Intertrochanteric fracture navigation method and system

Abstract

The invention discloses an intertrochanteric fracture navigation method and system, and relates to the technical field of intertrochanteric fracture navigation. The intertrochanteric fracture navigation method comprises the following steps: acquiring medical image data of a side hip joint area, and constructing a three-dimensional digital model; identifying a fracture line and a plurality of key bone anatomical mark points in the three-dimensional digital model; marking a near-end head and neck fracture block and a far-end backbone fracture block in the three-dimensional digital model based on a fracture line in the three-dimensional digital model; the method comprises the following steps: extracting a spatial attribute set between a near-end head and neck fracture block and a far-end backbone fracture block in a three-dimensional digital model on the basis of each key bony anatomical mark point in the three-dimensional digital model. The spatial attribute set between the near-end head and neck fracture block and the far-end backbone fracture block in the three-dimensional digital model is based on the spatial attribute set between the near-end head and neck fracture block and the far-end backbone fracture block in the three-dimensional digital model; and the target virtual implantation channel is automatically planned in the three-dimensional digital model in combination with the preset mechanical bearing area parameters so as to be used for navigation planning suggestions, so that the precision of preoperative navigation planning is realized.

Description

Technical Field

[0001] This invention relates to the field of intertrochanteric fracture navigation technology, specifically to an intertrochanteric fracture navigation method and system. Background Technology

[0002] Intertrochanteric fracture is a common type of hip fracture in orthopedic clinics, and it is more common in elderly people with osteoporosis. Because this fracture involves the core area of ​​the proximal femur's mechanical load-bearing structure, improper treatment can easily lead to serious complications such as nonunion, coxa vara deformity, and avascular necrosis of the femoral head, which seriously affect lower limb function and quality of life.

[0003] Currently, the mainstream treatment for intertrochanteric fractures is surgical internal fixation. The core objective is to achieve anatomical reduction and stable fixation of the fracture fragments through precise implantation of internal fixation devices, thereby ensuring early postoperative functional exercise. Commonly used surgical methods include intramedullary fixation and extramedullary fixation. Regardless of the surgical method, the implantation path planning of the internal fixation device is the key factor that determines the success or failure of the surgery. The biomechanical fit of the path directly affects the fixation stability, and the control of the safety boundaries of the path is related to whether the femoral head articular surface, cortical bone, and surrounding blood vessels and nerves are damaged.

[0004] The limitations of existing technologies include at least the following issues: Firstly, the lack of precise 3D digital models for preoperative planning makes it prone to deviations between the planned fracture reduction scheme and internal fixation path and the actual anatomical structure due to blind spots. Secondly, the extraction of key bony anatomical landmarks relies heavily on manual annotation, resulting in low automation. Furthermore, the technology lacks precise quantification capabilities for spatial attributes such as fracture fragment displacement and rotation, which are crucial for preoperative assessment, affecting the reliability of preoperative planning. Additionally, the lack of a standardized preoperative path safety and compatibility assessment system makes it difficult to balance the safety of internal fixation preoperatively. These limitations lead to low efficiency in preoperative planning, significant individual differences, and difficulty in providing reliable preoperative navigation suggestions for precise intraoperative navigation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a navigation method and system for intertrochanteric fractures, which solves the problem that existing technologies lack accurate three-dimensional models and standard assessments, making it difficult to provide reliable preoperative navigation recommendations.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a navigation method for intertrochanteric fractures, comprising the following steps: acquiring medical imaging data of the lateral hip joint region and constructing a three-dimensional digital model; identifying fracture lines and several key bony anatomical landmarks in the three-dimensional digital model; identifying the proximal head and neck fracture fragment and the distal shaft fracture fragment in the three-dimensional digital model based on the fracture lines in the three-dimensional digital model; extracting the spatial attribute set between the proximal head and neck fracture fragment and the distal shaft fracture fragment in the three-dimensional digital model based on each key bony anatomical landmark; and automatically planning a target virtual implantation channel in the three-dimensional digital model based on the spatial attribute set between the proximal head and neck fracture fragment and the distal shaft fracture fragment in the three-dimensional digital model, and in conjunction with preset mechanical bearing area parameters, for navigation planning suggestions.

[0007] Furthermore, the medical imaging data specifically refers to CT volume data, which includes a voxel array constituting a three-dimensional space, and includes the spatial coordinates and CT value corresponding to each voxel.

[0008] Furthermore, the specific steps for constructing the three-dimensional digital model are as follows: preprocessing the CT volume data of the lateral hip joint region; dividing the preprocessed CT volume data of the lateral hip joint region to obtain the bone tissue voxel set of the lateral hip joint region; and reconstructing the three-dimensional surface of the bone tissue voxel set of the lateral hip joint region to generate a three-dimensional digital model.

[0009] Furthermore, the specific steps for identifying fracture lines and several key bony anatomical landmarks in the three-dimensional digital model are as follows: surface gradient enhancement processing is performed on the three-dimensional digital model; pixel segmentation processing is performed on the three-dimensional digital model after surface gradient enhancement processing based on a pre-trained improved U-Net network to identify fracture lines; feature matching processing is performed on the three-dimensional digital model after surface gradient enhancement processing to identify several key bony anatomical landmarks.

[0010] Furthermore, the improved U-Net network includes an input layer, an encoding layer, an attention feature fusion layer, a decoding layer, and an output layer. The specific steps of pixel segmentation processing are as follows: the 3D digital model after surface gradient enhancement is converted to a new format and input into the input layer of the improved U-Net network; the encoding layer extracts shallow detail features and deep semantic features of the hip joint model from the converted 3D digital model and outputs several sets of bone tissue feature maps at different scales; the attention feature fusion layer performs channel and spatial weighting on each set of feature maps output by the encoding layer, filters and strengthens the fine features corresponding to the fracture line, and splices them to generate a fused feature map; the decoding layer uses the fused feature map as a basis, restores the size of the feature map by upsampling, and splices it with the feature map passed across layers from the encoding layer to complete the details and gradually reconstruct the feature map; the output layer maps the feature map reconstructed by the decoding layer into a binary segmentation map, performs morphological closing operation on the segmentation map to fill the small holes in the fracture line region, and extracts the center line of the fracture line based on the skeleton extraction algorithm to complete the fracture line recognition.

[0011] Furthermore, the specific steps for identifying the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the three-dimensional digital model are as follows: Based on the fracture lines in the three-dimensional digital model, construct the bone tissue separation boundary; based on the bone tissue separation boundary, segment the bone tissue voxel set in the three-dimensional digital model to obtain the bone tissue connected region set; based on each key bony anatomical landmark in the three-dimensional digital model, identify the bone tissue connected region set to identify the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the three-dimensional digital model.

[0012] Further, the specific steps for extracting the spatial attribute set between the proximal head and neck fracture fragment and the distal diaphysis fracture fragment in the 3D digital model are as follows: Based on the proximal head and neck fracture fragment and the distal diaphysis fracture fragment in the 3D digital model, construct corresponding local coordinate systems respectively; extract the rigid transformation parameters of the local coordinate system of the proximal head and neck fracture fragment relative to the local coordinate system of the distal diaphysis fracture fragment in the 3D digital model, including translation vectors and rotation matrices; based on the translation vectors, extract the displacement set of the proximal head and neck fracture fragment relative to the distal diaphysis fracture fragment in the 3D digital model; based on the rotation matrix, extract the rotation angle set of the proximal head and neck fracture fragment relative to the distal diaphysis fracture fragment in the 3D digital model.

[0013] Furthermore, the specific steps for automatically planning the target virtual implantation channel in the 3D digital model are as follows: Based on the distal bone fracture fragments in the 3D digital model, a set of path starting points is generated; based on the spatial attribute set between the proximal head and neck fracture fragments and the distal bone fracture fragments in the 3D digital model, and combined with the preset mechanical bearing zone parameters, the path ending point is determined; based on the set of path starting points and the path ending points, several candidate paths are constructed, and the adaptation evaluation value of each candidate path is extracted; based on the adaptation evaluation value of each candidate path, the optimal candidate path is selected and marked as the target virtual implantation channel.

[0014] Furthermore, the specific steps for extracting the adaptation evaluation value of each candidate path are as follows: Based on the three-dimensional digital model and the proximal head and neck fracture fragment and the distal diaphysis fracture fragment therein, extract the evaluation set of each candidate path, including the path length evaluation value, the tip safety distance evaluation value, and the cutting angle evaluation value; input the evaluation set of each candidate path into the pre-established path evaluation model, and analyze the adaptation evaluation value of each candidate path.

[0015] An intertrochanteric fracture navigation system includes: a data acquisition module for acquiring medical imaging data of the lateral hip joint region and constructing a three-dimensional digital model; a landmark recognition module for identifying fracture lines and several key bony anatomical landmarks in the three-dimensional digital model; a fracture fragment identification module for identifying proximal head and neck fracture fragments and distal shaft fracture fragments in the three-dimensional digital model based on fracture lines in the three-dimensional digital model; a fracture fragment attribute extraction module for extracting a set of spatial attributes between the proximal head and neck fracture fragments and the distal shaft fracture fragments in the three-dimensional digital model based on each key bony anatomical landmark; and an implantation channel planning and navigation module for automatically planning a target virtual implantation channel in the three-dimensional digital model based on the set of spatial attributes between the proximal head and neck fracture fragments and the distal shaft fracture fragments, combined with preset mechanical bearing zone parameters, for navigation planning suggestions.

[0016] The present invention has the following beneficial effects:

[0017] (1) This intertrochanteric fracture navigation method can fully restore the details of the hip joint anatomy and fracture pathology by acquiring medical imaging data of the lateral hip joint area and constructing a three-dimensional digital model. By accurately identifying the fracture line and key bony anatomical landmarks, it lays a solid data foundation for the division of fracture fragments and analysis of spatial relationships, ensuring the accuracy of fracture fragment identification. By extracting the spatial attribute set of the proximal head and neck fracture fragment and the distal shaft fracture fragment, it can quantify the core relationships such as the relative position, displacement and rotation between fracture fragments, providing data support for implantation channel planning. At the same time, by combining the preset mechanical bearing area parameters for automatic planning of the target virtual implantation channel, it can ensure that the channel is both adapted to the spatial morphology of the fracture fragment and conforms to the mechanical bearing characteristics of the proximal femur, taking into account both fixation safety and mechanical stability. This achieves the precision of preoperative navigation planning, helps to improve the quality and efficiency of surgical planning, and ensures the surgical effect.

[0018] (2) This intertrochanteric fracture navigation system achieves integrated closed-loop operation from medical image data processing to target virtual implantation channel planning through modular division of labor and full-process collaborative design. The functions of each module are focused and closely connected: the data acquisition module provides a precise three-dimensional model basis for subsequent analysis, the landmark recognition module completes the identification of fracture lines and key anatomical landmarks simultaneously, the fracture block identification module accurately divides the fracture area, the attribute extraction module quantifies the core spatial relationship, and the implantation channel planning and navigation module outputs navigation suggestions in combination with mechanical parameters. The overall architecture takes into account both functional targeting and process continuity, and can quickly adapt to the preoperative planning needs of different types of intertrochanteric fractures, providing surgeons with intuitive and systematic planning support, helping to efficiently formulate personalized surgical plans, and improving the convenience and adaptability of preoperative planning.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] Figure 1 This is a flowchart of a navigation method for intertrochanteric fractures according to the present invention.

[0021] Figure 2 This is a block diagram of an intertrochanteric fracture navigation system according to the present invention. Detailed Implementation

[0022] Please see Figure 1This invention provides a technical solution: a navigation method for intertrochanteric fractures, comprising the following steps: acquiring medical imaging data of the lateral hip joint region, the imaging data at least completely including the acetabulum, femoral head, femoral neck, and proximal femoral shaft, and constructing a three-dimensional digital model of the (hip joint) including fracture fragments, femoral head, femoral neck, greater trochanter, lesser trochanter, and part of the femoral shaft; identifying fracture lines and several key bony anatomical landmarks in the three-dimensional digital model; identifying the proximal head and neck fracture fragment and the distal shaft fracture fragment in the three-dimensional digital model based on the fracture lines in the three-dimensional digital model; extracting the spatial attribute set between the proximal head and neck fracture fragment and the distal shaft fracture fragment in the three-dimensional digital model based on each key bony anatomical landmark; and automatically planning the target virtual implantation channel of the internal fixation device in the three-dimensional digital model based on the spatial attribute set between the proximal head and neck fracture fragment and the distal shaft fracture fragment, combined with preset mechanical load-bearing zone parameters, for navigation planning suggestions.

[0023] Medical imaging data specifically refers to CT volume data, which includes a voxel array that constitutes a three-dimensional space, and includes the spatial coordinates and CT value corresponding to each voxel.

[0024] Specifically, the steps for constructing a three-dimensional digital model are as follows: preprocess the CT volume data of the lateral hip joint region, such as by using Gaussian filtering, median filtering or anisotropic diffusion filtering algorithms to smooth the voxel values ​​in the CT volume data in order to suppress random noise generated during image acquisition, while preserving the edge sharpness of bone tissue as much as possible.

[0025] The preprocessed CT volume data of the lateral hip joint region was segmented to obtain a bone tissue voxel set for the lateral hip joint region. Specifically, an adaptive threshold segmentation algorithm was used for segmentation, and a dynamic threshold was automatically extracted using the Otsu algorithm. Voxel points with CT values ​​≥ the threshold were initially identified as candidate bone tissue voxels. Morphological opening operations were performed using 3×3×3 cubic structuring elements (first erosion to remove small connected regions <50 voxels, then expansion to repair small defects on the bone tissue surface). Using the femoral head center region as the seed point, a region growth algorithm with a growth threshold of ±50 HU was used for purification to obtain a complete bone tissue voxel set containing bone tissue of the acetabulum, femoral head, femoral neck, and trochanter, but excluding soft tissue.

[0026] A 3D surface reconstruction of the bone tissue voxel set in the lateral hip joint region was performed to generate a 3D digital model. Specifically, the improved MarchingCubes algorithm was used to extract the surface of the bone tissue voxel set, with a reconstruction step size of 1 voxel point. By calculating the relationship between the vertex CT value of each voxel cube and the segmentation threshold, the position of the isosurface inside the cube was determined, thereby generating a set of triangular facets. The generated set of triangular facets was then optimized. First, an edge collapse algorithm was used to simplify the model mesh, controlling the number of facets to a preset number (e.g., 500,000-1,000,000), while maintaining the morphology of key anatomical structures of the bone tissue. Then, a Laplacian smoothing algorithm was used to smooth the model surface, with up to 3 iterations and a smoothing coefficient of 0.5, reducing the jagged protrusions on the model surface and making the surface roughness ≤0.05mm, thus generating a 3D digital model.

[0027] In this implementation scheme, the preprocessing stage employs multiple filtering algorithms to effectively suppress random noise in CT data while preserving the sharpness of bone tissue edges to the greatest extent, laying a high-quality data foundation for subsequent processing. Bone tissue voxel set segmentation is combined with adaptive threshold segmentation, morphological operations, and region growth purification to automatically screen candidate bone tissue voxels, remove minor impurities, and repair surface defects, ensuring that the bone tissue voxel set is pure and free from soft tissue interference, and completely covers the target anatomical region. The 3D reconstruction adopts an improved MarchingCubes algorithm and optimization strategy to reduce the surface roughness of the model while maintaining the morphology of key anatomical structures, balancing model accuracy and computational efficiency.

[0028] Specifically, the steps for identifying fracture lines and several key bony anatomical landmarks in a 3D digital model are as follows: Surface gradient enhancement is performed on the 3D digital model, i.e., based on the difference in voxel CT grayscale values, the spatial grayscale gradient amplitude is calculated and normalized, and the gradient features of bone tissue boundaries (such as the cortical bone and fracture lines) are enhanced through contrast enhancement; pixel segmentation is performed on the surface gradient-enhanced 3D digital model using a pre-trained improved U-Net network to identify fracture lines; feature matching is performed on the surface gradient-enhanced 3D digital model to identify several key bony anatomical landmarks, specifically:

[0029] The moving least squares method with a 5×5×5 voxel window is used to smooth and denoise the anatomical features and contours. Geometric features (surface normal vector (PCA algorithm, 20 neighborhood voxels), Gaussian / mean principal curvature (3×3×3 voxel window), shape index with values ​​from -1 to 1) and grayscale features (local CT value mean, variance, grayscale gradient magnitude calculated by Sobel operator) are extracted simultaneously and fused into a 12-dimensional feature vector to construct a global feature point cloud library.

[0030] Pre-defined anatomical morphology templates are constructed based on human hip joint anatomy standards. For key bony anatomical landmarks (including the apex of the greater trochanter, the tip of the lesser trochanter, the center of the femoral head, the proximal endpoint of the femoral neck midline, the distal endpoint of the femoral neck midline, and the apex of the superior lateral margin of the acetabulum), a dedicated anatomical morphology template is created. Each template includes core morphological constraints for the landmark (e.g., the greater trochanter apex template is defined as "the extreme point of surface protrusion, with the normal vector pointing outwards, and the average curvature within a 3mm radius being positive and within a certain range"). 50~150m⁻¹”; the femoral head center template is defined as “the center of a spherical region with a spherical fitting error ≤0.2mm and a local CT value of 800~1200HU”, the feature vector threshold range (based on statistical data from 100 normal hip joint CT cases, such as the gray-scale gradient amplitude threshold of the lesser trochanter tip ≥50HU / voxel) and spatial position association rules (such as the line connecting the two ends of the femoral neck midline must coincide with the femoral head center and the angle between the line and the proximal femoral shaft axis is 120°~135°);

[0031] Feature matching and candidate point selection: The nearest neighbor matching algorithm is used, with the Euclidean distance of the feature vector as the metric and a matching threshold of 0.15 set (cross-validation ensures an accuracy of ≥98%). Candidate points that meet the template constraints are selected. The top 5 candidate points with the smallest Euclidean distance of each marker point are retained. Invalid points are removed by combining the anatomical spatial distance rules to obtain the preliminary positioning coordinates.

[0032] Precise optimization and verification: For each initial coordinate, the K-nearest neighbor surface fitting algorithm is used to fit a quadratic surface (least square method, iteration ≤ 5 times) within a 10×10×10 voxel range. The extreme point or geometric center is taken as the precise coordinate (e.g., the maximum point is taken at the apex of the greater trochanter, and the center of the femoral head is taken as the center of the ball). The Euclidean distance between the precise coordinate and the gold standard is verified to be ≤ 0.3 mm. If it does not meet the requirement, it is re-optimized, and finally a complete and accurate recognition result is output.

[0033] The improved U-Net network consists of an input layer, an encoding layer, an attention feature fusion layer, a decoding layer, and an output layer. The encoding layer contains four encoding units, each of which consists of two 3×3×3 convolutional layers, a BatchNorm layer, a ReLU activation function, and a 2×2×2 max pooling layer (with a stride of 2). The number of convolutional kernels in each encoding unit is 64, 128, 256, and 512, respectively. The attention feature fusion layer uses the CBAM attention mechanism to perform channel attention and spatial attention weighted optimization on the multi-scale feature map output by the encoding layer (first generating channel descriptors through global average pooling, and then calculating channel weights through two fully connected layers). The feature map is then generated through a 7×7 convolutional layer, and the two are multiplied element-wise to obtain an enhanced feature map. The decoding layer contains four decoding units. Each decoding unit upsamples the feature map through a transposed convolutional layer (2×2×2, stride 2), and then concatenates and fuses it with the feature map of the corresponding scale of the encoding layer. This is then processed by two 3×3×3 convolutional layers, a BatchNorm layer, and a ReLU activation function. The final number of convolutional kernels in the decoding units is 256, 128, 64, and 32, respectively. The output layer uses a 1×1×1 convolutional layer to map the feature map into a binary segmentation map (fracture line regions are 1, non-fracture line regions are 0). The specific steps of pixel segmentation are as follows:

[0034] The 3D digital model enhanced by surface gradients undergoes format conversion (e.g., normalization to a 512×512×N tensor, where N is the number of voxel layers along the Z-axis of the model, and CT values ​​are standardized to the [0, 1] interval) and is then input into the input layer of the improved U-Net network. The encoding layer extracts shallow detail features (e.g., cortical texture) and deep semantic features (e.g., overall morphology of bone tissue) of the hip joint model from the format-converted 3D digital model (performing convolution, normalization, activation, and pooling operations), and outputs several sets of bone tissue feature maps at different scales. The attention feature fusion layer performs channel and spatial weighting on each set of feature maps output by the encoding layer, filters and enhances the fine features corresponding to the fracture line (e.g., gray-scale abrupt changes at the cortical bone interruption), and splices them together to generate a fused feature map.

[0035] The decoding layer, based on the fused feature map, restores the feature map size through upsampling and stitches it together with the feature map passed across layers from the encoding layer to complete the details, gradually reconstructing the feature map (the reconstructed size is consistent with the input tensor 512×512×N). The output layer maps the feature map reconstructed by the decoding layer into a binary segmentation map, performs a 3×3×3 morphological closing operation on the segmentation map (which can be iterated once, using cubic structuring elements, first dilution and then erosion, filling small holes of ≤3 voxels) to fill the small holes in the fracture line region, and extracts the fracture line centerline based on the skeleton extraction algorithm (using the Zhang-Suen thinning algorithm, iterating twice to obtain a single-pixel width centerline, retaining endpoints and branches), completing the fracture line recognition (the accuracy is verified by Dice coefficient ≥0.85 and Hausdorff distance ≤1.5mm after recognition).

[0036] The specific steps for identifying the proximal head and neck fracture fragments and the distal bone shaft fracture fragments in the three-dimensional digital model are as follows: Based on the fracture line in the three-dimensional digital model, a bone tissue separation boundary is constructed. Specifically, using the extracted fracture line centerline as a reference, a morphological expansion process (iteration 2 times) is performed on the fracture line using 3×3×3 cubic structural elements to expand the fracture line into a continuous voxel-level separation surface. This separation surface fits the fracture gap inside the bone tissue, forming a bone tissue separation boundary that completely wraps the fracture line and blocks the voxel connection between the proximal and distal bone tissues.

[0037] Based on the bone tissue separation boundary, the set of bone tissue voxels in the three-dimensional digital model is segmented to obtain a set of bone tissue connected regions. Specifically, the constructed bone tissue separation boundary is used as the voxel connectivity blocking condition. The three-dimensional 6-connected domain analysis algorithm is used to traverse the set of bone tissue voxels. Bone tissue voxels that have no connection relationship on both sides of the separation boundary are classified into different regions. Finally, a set of bone tissue connected regions composed of two independent bone tissue connected regions without voxel overlap is obtained.

[0038] Based on each key bony anatomical landmark in the 3D digital model, the set of connected bone tissue regions is identified to mark the proximal head and neck fracture fragment and the distal femoral shaft fracture fragment in the 3D digital model. Specifically, the voxel coordinates of each key bony anatomical landmark are determined to belong to the connected bone tissue region. The connected bone tissue region containing the center of the femoral head, the proximal endpoint of the femoral neck midline, and the distal endpoint of the femoral neck midline is marked as the proximal head and neck fracture fragment. The connected bone tissue region containing the tip of the lesser trochanter and the anatomical axis landmark of the proximal femoral shaft is marked as the distal femoral shaft fracture fragment, thus completing the accurate marking of the two fracture fragments.

[0039] In this implementation plan, multiple technologies are used in synergy to ensure the accuracy and reliability of fracture-related feature recognition and fracture fragment segmentation: surface gradient enhancement strengthens bone tissue boundary features, and an improved U-Net network combined with the CBAM attention mechanism accurately captures fine features of the fracture line. Morphological processing and skeleton extraction further improve the accuracy of fracture line recognition. Key landmark recognition integrates multi-dimensional features and anatomical template matching, combined with surface fitting optimization and accuracy verification to ensure accurate positioning. Based on the fracture line, a separation boundary is constructed, and combined with connected component analysis and landmark attribution judgment, accurate segmentation of proximal head and neck and distal bone shaft fracture fragments is achieved. This provides accurate basic data support for subsequent spatial attribute extraction and implantation channel planning, and improves the automation and accuracy of overall preoperative planning.

[0040] Specifically, the steps for extracting the spatial attribute set between the proximal head and neck fracture fragment and the distal femoral shaft fracture fragment in the 3D digital model are as follows: Based on the proximal head and neck fracture fragment and the distal femoral shaft fracture fragment in the 3D digital model, construct corresponding local coordinate systems respectively, that is: Construct the local coordinate system of the distal femoral shaft fracture fragment: with the tip of the lesser trochanter as the origin; the positive direction of the Z-axis is along the approximate direction of the femoral mechanical axis (the line connecting the tip of the lesser trochanter and the anatomical axis marker of the proximal femoral shaft points towards the femoral head); extract the principal inertial axis of the coronal plane using the PCA algorithm, and take the positive direction of the X-axis as perpendicular to the Z-axis and pointing towards the medial side of the femur; the positive direction of the Y-axis is obtained by the cross product of the X-axis and the Z-axis;

[0041] A local coordinate system for the proximal femoral head and neck fracture fragment is constructed: the origin is the center of the femoral head; the positive direction of the proximal Z-axis is along the mechanical axis of the femoral neck (the line connecting the center of the femoral head and the distal endpoint of the femoral neck midline, pointing towards the femoral neck); the principal inertial axis of the coronal plane is extracted using the PCA algorithm, and the positive direction of the proximal X-axis is taken as perpendicular to the proximal Z-axis and pointing towards the medial side of the femur; the positive direction of the proximal Y-axis is obtained by the cross product of the proximal X-axis and the Z-axis.

[0042] The rigid transformation parameters of the local coordinate system of the proximal head and neck fracture fragment relative to the local coordinate system of the distal femoral shaft fracture fragment in the three-dimensional digital model are extracted, including translation vectors and rotation matrices. Specifically, three sets of homologous paired points are selected from the key bony anatomical landmarks of the two fracture fragments (such as "femoral head center - lesser trochanter tip", "proximal endpoint of femoral neck midline - anatomical axis landmark of proximal femoral shaft", and "distal endpoint of femoral neck midline - extension point of proximal femoral shaft anatomical axis"), and the coordinates of each paired point in the global coordinate system are obtained respectively.

[0043] Calculate the centroid coordinates of the paired landmark groups for distal diaphysis fracture fragments and proximal head and neck fracture fragments, respectively. Then, subtract the centroid coordinates of the corresponding group from the coordinates of each landmark in both groups to perform coordinate decentering and construct the covariance matrix: multiply the decentered proximal head and neck side landmark coordinate vectors with the transposed corresponding coordinate vectors of the distal diaphysis side, sum the results to obtain the covariance matrix, and perform singular value decomposition on it to obtain matrices U, Σ, and V (matrix V needs to be transposed). Solve for the rotation matrix: calculate the determinant of the product matrix of U and the transposed V matrix. If the determinant is 1, directly multiply the V matrix with the transposed U matrix; if the determinant is -1, adjust the sign of the last column of the V matrix before multiplying to obtain a right-handed orthogonal rotation matrix (determinant is 1).

[0044] Translation vector: Subtract the product of the rotation matrix and the centroid coordinates of the paired points of the distal head and neck fracture fragments from the centroid coordinates of the paired points of the proximal head and neck fracture fragments. The result is the translation vector.

[0045] Based on translation vectors, the displacement set of the proximal head and neck fracture fragment relative to the distal shaft fracture fragment in the three-dimensional digital model is extracted in three-dimensional space. This includes the axial shortening distance along the femoral mechanical axis, the medial / lateral displacement distance, and the anterior / posterior displacement distance. That is, with the local coordinate system of the distal shaft fracture fragment as a fixed reference, the Z-axis corresponds to the femoral mechanical axis, the X-axis corresponds to the medial and lateral directions (positive direction medial, negative direction lateral), and the Y-axis corresponds to the anterior and posterior directions (positive direction anterior, negative direction posterior). Translation vector transformation decomposition: The translation vector in the global coordinate system is mapped to this local coordinate system to obtain the X, Y, and Z axis components. Displacement parameters are extracted: The absolute value of the Z-axis component is the axial shortening distance, and a negative value indicates shortening; the X-axis component corresponds to the medial and lateral displacement distance (positive value medial, negative value lateral), and the absolute value is the displacement amplitude; the Y-axis component corresponds to the anterior and posterior displacement distance (positive value anterior, negative value posterior), and the absolute value is the displacement amplitude.

[0046] Based on the rotation matrix, the set of rotation angles of the proximal head and neck fracture fragment relative to the distal diaphysis fracture fragment in the 3D digital model is extracted. This includes varus / valgus angles in the coronal plane, anteversion / posterior tilt angles in the sagittal plane, and internal / external rotation angles in the transverse plane. That is, the calculation benchmark and corresponding plane of the rotation angle are clearly defined. The local coordinate system of the distal diaphysis fracture fragment is used as the fixed benchmark, and the XZ plane of this coordinate system is defined as the coronal plane, the YZ plane as the sagittal plane, and the XY plane as the transverse plane. All rotation angles are defined as the rotation angles of the local coordinate system of the proximal head and neck fracture fragment relative to the local coordinate system of the distal diaphysis fracture fragment, and the calculation process follows the right-hand screw rule.

[0047] Next, the varus / valgus angle on the coronal plane is calculated: First, the elements of the rotation matrix representing the spatial posture relationship between the two fracture fragments are extracted in the corresponding directions on the coronal plane (XZ plane) to form a projection matrix that only reflects the rotation relationship on the coronal plane; then, the angle value is calculated by using the arctangent function through the ratio of the elements at two specific row and column positions in the projection matrix. If the angle value is positive, it means that the proximal head and neck fracture fragment is varus relative to the distal shaft fracture fragment.

[0048] Then, calculate the anterior / posterior tilt angle in the sagittal plane: similarly, first extract the elements in the corresponding direction of the sagittal plane (YZ plane) from the rotation matrix to construct the sagittal projection matrix; then, calculate the angle value using the arctangent function through the ratio of the elements at two specific row and column positions in this projection matrix. If the angle value is positive, it means that the proximal head and neck fracture fragment is tilted anteriorly relative to the distal shaft fracture fragment;

[0049] Next, the internal / external rotation angle on the cross section is calculated: the elements of the rotation matrix in the corresponding direction on the cross section (XY plane) are extracted to form the cross section projection matrix; the angle value is calculated by the arctangent function using the ratio of the elements at two specific row and column positions in the projection matrix. If the angle value is positive, it means that the proximal head and neck fracture fragment has internally rotated relative to the distal shaft fracture fragment.

[0050] The specific steps for automatically planning the target virtual implantation channel in the three-dimensional digital model are as follows: Based on the distal bone fracture fragment of the three-dimensional digital model, a set of path starting points is generated. Specifically, based on the preset fixation safety zone within the distal bone fracture fragment (i.e., the bone tissue area located within the distal bone fracture fragment, 2-5mm from the lateral cortex and 3-6mm from the medial cortex, avoiding the insertion point of the lesser trochanter muscle group and the course of blood vessels and nerves, ensuring the safety and strength of internal fixation), an initial set of starting points is generated within the fixation safety zone using a three-dimensional uniform grid sampling method. A sampling step size of 1mm is set along the X, Y, and Z axes to initially generate starting point coordinates in multiple global coordinate systems.

[0051] Based on the spatial attribute set between the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the 3D digital model, and combined with the preset mechanical bearing area parameters (located within the proximal head and neck fracture fragment, including the dense trabecular bone area of ​​the femoral head pressure, the femoral neck load-bearing core area, and related mechanical constraints), the path endpoint is determined. Specifically, based on the coronal and sagittal parameters of the rotation angle concentration, the direction of the femoral neck mechanical axis is adjusted to adapt to the spatial posture of the fracture fragments to ensure mechanical transmission. Within the proximal head and neck fracture fragment, the overlapping area between the dense trabecular bone area and the load-bearing core area is selected as the candidate endpoint area. If the area of ​​the overlapping area is less than the preset area (e.g., 20 mm²), the radius of the core area is expanded to 10 mm. The geometric center of the candidate area is solved using a weighted average method, with the weights allocated as trabecular bone density (the proportion of voxels with CT value variance ≤ 200 HU², 60%), distance from the femoral head center (30%), and spatial attribute adaptability (rotation direction matching degree, 10%), to obtain the unique endpoint coordinates, and this endpoint is located within the proximal head and neck fracture fragment.

[0052] Based on the set of path start points and path end points, several candidate paths are constructed, and the adaptation evaluation value of each candidate path is extracted. Based on the adaptation evaluation value of each candidate path, the optimal candidate path is selected and marked as the target virtual implantation channel. Specifically, the adaptation evaluation values ​​of each candidate path are sorted in descending order, and the candidate path in the first sequence is taken as the optimal candidate path and marked as the target virtual implantation channel.

[0053] The steps for constructing several candidate paths are as follows: connect all valid starting points in the path starting point set with the unique endpoint one by one with straight lines, and eliminate invalid paths that do not meet the requirements, such as paths that do not penetrate the two fracture fragments, have an angle with the femoral mechanical axis that exceeds the preset mechanical adaptation range (exemplary value 125°-135°), or have an angle with the fracture surface that exceeds the preset shear resistance range (exemplary value 30°-60°), so as to obtain several candidate paths.

[0054] The specific steps for extracting the adaptive evaluation values ​​for each candidate path are as follows: Based on the three-dimensional digital model and the proximal head and neck fracture fragments and distal shaft fracture fragments therein, an evaluation set for each candidate path is extracted, including path length evaluation values, tip safety distance evaluation values, and entry angle evaluation values, specifically:

[0055] Using the global coordinate system of the three-dimensional digital model as a reference, the coordinates of the candidate path start point (located in the safe fixation zone of the distal bone fracture fragment) and the end point (located in the mechanical bearing zone of the proximal head and neck fracture fragment) are read, and the straight-line distance between the two points is calculated using the Euclidean distance formula. This distance is the path length value.

[0056] Extract the tip safety distance value: The tip refers to the endpoint of the candidate path located on the end side of the proximal head and neck fracture fragment (close to the femoral head direction). Calculate the shortest Euclidean distance between this endpoint and the femoral head articular surface and the edge of the bone tissue of the proximal head and neck fracture fragment, and take the minimum value as the tip safety distance value to evaluate the degree of safety avoidance of the path endpoint to the articular surface and bone cortex.

[0057] Extracting the cutting angle value: Using the femoral mechanical axis (defined as the direction of the line connecting the tip of the lesser trochanter of the distal shaft fracture fragment and the center of the femoral head of the proximal head and neck fracture fragment) as the reference axis, obtain the spatial direction vector of the candidate path and the direction vector of the reference axis. Solve the angle between the two vectors using the vector angle calculation formula. This angle is the cutting angle value, which is used to characterize the degree of adaptation between the path and the direction of femoral mechanical transmission.

[0058] The path length, tip safety distance, and approach angle values ​​of each candidate path are standardized and mapped to path length evaluation values, tip safety distance evaluation values, and approach angle evaluation values ​​between 0 and 1. The evaluation set of each candidate path is input into a pre-established path evaluation model to analyze the adaptive evaluation value of each candidate path. The path evaluation model is as follows:

[0059] ;in, , , , The order is number 1 The candidate path's adaptation evaluation value, path length evaluation value, tip safety distance evaluation value, and cut-in angle evaluation value are all included. , , , The parameters are, in order, the path length adjustment coefficient, the tip safety adjustment coefficient, the cutting angle adjustment coefficient, and the cutting angle evaluation threshold stored in the database (in this implementation example, the values ​​can be 0.24, 0.35, 0.41, and 0.55, respectively). , , This represents the total number of candidate paths.

[0060] In this implementation plan, a dedicated local coordinate system is constructed for the two fracture fragments. After decentralization of the selected paired points, a covariance matrix is ​​constructed. The rigid transformation parameters are extracted through singular value decomposition and rotation direction correction to accurately quantify core spatial attributes such as axial shortening, medial / lateral / anteroposterior displacement, and multi-plane rotation angles. The path endpoints are determined based on the fixed safety zone and the mechanical bearing zone. Invalid paths are eliminated through multi-dimensional screening. The optimal channel is selected by combining a standardized evaluation set and a path evaluation model, taking into account both fixation safety and mechanical adaptability. This provides accurate data support for personalized preoperative navigation and improves the scientific nature and reliability of the planning.

[0061] Please see Figure 2 This invention provides a technical solution: an intertrochanteric fracture navigation system, comprising: a data acquisition module for acquiring medical imaging data of the lateral hip joint region and constructing a three-dimensional digital model; a landmark recognition module for identifying fracture lines and several key bony anatomical landmarks in the three-dimensional digital model; a fracture fragment identification module for identifying proximal head and neck fracture fragments and distal shaft fracture fragments in the three-dimensional digital model based on fracture lines in the three-dimensional digital model; a fracture fragment attribute extraction module for extracting a set of spatial attributes between the proximal head and neck fracture fragments and the distal shaft fracture fragments in the three-dimensional digital model based on each key bony anatomical landmark; and an implantation channel planning and navigation module for automatically planning a target virtual implantation channel in the three-dimensional digital model based on the set of spatial attributes between the proximal head and neck fracture fragments and the distal shaft fracture fragments in the three-dimensional digital model, combined with preset mechanical bearing area parameters, for navigation planning suggestions.

[0062] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0063] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A navigation method for intertrochanteric fractures, characterized in that, Includes the following steps: Acquire medical imaging data of the lateral hip joint region and construct a three-dimensional digital model; Identify fracture lines and several key bony anatomical landmarks in a three-dimensional digital model; Based on the fracture lines in the three-dimensional digital model, the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the three-dimensional digital model are identified. Based on each key bony anatomical landmark in the three-dimensional digital model, the spatial attribute set between the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the three-dimensional digital model is extracted. Based on the spatial attribute set between the proximal head and neck fracture fragments and the distal bone shaft fracture fragments in the three-dimensional digital model, and combined with the preset mechanical bearing zone parameters, a target virtual implantation channel is automatically planned in the three-dimensional digital model for navigation planning suggestions.

2. The intertrochanteric fracture navigation method according to claim 1, characterized in that, The medical imaging data specifically refers to CT volume data, which includes a voxel array constituting a three-dimensional space, and includes the spatial coordinates and CT value corresponding to each voxel.

3. The intertrochanteric fracture navigation method according to claim 2, characterized in that, The specific steps for constructing a 3D digital model are as follows: Preprocessing of CT volume data of the contralateral hip joint region; The preprocessed CT volume data of the lateral hip joint region was divided to obtain the bone tissue voxel set of the lateral hip joint region. Three-dimensional surface reconstruction of the bone tissue voxel set in the contralateral hip joint region was performed to generate a three-dimensional digital model.

4. The intertrochanteric fracture navigation method according to claim 1, characterized in that, The specific steps for identifying fracture lines and several key bony anatomical landmarks in a three-dimensional digital model are as follows: Surface gradient enhancement processing is applied to the three-dimensional digital model; A pre-trained improved U-Net network is used to perform pixel segmentation on a 3D digital model after surface gradient enhancement to identify fracture lines. Feature matching was performed on the three-dimensional digital model after surface gradient enhancement to identify several key bony anatomical landmarks.

5. The intertrochanteric fracture navigation method according to claim 4, characterized in that, The improved U-Net network includes an input layer, an encoding layer, an attention feature fusion layer, a decoding layer, and an output layer. The specific steps of pixel segmentation processing are as follows: The three-dimensional digital model enhanced with surface gradients is converted into a new format and then input into the input layer of the improved U-Net network. The encoding layer extracts shallow detail features and deep semantic features of the hip joint model from the format-converted 3D digital model and outputs several sets of bone tissue feature maps at different scales. The attention feature fusion layer performs channel and spatial weighting on each set of feature maps output by the encoding layer, filters and enhances the fine features corresponding to the fracture lines, and splices them to generate a fused feature map. The decoding layer is based on the fused feature map. It restores the size of the feature map by upsampling and splices it with the feature map passed across layers of the encoding layer to complete the details and gradually reconstruct the feature map. The output layer maps the feature map reconstructed by the decoding layer into a binary segmentation map, performs morphological closing operations on the segmentation map to fill the small holes in the fracture line region, and extracts the center line of the fracture line based on the skeleton extraction algorithm to complete the fracture line recognition.

6. The intertrochanteric fracture navigation method according to claim 1, characterized in that, The specific steps for identifying the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the 3D digital model are as follows: Based on the fracture line in the three-dimensional digital model, construct the bone tissue separation boundary; Based on the bone tissue partition boundary, the set of bone tissue voxels in the three-dimensional digital model is segmented to obtain the set of connected regions of bone tissue. Based on each key bony anatomical landmark in the 3D digital model, the set of connected regions of bone tissue is identified to distinguish the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the 3D digital model.

7. The intertrochanteric fracture navigation method according to claim 6, characterized in that, The specific steps for extracting the spatial attribute set between the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the 3D digital model are as follows: Based on the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the three-dimensional digital model, corresponding local coordinate systems are constructed respectively. Extract the rigid transformation parameters of the local coordinate system of the proximal head and neck fracture fragment relative to the local coordinate system of the distal diaphysis fracture fragment in the three-dimensional digital model, including translation vectors and rotation matrices; Based on translation vectors, the displacement set of the proximal head and neck fracture fragments relative to the distal diaphysis fracture fragments in the three-dimensional digital model is extracted; Based on the rotation matrix, the set of rotation angles of the proximal head and neck fracture fragments relative to the distal diaphysis fracture fragments in the three-dimensional digital model is extracted.

8. The intertrochanteric fracture navigation method according to claim 1, characterized in that, The specific steps for automatically planning the target virtual implantation channel in a 3D digital model are as follows: Based on the three-dimensional digital model of distal bone fracture fragments, a set of path starting points is generated; Based on the spatial attribute set between the proximal head and neck fracture fragments and the distal shaft fracture fragments in the three-dimensional digital model, and combined with the preset mechanical bearing zone parameters, the path endpoint is determined. Based on the set of path start points and path end points, several candidate paths are constructed, and the fitness evaluation value of each candidate path is extracted. Based on the adaptation evaluation value of each candidate path, the optimal candidate path is selected and marked as the target virtual implantation channel.

9. The intertrochanteric fracture navigation method according to claim 8, characterized in that, The specific steps for extracting the fitness evaluation value for each candidate path are as follows: Based on the three-dimensional digital model and the proximal head and neck fracture fragments and the distal diaphysis fracture fragments therein, an evaluation set is extracted for each candidate path, including path length evaluation value, tip safety distance evaluation value, and cutting angle evaluation value. The evaluation set of each candidate path is input into a pre-established path evaluation model to analyze the adaptive evaluation value of each candidate path.

10. A navigation system for intertrochanteric fractures, employing the intertrochanteric fracture navigation method according to any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire medical imaging data of the lateral hip joint region and construct a three-dimensional digital model; The landmark recognition module is used to identify fracture lines and several key bony anatomical landmarks in the three-dimensional digital model. The fracture fragment identification module is used to identify proximal head and neck fracture fragments and distal diaphysis fracture fragments in the three-dimensional digital model based on the fracture lines in the three-dimensional digital model. The fracture fragment attribute extraction module is used to extract the spatial attribute set between the proximal head and neck fracture fragments and the distal diaphysis fracture fragments in the three-dimensional digital model based on each key bony anatomical landmark in the three-dimensional digital model. The implantation channel planning and navigation module is used to automatically plan the target virtual implantation channel in the three-dimensional digital model based on the spatial attribute set between the proximal head and neck fracture fragments and the distal bone shaft fracture fragments in the three-dimensional digital model, and in combination with the preset mechanical bearing zone parameters, for navigation planning suggestions.

Images