional morphological skeletonization processing on the radius skeleton mask to extract the set of central skeleton points of the radius.

[0032] Step 42: Extract the radial skeleton edge of the radial central skeleton point set to form the radial skeleton edge point set, and determine the Euclidean distance between any two radial skeleton edge points in the radial skeleton edge point set;

[0033] Step 43: Take the line connecting the two radial skeleton edge points with the largest Euclidean distance as the radial centerline;

[0034] Step 44: Repeat steps 51-53 for the ulnar skeleton mask to obtain the ulnar centerline;

[0035] Step 45, obtain the center point C of the radial central line and the ulnar central line respectively. Rp and C ud To calculate the rotation axis v f ,

[0036] Furthermore, the radial centerline is projected onto the functional rotation axis to obtain a projection curve, and curvature analysis is performed on the projection curve to obtain the coordinates of the radial arch apex, including:

[0037] Initialize the projection operator P, P = Iv f (v f ) T I represents a 3×3 identity matrix;

[0038] Using the center point of the radial central line as a reference, project the radial central line onto a plane perpendicular to the functional rotation axis to obtain the projection curve, as shown below:

[0039]

[0040] in, Represents the projection curve, C R (s) represents the radial central line;

[0041] Projection curve Normalization is performed as follows include:

[0042] Projection curve Calculate the cumulative arc length ζ(s) from one endpoint of the projection curve to point s on the projection curve. in, represents the differential of the arc length of the projected curve at s′, where s′ represents the integral variable from one endpoint of the projected curve to s;

[0043] By accumulating the arc length ζ(s), along the path that causes Projection curve Normalization processing against Find the first derivative to obtain the tangent vector. For the tangent vector The curvature vector is obtained by taking the second derivative.

[0044] Based on tangent vector and curvature vector The curvature analysis of the projection curve is performed as follows:

[0045]

[0046] Where κ(t) represents the curvature of the projection curve at t, and ||·|| represents the modulus;

[0047] The location of the maximum curvature of the projection curve is determined based on κ(t), and the maximum location is converted into the location corresponding to the radial centerline through reparameterization mapping to obtain the coordinates of the radial arch top.

[0048] Furthermore, adaptive optimization is performed based on the radius arch top coordinates, fracture surface parameters, and Boolean bone matrix to determine the screw entry coordinates and screw entry direction, including:

[0049] The radial skeleton edge point with the smallest Euclidean distance to the radial arch apex is selected as the feature point. Using this feature point as the center, principal component analysis is performed on the 3×3×3 elements extracted from the Boolean skeleton matrix to obtain the radial external normal n. out ,

[0050] From the feature point along n out The distribution of element 1 in the radial skeleton mask is statistically analyzed, and the length of consecutive element 1 is calculated as the radial inlet cortical thickness t. R ; from the feature point along n out The distribution of element 1 in the ulnar skeleton mask is statistically analyzed, and the length of consecutive element 1 is calculated as the ulnar exit cortical thickness t. U ;

[0051] After taking the first derivative of the coordinates of the radial centerline at the top of the radial arch and normalizing it, the tangential vector t is obtained. b , C R (s)′ represents the first derivative of the coordinates of the radial centerline at the top of the radial arch;

[0052] The thickness-weighted coefficients α and β, and the fracture guidance coefficient γ, are determined as follows:

[0053]

[0054] Determine the screw inlet direction v s ,for:

[0055] Determine the screw entry coordinate P e , is: P e =A s +(0.5t R +∈)n out ∈ represents the safety margin.

[0056] Furthermore, based on the screw inlet coordinates, screw inlet direction, fracture surface parameters, and Boolean bone matrix, a multi-constraint search is performed to obtain the screw exit coordinates and screw length, including:

[0057] Starting from the screw inlet coordinates, construct a parametric line L(u) with the screw inlet direction, L(u) = P. e+uv s u represents a scalar parameter used to determine the position of a point on the parameterized line, u≥0;

[0058] Calculate the fracture intersection point u fr ,

[0059] Determine the search interval Ω, Ω={u|u≥u fr}; Determine the parameter u that minimizes the Euclidean distance between the search interval and any point on the radius skeleton edge in the set of radius skeleton edge points. * The screw length is u * The screw exit coordinate is P. x =L(u * ).

[0060] Furthermore, based on the screw inlet direction v s Screw entry coordinate P e Screw exit coordinate P x and screw length u * Perform forearm screw insertion.

[0061] The beneficial effects of this invention are as follows: By constructing a complete system that includes data acquisition, semantic recognition, fracture surface localization, functional rotation axis analysis, curvature calculation, and screw path optimization, the forearm functional rotation axis is used as the core constraint. Combined with the curvature analysis of the radial arch, the optimal insertion point with the thickest cortex is locked. Furthermore, a safe implantation channel is constructed through cross-bone cortical thickness weighting and multi-constraint search. This significantly improves the accuracy, mechanical stability, and functional protection of screw implantation. It effectively solves the problems of relying on experience, neglecting forearm rotation function protection, and insufficient screw holding force in traditional techniques. It provides a systematic solution for orthopedic surgery that takes into account anatomical rationality, functional compatibility, and clinical safety. Attached Figure Description

[0062] Figure 1 This is a block diagram of a system for constructing an orthopedic surgical analysis model according to the present invention. Detailed Implementation

[0063] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0064] like Figure 1As shown, a system for constructing an analysis model for orthopedic surgery includes:

[0065] The data acquisition module 101 is used to acquire raw tomographic scan data and resample and isotropically filter the raw tomographic scan data to obtain standard forearm three-dimensional image data.

[0066] Data processing module 102 is used to perform semantic recognition processing on standard forearm three-dimensional image data to obtain a Boolean skeleton matrix; wherein, element 1 in the Boolean skeleton matrix represents skeleton and element 0 represents non-skeleton.

[0067] The first analysis module 103 is used to perform fracture surface localization processing based on the Boolean skeleton matrix to extract fracture surface parameters, including: fracture surface normal vector and fracture surface passing point;

[0068] The second analysis module 104 is used to perform skeletonization processing on the Boolean skeleton matrix to extract the radial and ulnar central lines and determine the functional rotation axis based on the radial and ulnar central lines.

[0069] The third analysis module 105 is used to project the radial centerline onto the functional rotation axis to obtain the projection curve, and to perform curvature analysis on the projection curve to obtain the coordinates of the radial arch top.

[0070] The screw prediction module 106 is used to perform adaptive optimization based on the radius arch top coordinates, fracture surface parameters and Boolean skeleton matrix to determine the screw entry coordinates and screw entry direction; and to perform multi-constraint search processing based on the screw entry coordinates, screw entry direction, fracture surface parameters and Boolean skeleton matrix to obtain the screw exit coordinates and screw length.

[0071] It should be noted that resampling includes: sequentially resampling, trilinear interpolation, and scaling of the original forearm scan data to obtain original forearm scan data with consistent volumetric resolution;

[0072] Isotropic filtering includes: processing the original forearm scan data with a volume resolution uniformity through a preset three-dimensional isotropic Gaussian kernel convolution to obtain standard forearm scan data.

[0073] In one embodiment of the present invention, semantic recognition processing is performed on standard forearm three-dimensional image data to obtain a Boolean skeleton matrix, including:

[0074] Three-dimensional orthogonal slices of standard forearm 3D image data are extracted and used as three-channel inputs to the pre-trained UNet neural network; wherein, the three-dimensional orthogonal slices include: XY plane slices, YZ plane slices and XZ plane slices;

[0075] The pre-trained UNet neural network outputs a three-dimensional skeleton probability matrix; where the elements in the three-dimensional skeleton probability matrix take values ​​ranging from 0 to 1, representing the probability value of the corresponding pixel in the standard forearm three-dimensional image data belonging to the skeleton.

[0076] The 3D skeleton probability matrix is ​​binarized based on a probability threshold to obtain an initial Boolean skeleton matrix; the initial Boolean skeleton matrix is ​​then subjected to a morphological closing operation to obtain a smoothed Boolean skeleton matrix.

[0077] Connectivity analysis is performed on smoothed Boolean matrices to extract clusters of elements within them.

[0078] The element clusters are sorted from largest to smallest based on the number of elements in the clusters, forming a feature sort;

[0079] Extract the element clusters with the first and second positions in the feature sorting and use them as the radius skeleton mask and ulna skeleton mask, respectively.

[0080] A Boolean skeleton matrix is ​​formed by combining the radius skeleton mask and the ulna skeleton mask.

[0081] In detail, UNet is a neural network architecture commonly used for image segmentation. The training process of UNet is as follows: taking a large number of forearm tomographic image slices (sample data) as input, manually labeling each pixel to determine whether it belongs to the skeleton (sample label), minimizing the difference between the prediction result and the label through stochastic gradient descent (the loss function is the cross-entropy loss function), adjusting the UNet network parameters to enable it to recognize skeletal features, and finally obtaining a pre-trained model that can output the probability value of a pixel belonging to the skeleton.

[0082] Suppose that an element in the 3D skeleton probability matrix has a value of 0.6. If the probability threshold is set to 0.5, when the element value is ≥0.5, it is assigned a value of 1 (determined to be a skeleton); if the element value is <0.5, it is assigned a value of 0 (determined to be a non-skeleton).

[0083] Morphological closing operations are a common technique for matrix smoothing. By first dilating and then eroding, they can fill small holes inside bones, connect neighboring bone pixels, and smooth the contours of bone boundaries, making the bone structure more complete in the matrix and reducing the discrepancies or fragmentation caused by image noise or segmentation errors.

[0084] For a smooth Boolean matrix, if adjacent elements (up, down, left, right, front, back) are all 1 (representing bones), they are considered connected. By traversing the matrix, connected 1s are grouped into a cluster of elements (i.e., a connected region). If there are two independent skeletal regions in the matrix, analysis will result in two clusters of elements. After sorting by the number of elements from most to least, the top two clusters are used as skeletal masks for the radius and ulna, respectively, thus distinguishing the two-bone structure of the forearm.

[0085] In the anatomical structure of the forearm, the radius and ulna are the two main bones, which are relatively large and contain a large number of pixels. After performing connected component analysis on the smoothed Boolean matrix, the two clusters with the most elements, sorted from most to least numerous, correspond to the radius and ulna.

[0086] Both the radius and ulna skeleton masks are binary matrix structures. The elements in the matrix take values ​​of 0 or 1, where 1 indicates that the pixel at that position belongs to the skeleton region of the radius or ulna, and 0 indicates that the pixel at that position does not belong to the corresponding skeleton region.

[0087] In one embodiment of the present invention, fracture surface localization processing is performed based on a Boolean skeleton matrix to extract fracture surface parameters, including:

[0088] Edge detection is performed on the Boolean skeleton matrix to obtain a set of edge points;

[0089] Determine the gradient value of each edge point in the edge point set, and classify the edge points based on the gradient threshold, including: if the gradient value of an edge point is greater than or equal to the gradient threshold, then classify the edge point as a transition point to obtain a transition point set;

[0090] Connectivity analysis is performed on the jump points in the jump point set to extract the element clusters in the jump point set, and the element cluster with the largest number of element clusters is taken as the fracture surface point set.

[0091] Principal component analysis was used to perform planar fitting on the set of fracture surface points to obtain the fracture surface parameters NP, where NP = {n}. fr ,p fr}, n fr p represents the normal vector of the fracture surface. fr This indicates that the fracture surface has passed through a certain point.

[0092] In detail, 3D edge detection can use the Marching Cubes algorithm, which analyzes each cube unit in the 3D volume data and generates corresponding surface polygons based on the attributes of the unit vertices (such as whether they belong to the skeleton), thereby detecting the edge structure in the 3D data. It is often used for surface reconstruction and edge extraction of 3D models.

[0093] Transition points are special points further filtered from the set of edge points after edge detection of standard forearm 3D image data. Specifically, if the gradient value of an edge point is greater than or equal to a pre-set gradient threshold, it indicates that the pixel value at that point has undergone a significant transition. The element corresponding to that pixel in the Boolean skeleton matrix is ​​then classified as a transition point, forming a transition point set. The transition point set is the key feature point for locating the fracture surface.

[0094] A fracture surface is typically a large and continuous region within a skeletal structure. After performing connected component analysis on the set of transition points, multiple element clusters (sets of interconnected transition points) are formed. The cluster with the largest number of element clusters (containing the most transition points) best matches the characteristics of a fracture surface as a large, continuous structure due to its size and strong continuity; therefore, it is considered the set of fracture surface points.

[0095] Principal component analysis (PCA) is a method that simplifies data structure by analyzing the main characteristic directions of the data. When processing a set of fracture surface points, the distribution characteristics of these points in three-dimensional space are analyzed to identify the direction with the largest variance (principal component). Through fitting, a plane is determined that makes the set of fracture surface points as close to this plane as possible. The final fracture surface parameters are obtained, including the fracture surface normal vector and the fracture surface reference point. The fracture surface normal vector is a vector perpendicular to the fracture surface, representing the orientation and inclination of the fracture surface in three-dimensional space; the fracture surface reference point is a specific reference point on the fracture surface, representing the specific location of the fracture surface in space.

[0096] The fracture surface normal vector provides a key reference for planning the screw implantation direction, ensuring that the screw can pass through the fracture surface at a reasonable angle to achieve effective fixation, while avoiding the impact of directional deviation on the fixation effect or damage to surrounding tissues.

[0097] The fracture surface reference point serves as a benchmark for the spatial location of the fracture surface, assisting in the precise positioning of the fracture area and making subsequent screw path design, mechanical analysis, and other operations more closely aligned with the actual fracture situation.

[0098] In one embodiment of the present invention, a skeletonization process is performed on the Boolean skeleton matrix to extract the radial and ulnar central lines, and a functional rotation axis is determined based on the radial and ulnar central lines, including:

[0099] Step 41: Perform three-dimensional morphological skeletonization processing on the radius skeleton mask to extract the set of central skeleton points of the radius.

[0100] Step 42: Extract the radial skeleton edge of the radial central skeleton point set to form the radial skeleton edge point set, and determine the Euclidean distance between any two radial skeleton edge points in the radial skeleton edge point set;

[0101] Step 43: Take the line connecting the two radial skeleton edge points with the largest Euclidean distance as the radial centerline;

[0102] Step 44: Repeat steps 51-53 for the ulnar skeleton mask to obtain the ulnar centerline;

[0103] Step 45, obtain the center point C of the radial central line and the ulnar central line respectively. Rp and C udTo calculate the rotation axis v f ,

[0104] In one embodiment of the present invention, a three-dimensional morphological skeletonization process is performed on the radial skeleton mask to extract the set of central skeleton points of the radius, as follows:

[0105] A three-dimensional morphological erosion operation is performed on the radius skeleton mask to gradually remove pixels at the edge of the radius, but preserve its core skeleton structure.

[0106] Continue the above erosion process until no more pixels can be removed. The final set of pixels is the set of central skeletal points of the radius, which represents the core skeletal structure of the radius.

[0107] It should be noted that the three-dimensional morphological skeletonization process is an existing technology and will not be elaborated further.

[0108] In detail, each point in the set of radial central skeleton points is traversed, and its adjacent pixels (such as adjacent points above, below, left, right, front, and back in 3D space) are checked. If there are non-skeleton points (or they are on the matrix boundary) among the adjacent pixels of a skeleton point, then the skeleton point is marked as a skeleton edge point. Finally, all such points constitute the set of radial skeleton edge points.

[0109] The two points on the radial skeleton with the largest Euclidean distance span the longest possible distance of the radius in space. Connecting these two points represents the main direction of extension of the radius in space, serving as the radial centerline for subsequent analysis of the radius's spatial position and orientation.

[0110] Forearm rotation is the result of the coordinated movement of the radius and ulna. By obtaining the center point of their center lines and calculating the direction vector of the line connecting the two points, an axis that comprehensively reflects the overall rotational trend of the forearm can be obtained. The functional rotation axis provides a clear reference for the pronation and supination functions of the forearm. In orthopedic surgeries (such as screw implantation), ensuring that the operation does not deviate from this axis can effectively avoid postoperative forearm rotational function restriction and ensure the patient's normal limb movement.

[0111] In one embodiment of the present invention, the radial centerline is projected onto a functional rotation axis to obtain a projection curve, and the curvature analysis of the projection curve is performed to obtain the coordinates of the radial arch apex, including:

[0112] Initialize the projection operator P, P = Iv f (v f ) T I represents a 3×3 identity matrix;

[0113] Using the center point of the radial central line as a reference, project the radial central line onto a plane perpendicular to the functional rotation axis to obtain the projection curve, as shown below:

[0114]

[0115] in, Represents the projection curve, C R (s) represents the radial central line;

[0116] Projection curve Normalization is performed as follows include:

[0117] Projection curve Calculate the cumulative arc length ζ(s) from one endpoint of the projection curve to point s on the projection curve. in, represents the differential of the arc length of the projected curve at s′, where s′ represents the integral variable from one endpoint of the projected curve to s;

[0118] By accumulating the arc length ζ(s), along the path that causes Projection curve Normalization processing against Find the first derivative to obtain the tangent vector. For the tangent vector The curvature vector is obtained by taking the second derivative.

[0119] Based on tangent vector and curvature vector The curvature analysis of the projection curve is performed as follows:

[0120]

[0121] Where κ(t) represents the curvature of the projection curve at t, and ||·|| represents the modulus;

[0122] The location of the maximum curvature of the projection curve is determined based on κ(t), and the maximum location is converted into the location corresponding to the radial centerline through reparameterization mapping to obtain the coordinates of the radial arch top.

[0123] In detail, define P = Iv f (v f ) T This projection operator is used to eliminate vectors along the functional rotation axis v. f The directional component ensures that the radial centerline is subsequently projected onto a plane perpendicular to v. f On the plane.

[0124] Using the center point of the radial central line as a reference, the radial central line C is calculated using the formula. R (s) Projected onto the axis of rotation perpendicular to the function v fThe plane from which the projection curve is obtained. Dissect along the radial central line v f The directional component retains only the component perpendicular to v. f Shape information within a plane.

[0125] The cumulative arc length represents the cumulative arc length from the endpoint of the projected curve to point 's'. By adjusting the parameters, we can... Normalize the projected curve. After normalization, the parameter t is directly related to the arc length.

[0126] Using formula Calculate the curvature κ(t). The numerator is the magnitude of the cross product of the tangent vector and the curvature vector (reflecting the degree of perpendicularity and the magnitude of change between the vectors), and the denominator is the cube of the magnitude of the tangent vector. The final result represents the curvature of the projection curve at t. The more pronounced the curvature of κ(t), the better.

[0127] Find the location where κ(t) is maximum (i.e., the point where the projected curve is most curved), and through reparameterization mapping, convert this location into the actual location corresponding to the radial centerline. This location is the coordinate of the radial arch apex. The radial arch apex is the apex of the physiological curvature of the radius, and this point is located through curvature analysis.

[0128] In one embodiment of the present invention, adaptive optimization is performed based on the radial arch top coordinates, fracture surface parameters, and Boolean bone matrix to determine the screw entry coordinates and screw entry direction, including:

[0129] The radial skeleton edge point with the smallest Euclidean distance to the radial arch apex is selected as the feature point. Using this feature point as the center, principal component analysis is performed on the 3×3×3 elements extracted from the Boolean skeleton matrix to obtain the radial external normal n. out ,

[0130] From the feature point along n out The distribution of element 1 in the radial skeleton mask is statistically analyzed, and the length of consecutive element 1 is calculated as the radial inlet cortical thickness t. R ; from the feature point along n out The distribution of element 1 in the ulnar skeleton mask is statistically analyzed, and the length of consecutive element 1 is calculated as the ulnar exit cortical thickness t. U ;

[0131] After taking the first derivative of the coordinates of the radial centerline at the top of the radial arch and normalizing it, the tangential vector t is obtained. b , C R (s)′ represents the first derivative of the coordinates of the radial centerline at the top of the radial arch;

[0132] The thickness-weighted coefficients α and β, and the fracture guidance coefficient γ, are determined as follows:

[0133]

[0134] Determine the screw inlet direction v s ,for:

[0135] Determine the screw entry coordinate P e , is: P e =A s +(0.5t R +∈)n out ∈ represents the safety margin.

[0136] In detail, the point with the smallest Euclidean distance from the radial arch apex to the edge of the radial skeleton is selected as the feature point. Because the cortex at the radial arch apex is thick and has excellent mechanical properties, selecting this closest edge point ensures that the needle insertion point is within the ideal anatomical region.

[0137] Using feature points as the center, extract elements from the Boolean skeleton matrix and perform principal component analysis to obtain the radial external normal n. out PCA determines the outward normal of the radial surface at key points by analyzing the main characteristic directions of the data. This is used to characterize the outward orientation of the bone surface and provides a benchmark for statistical cortical thickness and determining the inlet direction.

[0138] From the feature point along the radial lateral normal n out The continuous distribution length of element "1" (representing a bone region) in the radius and ulna bone masks is statistically analyzed to obtain the cortical thickness at the radial inlet and ulna exit. This reflects the cortical thickness of the bones on both sides along the needle insertion path and is used for subsequent weighted calculations to ensure that the screw enters from the relatively more stable side (a larger thickness results in a higher weight), thereby improving mechanical stability.

[0139] After taking the first derivative of the radial centerline at the coordinates of the radial arch apex and normalizing it, the tangent vector is obtained. The tangent vector reflects the direction of the tangent at the arch apex of the radius, embodying the anatomical morphological characteristics of the radius, and ensuring that the screw path conforms to the natural shape of the radius when determining the entry direction.

[0140] α represents the weight of the ulnar exit thickness, and β represents the weight of the radial inlet thickness. If t U >t R If α > β, it means that the ulnar side has a greater influence on the inlet direction, ensuring that the screw direction prioritizes the stability of the thicker side.

[0141] If p fr Distance A s The farther away the fracture site is, the smaller the γ value, indicating that the fracture surface has a weaker influence on the direction of entry; conversely, the closer the fracture site is, the stronger the influence, making the direction of entry take into account the location of the fracture.

[0142] Through formula The functional rotation axis (ensuring uninterrupted forearm rotation), tangent vector (fitting the radius morphology), and fracture surface normal vector (adapting to fracture characteristics) are synthesized and normalized using weighted coefficients. Multi-objective optimization is achieved by comprehensively considering functional, anatomical, and fracture factors.

[0143] According to P e =A s +(0.5t R +∈)n out With feature point A s Based on n, along out Directional movement 0.5t R (Take half the thickness of the radial cortex to ensure an appropriate depth within the cortex) and add a safety margin. This ensures that the screw entry is in a stable cortical area while also providing a safety space to avoid being too deep or too shallow.

[0144] In one embodiment of the present invention, a multi-constraint search process is performed based on the screw entry coordinates, screw entry direction, fracture surface parameters, and Boolean bone matrix to obtain the screw exit coordinates and screw length, including:

[0145] Starting from the screw inlet coordinates, construct a parametric line L(u) with the screw inlet direction, L(u) = P. e +uv s u represents a scalar parameter used to determine the position of a point on the parameterized line, u≥0;

[0146] Calculate the fracture intersection point u fr ,

[0147] Determine the search interval Ω, Ω={u|u≥u fr}; Determine the parameter u that minimizes the Euclidean distance between the search interval and any point on the radius skeleton edge in the set of radius skeleton edge points. * The screw length is u * The screw exit coordinate is P. x =L(u * ).

[0148] In detail, a parameterized straight line is constructed to represent the extended path of the screw starting from the inlet. The position of each point on the straight line is determined by adjusting u, providing a path model for searching the exit coordinates.

[0149] Calculate the intersection parameter u of the parameterized straight fracture surface using vector dot product. fr . Molecule n fr (p fr -P e ) indicates to p fr -P e In n fr Projection of direction, divided into nfr v s Indicates v s In n fr The projection of the direction is divided to obtain the u-value when the straight line passes through the fracture surface, ensuring that the screw path passes through the fracture surface.

[0150] Limit the search range from u fr First, ensure that the screw exit is on the opposite side of the fracture surface to meet the requirement that the screw needs to pass through the fracture surface for fixation during surgery.

[0151] Find the parameter \(u^{*}\) that minimizes the Euclidean distance to any point in the set of points on the edge of the radius skeleton.

[0152] u * The screw length is designed to ensure that it passes through the fracture surface without excessively protruding from the radius (close to the edge of the radial skeleton but not excessively), balancing fixation effectiveness and safety. x =L(u * Substitute the parameterized straight line to determine the specific coordinates of the screw exit, so that the screw path can be accurately adapted to the bone structure.

[0153] In one embodiment of the invention, based on the screw entry direction v s Screw entry coordinate P e Screw exit coordinate P x and screw length u * Perform forearm screw insertion.

[0154] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A system for constructing orthopedic surgical analysis models, characterized in that, include: The data acquisition module 101 is used to acquire raw tomographic scan data and resample and isotropically filter the raw tomographic scan data to obtain standard forearm three-dimensional image data. Data processing module 102 is used to perform semantic recognition processing on standard forearm three-dimensional image data to obtain a Boolean skeleton matrix; wherein, element 1 in the Boolean skeleton matrix represents skeleton and element 0 represents non-skeleton. The first analysis module 103 is used to perform fracture surface localization processing based on the Boolean skeleton matrix to extract fracture surface parameters, including: fracture surface normal vector and fracture surface passing point; The second analysis module 104 is used to perform skeletonization processing on the Boolean skeleton matrix to extract the radial and ulnar central lines and determine the functional rotation axis based on the radial and ulnar central lines. The third analysis module 105 is used to project the radial centerline onto the functional rotation axis to obtain the projection curve, and to perform curvature analysis on the projection curve to obtain the coordinates of the radial arch top. The screw prediction module 106 is used to perform adaptive optimization based on the radius arch top coordinates, fracture surface parameters and Boolean skeleton matrix to determine the screw entry coordinates and screw entry direction; and to perform multi-constraint search processing based on the screw entry coordinates, screw entry direction, fracture surface parameters and Boolean skeleton matrix to obtain the screw exit coordinates and screw length.

2. The system for constructing an orthopedic surgical analysis model according to claim 1, characterized in that, Semantic recognition processing is performed on standard forearm 3D image data to obtain a Boolean skeleton matrix, including: Three-dimensional orthogonal slices of standard forearm 3D image data are extracted and used as three-channel inputs to the pre-trained UNet neural network; wherein, the three-dimensional orthogonal slices include: XY plane slices, YZ plane slices and XZ plane slices; The pre-trained UNet neural network outputs a three-dimensional skeleton probability matrix; where the elements in the three-dimensional skeleton probability matrix take values ​​ranging from 0 to 1, representing the probability value of the corresponding pixel in the standard forearm three-dimensional image data belonging to the skeleton. The 3D skeleton probability matrix is ​​binarized based on a probability threshold to obtain an initial Boolean skeleton matrix; the initial Boolean skeleton matrix is ​​then subjected to a morphological closing operation to obtain a smoothed Boolean skeleton matrix. Connectivity analysis is performed on smoothed Boolean matrices to extract clusters of elements within them. The element clusters are sorted from largest to smallest based on the number of elements in the clusters, forming a feature sort; Extract the element clusters with the first and second positions in the feature sorting and use them as the radius skeleton mask and ulna skeleton mask, respectively. A Boolean skeleton matrix is ​​formed by combining the radius skeleton mask and the ulna skeleton mask.

3. The system for constructing an orthopedic surgical analysis model according to claim 2, characterized in that, Fracture surface localization is performed based on a Boolean skeleton matrix to extract fracture surface parameters, including: Edge detection is performed on the Boolean skeleton matrix to obtain a set of edge points; Determine the gradient value of each edge point in the edge point set, and classify the edge points based on the gradient threshold, including: if the gradient value of an edge point is greater than or equal to the gradient threshold, then classify the edge point as a transition point to obtain a transition point set; Connectivity analysis is performed on the jump points in the jump point set to extract the element clusters in the jump point set, and the element cluster with the largest number of element clusters is taken as the fracture surface point set. Principal component analysis was used to perform planar fitting on the set of fracture surface points to obtain the fracture surface parameters NP, where NP = {n}. fr ,p fr }, n fr p represents the normal vector of the fracture surface. fr This indicates that the fracture surface has passed through a certain point.

4. The system for constructing an orthopedic surgical analysis model according to claim 3, characterized in that, Skeletalization is performed on the Boolean skeleton matrix to extract the radial and ulnar central lines, and the functional rotation axes are determined based on the radial and ulnar central lines, including: Step 41: Perform three-dimensional morphological skeletonization processing on the radius skeleton mask to extract the set of central skeleton points of the radius. Step 42: Extract the radial skeleton edge of the radial central skeleton point set to form the radial skeleton edge point set, and determine the Euclidean distance between any two radial skeleton edge points in the radial skeleton edge point set; Step 43: Take the line connecting the two radial skeleton edge points with the largest Euclidean distance as the radial centerline; Step 44: Repeat steps 51-53 for the ulnar skeleton mask to obtain the ulnar centerline; Step 45, obtain the center point C of the radial central line and the ulnar central line respectively. Rp and C ud To calculate the rotation axis v f , 5. The system for constructing an orthopedic surgical analysis model according to claim 4, characterized in that, The radial centerline is projected onto the functional rotation axis to obtain the projection curve, and curvature analysis is performed on the projection curve to obtain the coordinates of the radial arch apex, including: Initialize the projection operator P, P = Iv f (v f ) T I represents a 3×3 identity matrix; Using the center point of the radial central line as a reference, project the radial central line onto a plane perpendicular to the functional rotation axis to obtain the projection curve, as shown below: in, Represents the projection curve, C R (s) represents the radial central line; Projection curve Normalization is performed as follows include: Projection curve Calculate the cumulative arc length ζ(s) from one endpoint of the projection curve to point s on the projection curve. in, represents the differential of the arc length of the projected curve at s′, where s′ represents the integral variable from one endpoint of the projected curve to s; By accumulating the arc length ζ(s), along the path that causes Projection curve Normalization processing against Find the first derivative to obtain the tangent vector. For the tangent vector The curvature vector is obtained by taking the second derivative. Based on tangent vector and curvature vector The curvature analysis of the projection curve is performed as follows: Where κ(t) represents the curvature of the projection curve at t, and ||·|| represents the modulus; The location of the maximum curvature of the projection curve is determined based on κ(t), and the maximum location is converted into the location corresponding to the radial centerline through reparameterization mapping to obtain the coordinates of the radial arch top.

6. The system for constructing an orthopedic surgical analysis model according to claim 5, characterized in that, Adaptive optimization is performed based on the radial arch top coordinates, fracture surface parameters, and Boolean bone matrix to determine the screw entry coordinates and screw entry direction, including: The radial skeleton edge point with the smallest Euclidean distance to the radial arch apex is selected as the feature point. Using this feature point as the center, principal component analysis is performed on the 3×3×3 elements extracted from the Boolean skeleton matrix to obtain the radial external normal n. out , From the feature point along n out The distribution of element 1 in the radial skeleton mask is statistically analyzed, and the length of consecutive element 1 is calculated as the radial inlet cortical thickness t. R ; from the feature point along n out The distribution of element 1 in the ulnar skeleton mask is statistically analyzed, and the length of consecutive element 1 is calculated as the ulnar exit cortical thickness t. U ; After taking the first derivative of the coordinates of the radial centerline at the top of the radial arch and normalizing it, the tangential vector t is obtained. b , C R (s)′ represents the first derivative of the coordinates of the radial centerline at the top of the radial arch; The thickness-weighted coefficients α and β, and the fracture guidance coefficient γ, are determined as follows: Determine the screw inlet direction v s ,for: Determine the screw entry coordinate P e , is: P e =A s +(0.5t R +∈)n out ∈ represents the safety margin.

7. The system for constructing an orthopedic surgical analysis model according to claim 6, characterized in that, A multi-constraint search is performed based on screw inlet coordinates, screw inlet direction, fracture surface parameters, and Boolean bone matrix to obtain screw outlet coordinates and screw length, including: Starting from the screw inlet coordinates, construct a parametric line L(u) with the screw inlet direction, L(u) = P. e +uv s u represents a scalar parameter used to determine the position of a point on the parameterized line, u≥0; Calculate the fracture intersection point u fr , Determine the search interval Ω, Ω={u|u≥u fr }; Determine the parameter u that minimizes the Euclidean distance between the search interval and any point on the radius skeleton edge in the set of radius skeleton edge points. * The screw length is u * The screw exit coordinate is P. x =L(u * ).

8. The system for constructing an orthopedic surgical analysis model according to claim 7, characterized in that, Based on screw inlet direction v s Screw entry coordinate P e Screw exit coordinate P x and screw length u * Perform forearm screw insertion.

Images