Hip joint anatomical landmark point recognition method and system

Abstract

The application relates to a hip joint anatomical landmark point recognition method and system, which comprises the following steps: loading three-dimensional mask data containing femur and hip bone segmentation labels, extracting a bone surface point cloud and converting the same to a physical coordinate system; automatically screening a complete data side by comparing the number of bilateral point clouds; based on the complete femur point cloud, sequentially recognizing the femoral head center, lesser trochanter, femoral neck isthmus center and femoral shaft center point through geometric fitting and extreme value search; based on the same side hip bone point cloud, recognizing the acetabular fossa center, ischial tuberosity, pubic symphysis and anterior superior iliac spine point; finally, converting the landmark points to a three-dimensional model standard coordinate system and generating a visual model. The application realizes automatic and high-precision recognition of key anatomical landmark points of the hip joint, does not require manual intervention and a large amount of training data, has strong robustness, can be directly compatible with a three-dimensional surgical planning system, and significantly improves preoperative planning efficiency and accuracy.

Description

Technical Field

[0001] This application relates to the technical field of medical image processing, and in particular to a method and system for identifying anatomical landmarks of the hip joint. Background Technology

[0002] Total hip arthroplasty (THA) is the gold standard surgical treatment for end-stage hip arthritis, avascular necrosis of the femoral head, and developmental dysplasia of the hip. Its core objective is to reconstruct the anatomical structure and biomechanical function of the hip joint by implanting a prosthesis, thereby alleviating pain and restoring limb mobility. In the preoperative planning stage of THA, accurate identification of key anatomical landmarks of the hip joint (such as the femoral head center, acetabular fossa center, femoral neck isthmus, and ischial tuberosity) is a crucial prerequisite for subsequent prosthesis selection, setting of the acetabular cup abduction and anteversion angles, and planning of the femoral stem implantation depth and rotation angle. This directly determines the rationality of the prosthesis placement, thus affecting postoperative hip joint stability, long-term prosthesis survival rate, and the patient's postoperative limb function recovery. Currently, the acquisition of hip joint anatomical landmarks in clinical practice still mainly relies on manual annotation by physicians. This involves physicians manually locating and measuring landmarks based on two-dimensional DICOM tomographic images or three-dimensional reconstruction models, relying on clinical experience. This method has many unavoidable drawbacks in clinical application and is severely disconnected from the core clinical needs of THA preoperative planning.

[0003] 1. Significant subjective differences: The annotation results of different doctors vary significantly, affecting the consistency and reproducibility of surgical plans.

[0004] 2. Inefficient: Manual annotation is time-consuming and cannot meet the needs of rapid clinical diagnosis.

[0005] 3. Lack of three-dimensional spatial information: Annotation based on two-dimensional images cannot fully utilize the spatial information of three-dimensional anatomical structures, resulting in limited planning accuracy.

[0006] 4. Low degree of automation: Most existing automated identification methods are based on a specific coordinate system or a single algorithm, which is poorly adaptable to incomplete data (such as unilateral lesions) and has insufficient compatibility with commonly used three-dimensional data formats in clinical practice.

[0007] In existing technologies, such as CN119006465A, the automatic identification, segmentation, and measurement of the hip joint space are achieved through Faster R-CNN and an improved UNet network with an attention mechanism, and the condition is graded according to the BASRI-h scoring system. However, this method has several drawbacks: it relies on two-dimensional images and cannot obtain three-dimensional anatomical information, resulting in incomplete spatial measurements; the data processing flow is lengthy and the steps are tightly coupled, making it easy for errors to accumulate at multiple stages; the algorithm is highly dependent on image quality and annotation accuracy, and its generalization ability is limited in actual clinical images with artifacts, low contrast, or positional variations; it does not consider individual anatomical differences and pathological diversity, resulting in insufficient adaptability; the introduction of complex network modules leads to a heavy computational burden, which is not conducive to real-time deployment; and the grading only relies on a single imaging score and does not integrate multi-dimensional clinical information, thus limiting its value in assisting diagnosis. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for identifying anatomical landmarks of the hip joint, which partially solves or alleviates the above-mentioned deficiencies in the prior art and enables fully automatic and high-precision identification of key anatomical landmarks of the hip joint.

[0009] To solve the aforementioned technical problems, the present invention specifically adopts the following technical solution:

[0010] A first aspect of the present invention is to provide a method for identifying anatomical landmarks of the hip joint, comprising the following steps:

[0011] Load 3D mask data containing segmentation labels of the bones to be processed, and obtain 3D voxel data and corresponding spatial transformation parameters of the 3D mask data; the bones to be processed include the femur and hip bone;

[0012] The moving cube algorithm is used to extract the point cloud of the bone surface from the three-dimensional mask data, and the spatial transformation parameters are used to transform the voxel coordinate system of the point cloud to the medical image physical coordinate system.

[0013] The number of point clouds on the left and right sides of the skeleton is counted. Based on the ratio of the number of point clouds on the left side to the number of point clouds on the right side, the point clouds on the surface of the skeleton are divided into unilateral complete point clouds and / or unilateral incomplete point clouds.

[0014] The left / right femoral point cloud is obtained from the unilateral complete point cloud, and the femoral anatomical landmarks are identified, including the femoral head center point, lesser trochanter point, femoral neck isthmus center point and femoral shaft center point.

[0015] The left / right hip bone point cloud is obtained from the unilateral complete point cloud, and the hip anatomical landmarks are identified, including the center point of the acetabulum, the ischial tuberosity point, the pubic symphysis point, and the anterior superior iliac spine point.

[0016] The identified femoral and hip anatomical landmarks are transformed from the medical imaging physical coordinate system to the standard coordinate system of the three-dimensional model, and corresponding three-dimensional visualization models are generated.

[0017] Furthermore, the spatial transformation parameters include the affine matrix and the voxel spacing.

[0018] Furthermore, the step of extracting bone surface point clouds from the 3D mask data using the traveling cube algorithm and transforming the point cloud voxel coordinate system to the medical image physical coordinate system using the spatial transformation parameters includes:

[0019] For each bone segmentation label defined in the 3D mask data, extract the binarized mask data from its corresponding 3D mask region;

[0020] Based on the binarized mask data, all voxel units in the three-dimensional voxel data are traversed, and the crossing state of the isosurface is determined according to the comparison result of the corner value of each voxel unit with the first preset threshold.

[0021] Based on the crossing state of the isosurface, a triangular mesh representing the bone surface is generated by table lookup and linear interpolation.

[0022] Extract all vertices of the triangular mesh representing the bone surface to form the surface point cloud of the bone;

[0023] Affine matrices are used to transform each point in the surface point cloud of the skeleton from the point cloud voxel coordinate system to the medical image physical coordinate system.

[0024] Furthermore, the method of dividing the bone surface point cloud into unilateral complete point clouds and / or unilateral incomplete point clouds based on the ratio of the number of point clouds on the left bone to the number of point clouds on the right bone includes:

[0025] If the ratio of the number of point clouds on the side with fewer points to the number of point clouds on the side with more points is less than a second preset threshold, then the side with fewer points is determined to be a single-sided incomplete point cloud.

[0026] If the ratio of the number of point clouds on the side with fewer points to the number of point clouds on the side with more points is greater than or equal to the second preset threshold, then both point clouds are determined to be single-sided complete point clouds.

[0027] Furthermore, the identification of femoral anatomical landmarks includes:

[0028] The proximal femoral candidate region is determined based on the maximum Z-coordinate of the femoral point cloud. The point cloud within the proximal femoral candidate region is then fitted with a least-squares sphere, and the center of the fitted sphere is taken as the center point of the femoral head.

[0029] Based on the left and right sides of the femur, the medial direction vector is determined. In the femur point cloud, the Z and Y direction ranges are delineated with the center point of the femoral head as a reference. Within the range, the point with the largest projection along the medial direction vector is found and taken as the lesser trochanter point.

[0030] The greater trochanter point is identified, which is the point with the largest projection along the lateral direction of the proximal femur. The cervical axis direction is defined based on the femoral head center point, lesser trochanter point, and greater trochanter point. Multi-position sampling is performed along the cervical axis direction, and slice point clouds perpendicular to the cervical axis are extracted at each sampling position. The average distance from each slice point cloud to its geometric center is calculated. The slice position with the smallest average distance is selected, and the point cloud of the slice position with the smallest average distance is fitted with a least-squares circle. The center of the fitted circle is taken as the center point of the femoral neck isthmus.

[0031] Using the Z-coordinate of the small rotor point as a reference, at least two target cross-sectional heights on the femoral shaft are determined. Point clouds of the femoral shaft cross-section are extracted near each target cross-sectional height. The point clouds of the cross-section are fitted with least squares circles, and the center of the fitted circle is taken as the center point of the femoral shaft at the corresponding position.

[0032] Furthermore, when defining the cervical axis direction, the line connecting the center point of the femoral head with the midpoints of the greater trochanter and lesser trochanter is used as the initial cervical axis direction vector, and the initial cervical axis direction vector is normalized.

[0033] Furthermore, the step size for multi-position sampling along the neck axis is between 2mm and 5mm, and the total number of sampling positions is between 10 and 30.

[0034] Furthermore, the identification of anatomical landmarks of the hip bone includes:

[0035] Using the identified femoral head center point on the same side as the reference point, select points in the hip bone point cloud that are less than the third preset threshold to form a candidate point set. Perform spherical fitting on the candidate point set and use the center of the fitted sphere as the center point of the acetabulum.

[0036] In the hip bone point cloud, the point with the smallest Z coordinate is directly selected as the ischial tuberosity node;

[0037] Calculate the Z-coordinate range of the hip bone point cloud and determine the intermediate height value. Divide the point cloud into upper and lower parts using the intermediate height value as the boundary. Select the point with the largest Y-coordinate in the lower part of the point cloud as the pubic symphysis point, and select the point with the largest Y-coordinate in the upper part of the point cloud as the anterior superior iliac spine point.

[0038] Furthermore, generating the corresponding 3D visualization model includes:

[0039] A spherical mesh model with a predetermined radius is generated, centered on the transformed coordinates of each anatomical landmark, and output in STL format.

[0040] Secondly, this application also discloses a hip joint anatomical landmark recognition system, the system comprising:

[0041] The mask data loading module is configured to load three-dimensional mask data containing segmentation labels of the bones to be processed, and to acquire the three-dimensional voxel data of the three-dimensional mask data and the corresponding spatial transformation parameters; the bones to be processed include the femur and hip bone;

[0042] The bone surface point cloud extraction module is configured to extract bone surface point clouds from the three-dimensional mask data using the traveling cube algorithm, and to transform the point cloud voxel coordinate system to the medical image physical coordinate system using the spatial transformation parameters.

[0043] The complete point cloud determination module is configured to count the number of point clouds on the left and right sides of the skeleton. Based on the ratio of the number of point clouds on the left side to the number of point clouds on the right side, the point clouds on the surface of the skeleton are divided into unilateral complete point clouds and / or unilateral incomplete point clouds.

[0044] The femoral anatomical landmark recognition module is configured to acquire left / right femoral point clouds from a unilateral complete point cloud and identify femoral anatomical landmarks, including the femoral head center point, lesser trochanter point, femoral neck isthmus center point, and femoral shaft center point.

[0045] The hip anatomical landmark recognition module is configured to acquire the left / right hip point cloud from a unilateral complete point cloud and identify the hip anatomical landmarks, which include the center point of the acetabulum, the ischial tuberosity point, the pubic symphysis point, and the anterior superior iliac spine point.

[0046] The marker output module is configured to convert the identified femoral and hip anatomical markers from the medical image physical coordinate system to the standard coordinate system of the three-dimensional model and generate the corresponding three-dimensional visualization model.

[0047] Beneficial technical effects:

[0048] 1. This application directly processes high-precision surface point clouds through a geometric fitting algorithm based on anatomical rules (such as least squares sphere / circle fitting), achieving high recognition accuracy comparable to professional manual annotation, effectively overcoming the problems of large subjective differences and insufficient utilization of three-dimensional information in traditional methods.

[0049] 2. This application introduces an intelligent judgment mechanism for data integrity based on relative proportions and combines multiple parametric recognition steps based on dynamic anatomical features. It can robustly address common clinical challenges such as individual anatomical size differences, unilateral lesions, and incomplete data, ensuring that the recognition results remain reliable even under non-ideal data conditions.

[0050] 3. This application transforms the output results into a standard coordinate system and generates a visual model in STL format that can be directly used in a three-dimensional surgical planning system. This achieves seamless integration with existing clinical workflows and provides intuitive and precise quantitative tools for prosthesis selection, angle planning, and guide plate design, ultimately helping to improve surgical accuracy and patient prognosis.

[0051] 4. This application constructs a fully automated processing closed loop from 3D mask data input to standard format 3D model output by adopting the moving cube algorithm to achieve sub-voxel precision bone surface extraction, achieving intelligent judgment of data integrity based on relative proportion threshold, and combining anatomical prior geometric fitting and extreme value localization methods, which significantly improves the efficiency and automation level of clinical preoperative planning. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. The elements or parts in the drawings are not necessarily drawn to scale. Obviously, the drawings described below are some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0053] Figure 1 This is a flowchart of a method for identifying anatomical landmarks of the hip joint according to this application.

[0054] Figure 2 A schematic diagram of the projection of the identification results onto the coronal plane.

[0055] Figure 3 This is a schematic diagram of the module structure of a hip joint anatomical landmark recognition system in one embodiment of this application. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0057] In this document, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and have no specific meaning in themselves. Therefore, "module," "part," or "unit" may be used interchangeably.

[0058] In this document, the terms "upper," "lower," "inner," "outer," "front," "rear," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the present invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0059] In this document, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0060] In this document, "and / or" includes any and all combinations of one or more of the listed related items.

[0061] In this article, "multiple" means two or more, that is, it includes two, three, four, five, etc.

[0062] Figure 1 A flowchart of a method for identifying anatomical landmarks of the hip joint according to this application is shown. (Refer to...) Figure 1 The method specifically includes the following steps:

[0063] S101. Load the 3D mask data containing the segmentation labels of the bones to be processed, and obtain the 3D voxel data of the 3D mask data and the corresponding spatial transformation parameters; wherein, the bones to be processed include the femur and hip bone.

[0064] In some embodiments, spatial transformation parameters include the affine matrix and voxel spacing.

[0065] In one specific implementation, this step involves reading a pre-segmented 3D medical image mask file of the hip joint from an external storage medium or data transfer interface. The 3D mask data is typically stored in the NIfTI format, a common format in medical imaging, and its file defines discrete label values ​​representing different bone structures. Specifically, the bones to be processed include the left femur, right femur, left hip bone, and right hip bone. Each bone structure is assigned a unique and specific integer label value in the mask data; for example, label value 1 can be assigned to the left femur, label value 2 to the right femur, label value 3 to the left hip bone, and label value 4 to the right hip bone.

[0066] During data loading, the system or processor needs to read not only the 3D voxel data matrix that constitutes the 3D mask, but also the spatial transformation parameters closely associated with this data. The 3D voxel data matrix is ​​a 3D array, where each element corresponds to the bone label of that voxel location. The spatial transformation parameters are crucial, defining how to map integer index positions in the voxel coordinate system to a world coordinate system with physical dimensions (usually in millimeters). This set of parameters mainly includes a 4×4 affine transformation matrix and the actual spacing of the voxels in the three dimensions. The affine transformation matrix fully expresses the comprehensive transformation relationships of rotation, scaling, shearing, and translation, and is the mathematical foundation for all subsequent geometric calculations to convert from discrete voxel space to continuous physical space.

[0067] Performing this step using a professional medical image processing library (such as the nibabel library in a Python environment) ensures accurate parsing of the NIfTI file header information and extraction of the three-dimensional voxel data matrix and affine transformation matrix. This loaded data and parameters provide the necessary, formatted input for subsequent surface reconstruction, coordinate transformation, and precise anatomical geometric analysis.

[0068] S102. The moving cube algorithm is used to extract the point cloud of the bone surface from the three-dimensional mask data, and the point cloud voxel coordinate system is transformed to the medical image physical coordinate system using spatial transformation parameters.

[0069] In some embodiments, the core objective of this step is to transform the discrete, voxelized 3D segmentation mask into a continuous, high-precision 3D point cloud representation of the bone surface and place it in a coordinate system with clear physical meaning, laying the foundation for subsequent accurate identification of anatomical landmarks based on geometric features.

[0070] In some embodiments, the method of extracting bone surface point clouds from 3D mask data using the traveling cube algorithm and transforming the point cloud voxel coordinate system to the medical image physical coordinate system using spatial transformation parameters includes:

[0071] For each bone segmentation label defined in the 3D mask data, extract the binarized mask data from its corresponding 3D mask region;

[0072] Based on the binarized mask data, all voxel units in the three-dimensional voxel data are traversed, and the crossing state of the isosurface is determined according to the comparison result of the corner value of each voxel unit with the first preset threshold.

[0073] Based on the crossing state of the isosurface, a triangular mesh representing the bone surface is generated by table lookup and linear interpolation.

[0074] Extract all vertices of the triangular mesh representing the bone surface to form the surface point cloud of the bone;

[0075] Affine matrices are used to transform each point in the surface point cloud of the skeleton from the point cloud voxel coordinate system to the medical image physical coordinate system.

[0076] In one specific implementation, a binary mask is first generated for each specific bone label (e.g., left femur) in the loaded 3D voxel data matrix. The binary mask is a 3D Boolean matrix of the same size as the original data, where positions with voxel values ​​corresponding to specific labels are marked as "1," indicating that the voxel belongs to the target bone; other positions are marked as "0." Next, the generated binary mask is subjected to isosurface extraction using the traveling cube algorithm. This algorithm is a classic 3D scalar field isosurface triangulation method, the process of which includes: systematically traversing all cube cells in the 3D voxel data; determining the possible orientation of the isosurface within the cell based on the binary state (0 or 1) at the eight corner points of the cell by searching a predefined topological configuration table; calculating the coordinates of the intersection points between the isosurface and the cell edges using linear interpolation; and finally, connecting these intersection points to form a triangular mesh. The set of vertices of all triangular meshes constitutes the preliminary surface point cloud of the bone in voxel coordinates, with coordinates still based on integer indices from the voxel mesh.

[0077] However, point clouds in voxel coordinates only contain discrete grid position information, lacking true physical dimensions and spatial orientation, and cannot be directly used for geometric calculations based on physical quantities such as distance and angle. Therefore, it is necessary to utilize the spatial transformation parameters obtained in step S101, especially the affine transformation matrix, to transform the point cloud coordinates from the voxel coordinate system to the standard medical imaging physical coordinate system (usually the RAS coordinate system: right-front-top). This transformation simultaneously considers the anisotropic spacing of voxels, the orientation of the image, and possible offsets, ensuring that the transformed point cloud can accurately reflect the actual shape, size, and position of the bone in the patient's anatomical space. Through this step, the original discrete segmentation data is transformed into a bone surface point cloud with sub-voxel accuracy in continuous physical space.

[0078] S103. Count the number of point clouds of the left and right bones. Based on the ratio of the number of point clouds of the left bone to the number of point clouds of the right bone, divide the point clouds on the bone surface into unilateral complete point clouds and / or unilateral incomplete point clouds.

[0079] In some embodiments, classifying the point cloud on the bone surface into unilateral complete point clouds and / or unilateral incomplete point clouds based on the ratio of the number of point clouds on the left bone to the number of point clouds on the right bone includes:

[0080] If the ratio of the number of point clouds on the side with fewer points to the number of point clouds on the side with more points is less than the second preset threshold, then the side with fewer points is determined to be a single-sided incomplete point cloud.

[0081] If the ratio of the number of point clouds on the side with fewer points to the number of point clouds on the side with more points is greater than or equal to the second preset threshold, then both point clouds are determined to be single-sided complete point clouds.

[0082] In one specific implementation, this step aims to automatically assess the integrity status of the bilateral skeleton in the input data, intelligently identify and handle data incompleteness caused by common clinical situations (such as unilateral lesions, insufficient scanning range, or failure of the segmentation algorithm on one side), thereby significantly enhancing the clinical applicability and robustness of the entire identification method.

[0083] In practice, the system performs point cloud counts on the surface point cloud sets representing the left skeleton (including the left femur and left hip bone) and the right skeleton (including the right femur and right hip bone). The point cloud count refers to the total number of all three-dimensional points constituting the surface model of that side of the skeleton after extraction using the traveling cubes algorithm and transformation to the physical coordinate system. After the count is completed, the system calculates the ratio between the total number of point clouds on the left and right sides of the skeleton. To quantify this, an empirical threshold is pre-set; in this embodiment, this threshold is preferably 20%, but its specific value can be adjusted based on the statistical characteristics of the clinical dataset.

[0084] Based on the calculated ratio, the system executes the following classification logic: If the number of point clouds on one side of the skeleton (e.g., the left side) is less than the ratio threshold of the number of point clouds on the opposite side (right side) (i.e., left side number / right side number < 20%), the system determines that the data on that side (left side) is a unilateral incomplete point cloud. Conversely, if the ratio of the number of point clouds on both sides is higher than or equal to the threshold, the system determines that the data on both sides are relatively complete, each being a unilateral complete point cloud. This judgment process is applied independently to the femur and hip bones, but the number of point clouds on the femur is usually used as the main judgment criterion because its structure is usually more continuous and complete in the mask.

[0085] After classification, in the subsequent anatomical landmark recognition process, the system will only process the point cloud data of the unilateral or bilateral skeletons marked as complete, while automatically ignoring or skipping the side marked as incomplete. This mechanism prevents the algorithm from forcibly performing unreliable geometric fitting and extreme value search on the side with missing information when the data quality is significantly asymmetrical, thereby avoiding erroneous recognition results and ensuring that the accuracy of the final output landmarks is based solely on reliable and complete anatomical structural information.

[0086] S104. Obtain the left / right femoral point cloud from the unilateral complete point cloud and identify the femoral anatomical landmarks, including the femoral head center point, lesser trochanter point, femoral neck isthmus center point, and femoral shaft center point.

[0087] In some embodiments, identifying femoral anatomical landmarks includes:

[0088] The proximal femur candidate region is determined based on the maximum Z-coordinate of the femoral point cloud. The point cloud within the proximal femur candidate region is fitted with a least-squares sphere, and the center of the fitted sphere is used as the center point of the femoral head.

[0089] Based on the left and right sides of the femur, determine the medial direction vector, delineate the Z and Y direction ranges in the femoral point cloud with the center point of the femoral head as a reference, and find the point with the largest projection along the medial direction vector within the range as the lesser trochanter point.

[0090] The greater trochanter point is identified as the point with the largest projection along the lateral direction of the proximal femur. The cervical axis direction is defined based on the femoral head center point, lesser trochanter point, and greater trochanter point. Multiple sampling is performed along the cervical axis direction, and slice point clouds perpendicular to the cervical axis are extracted at each sampling position. The average distance from each slice point cloud to its geometric center is calculated, and the slice position with the smallest average distance is selected. Least square circle fitting is performed on the point cloud of the slice position with the smallest average distance, and the center of the fitted circle is taken as the center point of the femoral neck isthmus.

[0091] Using the Z-coordinate of the minor trochanter point as a reference, at least two target section heights on the femoral shaft are determined. Point clouds of the femoral shaft cross-section are extracted near each target section height. Least square circle fitting is performed on the cross-sectional point clouds, and the center of the fitted circle is taken as the center point of the femoral shaft at the corresponding position.

[0092] In some embodiments, when defining the cervical axis direction, the line connecting the center point of the femoral head with the midpoints of the greater trochanter and lesser trochanter is used as the initial cervical axis direction vector, and the initial cervical axis direction vector is normalized.

[0093] In some embodiments, the step size for multi-position sampling along the neck axis is between 2 mm and 5 mm, and the total number of sampling positions is between 10 and 30.

[0094] S105. Obtain the left / right hip bone point cloud from the unilateral complete point cloud and identify the hip bone anatomical landmarks, including the center point of the acetabulum, the ischial tuberosity point, the pubic symphysis point, and the anterior superior iliac spine point.

[0095] In some embodiments, identifying hip anatomical landmarks includes:

[0096] Using the identified femoral head center point on the same side as the reference point, select points in the hip bone point cloud that are less than the third preset threshold to form a candidate point set. Perform sphere fitting on the candidate point set and use the center of the fitted sphere as the center point of the acetabulum.

[0097] In the hip bone point cloud, the point with the smallest Z coordinate is directly selected as the ischial tuberosity node;

[0098] Calculate the Z-coordinate range of the hip bone point cloud and determine the intermediate height value. Divide the point cloud into upper and lower parts using the intermediate height value as the boundary. Select the point with the largest Y-coordinate in the lower part of the point cloud as the pubic symphysis point, and select the point with the largest Y-coordinate in the upper part of the point cloud as the anterior superior iliac spine point.

[0099] S106. The identified femoral and hip anatomical landmarks are converted from the medical imaging physical coordinate system to the three-dimensional model standard coordinate system, and the corresponding three-dimensional visualization models are generated.

[0100] In some embodiments, generating the corresponding 3D visualization model includes:

[0101] A spherical mesh model with a predetermined radius is generated, centered on the transformed coordinates of each anatomical landmark, and output in STL format.

[0102] The core objective of this step is to seamlessly convert the coordinates of the marker points, which were previously calculated using the physical coordinate system commonly used in the field of medical imaging, into a standard coordinate system format that is widely accepted and followed by downstream 3D surgical planning, computer-aided design, and navigation software. Based on this, an intuitive 3D visualization graphic model can be generated that can be directly used for clinical interaction and planning decisions.

[0103] In practice, the system first aggregates and processes the three-dimensional coordinate data of all femoral and hip anatomical landmarks identified in the standard physical coordinate system of medical imaging. It should be noted that medical imaging systems, such as the widely adopted DICOM standard and its derivative formats, typically use the RAS coordinate system to describe physical spatial locations by default. However, most professional software platforms used for 3D modeling, engineering analysis, and surgical planning use the LPS coordinate system or an equivalent coordinate system as their internal spatial reference. This difference between coordinate systems is mainly reflected in the completely opposite definitions of the horizontal and axial directions. Without a unified coordinate transformation, the directly imported identification results will exhibit mirror errors in 3D space, resulting in left-right reversals and front-back confusion, making them completely unusable for accurate preoperative measurement and planning. Therefore, performing a precise coordinate system transformation is an indispensable step in this process.

[0104] The system applies a defined set of coordinate axis mapping and direction reversal rules to the coordinates of each identified anatomical landmark. Specifically, the X-coordinate value in the original RAS coordinate system is inverted to obtain the new X-coordinate; the original Y-coordinate value is inverted to obtain the new Y-coordinate; while the original Z-coordinate value remains unchanged. Through this set of linear operations, the spatial position of all landmarks is accurately transformed to the standard coordinate system of the three-dimensional model.

[0105] After successfully completing the coordinate system transformation, the system generates a corresponding 3D visualization model. To transform abstract spatial coordinate points into tangible objects that surgeons and planning engineers can intuitively perceive and easily manipulate, the system creates a unique 3D graphic identifier for each transformed marker point. Specifically, a triangularly patterned spherical mesh model with a fixed display radius is programmatically generated, centered on the point's 3D coordinates. Each spherical model acts like a tiny, colored 3D landmark, its spatial position precisely corresponding to the key anatomical point it represents. These spherical models are then integrated and encoded and output using standard 3D exchange formats or standard triangulation language formats used in the industrial and medical fields.

[0106] The following disclosure provides a specific embodiment of a hip joint anatomical landmark identification method applicable to this application, which specifically includes the following steps:

[0107] Step S1: Load 3D mask data.

[0108] The input data is a pre-segmented 3D mask file of the hip joint in the NIfTI format, a common format in medical imaging. This file contains segmentation labels for the left and right femurs and the left and right hip bones. The file is loaded using a medical image processing library (such as Python's nibabel library), which not only reads the 3D voxel data matrix... It is also necessary to obtain its key spatial transformation matrix (affine matrix) and voxel spacing ( , , The affine matrix defines the coordinates from the voxel coordinates (). ) to physical space coordinates ( The linear transformation relationship (unit: millimeter, usually in RAS coordinate system) is the basis for subsequent coordinate transformations.

[0109] Step S2: Extraction of point cloud from bone surface.

[0110] For each bone label defined in the mask, first extract its binary mask B:

[0111] when It equals the value of the label, otherwise it is 0.

[0112] Subsequently, the Marching Cubes algorithm was applied to the binary mask B to extract its isosurfaces. The core of this algorithm is to traverse all voxel units in the 3D data (each unit consists of 8 adjacent voxel corner points), determining whether the isosurface passes through the unit based on whether the corner point value exceeds a preset threshold, and generating triangular facets through table lookup and linear interpolation. Ultimately, the vertices of all the triangular facets constitute a high-precision surface point cloud of the skeleton. (Coordinates are still voxel integer indices).

[0113] Finally, using the affine matrix A' obtained in step S1, the point cloud is... Each point in the graph is transformed from the voxel coordinate system to the physical coordinate system (RAS):

[0114]

[0115] After transformation, the point cloud in physical coordinates is obtained. This is used for all subsequent geometric calculations.

[0116] Step S3: Single-sided data processing and judgment.

[0117] To enhance the algorithm's robustness to common clinical unilateral lesions or incomplete scan cases, this step introduces judgment logic. The number of bone point clouds on both the left and right sides is calculated separately:

[0118] A threshold ratio is set (preferably 20% in this embodiment). The judgment logic is as follows: if the number of point clouds on the left femur is less than 20% of the number of point clouds on the right femur, the left femur data is considered incomplete or missing, and subsequent processing will skip the left femur; the judgment of the right femur and hip bones is the same. This mechanism ensures that the algorithm accurately identifies only the side with relatively complete data, avoiding errors caused by incomplete data.

[0119] Step S4: Identification of femoral anatomical landmarks.

[0120] For the femur that is determined to be intact through step S3, its key landmarks are identified in sequence.

[0121] S41, Femoral head center point A: In the physical coordinate system, the femoral head is usually located at the top (maximum Z-coordinate). First, find the maximum Z-coordinate value of the point cloud. ,Pick Points within the range are used as the candidate point set for the femoral head. Perform a least-squares sphere fit on this point set.

[0122] Construct the objective function: .

[0123] in belong The center of the sphere is obtained by solving a system of linear equations. , which represents the coordinates of the center point A of the femoral head, and R is the fitting radius.

[0124] S42, Lesser trochanter point P / Q: Determine the anatomical medial direction vector based on whether the femur is located on the left or right side. (Left femur is [+1,0,0], right femur is [-1,0,0]). Define a 3D search region in the point cloud: using point A as the reference, its Z-coordinate is between... And the Y coordinate is less than (To avoid mistakenly selecting the front structure). Within this region, calculate all points along... The projection value of the direction, the point with the largest projection is the innermost point P (left) or Q (right) of the small rotor.

[0125] S42, Center point B of the femoral neck isthmus:

[0126] First, identify the greater trochanter (GT): In the cloud of the proximal end of the femur, look for the point along the lateral direction (…). The point with the largest projection.

[0127] Define the initial neck axis direction: And normalize to obtain unit vector Where P is the innermost point of the small rotor (left); GT is the point of the large rotor.

[0128] Perform equidistant sampling (e.g., step size) along the neck axis from near point A to the distal end. (Sampling 20 points). For each sampling location Extract a perpendicular to Thin-layer slice point cloud (typically with a thickness of ).

[0129] Calculate the average distance from each slice of point cloud to its own geometric center, and use this as the approximate radius of the neck at that location. Find the The smallest sampling location is the narrowest part of the neck isthmus.

[0130] At this narrowest point, perform least-squares circle fitting on the slice point cloud. First, project the 3D points onto a plane perpendicular to... A two-dimensional point set is obtained on the plane, and the center of the circle is obtained by fitting a circle in the two-dimensional space. and radius Finally By back-projecting back into three-dimensional space, the three-dimensional coordinates of the center point B of the femoral neck isthmus are obtained.

[0131] S44. Femoral shaft center points C and D: using the Z coordinates of the lesser trochanter point P. For reference, two backbone section locations are defined: In respectively and Nearby location Extract the point cloud of the femoral shaft cross-section and fit the cross-section circle using the same circle fitting method as in S43 (where the plane normal vector is [0,0,1]). The center of the circle is the femoral shaft center point C and D. During fitting, the Z coordinate of the circle center can be constrained to... or .

[0132] Step S5: Identification of anatomical landmarks of the hip bone.

[0133] S51, Acetabular fossa center point O: Using the previously identified femoral head center point A on the same side as a reference, select all points in the hip bone point cloud whose Euclidean distance from point A is less than 30mm to form a candidate point set for the acetabular fossa. Perform the same sphere fitting on this point set as on S41, and the center of the fitted sphere is the center point O of the acetabulum.

[0134] S52, Ischial tuberosity nodes M / N: The ischial tuberosity is the lowest point of the hip bone. Directly find the point with the smallest Z-coordinate in the hip bone point cloud; this point is the ischial tuberosity node M (left) or N (right).

[0135] S53, pubic symphysis point S and anterior superior iliac spine point I:

[0136] Calculate the range of the hip bone point cloud in the Z direction: minimum value and maximum value , midpoint .

[0137] Point cloud It is divided into upper and lower parts by a boundary.

[0138] Point cloud in the lower half ( In the ), find the point with the largest Y coordinate, which is the most anterior protruding point, and define it as the pubic symphysis point S.

[0139] Point clouds in the upper half ( Similarly, find the point with the largest Y-coordinate, which is the anterior superior iliac spine point I.

[0140] Step S6: Coordinate system transformation and 3D model generation.

[0141] Medical imaging systems (such as DICOM and NIfTI) typically use the RAS (Right-Anterior-Superior) coordinate system, while many 3D modeling and surgical planning software programs use the LPS (Left-Posterior-Superior) coordinate system. To ensure the recognition results can be directly applied to downstream 3D systems, coordinate system transformation is necessary. The transformation formula is: Convert the coordinates of all identified marker points using this formula.

[0142] Subsequently, a triangular mesh sphere model with a specific radius (e.g., 5.0 mm) is generated, centered on each transformed LPS coordinate. This model can be exported using the standard Triangle Language (STL) format. Figure 2 As shown, all identified landmarks can also be projected and visualized on a coronal 2D view, facilitating quick verification by users. Ultimately, these STL-formatted landmark spherical models can be seamlessly imported into a 3D surgical planning system for precise alignment with the patient's 3D skeletal model, aiding in preoperative measurements and guide design.

[0143] Further reference Figure 3 As an implementation of the above-described method, this application provides an embodiment of a hip joint anatomical landmark recognition method system, which is similar to... Figure 1 Corresponding to the method embodiments shown, the system can be specifically applied to various electronic devices.

[0144] refer to Figure 3 A method system for identifying anatomical landmarks of the hip joint, comprising:

[0145] The mask data loading module 201 is configured to load three-dimensional mask data containing segmentation labels of the bones to be processed, and to obtain the three-dimensional voxel data and corresponding spatial transformation parameters of the three-dimensional mask data; the bones to be processed include the femur and hip bone;

[0146] The bone surface point cloud extraction module 202 is configured to extract bone surface point clouds from the three-dimensional mask data using the traveling cube algorithm, and to transform the point cloud voxel coordinate system to the medical image physical coordinate system using the spatial transformation parameters.

[0147] The complete point cloud judgment module 203 is configured to count the number of point clouds on the left and right bones, and divide the point cloud on the bone surface into a single-sided complete point cloud and / or a single-sided incomplete point cloud based on the ratio of the number of point clouds on the left bone to the number of point clouds on the right bone.

[0148] The femoral anatomical landmark recognition module 204 is configured to acquire the left / right femoral point cloud from a unilateral complete point cloud and identify the femoral anatomical landmarks, including the femoral head center point, lesser trochanter point, femoral neck isthmus center point and femoral shaft center point.

[0149] The hip anatomical landmark recognition module 205 is configured to acquire the left / right hip point cloud from a unilateral complete point cloud and identify the hip anatomical landmarks, which include the center point of the acetabulum, the ischial tuberosity point, the pubic symphysis point, and the anterior superior iliac spine point.

[0150] The marker output module 206 is configured to convert the identified femoral anatomical markers and hip anatomical markers from the medical image physical coordinate system to the three-dimensional model standard coordinate system, and generate the corresponding three-dimensional visualization model.

[0151] In another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the following... Figure 1 The method shown.

[0152] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0153] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a computer terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0154] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for identifying anatomical landmarks of the hip joint, characterized in that, Includes the following steps: Load 3D mask data containing segmentation labels of the bones to be processed, and obtain 3D voxel data and corresponding spatial transformation parameters of the 3D mask data; the bones to be processed include the femur and hip bone; The moving cube algorithm is used to extract the point cloud of the bone surface from the three-dimensional mask data, and the spatial transformation parameters are used to transform the voxel coordinate system of the point cloud to the medical image physical coordinate system. The number of point clouds on the left and right sides of the skeleton is counted. Based on the ratio of the number of point clouds on the left side to the number of point clouds on the right side, the point clouds on the surface of the skeleton are divided into unilateral complete point clouds and / or unilateral incomplete point clouds. The left / right femoral point cloud is obtained from the unilateral complete point cloud, and the femoral anatomical landmarks are identified, including the femoral head center point, lesser trochanter point, femoral neck isthmus center point and femoral shaft center point. The left / right hip bone point cloud is obtained from the unilateral complete point cloud, and the hip anatomical landmarks are identified, including the center point of the acetabulum, the ischial tuberosity point, the pubic symphysis point, and the anterior superior iliac spine point. The identification of anatomical landmarks of the hip bone includes: Using the identified femoral head center point on the same side as the reference point, select points in the hip bone point cloud that are less than the third preset threshold to form a candidate point set. Perform spherical fitting on the candidate point set and use the center of the fitted sphere as the center point of the acetabulum. In the hip bone point cloud, the point with the smallest Z coordinate is directly selected as the ischial tuberosity node; Calculate the Z-coordinate range of the hip bone point cloud and determine the intermediate height value. Divide the point cloud into upper and lower parts using the intermediate height value as the boundary. Select the point with the largest Y-coordinate in the lower part of the point cloud as the pubic symphysis point, and select the point with the largest Y-coordinate in the upper part of the point cloud as the anterior superior iliac spine point. The identified femoral and hip anatomical landmarks are transformed from the medical imaging physical coordinate system to the standard coordinate system of the three-dimensional model, and corresponding three-dimensional visualization models are generated.

2. The method for identifying anatomical landmarks of the hip joint according to claim 1, characterized in that, The spatial transformation parameters include the affine matrix and the voxel spacing.

3. The method for identifying anatomical landmarks of the hip joint according to claim 2, characterized in that, The step of extracting bone surface point clouds from the 3D mask data using the moving cube algorithm and transforming the point cloud voxel coordinate system to the medical image physical coordinate system using the spatial transformation parameters includes: For each bone segmentation label defined in the 3D mask data, extract the binarized mask data from its corresponding 3D mask region; Based on the binarized mask data, all voxel units in the three-dimensional voxel data are traversed, and the crossing state of the isosurface is determined according to the comparison result of the corner value of each voxel unit with the first preset threshold. Based on the crossing state of the isosurface, a triangular mesh representing the bone surface is generated by table lookup and linear interpolation. Extract all vertices of the triangular mesh representing the bone surface to form the surface point cloud of the bone; Affine matrices are used to transform each point in the surface point cloud of the skeleton from the point cloud voxel coordinate system to the medical image physical coordinate system.

4. The method for identifying anatomical landmarks of the hip joint according to claim 1, characterized in that, The method of dividing the bone surface point cloud into unilateral complete point clouds and / or unilateral incomplete point clouds based on the ratio of the number of point clouds on the left bone to the number of point clouds on the right bone includes: If the ratio of the number of point clouds on the side with fewer points to the number of point clouds on the side with more points is less than a second preset threshold, then the side with fewer points is determined to be a single-sided incomplete point cloud. If the ratio of the number of point clouds on the side with fewer points to the number of point clouds on the side with more points is greater than or equal to the second preset threshold, then both point clouds are determined to be single-sided complete point clouds.

5. The method for identifying anatomical landmarks of the hip joint according to claim 1, characterized in that, The identified femoral anatomical landmarks include: The proximal femoral candidate region is determined based on the maximum Z-coordinate of the femoral point cloud. The point cloud within the proximal femoral candidate region is then fitted with a least-squares sphere, and the center of the fitted sphere is taken as the center point of the femoral head. Based on the left and right sides of the femur, the medial direction vector is determined. In the femur point cloud, the Z and Y direction ranges are delineated with the center point of the femoral head as a reference. Within the range, the point with the largest projection along the medial direction vector is found and taken as the lesser trochanter point. The greater trochanter point is identified, which is the point with the largest projection along the lateral direction of the proximal femur. The cervical axis direction is defined based on the femoral head center point, lesser trochanter point, and greater trochanter point. Multi-position sampling is performed along the cervical axis direction, and slice point clouds perpendicular to the cervical axis are extracted at each sampling position. The average distance from each slice point cloud to its geometric center is calculated. The slice position with the smallest average distance is selected, and the point cloud of the slice position with the smallest average distance is fitted with a least-squares circle. The center of the fitted circle is taken as the center point of the femoral neck isthmus. Using the Z-coordinate of the small rotor point as a reference, at least two target cross-sectional heights on the femoral shaft are determined. Point clouds of the femoral shaft cross-section are extracted near each target cross-sectional height. The point clouds of the cross-section are fitted with least squares circles, and the center of the fitted circle is taken as the center point of the femoral shaft at the corresponding position.

6. The method for identifying anatomical landmarks of the hip joint according to claim 5, characterized in that: When defining the cervical axis direction, the line connecting the center point of the femoral head with the midpoints of the greater trochanter and lesser trochanter is used as the initial cervical axis direction vector, and the initial cervical axis direction vector is normalized.

7. The method for identifying anatomical landmarks of the hip joint according to claim 6, characterized in that, The step size for multi-position sampling along the neck axis is between 2mm and 5mm, and the total number of sampling positions is between 10 and 30.

8. A method for identifying anatomical landmarks of the hip joint according to any one of claims 1-7, characterized in that, The generation of the corresponding 3D visualization model includes: A spherical mesh model with a predetermined radius is generated, centered on the transformed coordinates of each anatomical landmark, and output in STL format.

9. A hip joint anatomical landmark recognition system, characterized in that, The system includes: The mask data loading module is configured to load three-dimensional mask data containing segmentation labels of the bones to be processed, and to acquire the three-dimensional voxel data of the three-dimensional mask data and the corresponding spatial transformation parameters; the bones to be processed include the femur and hip bone; The bone surface point cloud extraction module is configured to extract bone surface point clouds from the three-dimensional mask data using the traveling cube algorithm, and to transform the point cloud voxel coordinate system to the medical image physical coordinate system using the spatial transformation parameters. The complete point cloud determination module is configured to count the number of point clouds on the left and right sides of the skeleton. Based on the ratio of the number of point clouds on the left side to the number of point clouds on the right side, the point clouds on the surface of the skeleton are divided into unilateral complete point clouds and / or unilateral incomplete point clouds. The femoral anatomical landmark recognition module is configured to acquire left / right femoral point clouds from a unilateral complete point cloud and identify femoral anatomical landmarks, including the femoral head center point, lesser trochanter point, femoral neck isthmus center point, and femoral shaft center point. The hip anatomical landmark recognition module is configured to acquire the left / right hip point cloud from a unilateral complete point cloud and identify the hip anatomical landmarks, which include the center point of the acetabulum, the ischial tuberosity point, the pubic symphysis point, and the anterior superior iliac spine point. The identification of anatomical landmarks of the hip bone includes: Using the identified femoral head center point on the same side as the reference point, select points in the hip bone point cloud that are less than the third preset threshold to form a candidate point set. Perform spherical fitting on the candidate point set and use the center of the fitted sphere as the center point of the acetabulum. In the hip bone point cloud, the point with the smallest Z coordinate is directly selected as the ischial tuberosity node; Calculate the Z-coordinate range of the hip bone point cloud and determine the intermediate height value. Divide the point cloud into upper and lower parts using the intermediate height value as the boundary. Select the point with the largest Y-coordinate in the lower part of the point cloud as the pubic symphysis point, and select the point with the largest Y-coordinate in the upper part of the point cloud as the anterior superior iliac spine point. The marker output module is configured to convert the identified femoral and hip anatomical markers from the medical image physical coordinate system to the standard coordinate system of the three-dimensional model and generate the corresponding three-dimensional visualization model.

Images