Preoperative planning method for hip replacement surgery and electronic device

Abstract

This application relates to the field of medical technology and provides a preoperative planning method and electronic device for hip replacement surgery. The method includes: constructing a target three-dimensional model of the target bone tissue based on the multimodal medical imaging data of the target object; obtaining a semantic segmentation mask of the target bone tissue using a preset bone tissue segmentation model; predicting the three-dimensional coordinates of each preset anatomical marker point on the target bone tissue under the constraint of the semantic segmentation mask by a preset heatmap regression model; calculating the femoral cortical index based on the three-dimensional coordinates of the target anatomical marker points, and determining the medullary canal type based on the femoral cortical index; inputting the three-dimensional coordinates of each preset anatomical marker point and the medullary canal type into a preset reinforcement learning model for processing to obtain an initial hip joint prosthesis planning scheme; displaying the target three-dimensional model and the target hip joint prosthesis model virtually installed on the target three-dimensional model in the initial installation pose, which can improve the efficiency and accuracy of preoperative planning for hip replacement surgery.

Description

Technical Field

[0001] This application belongs to the field of medical technology, and in particular relates to a preoperative planning method and electronic equipment for hip replacement surgery. Background Technology

[0002] Hip replacement surgery involves placing a hip prosthesis onto healthy bone in the correct location to replace a diseased hip joint, thereby restoring normal hip function. Before performing hip replacement surgery, it is necessary to plan the type and placement of the hip prosthesis that best suits the patient. This preoperative planning typically involves selecting multiple anatomical landmarks on the pelvis and femur for reference.

[0003] Traditional preoperative planning for hip replacement surgery typically involves surgeons manually comparing X-ray images to determine the type and placement of the hip prosthesis, or manually selecting anatomical landmarks from computed tomography (CT) data. However, manual methods are less accurate and time-consuming, reducing the efficiency and accuracy of preoperative planning for hip replacement surgery. Summary of the Invention

[0004] In view of this, embodiments of this application provide a preoperative planning method and electronic device for hip replacement surgery to solve the technical problems of low efficiency and accuracy of preoperative planning in traditional hip replacement surgery.

[0005] In a first aspect, embodiments of this application provide a preoperative planning method for hip replacement surgery, including: Acquire multimodal medical imaging data of the target object, and construct a target three-dimensional model of the target bone tissue of the target object based on the multimodal medical imaging data; the multimodal medical imaging data includes three-dimensional image data and two-dimensional image data; the target bone tissue includes the pelvis and bilateral femurs; The three-dimensional image data is processed using a preset bone tissue segmentation model to obtain a semantic segmentation mask of the target bone tissue. Under the constraint of the semantic segmentation mask, the three-dimensional coordinates of each preset anatomical marker point on the target bone tissue are predicted by a preset heatmap regression model. The femoral cortical index of the target object is calculated based on the three-dimensional coordinates of the target anatomical markers on both femurs, and the medullary canal type of the target object is determined based on the femoral cortical index; The three-dimensional coordinates of each of the preset anatomical markers and the medullary canal classification are input into a preset reinforcement learning model for processing to obtain an initial hip joint prosthesis planning scheme; the initial hip joint prosthesis planning scheme includes the model and initial installation pose of the target hip joint prosthesis model adapted to the target object; Display the target 3D model and the target hip joint prosthesis model virtually installed on the target 3D model in the initial installation pose.

[0006] In an optional implementation of the first aspect, the initial hip joint prosthesis planning scheme further includes the current values ​​and adjustable ranges of each preset planning parameter; correspondingly, after obtaining the initial hip joint prosthesis planning scheme, the method further includes: Displays the current value and adjustable range of each of the preset planning parameters, and displays the current value of each preset evaluation parameter used to evaluate the installation effect of the target hip joint prosthesis model; In response to the user's adjustment of the current value of any of the preset planning parameters, the installation pose of the target hip joint prosthesis model and the current values ​​of each of the preset evaluation parameters are updated synchronously.

[0007] In one optional implementation of the first aspect, the three-dimensional image data is supine three-dimensional image data; the two-dimensional image data is standing two-dimensional image data; correspondingly, constructing a target three-dimensional model of the target bone tissue of the target object based on the multimodal medical image data includes: Based on the supine three-dimensional image data, an initial three-dimensional surface model of the target bone tissue is constructed when the target object is in a supine position. The pelvic tilt angle of the target object when it is in a standing position is determined based on the standing two-dimensional image data. The initial three-dimensional surface model is rotated based on the pelvic tilt angle to obtain the target three-dimensional model of the target bone tissue.

[0008] In one optional implementation of the first aspect, constructing an initial three-dimensional surface model of the target bone tissue when the target object is in a supine position based on the supine three-dimensional image data includes: The supine 3D image data is processed into a corresponding 3D voxel mesh; the 3D voxel mesh includes a number of voxels, each voxel having N vertices, where N is an integer greater than 3; For each voxel in the three-dimensional voxel mesh, the vertex status code of the voxel is determined according to the relationship between the gray values ​​of the N vertices of the voxel and a preset gray threshold. The vertex status code is an N-bit binary number, and the values ​​of the N bits of the N-bit binary number correspond one-to-one with the N vertices. The value of each bit of the N-bit binary number is used to represent the spatial positional relationship between the corresponding vertex and the target bone tissue. For each target voxel in the three-dimensional voxel mesh whose vertex status code is neither all 0s nor all 1s, the three-dimensional coordinates of each target isopleth point on the isopleth surface within the target voxel are determined based on the vertex status code of the target voxel, and at least one triangular facet corresponding to the target voxel is generated based on the three-dimensional coordinates of all the target isopleth points on the target voxel; the isopleth surface refers to the geometric sectional surface formed inside the target voxel when the surface of the target bone tissue passes through the target voxel, and the target isopleth point refers to the intersection of the isopleth surface and the edge line of the target voxel; The three-dimensional surface model composed of the triangular facets corresponding to all the target voxels is determined as the initial three-dimensional surface model of the target bone tissue.

[0009] In one optional implementation of the first aspect, the three-dimensional coordinates of each target isopleth point on the isopleth surface within the target voxel are determined based on the vertex status code of the target voxel, and at least one triangular facet corresponding to the target voxel is generated based on the three-dimensional coordinates of all the target isopleth points on the target voxel, including: The target topological configuration of the isosurface within the target voxel is determined based on the vertex status code of the target voxel; the target topological configuration is used to describe the geometry of the isosurface. For each target edge line of the target voxel, the three-dimensional coordinates of the target isopleth points on the target edge line are determined based on the three-dimensional coordinates of two vertices on the target edge line, the grayscale values ​​of the two vertices, and the preset grayscale threshold; the target edge line refers to the edge line that intersects with the isopleth surface, and the grayscale value of the target isopleth point is equal to the preset grayscale threshold. According to the connection rules corresponding to the target topology, the target isopleths on each of the target edges are connected to obtain at least one triangular facet corresponding to the target voxel.

[0010] In one optional implementation of the first aspect, the three-dimensional image data is processed using a preset bone tissue segmentation model to obtain a semantic segmentation mask for the target bone tissue, including: The three-dimensional image data is downsampled to a reduced-dimensional three-dimensional image data with isotropic properties of the first size; The reduced-dimensional 3D image data is input into a preset bone tissue segmentation model for processing to obtain a semantic segmentation mask for the target bone tissue.

[0011] In one optional implementation of the first aspect, under the constraint of the semantic segmentation mask, the three-dimensional coordinates of each preset anatomical marker point on the target bone tissue are predicted by a preset heatmap regression model, including: The semantic segmentation mask is used as a spatial attention constraint for a preset heatmap regression model, so that the preset heatmap regression model outputs a three-dimensional Gaussian heatmap corresponding to each preset anatomical marker point on the target bone tissue under the spatial attention constraint; each three-dimensional Gaussian heatmap includes the confidence score and three-dimensional coordinates of multiple points on the target bone tissue, and the confidence score of each point is used to represent the probability that the point belongs to the corresponding preset anatomical marker point; The three-dimensional coordinates of the target point with the highest confidence in each of the three-dimensional Gaussian heatmaps are respectively determined as the three-dimensional coordinates of the corresponding preset anatomical marker points.

[0012] In one optional implementation of the first aspect, the bilateral femurs include the affected femur and the contralateral femur; the target anatomical landmarks on the bilateral femurs are the lesser trochanter landmarks on the affected femur; correspondingly, the femoral cortical index of the target object is calculated based on the three-dimensional coordinates of the target anatomical landmarks on the bilateral femurs, including: At a target position on the affected femur at a first preset distance from the lesser trochanter marker, measure the femoral outer diameter and medullary canal inner diameter of the affected femur; The femoral cortical index of the target object is determined based on the outer diameter of the femur and the inner diameter of the medullary canal.

[0013] In one alternative implementation of the first aspect, determining the medullary canal classification of the target object based on the femoral cortical index includes: If the femoral cortical index is less than the first index threshold, the medullary canal type of the target object is determined to be cylindrical. If the femoral cortical index is greater than the second index threshold, the medullary canal type of the target object is determined to be funnel-shaped; If the femoral cortical index is greater than or equal to the first index threshold and less than or equal to the second index threshold, the medullary canal type of the target object is determined to be between the cylindrical type and the funnel type.

[0014] In a second aspect, embodiments of this application provide an electronic device, including a memory and a computer program stored in the memory and executable on a processor, wherein the processor executes the computer program to implement the steps of the method described in any optional implementation of the first aspect above.

[0015] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in any optional implementation of the first aspect above.

[0016] Fourthly, embodiments of this application provide a computer program product that, when run on an electronic device, causes the electronic device to implement the steps of the method described in any optional implementation of the first aspect.

[0017] The preoperative planning method, electronic device, computer-readable storage medium, and computer program product for hip replacement surgery provided in the embodiments of this application have the following beneficial effects: The preoperative planning method for hip replacement surgery provided in this application constructs a target three-dimensional model of the target bone tissue based on multimodal medical imaging data of the target object. Since multimodal medical imaging data provides more comprehensive information, the constructed target three-dimensional model can intuitively and accurately reproduce the true morphology of the target bone tissue. By processing the three-dimensional image data in the multimodal medical imaging data using a preset bone tissue segmentation model, a semantic segmentation mask for the target bone tissue is obtained. Under the constraint of this semantic segmentation mask, a preset heatmap regression model predicts the three-dimensional coordinates of each preset anatomical marker point on the target bone tissue. This not only limits the region of interest of the preset heatmap regression model to the target bone tissue, narrowing the search range of anatomical marker points and improving the extraction efficiency of anatomical marker points, but also overcomes the errors and time costs caused by manual marking, achieving automated extraction of anatomical marker points and improving the extraction accuracy. The method calculates the femoral cortical index of the target object based on the three-dimensional coordinates of the target anatomical marker points on both femurs, and... The femoral cortical index is used to determine the medullary canal type of the target patient. Since the femoral cortical index can quantify the bone quality represented by the cortical bone thickness of the target patient, and the medullary canal type can characterize the morphological features of the femoral medullary canal, the individual anatomical characteristics and bone quality of the target patient can be transformed into quantitative parameters that can be used for reference. By inputting the three-dimensional coordinates of each preset anatomical marker point and the medullary canal type into a preset reinforcement learning model for processing, an initial hip joint prosthesis planning scheme is obtained. Since the preset reinforcement learning model can automatically output the model and initial installation position of the target hip joint prosthesis that is compatible with the target patient, it can improve the efficiency, accuracy and personalization of preoperative planning for hip replacement surgery, and overcome the problem that traditional preoperative planning is less efficient and accurate because it relies entirely on the doctor's subjective experience. By displaying the target three-dimensional model and the target hip joint prosthesis model virtually installed on the target three-dimensional model in the initial installation position, a visual reference can be provided to the doctor, which can facilitate the doctor to quickly confirm or adjust the planning scheme, thereby reducing surgical risks. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A schematic flowchart illustrating a preoperative planning method for hip replacement surgery provided in this application embodiment; Figure 2 This is a schematic diagram illustrating the implementation process of S12 in a preoperative planning method for hip replacement surgery provided in an embodiment of this application. Figure 3 A schematic flowchart illustrating a preoperative planning method for hip replacement surgery, provided for another embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] The following embodiments are only used to illustrate the technical solutions of this application more clearly, and are therefore only examples and should not be used to limit the scope of protection of this application.

[0021] In the description of the embodiments of this application, the technical terms "comprising," "including," "having," and any variations thereof all mean "including but not limited to," unless otherwise specifically emphasized. In the description of the embodiments of this application, unless otherwise stated, the technical term "multiple" refers to two or more, and the technical terms "at least one" or "one or more" refer to one, two, or more than two. The technical terms "first," "second," etc., are only used to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary / secondary relationship of the indicated technical features. The technical term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0022] This application first provides a preoperative planning method for hip replacement surgery, which can be applied to an electronic device. This electronic device can be, for example, a mobile phone, tablet computer, laptop computer, or desktop computer; this application does not limit the type of electronic device.

[0023] Figure 1 This is a schematic flowchart illustrating a preoperative planning method for hip replacement surgery provided in an embodiment of this application. Figure 1 As shown, the method may include S11 to S15, as detailed below: S11, acquire multimodal medical image data of the target object, and construct a target three-dimensional model of the target bone tissue of the target object based on the multimodal medical image data.

[0024] The target group can be individuals who require hip replacement surgery (such as humans or other animals).

[0025] Multimodal medical imaging data includes at least two different types of medical imaging data.

[0026] Optionally, multimodal medical imaging data may include three-dimensional imaging data and two-dimensional imaging data.

[0027] The three-dimensional imaging data can be supine three-dimensional imaging data, such as supine computed tomography (CT) data. The supine three-dimensional imaging data must include data of the complete pelvis and both femurs (including the affected femur and the contralateral femur).

[0028] Two-dimensional imaging data can be standing two-dimensional imaging data, such as standing X-ray images. Standing two-dimensional imaging data must include data on the bilateral anterior superior iliac spines and the pubic symphysis center.

[0029] It should be understood that because hip replacement surgery typically involves the pelvis and both femurs, preoperative planning for hip replacement surgery requires constructing three-dimensional models of the pelvis and both femurs. Based on this, the target bone tissue for the target patient can include the pelvis and both femurs.

[0030] In one specific implementation, the electronic device can construct a target three-dimensional model of the target bone tissue through the following steps 1.1 to 1.3: Step 1.1: Construct an initial three-dimensional surface model of the target bone tissue when the target object is in a supine position based on the three-dimensional image data of the supine position.

[0031] The initial three-dimensional surface model refers to the three-dimensional surface model directly constructed based on supine three-dimensional image data.

[0032] Optionally, the electronic device can use a moving cube algorithm to construct an initial three-dimensional surface model of the target bone tissue when the target object is in a supine position, based on the supine three-dimensional image data.

[0033] Specifically, step 1.1 may include the following steps 1.11 to 1.14: Step 1.11: Process the supine 3D image data into the corresponding 3D voxel mesh.

[0034] It should be understood that three-dimensional image data usually exists in the form of a sequence of two-dimensional tomographic images, and the header file of three-dimensional image data usually includes spatial location metadata (including inter-slice spacing and pixel spacing) of each two-dimensional tomographic image. Therefore, electronic devices can process the supine three-dimensional image data into a corresponding three-dimensional voxel mesh by parsing the spatial location metadata of each two-dimensional tomographic image in the header file of the supine three-dimensional image data and based on the spatial location metadata of each two-dimensional tomographic image.

[0035] It should also be understood that a 3D voxel mesh comprises several voxels, each of which can have N vertices. Here, N is an integer greater than 3. For example, each voxel can be a cube, in which case each voxel can have 8 vertices.

[0036] Step 1.12: For each voxel in the 3D voxel mesh, determine the vertex status code of the voxel based on the relationship between the gray values ​​of the N vertices of the voxel and the preset gray threshold.

[0037] The preset grayscale threshold is a grayscale boundary value used to distinguish bone tissue from other tissues. The preset grayscale threshold can be set according to actual needs, and this application embodiment does not limit it.

[0038] The vertex status code of a voxel can be used to represent the spatial relationship between the voxel and the target bone tissue. For example, the spatial relationship between the voxel and the target bone tissue can include the following: the voxel is inside the target bone tissue; the voxel is outside the target bone tissue; the surface of the target bone tissue passes through the voxel (i.e., part of the voxel is inside the target bone tissue and another part is outside the target bone tissue).

[0039] Optionally, the vertex status code of a voxel can be an N-bit binary number. Based on this, the values ​​of the N bits in the vertex status code (N-bit binary number) of each voxel can correspond one-to-one with the N vertices of that voxel. The value of each bit in the vertex status code (N-bit binary number) of each voxel can be used to represent the spatial relationship between its corresponding vertex and the target bone tissue. For example, the spatial relationship between the vertex and the target bone tissue can include the following: the vertex is inside the target bone tissue; the vertex is outside the target bone tissue; the vertex is on the surface of the target bone tissue.

[0040] Optionally, for each vertex of each voxel, if the gray value of the vertex is greater than a preset gray value threshold, it means that the vertex is inside the target bone tissue; if the gray value of the vertex is less than the preset gray value threshold, it means that the vertex is outside the target bone tissue; if the gray value of the vertex is equal to the preset gray value threshold, it means that the vertex is on the surface of the target bone tissue.

[0041] In some embodiments, for each vertex of each voxel, if the grayscale value of the vertex is greater than or equal to a preset grayscale threshold, the electronic device can determine the value of the binary bit corresponding to the vertex as 1; if the grayscale value of the vertex is less than the preset grayscale threshold, the electronic device can determine the value of the binary bit corresponding to the vertex as 0.

[0042] In other embodiments, for each vertex of each voxel, if the grayscale value of the vertex is greater than or equal to a preset grayscale threshold, the electronic device can determine the value of the binary bit corresponding to the vertex as 0; if the grayscale value of the vertex is less than the preset grayscale threshold, the electronic device can determine the value of the binary bit corresponding to the vertex as 1.

[0043] Optionally, for each voxel, after determining the values ​​of the binary bits corresponding to the N vertices of the voxel, the electronic device can determine the N-bit binary number obtained by combining the values ​​of the binary bits corresponding to the N vertices of the voxel as the vertex status code of the voxel.

[0044] Step 1.13: For each target voxel in the 3D voxel mesh whose vertex status code is neither all 0 nor all 1, determine the 3D coordinates of each target isopleth point on the isopleth surface within the target voxel based on the vertex status code of the target voxel, and generate at least one triangular facet corresponding to the target voxel based on the 3D coordinates of all target isopleth points on the target voxel.

[0045] In this context, a vertex status code that is neither all 0s nor all 1s means that the vertex status code is neither (0,0,0,0,0,0,0,0) nor (1,1,1,1,1,1,1,1,1). For example, if the vertex status code of a voxel is (0,0,0,0,1,1,0,1), it means that the vertex status code of that voxel is neither all 0s nor all 1s, i.e., that voxel is the target voxel. It should be understood that if the vertex status code of a voxel is all 0s, it means that the voxel is outside the target bone tissue; if the vertex status code of a voxel is all 1s, it means that the voxel is inside the target bone tissue; if the vertex status code of a voxel is neither all 0s nor all 1s, it means that the surface of the target bone tissue passes through the voxel, i.e., there is an isosurface within the voxel corresponding to the surface of the target bone tissue.

[0046] The isosurface refers to the geometric cross-section formed inside the target voxel when the surface of the target bone tissue passes through it. It is used to separate the target bone tissue from other tissues. It should be understood that the gray value of each point on the isosurface (i.e., the isopoint) is equal to a preset gray value threshold.

[0047] Target iso-points refer to the intersections of isosurfaces and the edges of target voxels. It should be understood that regardless of how the surface of the target bone tissue crosses the target voxel, the surface of the target bone tissue (i.e., the isosurface) will intersect at least 3 edges of the target voxel, meaning that each isosurface within a target voxel contains at least 3 target iso-points.

[0048] Optionally, for each target voxel, the electronic device may use the following steps 1.131 to 1.133 to determine the three-dimensional coordinates of each target isopleth point on the isosurface within the target voxel: Step 1.131: Determine the target topological configuration of the isosurface within the target voxel based on the vertex status code of the target voxel.

[0049] The target topological configuration describes the geometry of the isosurface. The geometry of the isosurface can be, for example, a triangle, quadrilateral, or pentagon. It should be understood that the topological configurations of isosurfaces within target voxels with different vertex status codes may be different or the same; this embodiment does not limit this.

[0050] In some embodiments, the electronic device may store a correspondence between multiple vertex status codes (not all 0s or all 1s) and multiple isosurface topological configurations. Based on this, for each target voxel, the electronic device may, based on the vertex status code of the target voxel, query the topological configuration of the isosurface corresponding to the vertex status code of the target voxel from the above correspondence, and determine the topological configuration as the target topological configuration of the isosurface within the target voxel.

[0051] Step 1.132: For each target edge line of the target voxel, determine the three-dimensional coordinates of the target isopleth points on the target edge line based on the three-dimensional coordinates of the two vertices on the target edge line, the grayscale values ​​of the two vertices, and the preset grayscale threshold.

[0052] Among them, the target edge line of the target voxel refers to the edge line on the target voxel that intersects with the isosurface.

[0053] It should be understood that the grayscale value of the target isopleth is equal to the preset grayscale threshold.

[0054] Specifically, for each target edge of a target voxel, the electronic device can determine the three-dimensional coordinates of the target isopleth points on the target edge using the following formula (1) based on the three-dimensional coordinates of the two vertices of the target edge, the grayscale values ​​of the two vertices, and a preset grayscale threshold: P = P A +[( T - V A ) / ( V B -V A )]•( P B - P A ); formula (1) in, P These are the three-dimensional coordinates of the target contour points on the target's edge. P A and P B These are the three-dimensional coordinates of two vertices on the edge of the target. V A and V B These are the grayscale values ​​of two vertices on the edge of the target, respectively. T This is the preset grayscale threshold.

[0055] Step 1.133: According to the connection rules corresponding to the target topology, connect the target isopleths on each target edge to obtain at least one triangular facet corresponding to the target voxel.

[0056] It should be understood that the connection rules for triangular facets differ depending on the topological configuration. Electronic devices can store the correspondence between multiple isosurface topological configurations and the connection rules for multiple triangular facets.

[0057] Based on this, for each target voxel, the electronic device can look up the connection rule corresponding to the target topology configuration from the correspondence based on the target topology configuration of the isosurface within the target voxel.

[0058] Step 1.14: The three-dimensional surface model composed of triangular facets corresponding to all target voxels is determined as the initial three-dimensional surface model of the target bone tissue.

[0059] It should be understood that the initial three-dimensional surface model of the target bone tissue is composed of several triangular facets. After the electronic device determines the triangular facets corresponding to all target voxels, the three-dimensional surface model naturally formed by these triangular facets is the initial three-dimensional surface model of the target bone tissue.

[0060] Step 1.2: Determine the pelvic tilt angle of the target object when it is in a standing position based on the two-dimensional image data of the standing position.

[0061] It should be understood that since the standing two-dimensional image data includes at least the data of the bilateral anterior superior iliac spines and the center of the pubic symphysis, the electronic device can first determine the midpoint of the bilateral anterior superior iliac spines in the standing two-dimensional image data, and then determine the angle between the line connecting the midpoint and the center of the pubic symphysis and the target vertical line as the pelvic tilt angle of the target object when it is in a standing position.

[0062] The target vertical line refers to the line that passes through the center of the pubic symphysis and is perpendicular to the horizontal plane.

[0063] Step 1.3: Rotate the initial three-dimensional surface model based on the pelvic tilt angle to obtain the target three-dimensional model of the target bone tissue.

[0064] Optionally, the standing two-dimensional image data may also include data on both acetabula.

[0065] Based on this, the electronic device can apply a rotation to the initial three-dimensional surface model of the target bone tissue with the line connecting the centers of the two acetabula as the axis of rotation based on the pelvic tilt angle of the target object when it is in a standing position, and can determine the rotated initial three-dimensional surface model as the target three-dimensional model of the target bone tissue.

[0066] S12, the three-dimensional image data is processed using a preset bone tissue segmentation model to obtain a semantic segmentation mask of the target bone tissue, and under the constraint of the semantic segmentation mask, the three-dimensional coordinates of each preset anatomical marker point on the target bone tissue are predicted by a preset heatmap regression model.

[0067] The preset bone tissue segmentation model refers to a pre-trained bone tissue segmentation model. For example, the bone tissue segmentation model can be a three-dimensional U-shaped convolutional neural network (U-Net) model, or other types of convolutional network models. This application embodiment does not limit the type of bone tissue segmentation model.

[0068] Optionally, the preset bone tissue segmentation model can be trained by the electronic device using a deep learning algorithm based on a first sample dataset. The first sample dataset may include several first sample data points, each of which can consist of a 3D image (or reduced-dimensional 3D image) of a sample object and its semantic segmentation mask for the target bone tissue. Therefore, when training the bone tissue segmentation model, the electronic device can use the 3D image data of the sample object in each first sample data point as the input to the bone tissue segmentation model, and the semantic segmentation mask in each first sample data point as the output of the bone tissue segmentation model. This allows the bone tissue segmentation model to learn the correspondence between the 3D image data of various objects and their semantic segmentation masks during training. After the bone tissue segmentation model is trained, the electronic device can define the trained bone tissue segmentation model as the preset bone tissue segmentation model.

[0069] Optionally, when training the bone tissue segmentation model, the electronic device can use the DICE coefficient corresponding to the Sorenson-Dice loss function to evaluate the training effect of the bone tissue segmentation model. For example, when the DICE coefficient is greater than or equal to a first coefficient threshold (e.g., 0.95), the electronic device can confirm that the bone tissue segmentation model has achieved the expected training effect, at which point the training of the bone tissue segmentation model can be stopped, and the currently trained bone tissue segmentation model can be identified as the preset bone tissue segmentation model. Alternatively, the electronic device can continue to train the bone tissue segmentation model when the DICE coefficient is less than the first coefficient threshold until the expected training effect is achieved (i.e., the DICE coefficient is greater than or equal to the first coefficient threshold).

[0070] Optionally, to improve the efficiency of semantic segmentation mask acquisition and thus the preoperative planning efficiency of hip replacement surgery, the electronic device can store the preset bone tissue segmentation model in local memory after training, so that it can be quickly retrieved from local memory for subsequent use. Of course, in other embodiments, to avoid occupying local storage space, the electronic device can also store the preset bone tissue segmentation model in a connected external storage device or server, and retrieve it from the external memory or server when needed.

[0071] Optionally, the preset bone tissue segmentation model may include a 3D encoder and a 3D decoder connected sequentially between its input and output ends. The 3D encoder can be used to extract features from the 3D image data input to the preset bone tissue segmentation model to obtain deep semantic features of the 3D image data. The 3D decoder can be used to upsample these deep semantic features layer by layer to obtain a semantic segmentation mask for the target bone tissue.

[0072] For example, a 3D encoder may include one or more 3D encoding networks, each of which may include M×M×M encoding convolutional kernels. Correspondingly, a 3D decoder may also include one or more 3D decoding networks, each of which may include M×M×M decoding convolutional kernels. The total number of 3D encoding networks in the 3D encoder and the total number of 3D decoding networks in the 3D decoder may be equal. M is an integer greater than 1, and the value of M can be set according to actual needs; this embodiment does not limit its value.

[0073] Optionally, in order to reduce the data processing volume of the preset bone tissue segmentation model, improve the efficiency of semantic segmentation mask acquisition, and further improve the preoperative planning efficiency of hip replacement surgery, the electronic device can first downsample the three-dimensional image data of the target object to a dimensionally reduced three-dimensional image data with isotropic first size, and then input the dimensionally reduced three-dimensional image data into the preset bone tissue segmentation model for processing to obtain the semantic segmentation mask of the target bone tissue.

[0074] Isotropy is used to describe that the dimensions of the reduced 3D image data are the same in all directions (i.e., the X-axis, Y-axis, and Z-axis). Based on this, isotropy as the first dimension specifically means that the dimensions of the reduced 3D image data in all directions are the first dimension. The first dimension can be set according to actual needs, and the embodiments of this application do not limit its value. For example, the first dimension can be 4 millimeters (mm).

[0075] The preset heatmap regression model refers to a pre-trained heatmap regression model. For example, the heatmap regression model can be a convolutional neural network model based on heatmap regression. The type of convolutional neural network model can be set according to actual needs; this application embodiment does not limit its type.

[0076] Optionally, the preset heatmap regression model can be trained by the electronic device using a deep learning algorithm based on a second sample dataset. The second sample dataset can include several sets of second sample data, each set consisting of a semantic segmentation mask of the target bone tissue of a sample object, 3D image data, and the 3D coordinates of various preset anatomical markers on the target bone tissue. Therefore, when training the preset heatmap regression model, the electronic device can use the semantic segmentation mask and 3D image data of the target bone tissue of each sample object in each set of second sample data as input to the heatmap regression model, and use the 3D coordinates of various preset anatomical markers on the target bone tissue of each sample object in each set of second sample data as output. This allows the heatmap regression model to learn the correspondence between the semantic segmentation mask and 3D image data of the target bone tissue of various objects and the 3D coordinates of various preset anatomical markers on the target bone tissue during training. After the heatmap regression model is trained, the electronic device can define the trained heatmap regression model as the preset heatmap regression model.

[0077] Optionally, to improve the efficiency of acquiring the three-dimensional coordinates of preset anatomical markers and thus further improve the preoperative planning efficiency of hip replacement surgery, the electronic device can store the preset heatmap regression model in local memory after training, so that it can be quickly retrieved from local memory for subsequent use. Of course, in other embodiments, to avoid occupying local storage space, the electronic device can also store the preset heatmap regression model in a connected external storage device or server, and retrieve it from the external memory or server when needed.

[0078] It should be understood that, under the constraints of the semantic segmentation mask, the region of interest (i.e., attention) of the pre-defined heatmap regression model will be limited to the target bone tissue.

[0079] Optionally, the semantic segmentation mask can be a 3D binarized model of the same size as the 3D image data. Each voxel in the semantic segmentation mask has a value of either 0 or 1. For example, a voxel value of 1 in the semantic segmentation mask indicates that the corresponding voxel in the 3D image data is located on the target bone tissue; a voxel value of 0 indicates that the corresponding voxel in the 3D image data is located on other tissues and not on the target bone tissue.

[0080] Optionally, preset anatomical landmarks on the target bone tissue can be selected according to actual needs. For example, preset anatomical landmarks may include lesser trochanter landmarks on both femurs, anterior superior iliac spine landmarks on both femurs, pubic bone landmarks on both femurs, central landmarks of both acetabulums on both femurs, lower edge landmarks of both femoral teardrops on both femurs, central landmarks of both femoral heads on both femurs, and reference landmarks for the axis of the medullary canal on both femurs.

[0081] In some embodiments, predicting the three-dimensional coordinates of various preset anatomical markers on the target bone tissue using a preset heatmap regression model in S12 may include, for example: Figure 2 S121 to S122 are described in detail below: S121, the semantic segmentation mask is used as the spatial attention constraint of the preset heatmap regression model, so that the preset heatmap regression model outputs the three-dimensional Gaussian heatmap corresponding to each preset anatomical marker point on the target bone tissue under the spatial attention constraint.

[0082] It should be understood that the purpose of using the semantic segmentation mask of the target bone tissue as a spatial attention constraint for the preset heatmap regression model is to limit the region of interest of the preset heatmap regression model to the target bone tissue, thereby reducing the computational load of the preset heatmap regression model while improving the extraction accuracy of preset anatomical markers.

[0083] Each preset anatomical marker's 3D Gaussian heatmap includes the confidence scores and 3D coordinates of multiple points on the target bone tissue (i.e., multiple points surrounding the preset anatomical marker). The confidence score of each point can be used to represent the probability that the point belongs to the corresponding preset anatomical marker.

[0084] S122, the three-dimensional coordinates of the target point with the highest confidence in each three-dimensional Gaussian heatmap are determined as the three-dimensional coordinates of the corresponding preset anatomical marker point.

[0085] It should be understood that the target point with the highest confidence in the three-dimensional Gaussian heat map of each preset anatomical marker is most likely to be the preset anatomical marker. Therefore, for each preset anatomical marker's three-dimensional Gaussian heat map, the electronic device can determine the three-dimensional coordinates of the target point with the highest confidence in the three-dimensional Gaussian heat map of the preset anatomical marker as the three-dimensional coordinates of the preset anatomical marker.

[0086] In practical applications, if the confidence level of the point with the highest confidence level in the 3D Gaussian heat map of a certain preset anatomical marker is less than the preset confidence threshold (i.e., the confidence level of all points in the 3D Gaussian heat map is less than the preset confidence threshold), it indicates that the accuracy of the 3D Gaussian heat map predicted by the electronic device for the preset anatomical marker is low. In this case, the user can manually correct the 3D coordinates of the preset anatomical marker.

[0087] S13, calculate the femoral cortical index of the target object based on the three-dimensional coordinates of the target anatomical markers on both femurs, and determine the medullary canal type of the target object based on the femoral cortical index.

[0088] The target anatomical landmark on both femurs can be the lesser trochanter landmark on the affected femur.

[0089] Based on this, optionally, the electronic device can calculate the femoral cortical index of the target object through the following steps 2.1 to 2.2: Step 2.1: At the target position on the affected femur at a first preset distance from the lesser trochanter marker, measure the outer diameter of the femur and the inner diameter of the medullary canal.

[0090] The first preset distance can be set according to actual needs. This application embodiment does not limit it. For example, the first preset distance can be 10 centimeters (cm).

[0091] Specifically, the target location can refer to the location reached after extending a first preset distance from the lesser trochanter marker point toward the distal femur.

[0092] Step 2.2: Determine the femoral cortical index of the target subject based on the femoral outer diameter and medullary canal inner diameter.

[0093] Specifically, the electronic device can calculate the femoral cortical index of the target object using the following formula (2) based on the femoral outer diameter and medullary canal inner diameter at the target location: FCI =( D - d ) / D ;Formula (2) in, FCI For the target femoral cortex index, D The outer diameter of the femur at the target location. d This is the inner diameter of the medullary cavity at the target location.

[0094] Optionally, after determining the femoral cortical index of the target object, the electronic device can determine the medullary canal classification of the target object based on the following steps 2.3 to 2.5: Step 2.3: If the femoral cortical index of the target object is less than the first index threshold, the medullary canal type of the target object is determined to be straight.

[0095] Step 2.4: If the femoral cortical index of the target object is greater than the second index threshold, the medullary canal type of the target object is determined to be funnel-shaped.

[0096] Step 2.5: If the femoral cortical index of the target object is greater than or equal to the first index threshold and less than or equal to the second index threshold, the medullary canal type of the target object is determined to be between the cylindrical type and the funnel type.

[0097] Both the first and second exponential thresholds can be determined according to actual conditions, and the embodiments of this application do not limit them. For example, the first exponential threshold can be 0.5 and the second exponential threshold can be 0.6.

[0098] It should be understood that if the femoral cortical index of the target subject is less than the first index threshold, it indicates that the target subject's cortical bone is thin and there may be a risk of osteoporosis; if the femoral cortical index of the target subject is greater than the second index threshold, it indicates that the target subject's cortical bone is thick and the bone quality is good; if the femoral cortical index of the target subject is greater than or equal to the first index threshold and less than or equal to the second index threshold, it indicates that the target subject's cortical bone and bone quality are relatively normal.

[0099] S14. Input the three-dimensional coordinates of each preset anatomical marker and the medullary canal classification into the preset reinforcement learning model for processing to obtain the initial hip joint prosthesis planning scheme.

[0100] Here, the preset reinforcement learning model refers to a pre-trained reinforcement learning model. For example, the reinforcement learning model can be a hierarchical reinforcement learning model or other types of reinforcement learning models. This application embodiment does not limit the type of reinforcement learning model.

[0101] Optionally, the preset reinforcement learning model can be trained by the electronic device using a deep learning algorithm based on a third sample dataset. The third sample dataset may include several sets of third sample data, each set consisting of the 3D coordinates of various preset anatomical markers, medullary canal classification, and an initial hip joint prosthesis planning scheme for a sample object. Therefore, when training the reinforcement learning model, the electronic device can use the 3D coordinates and medullary canal classification of each preset anatomical marker in each set of third sample data as input to the reinforcement learning model, and the initial hip joint prosthesis planning scheme from each set of third sample data as output. This allows the reinforcement learning model to learn the correspondence between the 3D coordinates and medullary canal classification of various preset anatomical markers and their initial hip joint prosthesis planning schemes during training. After the reinforcement learning model is trained, the electronic device can define the trained reinforcement learning model as the preset reinforcement learning model.

[0102] Optionally, when training a reinforcement learning model, the electronic device can evaluate the training effect of the model based on a preset three-level reward function. The three-level reward function can include a first-level reward function, a second-level reward function, and a third-level reward function. The first-level reward function can be used for safety assessment, the second-level reward function can be used for force line assessment, and the third-level reward function can be used for functional assessment.

[0103] For example, the first-level reward function can be represented by the following formula (3): R 1= w cov •max(0,( Cov -0.8) / 0.05)+ w fit • FitScore ;Formula (3) in, R 1 represents the first evaluation value corresponding to the first-level reward function. w cov The first safety weight corresponding to the bone coverage of the acetabular cup model is... Cov For the bone coverage of the acetabular cup model, w fit The second safety weight is the score corresponding to the matching degree between the affected femoral stem model and three preset points within the medullary canal of the affected femur. FitScore The matching degree between the affected femoral stem model and three preset points within the medullary canal of the affected femur is scored.

[0104] The bone coverage of the acetabular cup model can be used to describe the degree to which the acetabular cup model is covered by the affected acetabular bone tissue in the target bone tissue. It should be understood that the greater the bone coverage of the acetabular cup model, the larger the contact area between the acetabular cup model and the affected acetabular bone tissue, and the higher the installation stability of the acetabular cup model.

[0105] The three preset points within the medullary canal of the affected femur can be set according to actual needs. For example, these three points may include a first point on the medial side of the proximal end of the medullary canal of the affected femur, a second point on the lateral side of the proximal end of the medullary canal of the affected femur, and a third point on the isthmus of the distal end of the medullary canal of the affected femur.

[0106] The matching score described above can be used to describe the degree of fit between the affected femoral stem model and the medullary canal wall of the affected femur at the three preset points. It should be understood that the higher the matching score, the tighter the fit between the affected femoral stem model and the medullary canal wall at the three preset points.

[0107] The second-level reward function can be expressed by the following formula (4): R 2= w offset •max(0,(5-Δo ffset ) / 5)+ w leg •max(0,(5-Δ leg ) / 5); formula (4) in, R 2 represents the second evaluation value corresponding to the second-level reward function, Δo. ffset The eccentricity deviation of the bilateral femurs, Δ leg For the bilateral lower limb length deviation of the target object, w offset The weight of the first force line corresponding to the eccentricity deviation of the bilateral femurs. w leg The weight of the second force line corresponding to the bilateral lower limb length deviation.

[0108] Bilateral femoral eccentricity deviation refers to the difference between the eccentricity of the affected femur and the eccentricity of the contralateral femur, which can be used to describe the degree of symmetry of the bilateral hip joint eccentricity of the target object.

[0109] Bilateral lower limb length deviation refers to the difference in length between the affected lower limb and the contralateral lower limb, and it can be used to describe the degree of symmetry in the bilateral lower limb length of a target object.

[0110] The third-level reward function can be expressed by the following formula (5): R 3= w ant •max(0,(60-| CA -40|) / 20); formula (5) in, R 3 represents the third evaluation value corresponding to the third-level reward function. w ant To determine the functional weights corresponding to the combined pitch angle, CA This refers to the combined forward tilt angle.

[0111] The combined anteversion angle can be the sum of the anteversion angle of the acetabular cup model and the anteversion angle of the femoral stem model, and it can be used to describe the overall anteversion state of the hip joint prosthesis model.

[0112] The first safety weight, the second safety weight, the first force line weight, the second force line weight, and the functional weight mentioned above can all be set according to actual needs, and the embodiments of this application do not limit them.

[0113] Specifically, electronic devices can evaluate the training effect of reinforcement learning models based on a weighted sum of the first evaluation value corresponding to the first-level reward function, the second evaluation value corresponding to the second-level reward function, and the third evaluation value corresponding to the third-level reward function. The first weighting coefficient corresponding to the first evaluation value can be significantly larger than the second weighting coefficient corresponding to the second evaluation value, and the second weighting coefficient corresponding to the second evaluation value can be significantly larger than the third weighting coefficient corresponding to the third evaluation value. This allows the reinforcement learning model to prioritize safety during training; that is, after ensuring that the safety of the prosthesis implantation meets the minimum requirements, force line optimization is performed, and functional optimization is then carried out based on the optimized force line.

[0114] Optionally, the electronic device can confirm that the reinforcement learning model has achieved the expected training effect when the above weighted sum meets the preset requirements. At this time, the training of the reinforcement learning model can be stopped, and the reinforcement learning model obtained from the current training can be determined as the preset reinforcement learning model.

[0115] Optionally, to improve the efficiency of obtaining the initial hip joint prosthesis planning scheme, and thus further improve the preoperative planning efficiency of hip replacement surgery, the electronic device can store the preset reinforcement learning model in local memory after training, so that it can be quickly retrieved from local memory for subsequent use. Of course, in other embodiments, to avoid occupying local storage space, the electronic device can also store the preset reinforcement learning model in a connected external storage device or server, and retrieve it from the external memory or server when needed.

[0116] In some embodiments, the initial hip prosthesis planning scheme may include the model and initial installation position of the target hip prosthesis model adapted to the target object.

[0117] S15 displays the target 3D model and the target hip joint prosthesis model virtually installed on the target 3D model in the initial installation pose.

[0118] In some embodiments, when the initial hip prosthesis planning scheme includes the model and initial installation pose of a target hip prosthesis model adapted to the target object, the electronic device can display the target 3D model and the target hip prosthesis model virtually installed on the target 3D model in the initial installation pose.

[0119] Optionally, the display interface of the electronic device may include a view area. Based on this, the electronic device can display the target 3D model and the target hip joint prosthesis model virtually mounted on the target 3D model in the view area of ​​the display interface.

[0120] In other embodiments, the initial hip prosthesis planning scheme may include, in addition to the model and initial installation pose of the target hip prosthesis model adapted to the target object, the current values ​​and adjustable ranges of various preset planning parameters. Based on this, as... Figure 3 As shown, the preoperative planning methods for hip replacement surgery may also include S16-S17, which are detailed below: S16 displays the current value and adjustable range of each preset planning parameter, and displays the current value of each preset evaluation parameter used to evaluate the installation effect of the target hip joint prosthesis model.

[0121] For example, the preset planning parameters may include preset acetabular cup model planning parameters and preset femoral stem model planning parameters. The preset acetabular cup model planning parameters may include the model type, insertion depth, abduction angle, and / or anteversion angle of the acetabular cup model. The preset femoral stem model planning parameters may include the model type and / or insertion depth of the femoral stem model.

[0122] For example, preset evaluation parameters may include the abduction angle of the acetabular cup model, the anteversion angle of the acetabular cup model, the bone coverage of the acetabular cup model, the combined anteversion angle, the bilateral lower limb length deviation and / or the bilateral femoral eccentricity deviation, etc.

[0123] Optionally, the electronic device's display interface may also include a parameter adjustment area. Based on this, the electronic device can display the current value and adjustable range of each preset planning parameter in the parameter adjustment area of ​​the display interface.

[0124] Optionally, the electronic device's display interface may also include a parameter evaluation area. Based on this, the electronic device can display the current values ​​of various preset evaluation parameters used to assess the installation effect of the target hip joint prosthesis model in the parameter evaluation area of ​​the display interface.

[0125] In practical applications, after the electronic device displays the adjustable range of each preset planning parameter, the user can adjust the current value of any preset planning parameter within its adjustable range according to actual needs. For example, the electronic device's display interface can be configured with adjustment controls for adjusting the current value of each preset planning parameter. These controls can be, for example, slider controls or numeric input controls. The user can adjust the current value of the preset planning parameter using these controls.

[0126] S17, in response to the user's adjustment of the current value of any preset planning parameter, synchronously update the installation pose of the target hip joint prosthesis model and the current values ​​of each preset evaluation parameter.

[0127] It should be understood that when the current value of any preset planning parameter changes, the current values ​​of some or all preset evaluation parameters will also change accordingly.

[0128] It should be understood that after the electronic device synchronously updates the installation pose of the target hip joint prosthesis model and the current values ​​of each preset evaluation parameter, it will display the target hip joint prosthesis model with the updated installation pose and the current values ​​of each preset evaluation parameter.

[0129] In some embodiments, the electronic device's display interface may also be configured with a planning report generation control. Based on this, when the user believes that the current values ​​of the preset evaluation parameters meet the expected requirements, they can obtain a planning report by clicking the planning report generation control. When the electronic device detects that the planning report generation control has been clicked, it can generate and output a final planning report based on the basic information of the target object, the model number of the target hip joint prosthesis, and the current values ​​of the preset planning parameters (i.e., the values ​​when the planning report generation control is clicked). The final planning report may include the basic information of the target object, the model number of the target hip joint prosthesis, and the current values ​​of the preset planning parameters.

[0130] As can be seen from the above, the embodiments of this application construct a target three-dimensional model of the target bone tissue based on the multimodal medical image data of the target object. Since multimodal medical image data can provide more comprehensive information, the constructed target three-dimensional model can intuitively and accurately restore the real morphology of the target bone tissue. By using a preset bone tissue segmentation model to process the three-dimensional image data in the multimodal medical image data, a semantic segmentation mask of the target bone tissue is obtained. Under the constraint of this semantic segmentation mask, the three-dimensional coordinates of each preset anatomical marker on the target bone tissue are predicted by a preset heatmap regression model. This not only limits the region of interest of the preset heatmap regression model to the target bone tissue, narrows the search range of anatomical markers, and improves the extraction efficiency of anatomical markers, but also overcomes the errors and time costs caused by manual marking, realizes the automated extraction of anatomical markers, and improves the extraction accuracy of anatomical markers. By calculating the femoral cortical index of the target object based on the three-dimensional coordinates of the target anatomical markers on both femurs, and based on the femoral cortical index... The method determines the medullary canal type of the target patient. Since the femoral cortical index can quantify the bone quality represented by the cortical bone thickness of the target patient, and the medullary canal type can characterize the morphological features of the femoral medullary canal, the individual anatomical features and bone quality of the target patient can be transformed into quantitative parameters for reference. By inputting the three-dimensional coordinates of each preset anatomical marker point and the medullary canal type into a preset reinforcement learning model for processing, an initial hip joint prosthesis planning scheme is obtained. Since the preset reinforcement learning model can automatically output the model and initial installation position of the target hip joint prosthesis that are adapted to the target patient, it can improve the efficiency, accuracy and personalization of preoperative planning for hip replacement surgery, overcoming the problem of low planning efficiency and accuracy caused by traditional preoperative planning relying entirely on the doctor's subjective experience. By displaying the target three-dimensional model and the target hip joint prosthesis model virtually installed on the target three-dimensional model in the initial installation position, a visual reference can be provided to the doctor, making it easier for the doctor to quickly confirm or adjust the planning scheme, thereby reducing surgical risks.

[0131] It is understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0132] This application also provides an electronic device. Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4As shown, the electronic device may include: a processor 40, a memory 41, and a computer program 42 stored in the memory 41 and executable on the processor 40, such as a program corresponding to a preoperative planning method for hip replacement surgery. When the processor 40 executes the computer program 42, it implements the steps in the above-described embodiments of the preoperative planning method for hip replacement surgery, for example... Figure 1 S11~S15 are shown.

[0133] For example, computer program 42 may be divided into one or more modules / units, one or more of which are stored in memory 41 and executed by processor 40 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 42 in electronic device 4.

[0134] Those skilled in the art will understand that Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or combine certain components, or use different components.

[0135] The processor 40 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0136] The memory 41 can be an internal storage unit of the electronic device 4, such as a hard disk or RAM. The memory 41 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, or flash card. Furthermore, the memory 41 can include both internal and external storage units of the electronic device 4. The memory 41 is used to store computer programs and other programs and data required by the electronic device. The memory 41 can also be used to temporarily store data that has been output or will be output.

[0137] This application also provides a computer-readable storage medium that can store a computer program, which, when executed by a processor, can implement the steps in the above method embodiments.

[0138] This application also provides a computer program product that, when run on an electronic device, enables the electronic device to perform the steps described in the above method embodiments.

[0139] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, refer to the relevant descriptions of other embodiments.

[0140] It should be noted that, unless otherwise specified, all technical terms used in the embodiments of this application have the same meaning as commonly understood by those skilled in the art to which this application belongs. The technical terms used in the embodiments of this application are only used to explain specific embodiments of this application and are not intended to limit this application.

[0141] The term "embodiment" as used in the description of embodiments in this application means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0142] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0143] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A preoperative planning method for hip replacement surgery, characterized in that, include: Acquire multimodal medical image data of the target object, and construct a target three-dimensional model of the target bone tissue of the target object based on the multimodal medical image data; The multimodal medical imaging data includes three-dimensional imaging data and two-dimensional imaging data; the target bone tissue includes the pelvis and bilateral femurs; The three-dimensional image data is processed using a preset bone tissue segmentation model to obtain a semantic segmentation mask of the target bone tissue. Under the constraint of the semantic segmentation mask, the three-dimensional coordinates of each preset anatomical marker point on the target bone tissue are predicted by a preset heatmap regression model. The femoral cortical index of the target object is calculated based on the three-dimensional coordinates of the target anatomical markers on both femurs, and the medullary canal type of the target object is determined based on the femoral cortical index; The three-dimensional coordinates of each of the preset anatomical markers and the medullary canal classification are input into a preset reinforcement learning model for processing to obtain an initial hip joint prosthesis planning scheme. The initial hip joint prosthesis planning scheme includes the model and initial installation pose of the target hip joint prosthesis model adapted to the target object, and the initial hip joint prosthesis planning scheme also includes the current value and adjustable range of each preset planning parameter. Display the target 3D model and the target hip joint prosthesis model virtually installed on the target 3D model in the initial installation pose; Displays the current value and adjustable range of each of the preset planning parameters, and displays the current value of each preset evaluation parameter used to evaluate the installation effect of the target hip joint prosthesis model; In response to the user's adjustment of the current value of any of the preset planning parameters, the installation pose of the target hip joint prosthesis model and the current values ​​of each of the preset evaluation parameters are updated synchronously.

2. The method according to claim 1, characterized in that, The three-dimensional image data is supine three-dimensional image data; the two-dimensional image data is standing two-dimensional image data; correspondingly, a target three-dimensional model of the target bone tissue of the target object is constructed based on the multimodal medical image data, including: Based on the supine three-dimensional image data, an initial three-dimensional surface model of the target bone tissue is constructed when the target object is in a supine position. The pelvic tilt angle of the target object when it is in a standing position is determined based on the standing two-dimensional image data. The initial three-dimensional surface model is rotated based on the pelvic tilt angle to obtain the target three-dimensional model of the target bone tissue.

3. The method according to claim 2, characterized in that, Based on the supine three-dimensional image data, an initial three-dimensional surface model of the target bone tissue when the target object is in a supine position is constructed, including: The supine three-dimensional image data is processed into a corresponding three-dimensional voxel mesh; the three-dimensional voxel mesh includes a number of voxels, each of which has N vertices, where N is an integer greater than 3; For each voxel in the three-dimensional voxel mesh, the vertex status code of the voxel is determined according to the relationship between the gray values ​​of the N vertices of the voxel and a preset gray threshold. The vertex status code is an N-bit binary number, and the values ​​of the N bits of the N-bit binary number correspond one-to-one with the N vertices. The value of each bit of the N-bit binary number is used to represent the spatial positional relationship between the corresponding vertex and the target bone tissue. For each target voxel in the three-dimensional voxel mesh whose vertex status code is neither all 0s nor all 1s, the three-dimensional coordinates of each target isopleth point on the isopleth surface within the target voxel are determined based on the vertex status code of the target voxel, and at least one triangular facet corresponding to the target voxel is generated based on the three-dimensional coordinates of all the target isopleth points on the target voxel; the isopleth surface refers to the geometric sectional surface formed inside the target voxel when the surface of the target bone tissue passes through the target voxel, and the target isopleth point refers to the intersection of the isopleth surface and the edge line of the target voxel; The three-dimensional surface model composed of the triangular facets corresponding to all the target voxels is determined as the initial three-dimensional surface model of the target bone tissue.

4. The method according to claim 3, characterized in that, The three-dimensional coordinates of each target isopleth point on the isopleth surface within the target voxel are determined based on the vertex status code of the target voxel, and at least one triangular facet corresponding to the target voxel is generated based on the three-dimensional coordinates of all the target isopleth points on the target voxel, including: The target topological configuration of the isosurface within the target voxel is determined based on the vertex status code of the target voxel; the target topological configuration is used to describe the geometry of the isosurface. For each target edge line of the target voxel, the three-dimensional coordinates of the target isopleth points on the target edge line are determined based on the three-dimensional coordinates of two vertices on the target edge line, the grayscale values ​​of the two vertices, and the preset grayscale threshold; the target edge line refers to the edge line that intersects with the isopleth surface, and the grayscale value of the target isopleth point is equal to the preset grayscale threshold. According to the connection rules corresponding to the target topology, the target isopleths on each of the target edges are connected to obtain at least one triangular facet corresponding to the target voxel.

5. The method according to any one of claims 1-4, characterized in that, The three-dimensional image data is processed using a preset bone tissue segmentation model to obtain a semantic segmentation mask for the target bone tissue, including: The three-dimensional image data is downsampled to a reduced-dimensional three-dimensional image data with isotropic properties of the first size; The reduced-dimensional 3D image data is input into a preset bone tissue segmentation model for processing to obtain a semantic segmentation mask for the target bone tissue.

6. The method according to any one of claims 1-4, characterized in that, Under the constraints of the semantic segmentation mask, the three-dimensional coordinates of each preset anatomical marker point on the target bone tissue are predicted using a preset heatmap regression model, including: The semantic segmentation mask is used as a spatial attention constraint for a preset heatmap regression model, so that the preset heatmap regression model outputs a three-dimensional Gaussian heatmap corresponding to each preset anatomical marker point on the target bone tissue under the spatial attention constraint; each three-dimensional Gaussian heatmap includes the confidence score and three-dimensional coordinates of multiple points on the target bone tissue, and the confidence score of each point is used to represent the probability that the point belongs to the corresponding preset anatomical marker point; The three-dimensional coordinates of the target point with the highest confidence in each of the three-dimensional Gaussian heatmaps are respectively determined as the three-dimensional coordinates of the corresponding preset anatomical marker points.

7. The method according to any one of claims 1-4, characterized in that, The bilateral femurs include the affected femur and the contralateral femur; the target anatomical landmarks on the bilateral femurs are the lesser trochanter landmarks on the affected femur; correspondingly, the femoral cortical index of the target object is calculated based on the three-dimensional coordinates of the target anatomical landmarks on the bilateral femurs, including: At a target position on the affected femur at a first preset distance from the lesser trochanter marker, measure the femoral outer diameter and medullary canal inner diameter of the affected femur; The femoral cortical index of the target object is determined based on the outer diameter of the femur and the inner diameter of the medullary canal.

8. The method according to any one of claims 1-4, characterized in that, Determining the medullary canal classification of the target subject based on the femoral cortical index includes: If the femoral cortical index is less than the first index threshold, the medullary canal type of the target object is determined to be cylindrical. If the femoral cortical index is greater than the second index threshold, the medullary canal type of the target object is determined to be funnel-shaped; If the femoral cortical index is greater than or equal to the first index threshold and less than or equal to the second index threshold, the medullary canal type of the target object is determined to be between the cylindrical type and the funnel type.

9. An electronic device, characterized in that, It includes a memory and a computer program stored in the memory and executable on a processor, wherein the processor, when executing the computer program, implements the steps of the method as described in any one of claims 1-8.

Images