Target feature point extraction method and device, computer device and storage medium

Abstract

The application relates to a target feature point extraction method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a to-be-processed image; acquiring a template corresponding to the to-be-processed image and acquiring the position relationship between the template and standard feature points; the template is an image generated based on a sample image, and the standard feature points are located in the template; registering the template to the to-be-processed image; according to the position relationship between the template and the standard feature points and the position relationship between the registered template and the to-be-processed image, determining target feature points corresponding to the standard feature points in the to-be-processed image. The method can realize automatic extraction of target feature points.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, apparatus, computer device, and storage medium for extracting target feature points. Background Technology

[0002] With the development of image processing technology, the automatic extraction of skeletal anatomical feature points from CT or MR images can be widely applied in medical image-based assisted diagnosis and treatment. At the same time, people's demands for the quality of medical services are increasing, and the precision and digitalization of bone surgery have become a global trend in medical development. Therefore, the automatic extraction of feature points can assist doctors in surgical planning, improve surgical efficiency, and also help patients in areas with relatively scarce medical resources enjoy better surgical outcomes.

[0003] In traditional techniques, skeletal anatomical landmarks are usually manually selected by experienced doctors. Since the selection of the location of skeletal anatomical landmarks is a key step in preoperative planning, this process requires doctors to have a high level of knowledge in anatomy and imaging, as well as clinical experience. Furthermore, manually obtaining landmarks requires doctors to spend a lot of time and energy and is a complex operation. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer device, and storage medium for automatically locating target feature points to address the aforementioned technical problems.

[0005] Firstly, this application provides a method for extracting target feature points, the method comprising:

[0006] Obtain the image to be processed;

[0007] Obtain a template corresponding to the image to be processed, and obtain the positional relationship between the template and the standard feature points; the template is an image generated based on the sample image, and the standard feature points are located in the template;

[0008] The template is registered to the image to be processed;

[0009] Based on the positional relationship between the template and the standard feature points, and combined with the positional relationship between the registered template and the image to be processed, the target feature points in the image to be processed corresponding to the standard feature points are determined.

[0010] In one embodiment, before obtaining the template corresponding to the image to be processed, the template is further generated by the following steps:

[0011] Acquire several sample images;

[0012] An initial template is selected from several sample images, and the initial template is registered to the remaining sample images to obtain registered images;

[0013] Calculate the statistical image corresponding to the registered image;

[0014] When the similarity between the statistical image and the initial template meets the requirements, the statistical image is used as the template; otherwise, the statistical image is used as the new initial template, and the process returns to the step of registering the initial template with the remaining sample images to obtain registered images, until the similarity between the statistical image and the initial template meets the requirements.

[0015] In one embodiment, calculating the statistical image corresponding to the registered image includes:

[0016] Obtain the initial position of the corresponding point in each of the registered images;

[0017] The average value of each initial position of the corresponding point is calculated as the target position of the corresponding point, and the statistical image is generated based on the target position of the corresponding point.

[0018] In one embodiment, after generating the template, the method further includes:

[0019] Receive standard feature point configuration instructions for the template;

[0020] Configure the corresponding standard feature points in the template according to the standard feature point configuration instructions.

[0021] In one embodiment, after calculating the statistical image corresponding to the registered image, the method further includes:

[0022] Calculate the distance between the statistical image and the corresponding point in the initial template;

[0023] The similarity between the statistical image and the initial template is calculated based on the distances between all corresponding points.

[0024] In one embodiment, the image to be processed and the sample image are three-dimensional mesh point cloud images; before registering the initial template to the remaining sample images, the method further includes: preprocessing the preprocessed image; and / or

[0025] Before registering the template to the image to be processed, the method further includes:

[0026] The sample image is preprocessed; the preprocessing includes at least one of surface point cloud extraction, point cloud downsampling, and normalization.

[0027] The extraction of surface point cloud involves extracting the vertices of all grids in the image to be processed and the sample image to obtain surface point cloud;

[0028] The point cloud downsampling involves dividing the image to be processed into at least one processing region and sampling the point in the processing region that is closest to the center of the image to be processed as the sampling point of the processing region; the normalization involves aligning the points in the image to be processed to the same coordinate space.

[0029] In one embodiment, registering the template to the image to be processed includes:

[0030] Obtain the registration function and initialize the registration function;

[0031] The image to be processed and the template are input into the registration function to optimize the parameters in the registration function;

[0032] When the change in the parameters of the registration function after optimization is less than a preset standard, it is determined that the template and the image to be processed have been successfully registered; otherwise, the image to be processed and the template are input into the registration function after parameter optimization to further optimize the parameters in the registration function.

[0033] In one embodiment, determining the target feature points in the image to be processed corresponding to the standard feature points based on the positional relationship between the template and the standard feature points, and in conjunction with the positional relationship between the registered template and the image to be processed, includes:

[0034] When the standard feature points are on the template surface, obtain the normal vectors of the standard feature points in the template after registration with the image to be processed;

[0035] When the normal vector intersects with the registered image to be processed, the distance between the intersection point and the standard feature point is calculated.

[0036] When the distance between the intersection point and the standard feature point is less than a preset distance, the intersection point is taken as the target feature point;

[0037] If there is no intersection point or the distance between the intersection point and the standard feature point is greater than the preset distance, then the point closest to the standard feature point in the registered template is selected from the image to be processed as the target feature point.

[0038] In one embodiment, determining the target feature point in the image to be processed corresponding to the standard feature point based on the positional relationship between the template and the standard feature point, and in conjunction with the positional relationship between the registered template and the image to be processed, further includes:

[0039] When the standard feature point is not on the surface of the standard plate, a preset number of points are selected from the surface of the template as associated points based on the standard feature point.

[0040] Based on the registration relationship between the template and the image to be processed, the target point of the associated point in the registered image is determined;

[0041] Calculate the target feature points of the image to be processed based on the target points.

[0042] In one embodiment, the image to be processed is a skeletal image, and the standard feature points are skeletal feature points, including at least one of femoral feature points and tibial feature points.

[0043] A method for processing skeletal data, the method comprising:

[0044] Obtain the skeleton image to be processed;

[0045] The skeletal image to be processed is processed to obtain skeletal feature points according to the target feature point extraction method in any of the above embodiments;

[0046] The skeletal feature points are processed according to preset rules.

[0047] In one embodiment, processing the skeletal feature points according to preset rules includes:

[0048] At least one of the femoral mechanical axis, the femoral condyle line, and the tibial mechanical axis is calculated based on the skeletal feature points.

[0049] In one embodiment, processing the skeletal feature points according to preset rules includes:

[0050] The skeletal feature points are optimized according to preset rules.

[0051] Secondly, this application also provides a target feature point extraction device, the device comprising:

[0052] The data acquisition module is used to acquire the image to be processed;

[0053] The template query module is used to obtain a template corresponding to the image to be processed, and to obtain the positional relationship between the template and the target feature points;

[0054] A registration module is used to register the template to the image to be processed;

[0055] The target extraction module is used to determine the target feature points in the image to be processed that correspond to the standard feature points, based on the positional relationship between the template and the target feature points, and in conjunction with the positional relationship between the registered template and the image to be processed.

[0056] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0057] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0058] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method.

[0059] The aforementioned target feature point extraction method, apparatus, computer equipment, and storage medium can register a template with an image to be processed, and extract target feature points from the image to be processed based on the positional relationship between the template and standard feature points, as well as the positional relationship between the registered template and the image to be processed, thereby achieving automatic extraction of target feature points and improving efficiency. Attached Figure Description

[0060] Figure 1 This is an application environment diagram of the target feature point extraction method in one embodiment;

[0061] Figure 2 This is a flowchart illustrating a target feature point extraction method in one embodiment;

[0062] Figure 3 This is a schematic diagram of lower limb bone segmentation and surface reconstruction in one embodiment;

[0063] Figure 4 This is a flowchart illustrating the template generation method in another embodiment;

[0064] Figure 5 This is a schematic diagram of statistical skeletal image acquisition in one embodiment;

[0065] Figure 6 A schematic diagram of standard feature point configuration in one embodiment;

[0066] Figure 7 This is a schematic diagram illustrating the comparison of similarity between two skeletal images in one embodiment;

[0067] Figure 8This is a schematic diagram of a point cloud downsampling method in one embodiment;

[0068] Figure 9 This is a schematic diagram of data preprocessing in one embodiment;

[0069] Figure 10 This is a schematic diagram of skeletal point cloud data normalization in one embodiment;

[0070] Figure 11 This is a schematic diagram of non-rigid registration in one embodiment;

[0071] Figure 12 This is a schematic diagram illustrating the process of optimizing the registration function using the Expectation-Maximization (EM) algorithm in one embodiment;

[0072] Figure 13 This is a schematic diagram illustrating the principle of surface feature point extraction in one embodiment;

[0073] Figure 14 This is a schematic diagram illustrating the principle of surface feature point extraction in another embodiment;

[0074] Figure 15 This is a schematic diagram illustrating the principle of non-surface skeletal feature point extraction in one embodiment;

[0075] Figure 16 This is a flowchart illustrating a skeletal data processing method in one embodiment;

[0076] Figure 17 A schematic diagram of feature point location optimization in one embodiment;

[0077] Figure 18 This is a structural block diagram of a target feature point extraction device in one embodiment;

[0078] Figure 19 This is a structural block diagram of a skeletal data processing device in one embodiment;

[0079] Figure 20 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0080] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0081] The target feature point extraction method provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with medical imaging device 104 via a network. Terminal 102 receives 3D images scanned by medical imaging device 104 and stored using a 3D matrix. It then performs 3D reconstruction on these images to obtain the image to be processed, and acquires a pre-generated template corresponding to the image to be processed. The template and the image to be processed are registered. Based on the positional relationship between the template and standard feature points, and considering the position between the registered template and the image to be processed, the target feature points in the image to be processed corresponding to the standard feature points are determined. Since the template includes standard feature points, after registration, it can automatically extract the target feature points corresponding to the standard feature points in the image to be processed through mapping based on the registered template. This eliminates the need for manual extraction of target feature points from the image to be processed, saving significant time and improving efficiency.

[0082] The terminal 102 may be, but is not limited to, various personal computers, laptops, smartphones, tablets, portable wearable devices, as well as functional modules and dedicated circuits of the medical imaging equipment itself. The medical imaging equipment 104 includes, but is not limited to, various imaging devices, such as CT imaging equipment (CT: Computed Tomography, which uses precisely collimated X-ray beams and highly sensitive detectors to perform a series of cross-sectional scans around a part of the human body, and can reconstruct precise three-dimensional images of tumors, etc.), magnetic resonance imaging equipment (a type of tomographic imaging that uses magnetic resonance to obtain electromagnetic signals from the human body and reconstruct human information images), positron emission tomography (PET / MR) equipment, and PET / MR systems, etc.

[0083] In one embodiment, such as Figure 2 As shown, a target feature point extraction method is provided, which can be applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps:

[0084] S202, Obtain the image to be processed.

[0085] Specifically, the image to be processed is preferably three-dimensional surface mesh data, which can be obtained by three-dimensional reconstruction of three-dimensional images acquired by medical imaging equipment. In other embodiments, when the medical imaging equipment acquires three-dimensional surface mesh data, it is not necessary to perform three-dimensional reconstruction.

[0086] In the field of medical imaging, 3D scans such as CT or MR are generally 3D images, which are medical image data stored in the form of a 3D matrix. The 3D image includes the target to be processed, such as the target bone or organ where the target feature points are located.

[0087] Specifically, 3D reconstruction can include: first, the terminal uses image segmentation technology to segment the target in the 3D matrix to obtain mask data stored in the form of a 3D matrix; then, 3D reconstruction is performed on the mask data to obtain the image to be processed. Image segmentation technology includes, but is not limited to, image segmentation technology based on deep learning fully convolutional networks, or based on traditional machine learning (such as random forests), or based on segmentation techniques such as clustering, region growing, active contours, level sets, and thresholding. Methods for 3D reconstruction of the mask data include, but are not limited to, the Marching Cube algorithm, interpolation reconstruction using the Marching Cube algorithm based on surface thresholds near the contour, and Poisson surface reconstruction algorithms. Figure 3 This is a schematic diagram illustrating the segmentation and surface reconstruction of the lower limb bones in one embodiment. The terminal uses image segmentation technology to segment CT images, obtaining skeletal data in the sectional, sagittal, and coronal planes. This involves extracting the bone pixels from the image to be processed. Then, a surface reconstruction method is used to represent the surface of the segmented bone pixels in the form of mesh data, thus obtaining the bone image to be processed. This bone image can be used for subsequent registration.

[0088] S204, obtain the template corresponding to the image to be processed, and obtain the positional relationship between the template and the standard feature points; the template is an image generated based on the sample image, and the standard feature points are located in the template.

[0089] Specifically, the template is generated in advance based on sample images and is used to represent the standard morphology of bones or organs corresponding to the image to be processed. The template can be generated based on sample images of users collected before surgery, or it can be generated based on a large number of sample images of different users to suit a large number of users, such as obtaining it based on the average image of the sample images. This way, it is not necessary to generate a template for each user before surgery.

[0090] Standard feature points are feature points of bones or organs selected in the template. These standard feature points can be manually selected by doctors or others in the template. It should be noted that feature lines, feature surfaces, and feature regions can all be considered as being composed of feature points. These standard feature points correspond to the target feature points to be extracted from the image to be processed.

[0091] The positional relationship between the template and the standard feature points is data used to characterize the position of the standard feature points within the template, and this positional relationship can be determined in the image coordinate system where the template is located. Optionally, the template and the positional relationship between the standard feature points and the template are pre-generated. In other embodiments, when it is necessary to add new standard feature points, the new standard feature points can also be marked in the template in real time, without specific limitations.

[0092] Optionally, the template data can be stored according to the type of bone. This way, after obtaining the image to be processed, the corresponding template that has been stored can be selected according to the bone type of the image to be processed.

[0093] S206, Register the template to the image to be processed.

[0094] Preferably, the registration here refers to surface registration, which unifies the 3D surface mesh data in the template and the 3D surface mesh data of the image to be processed into the same coordinate system. Registration establishes a one-to-one correspondence between the positions of the 3D surface mesh data in the template and the positions of the 3D surface mesh data in the image to be processed, thus laying the foundation for obtaining target feature points in the image to be processed. Surface registration can include, but is not limited to, non-rigid registration algorithms.

[0095] S208. Based on the positional relationship between the template and the standard feature points, and combined with the positional relationship between the registered template and the image to be processed, determine the target feature points in the image to be processed that correspond to the standard feature points.

[0096] In this context, the target feature point refers to the feature point on the image to be processed after the template is registered with the image to be processed. For example, the template is registered with the image to be processed, so that the grid in the template corresponds one-to-one with the grid in the image to be processed. In this way, the standard feature points in the template also correspond to the target feature points in the image to be processed. These target feature points are the feature points to be extracted.

[0097] One point to note is that the number of standard feature points is not limited here. In one target feature point extraction, the terminal can extract the target feature points corresponding to multiple standard feature points in parallel, thereby improving the efficiency of target feature point extraction.

[0098] In other embodiments, after acquiring the target feature points, the terminal can also output the target feature points for doctors and others to examine. When the doctor confirms that the extracted target feature points are correct, the extracted target feature points are correct. If there is a problem, the terminal can receive an adjustment instruction for the target feature points and fine-tune the target feature points according to the adjustment instruction to ensure the accuracy of the output target feature points.

[0099] In the above embodiments, the template can be registered with the image to be processed, and the target feature points can be extracted from the image to be processed based on the positional relationship between the template and the standard feature points and the positional relationship between the registered template and the image to be processed, thereby realizing the automatic extraction of target feature points and improving efficiency.

[0100] In one embodiment, such as Figure 4 The flowchart shown illustrates a template generation method in one embodiment. This template generation method may include:

[0101] S402: Acquire several sample images.

[0102] Specifically, the sample image is preferably three-dimensional surface mesh data, which can be obtained by reconstructing three-dimensional images acquired by medical imaging equipment. The specific three-dimensional reconstruction method can be found above. Taking the lower limb bone as an example, the terminal first collects a large amount of lower limb bone medical image data from different patients as a training set. Then, the medical image data in the training set is segmented and reconstructed according to the three-dimensional reconstruction method described above to obtain the sample image.

[0103] S404: Select an initial template from several sample images.

[0104] Specifically, the initial template can be any one of the sample images selected from several sample images. It should be noted that when there is only one set of sample images, it is directly used as the template. If there are at least two sets of sample images, then any one of the sample images is selected as the initial template.

[0105] S406: Register the initial template to the remaining sample images to obtain registered images.

[0106] Specifically, the registered image is obtained by registering the initial template to the remaining sample images using a registration algorithm. For example, a non-rigid registration algorithm can be used to map the initial template to other remaining sample images to obtain the registered image.

[0107] S408: Calculate the statistical image corresponding to the registered image.

[0108] The statistical image is calculated from the registered image according to certain rules, such as averaging, finding the maximum and minimum values, finding the median, etc., of the positions of the points in the registered image. The statistical image can serve as a representative of the overall situation of this registration. In one embodiment, the statistical image can be used to compare the similarity with the initial template to further obtain the similarity between the statistical image and the initial template.

[0109] S410: When the similarity between the statistical image and the initial template meets the requirements, the statistical image is used as the template; otherwise, the statistical image is used as the new initial template, and the process of registering the initial template to the remaining sample images is returned until the similarity between the statistical image and the initial template meets the requirements.

[0110] Specifically, similarity is a quantitative value that reflects the similarity between a statistical image and an initial template. The higher the similarity between the statistical image and the initial template, the more similar they are; conversely, the lower the similarity, the less similar they are. The similarity can be calculated based on the distance between corresponding points in the statistical image and the initial template.

[0111] Specifically, when the similarity between the statistical image and the initial template meets the requirement (meaning the similarity between the statistical image and the initial template is greater than or equal to a preset threshold), the terminal considers the statistical image sufficiently similar to the initial template and uses the statistical image as the template corresponding to the sample image. For example, if the sample image is a patient's lower limb bone image, the template is the lower limb bone template. If the similarity between the statistical image and the initial template is less than the preset threshold, the currently obtained statistical image is used as the initial template for the next iteration, and the current initial template is registered with the sample image to obtain a registered image. The statistical image corresponding to the registered image is then calculated until the similarity between the statistical image and the initial template is greater than or equal to the preset threshold, resulting in the final template. The similarity threshold can be adjusted according to the actual situation.

[0112] In the above embodiments, the template is obtained by iteratively registering the sample image with the initial template, and by calculating the statistical image corresponding to the registered image and comparing the similarity between the statistical image and the initial template. The template obtained in this way is more realistic and accurate, which can lay a good foundation for subsequent registration between the image to be processed and the standard image to obtain the target feature points.

[0113] In one embodiment, calculating the statistical image corresponding to the registered image includes: obtaining the initial position of the corresponding point in each registered image; calculating the average value of each initial position of the corresponding point as the target position of the corresponding point; and generating the statistical image based on the target position of the corresponding point.

[0114] Specifically, obtaining the initial position of the corresponding point in the registration image means that after the sample image in the training set is registered with the initial template, the terminal can obtain the position of the corresponding point in the grid data in the registration image, then average the position of the point to obtain the average position, and generate a statistical image based on the average position of all points.

[0115] Specifically, in combination Figure 5 As shown, Figure 5 This is a schematic diagram illustrating the acquisition of statistical skeletal images in one embodiment. Figure 5The training set contains N skeleton images to be processed, which are obtained by 3D reconstruction of 3D images acquired by medical imaging equipment. First, the terminal arbitrarily selects one skeleton image from the training set as the initial skeleton template. Then, a surface registration algorithm is used to register the initial skeleton template to the remaining N-1 skeleton images to be processed, resulting in N-1 registered skeleton images. Figure 5 The solid lines represent the points P(i) on the corresponding initial skeleton template, which are mapped to P1(i)...P2. N-1 (i), i = 1, 2, 3, ..., P N-1 (i) refers to the corresponding points in the mesh data of the registered skeleton image, and then for a set of corresponding points P1(i)...P N-1 (i), find its average value P'(i)=(P1(i)+P2(i)+...+P N-1 (i)) / (N-1), and generate the corresponding statistical skeleton image based on P'(i).

[0116] In the above embodiments, by calculating the corresponding points of the grid data in the registered image, the corresponding statistical image can be accurately obtained.

[0117] In one embodiment, after generating the template, the method further includes: receiving a standard feature point configuration instruction for the template; and configuring the corresponding standard feature points in the template according to the standard feature point configuration instruction.

[0118] Specifically, the standard feature point configuration instruction is a computer instruction used to obtain standard feature points on a template. It can be input by the user according to the application scenario. For example, the standard feature point configuration instruction can be an instruction from a doctor to select anatomical feature points, feature surfaces, or feature lines in a standard bone template. According to the configuration instruction, the corresponding standard feature points are configured in the template. Specifically, after the terminal receives the standard feature point configuration instruction, it marks the corresponding feature points, feature surfaces, or feature lines on the template according to the standard feature point configuration instruction. For example, on the femur, feature points such as the distal end of the lateral femoral condyle and the distal end of the medial femoral condyle are marked.

[0119] Specifically, in combination Figure 6 As shown, Figure 6This is a schematic diagram of standard feature point configuration in one embodiment. The standard feature point configuration instructions are for configuring feature points of anatomical feature points in standard lower limb bones. The corresponding anatomical feature points configured by the terminal in standard lower limb bones according to the standard feature point configuration instructions include: hip joint center 1, lateral femoral condyle 2, medial femoral condyle 3, intercondylar fossa of femur 4, distal end of lateral femoral condyle 5, distal end of medial femoral condyle 6, posterior end of lateral femoral condyle 7, posterior end of medial femoral condyle 8, lateral tibial plateau 9, medial tibial plateau 10, tibial spine 11, tibial tuberosity 12, lateral ankle 13, medial ankle 14, and midpoint of ankle 15, any one or more of these.

[0120] In the above embodiments, the required features can be obtained on the template through standard feature point configuration instructions. These feature points can be used to subsequently determine the corresponding feature points in the image to be processed.

[0121] In one embodiment, after calculating the statistical image corresponding to the registered image, the method further includes: calculating the distance between the statistical image and the corresponding points in the initial template; and calculating the similarity between the statistical image and the initial template based on the distances between all corresponding points.

[0122] Specifically, the terminal first calculates the distance between the statistical image and the corresponding points of each set of grid data in the initial template, and then calculates the similarity between the statistical image and the initial template based on the distance between the corresponding points of each set of grid data. In other embodiments, the similarity can be expressed as the reciprocal of the average distance between all corresponding points, combined with... Figure 7 As shown, Figure 7 This is a schematic diagram illustrating the similarity between two skeletal images in one embodiment. The terminal calculates the distance d between each point P'(i) and the corresponding point P(i) in the initial skeletal template. i And the corresponding similarity, where the similarity can be expressed as:

[0123]

[0124] Where m is the number of points contained in the initial skeleton template, d m This represents the distance between the m-th corresponding points. The smaller the average distance between the corresponding points in each set of grid data of the statistical skeleton image and the initial skeleton template, the greater the similarity between the statistical skeleton image and the initial skeleton template. When the similarity exceeds a certain threshold, the statistical skeleton image is considered sufficiently similar to the standard skeleton template, and the statistical skeleton image can then be used as the standard model. The similarity threshold can be adjusted according to the actual situation.

[0125] In the above embodiments, the standard model corresponding to the sample image can be accurately obtained by calculating the similarity between the statistical image and the initial template.

[0126] In one embodiment, the image to be processed and the sample image are three-dimensional mesh point cloud images; before registering the initial template to the remaining sample images, the method further includes: preprocessing the image to be processed; and or before registering the template to the image to be processed, the method further includes: preprocessing the sample images; the preprocessing includes at least one of surface point cloud extraction, point cloud downsampling, and normalization; surface point cloud extraction is to extract the vertices of all meshes in the image to be processed and the sample images to obtain surface point clouds; point cloud downsampling is to divide the image to be processed into at least one processing region, and sample the point in the processing region that is closest to the center of the processing region as the sampling point of the processing region; normalization is to align the points in the image to be processed to the same coordinate space.

[0127] The processing area refers to the equidistant division of the entire space according to a preset interval. The preset interval can be determined based on the actual application scenario. Optionally, if the preset interval is L, the entire space can be divided into several processing areas with an interval of L. The sampling point refers to a point selected from the image to be processed according to a preset rule. For example, the sampling point can be obtained by dividing the image to be processed into processing areas and selecting the point closest to the center of the processing area.

[0128] Specifically, surface point cloud extraction refers to extracting all grid vertices to obtain the surface point cloud of the image to be processed and / or the sample image. Point cloud downsampling involves dividing the space containing the input point cloud into several small cubic spaces (processing regions) at a certain interval L. Each small cubic space may or may not contain points from the surface point clouds of the image to be processed and the sample image. If a small cubic space containing points from the surface point cloud contains only one point, it is directly retained; otherwise, the distance from each point to the center point of the small cubic space is calculated, and only the point closest to the center is retained as the sampling point, while the rest are removed. Finally, a sparse point cloud with a spatial distribution basically the same as the original point cloud but with fewer points is obtained. Figure 8 As shown, Figure 8 This is a schematic diagram of a point cloud downsampling method in one embodiment. Solid points represent the points closest to the grid center, while hollow points represent other points. After point cloud downsampling, only the points closest to the cube center (the solid points in the diagram) remain. In another embodiment, combined with... Figure 9 As shown, Figure 9 This is a schematic diagram of data preprocessing in one embodiment. A sparse point cloud can be obtained from a standard bone template and a patient bone image after surface point cloud extraction and point cloud downsampling.

[0129] Specifically, normalization refers to converting the sample images and the images to be processed to the same coordinate system, making subsequent data processing more convenient. For example, if all the sample images, the images to be processed, and their corresponding templates are not taken from the same position and perspective, that is, if all the sample images, the images to be processed, and their corresponding templates are not in the same coordinate space, then it is preferable to align all the sample images, the images to be processed, and their corresponding templates to the same coordinate space, for example, aligning the sampling points to the same coordinate space. In other embodiments, the normalization process for the sample images and the images to be processed first calculates the centroid coordinates C (the center position of all points) of all the sample images, the images to be processed, and their corresponding templates, then translates the point cloud by -C so that its centroid coincides with the origin of the coordinate system, and then calculates the variance Var of the point cloud coordinates after translation, dividing the coordinates of each point in the point cloud by... This yields normalized point cloud data. Combined with… Figure 10 As shown, Figure 10 This is a schematic diagram of the normalization of skeletal point cloud data in one embodiment. The terminal first adjusts the mean of the skeletal point cloud to 0, and then adjusts the variance of the point cloud to 1 to obtain the normalized femoral point cloud data.

[0130] In the above embodiments, preprocessing the sample image and the image to be processed can accelerate the calculation speed and convergence speed of subsequent registration operations.

[0131] In one embodiment, registering a template to an image to be processed includes: obtaining a registration function and initializing the registration function; inputting the image to be processed and the template into the registration function to optimize the parameters in the registration function; when the change in the parameters of the registration function after optimization is less than a preset standard, determining that the template and the image to be processed have been registered; otherwise, continuing to input the image to be processed and the template into the parameter-optimized registration function to optimize the parameters in the registration function.

[0132] Specifically, a registration function is a program that registers a template to an image to be processed. After inputting the template and the image to be processed into the registration function, a registered template is obtained. The terminal first obtains and initializes the corresponding registration function, including initializing its parameters. The image to be processed and the template are then input into the registration function to optimize the parameters and obtain the corresponding registered image. Specifically, based on the current parameters of the registration function and the input template and image to be processed, the terminal calculates the posterior probability matrix using Bayes' theorem and calculates the optimization direction of the registration function. The corresponding parameters are updated according to the optimization direction. Then, it is determined whether the change in parameters before and after optimization is less than a preset standard, which can be adjusted according to actual conditions. If the change is less than the preset standard, the template and the image to be processed are considered successfully registered, and the registered template and the image to be processed are output. Otherwise, the optimized parameters are used as the current parameters of the registration function, and the above operations continue until the template and the image to be processed are successfully registered. In other embodiments, the image to be processed and the template input to the registration function are preprocessed data, wherein the preprocessing includes at least one of surface point cloud extraction, point cloud downsampling, and normalization, which can speed up the calculation of the registration function.

[0133] Specifically, in combination Figure 11 As shown, Figure 11 This is a schematic diagram of non-rigid registration in one embodiment. Figure 11 M circular points y1…y M For points on a standard skeletal template, there are N triangle points x1…x N Let Y be the points on the skeleton image to be processed. Here, Y is the point set composed of points from the standard skeleton template. Mx3 =(y1,…y M ) T A Gaussian mixture model (GMM) is established for the mean, with variance σ. 2 The point set X in the skeleton image to be processed Nx3 =(x1,…x N ) T It is considered to be generated by the GMM. The probability density of the GMM is:

[0134]

[0135] Among them, P(m)=1 / M.

[0136] If noise, i.e., out-of-field noise, is considered, then an additional uniform distribution is added, for

[0137]

[0138] Where ω is the probability of an exterior point.

[0139] The purpose of registration is to maximize the probability of X in the GMM by transforming the mean Y of the GMM. Assuming the mean Y of the GMM is transformed by parameter θ, the registration function to be optimized is:

[0140]

[0141] Specifically, in combination Figure 12 As shown, Figure 12 This diagram illustrates the process of optimizing the registration function using the Expectation-Maximization (EM) algorithm in one embodiment. First, the standard bone template and patient bone are preprocessed to obtain normalized point cloud data. Then, reasonable initial values ​​are set for the parameters θ and σ of the registration function to be optimized. Based on the current parameter values ​​and the input standard bone template and patient bone data, the posterior probability matrix is ​​calculated using Bayes' theorem. The optimization direction of the registration function in this iteration is calculated. The values ​​of θ and σ are updated according to the optimization direction of the registration function. The iteration converges based on whether the change in parameters is less than a certain threshold. If convergence is achieved or the number of iterations reaches the set maximum number of iterations, the iteration stops, and the deformed standard bone template becomes the registration result. Otherwise, the optimized parameters are used as the current parameters of the registration function, and the above operations continue until the standard bone template and patient bone are successfully registered.

[0142] In the above embodiments, the standard bone template and the bone image to be processed can be registered using the registration function to obtain the corresponding registered standard bone template.

[0143] In one embodiment, based on the positional relationship between the template and the standard feature points, and combined with the positional relationship between the registered template and the image to be processed, the target feature point in the image to be processed corresponding to the standard feature point is determined. This includes: when the standard feature point is on the template surface, obtaining the normal vector of the standard feature point in the registered template; when the normal vector intersects with the registered image to be processed, calculating the distance between the intersection point and the standard feature point; when the distance between the intersection point and the standard feature point is less than a preset distance, taking the intersection point as the target feature point; when there is no intersection point or the distance between the intersection point and the standard feature point is greater than the preset distance, selecting the point in the image to be processed that is closest to the standard feature point in the registered template as the target feature point.

[0144] Specifically, before registering the template with the image to be processed, it is first determined whether the standard feature points are on the surface of the standard skeleton template. If the standard feature points are on the template surface, registration is performed directly. After the template and the image to be processed are registered, the normal vectors of the registered template and the standard feature points are plotted, and a straight line is drawn along these normal vectors. Then, it is determined whether this straight line intersects with the image to be processed, and if an intersection exists, whether the distance between the intersection point and the standard feature points is less than a preset distance. Different operations are then performed according to different situations to extract the target feature points in the image to be processed. The preset distance can be adjusted according to the actual application scenario.

[0145] Specifically, in combination Figure 13 As shown, Figure 13 This is a schematic diagram illustrating the principle of surface feature point extraction in one embodiment. Figure 13 This indicates that the line containing the normal vector of the standard bone template intersects the bone image to be processed, and the distance between this intersection point and the standard bone feature point is less than a preset distance. Figure 13 The midpoint P represents the standard bone feature point in the registered standard bone template. i and P j Taking these two skeletal feature points as examples, P is used. i and P j Draw a straight line along the normal vector of the standard bone template and calculate the distance between the intersection of the straight line and the bone image to be processed and the standard bone feature point. If the distance between the intersection of the straight line and the standard bone feature point is less than the preset distance, then the intersection of the straight line and the bone image to be processed, i.e. the triangle in the figure, is the bone feature point Pi' and Pj' of the bone image to be processed.

[0146] Specifically, Figure 14 This is a schematic diagram illustrating the principle of surface feature point extraction in another embodiment. Figure 14 This indicates that the line containing the normal vectors of the standard skeletal feature points and the standard skeletal template does not intersect with the skeletal image to be processed, denoted by P. k Taking this skeletal feature point as an example, P k If the line containing the normal vector of the standard bone template does not intersect with the bone image to be processed, then the point closest to the registered standard bone feature point in the bone image to be processed is selected as the bone feature point of the bone image to be processed, i.e., P, where the triangle in the figure is located. k '.

[0147] Specifically, when the line containing the normal vector of the standard bone feature point and the standard bone template intersects with the bone image to be processed, and the distance between the intersection point and the standard bone feature point is greater than a preset distance, the point closest to the registered standard bone feature point and the bone image to be processed is selected as the bone feature point of the bone image to be processed.

[0148] In the above embodiments, after registering the standard image with the image to be processed, different operations can be performed according to different situations to accurately obtain the target feature points of the image to be processed on the surface.

[0149] In one embodiment, the method of determining the target feature points in the image to be processed corresponding to the standard feature points based on the positional relationship between the template and the standard feature points, combined with the positional relationship between the registered template and the image to be processed, further includes: when the standard feature points are not on the surface of the standard plate, selecting a preset number of points from the template surface as associated points based on the standard feature points; determining the target points of the associated points in the registered image based on the registration relationship between the template and the image to be processed; and calculating the target feature points of the image to be processed based on the target points.

[0150] Specifically, when the standard feature points are not on the template surface, it is first necessary to select points on the nearby surface as associated points based on the structural feature points around the standard feature points. Associated points refer to points on the template surface that can reflect the standard feature points inside the template. For example, for the lower limb bone, the center of the sphere fitted by the bone associated points is the bone feature point inside the bone. After determining the associated points, the template is registered with the image to be processed, and the position of the target point corresponding to the associated point is obtained. The determination of the target point position can refer to the processing when the standard feature points are on the template surface. Then, the target feature points of the image to be processed are calculated based on the target point position. Continuing with the lower limb bone as an example, the target points are fitted into a sphere, and the center of the sphere is the bone feature point of the bone image to be processed.

[0151] Specifically, in combination Figure 15 As shown, Figure 15 This is a schematic diagram illustrating the principle of non-surface skeletal feature point extraction in one embodiment. The left image shows the standard skeletal template, and the right image shows the registered standard skeletal template. The registered standard skeletal template is obtained through non-rigid registration. Taking the femoral head center point C as an example, N points P1…P2 on the nearby surface are selected on the standard skeletal template. N As related points, P1…P N These N points can be used to approximate the center point C of the sphere. After determining the associated points, non-rigid registration is used to register the standard skeleton template with the skeleton image to be processed. Through registration, these N associated points are mapped onto the skeleton image to be processed, obtaining the corresponding target points P1'...P N ', then the center of the femoral head in the bone image to be processed can be obtained through P1'...P N The fitted center position of the sphere is obtained. Figure 15 C' in the image represents the center of the femoral head in the bone image to be processed.

[0152] In the above embodiments, feature points on the surface near the standard feature points can be selected as association points, and the standard feature points of the image to be processed that are not on the surface can be obtained through the registered association points, thus solving the problem that it is difficult to obtain feature points inside the template through surface registration.

[0153] In one embodiment, the image to be processed is a skeletal image, and the standard feature points are skeletal feature points, which include at least one of femoral feature points and tibial feature points.

[0154] Specifically, the bone image to be processed is obtained by three-dimensional reconstruction of a patient's bone from a three-dimensional image acquired by a medical imaging device. The standard feature points are anatomical feature points on a pre-generated standard bone template. Taking the lower limb bone as an example, the following combination is further used... Figure 6 Among them, the distal end of the lateral femoral condyle (5), the distal end of the medial femoral condyle (6), the posterior end of the lateral femoral condyle (7), and the posterior end of the medial femoral condyle (8) are characteristic points on the femoral side, while the lateral tibial plateau (9) and the medial tibial plateau (10) are characteristic points on the tibial side and serve as reference points for measuring osteotomy volume during knee replacement.

[0155] In one embodiment, the skeletal data processing method includes: acquiring a skeletal image to be processed; processing the skeletal image to be processed according to the target feature extraction method of any of the above embodiments to obtain skeletal feature points; and processing the skeletal feature points according to preset rules.

[0156] Specifically, in combination Figure 16 As shown, Figure 16 The diagram illustrates the process of bone data processing in one embodiment. The terminal first acquires a bone image to be processed, which is a three-dimensional image of a patient's bones captured by a medical imaging device and reconstructed in three dimensions. The method for extracting target features according to any of the above embodiments, processing the bone image to obtain bone feature points, includes: after acquiring the patient's bone image, the terminal queries a standard bone template corresponding to the patient's bone image and obtains the relationship between the standard bone template and standard bone feature points. The standard bone template can include bones from any part of the human body, and its generation method can follow the template generation method described above. The standard bone feature points can be manually selected by a doctor from the standard bone template. The terminal registers the standard bone template to the patient's bone image to obtain a registered standard bone template, i.e., a deformed standard bone template. The registration method can be, but is not limited to, a non-rigid registration algorithm. Based on the position of the standard feature points in the template, combined with the deformed standard bone template, the bone feature points corresponding to the standard bone feature points in the patient's bone image can be determined. Finally, the terminal processes the obtained bone feature points according to preset rules to obtain optimized bone feature points, thereby improving the positional accuracy of the bone feature points.

[0157] In the above embodiments, by registering the standard bone template with the patient's bone image, the standard bone feature points on the standard bone template can be mapped to the patient's bones, thereby realizing the automatic extraction of the patient's bone feature point positions.

[0158] In one embodiment, the skeletal feature points are processed according to preset rules, including: calculating at least one of the femoral mechanical axis, the femoral condyle line, and the tibial mechanical axis based on the skeletal feature points.

[0159] Specifically, continue to combine Figure 6 The femoral mechanical axis can be determined based on the hip joint center-1 and the intercondylar notch of the femur-4. The femoral condylar line can be determined based on the lateral femoral condyle-2 and the medial femoral condyle-3. The tibial mechanical axis can be determined based on the tibial spine-11 and the midpoint of the ankle-15.

[0160] In the above embodiments, the corresponding lower limb physiological axes of the skeletal feature points can be calculated through the skeletal feature points, and these physiological axes can further determine the placement angle of the joint prosthesis.

[0161] In one embodiment, processing the skeletal feature points according to preset rules further includes: calculating the placement angle of the joint prosthesis based on the femoral mechanical axis, the femoral condyle line, and the tibial mechanical axis.

[0162] In one embodiment, processing skeletal feature points according to preset rules includes: optimizing skeletal feature points according to preset rules.

[0163] Specifically, for some geometrically significant feature points, the skeletal feature points obtained through any of the above embodiments may not be very accurate. Therefore, it is necessary to further optimize the obtained skeletal feature points to improve their positional accuracy. For example, the skeletal feature points can be projected onto the corresponding physiological axis, and projection points within a certain range on the physiological axis can be selected as the optimized skeletal feature points in the image to be processed.

[0164] Specifically, in combination Figure 17 As shown, Figure 17 This is a schematic diagram of feature point position optimization in one embodiment. Taking the distal tangent point of the femur as an example, after registration to obtain the corresponding skeletal feature point position, the skeletal feature point is projected onto the femoral mechanical axis according to the definition. Within a certain range, the point projected as the farthest point on the femoral mechanical axis is selected as the optimized skeletal feature point. At this time, the line connecting the two distal tangent points, i.e. the optimized skeletal feature points, is tangent to the femur.

[0165] In the above embodiments, by optimizing some geometrically significant feature points, more accurate skeletal feature point positions can be obtained in the image to be processed.

[0166] It should be understood that, although Figure 2 , Figure 4 and Figure 13 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated in this document, there is no strict order in which these steps are executed; they can be performed in other orders. Figure 2 , Figure 4 and Figure 13 At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0167] In one embodiment, such as Figure 18 As shown, a target feature point extraction device is provided, including: a data acquisition module 100, a template query module 200, a registration module 300, and a target extraction module 400, wherein:

[0168] The data acquisition module 100 is used to acquire the image to be processed.

[0169] The template query module 200 is used to obtain the template corresponding to the image to be processed and to obtain the positional relationship between the template and the target feature points; the template is an image generated based on the sample image, and the standard feature points are located in the template.

[0170] The registration module 300 is used to register the template to the image to be processed.

[0171] The target extraction module 400 is used to determine the target feature points in the image to be processed that correspond to the standard feature points, based on the positional relationship between the template and the target feature points, and in combination with the positional relationship between the registered template and the image to be processed.

[0172] In one embodiment, the target feature point extraction device may further include:

[0173] Sample acquisition template, used to acquire several sample images.

[0174] The sample registration module is used to select an initial template from several sample images and register the initial template to the remaining sample images to obtain registered images.

[0175] The statistical image calculation module is used to calculate the statistical image corresponding to the registered image.

[0176] The similarity judgment module is used to use the statistical image as the template when the similarity between the statistical image and the initial template meets the requirements; otherwise, it uses the statistical image as the new initial template and returns to the step of registering the initial template to the remaining sample images to obtain the registered images, until the similarity between the statistical image and the initial template meets the requirements.

[0177] In one embodiment, the above-mentioned statistical image calculation module may include:

[0178] The position acquisition unit is used to acquire the initial position of the corresponding point in each registration image.

[0179] The statistical image generation unit is used to calculate the average value of the initial positions of the corresponding points as the target positions of the corresponding points, and to generate a statistical image based on the target positions of the corresponding points.

[0180] In one embodiment, the target feature point extraction device may further include:

[0181] The instruction acquisition module is used to receive standard feature point configuration instructions for the template.

[0182] The feature acquisition module is used to configure the corresponding standard feature points in the template according to the standard feature point configuration instructions.

[0183] In one embodiment, the target feature point extraction device may further include:

[0184] The corresponding point distance calculation module is used to calculate the distance between the statistical image and the corresponding points in the initial template.

[0185] The similarity calculation module is used to calculate the similarity between the statistical image and the initial template based on the distance of all corresponding points.

[0186] In one embodiment, the target feature point extraction device may further include:

[0187] The first preprocessing module is used to preprocess the image to be processed. The preprocessing includes at least one of a surface point cloud extraction unit, a point cloud downsampling unit, and a normalization unit.

[0188] The second preprocessing module is used to preprocess the sample image. The preprocessing includes at least one of a surface point cloud extraction unit, a point cloud downsampling unit, and a normalization unit.

[0189] The surface point cloud extraction unit is used to extract the vertices of all grids in the image to be processed and the sample image to obtain the surface point cloud.

[0190] The point cloud downsampling unit is used to divide the image to be processed into at least one processing region, and to sample the point in the processing region that is closest to the center of the region to be processed as the sampling point of the processing region.

[0191] The normalization unit is used to align points in the image to be processed to the same coordinate space.

[0192] In one embodiment, the registration module 300 may further include:

[0193] The registration function acquisition unit is used to acquire and initialize the registration function.

[0194] The registration function optimization unit is used to input the image to be processed and the template into the registration function to optimize the parameters in the registration function.

[0195] The registration function judgment unit is used to determine that the template and the image to be processed have been registered when the change in the parameters of the registration function after optimization is less than the preset standard. Otherwise, the image to be processed and the template are input into the registration function after parameter optimization to optimize the parameters in the registration function.

[0196] In one embodiment, the target extraction module 400 further includes:

[0197] The normal vector acquisition unit is used to acquire the normal vectors of the standard feature points in the template after registration with the image to be processed, when the standard feature points are on the template surface.

[0198] The distance calculation unit is used to calculate the distance between the intersection point and the standard feature point when the normal vector intersects with the registered skeleton image to be processed.

[0199] The first skeletal feature determination unit is used to determine the intersection point as the target feature point when the distance between the intersection point and the standard feature point is less than a preset distance.

[0200] The second skeletal feature determination unit is used to select the point in the image to be processed that is closest to the standard feature point in the registered template as the target feature point when there is no intersection or the distance between the intersection and the standard feature point is greater than a preset distance.

[0201] In one embodiment, the target extraction module 400 further includes:

[0202] The associated point acquisition unit is used to select a preset number of points from the template surface as associated points when the standard feature points are not on the surface of the standard plate.

[0203] Target point acquisition unit: used to determine the target point of the associated point in the registered image based on the registration relationship between the template and the image to be processed.

[0204] The third skeletal feature determination unit is used to calculate the target feature points of the image to be processed based on the target points.

[0205] In one embodiment, the target feature point extraction device further includes:

[0206] A standard skeletal feature acquisition module is used to acquire at least one of femoral feature points and tibial feature points.

[0207] In one embodiment, such as Figure 19 As shown, a skeletal data processing device is provided, including: a skeletal image acquisition module 500, a skeletal feature extraction module 600, and a skeletal feature processing module 700, wherein:

[0208] The skeleton image acquisition module 500 is used to acquire the skeleton image to be processed.

[0209] The skeletal feature extraction module 600 is used to process the skeletal image to be processed according to any one of the above embodiments of the target feature point extraction device to obtain skeletal feature points.

[0210] The skeletal feature processing module 700 processes skeletal feature points according to preset rules.

[0211] In one embodiment, the skeletal feature processing module 700 further includes:

[0212] The axis calculation unit is used to calculate at least one of the femoral mechanical axis, the femoral condyle line, and the tibial mechanical axis based on skeletal feature points.

[0213] In one embodiment, the skeletal feature extraction module 600 further includes:

[0214] The skeletal feature optimization unit is used to optimize skeletal feature points according to preset rules.

[0215] Specific limitations regarding the target feature point extraction device can be found in the limitations of the target feature point extraction method described above, and will not be repeated here. Each module in the aforementioned target feature point extraction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0216] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 20As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a target feature point extraction method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0217] Those skilled in the art will understand that Figure 20 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0218] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0219] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0220] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0221] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0222] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0223] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for extracting target feature points, characterized in that, The target feature point extraction method includes: Obtain the image to be processed; Obtain a template corresponding to the image to be processed, and obtain the positional relationship between the template and the standard feature points; the template is an image generated based on the sample image, and the standard feature points are located in the template; The template is registered to the image to be processed; Based on the positional relationship between the template and the standard feature points, and combined with the positional relationship between the registered template and the image to be processed, the target feature points in the image to be processed corresponding to the standard feature points are determined; The step of determining the target feature point in the image to be processed corresponding to the standard feature point based on the positional relationship between the template and the standard feature point, and in conjunction with the positional relationship between the registered template and the image to be processed, further includes: When the standard feature point is not on the surface of the standard plate, a preset number of points are selected from the surface of the template as associated points based on the standard feature point. Based on the registration relationship between the template and the image to be processed, the target point of the associated point in the registered image is determined; Calculate the target feature points of the image to be processed based on the target points.

2. The target feature point extraction method according to claim 1, characterized in that, Before obtaining the template corresponding to the image to be processed, the template is generated by the following steps: Acquire several sample images; An initial template is selected from several sample images, and the initial template is registered to the remaining sample images to obtain registered images; Calculate the statistical image corresponding to the registered image; When the similarity between the statistical image and the initial template meets the requirements, the statistical image is used as the template; otherwise, the statistical image is used as the new initial template, and the process returns to the step of registering the initial template with the remaining sample images to obtain registered images, until the similarity between the statistical image and the initial template meets the requirements.

3. The target feature point extraction method according to claim 2, characterized in that, The calculation of the statistical image corresponding to the registered image includes: Obtain the initial position of the corresponding point in each of the registered images; The average value of each initial position of the corresponding point is calculated as the target position of the corresponding point, and the statistical image is generated based on the target position of the corresponding point.

4. The method according to claim 2, characterized in that, After generating the template, the process also includes: Receive standard feature point configuration instructions for the template; Configure the corresponding standard feature points in the template according to the standard feature point configuration instructions.

5. The target feature point extraction method according to claim 2, characterized in that, After calculating the statistical image corresponding to the registered image, the method further includes: Calculate the distance between the statistical image and the corresponding point in the initial template; The similarity between the statistical image and the initial template is calculated based on the distances between all corresponding points.

6. The target feature point extraction method according to claim 2, characterized in that, The image to be processed and the sample image are three-dimensional mesh point cloud images; before registering the initial template to the remaining sample images, the method further includes: preprocessing the image to be processed; and / or Before registering the template to the image to be processed, the method further includes: The sample image is preprocessed; the preprocessing includes at least one of surface point cloud extraction, point cloud downsampling, and normalization. The extraction of surface point cloud involves extracting the vertices of all grids in the image to be processed and the sample image to obtain surface point cloud; The point cloud downsampling involves dividing the image to be processed into at least one processing region, and sampling the point in the processing region that is closest to the center of the processing region as the sampling point of the processing region. The normalization process involves aligning the points in the image to be processed to the same coordinate space.

7. The target feature point extraction method according to claim 1, characterized in that, The step of registering the template to the image to be processed includes: Obtain the registration function and initialize the registration function; The image to be processed and the template are input into the registration function to optimize the parameters in the registration function; When the change in the parameters of the registration function after optimization is less than a preset standard, it is determined that the template and the image to be processed have been registered. Otherwise, the image to be processed and the template are input into the registration function after parameter optimization to optimize the parameters in the registration function.

8. The target feature point extraction method according to claim 1, characterized in that, The step of determining the target feature points in the image to be processed corresponding to the standard feature points based on the positional relationship between the template and the standard feature points, and in conjunction with the positional relationship between the registered template and the image to be processed, includes: When the standard feature points are on the template surface, obtain the normal vectors of the standard feature points in the template after registration with the image to be processed; When the normal vector intersects with the registered image to be processed, the distance between the intersection point and the standard feature point is calculated. When the distance between the intersection point and the standard feature point is less than a preset distance, the intersection point is taken as the target feature point; If there is no intersection point or the distance between the intersection point and the standard feature point is greater than the preset distance, then the point closest to the standard feature point in the registered template is selected from the image to be processed as the target feature point.

9. The target feature point extraction method according to any one of claims 1 to 8, characterized in that, The image to be processed is a skeletal image, and the standard feature points are skeletal feature points, including at least one of femoral feature points and tibial feature points.

10. A method for processing skeletal data, characterized in that, The skeletal data processing method includes: Obtain the skeleton image to be processed; The target feature point extraction method according to any one of claims 1 to 9 processes the skeletal image to be processed to obtain skeletal feature points; The skeletal feature points are processed according to preset rules.

11. The skeletal data processing method according to claim 10, characterized in that, The process of processing the skeletal feature points according to preset rules includes: At least one of the femoral mechanical axis, the femoral condyle line, and the tibial mechanical axis is calculated based on the skeletal feature points.

12. The skeletal data processing method according to claim 10, characterized in that, The step of processing the skeletal feature points according to preset rules further includes: The skeletal feature points are optimized according to preset rules.

13. A target feature point extraction device, characterized in that, The device includes: The data acquisition module is used to acquire the image to be processed; The template query module is used to obtain a template corresponding to the image to be processed, and to obtain the positional relationship between the template and the standard feature points; the template is an image generated based on the sample image, and the standard feature points are located in the template; A registration module is used to register the template to the image to be processed; The target extraction module is used to determine the target feature points in the image to be processed that correspond to the standard feature points, based on the positional relationship between the template and the standard feature points, and in combination with the positional relationship between the registered template and the image to be processed. The target extraction module includes: The association point acquisition unit is used to select a preset number of points as association points from the template surface based on the standard feature points when the standard feature points are not on the surface of the standard plate. The target point acquisition unit is used to determine the target point of the associated point in the registered image based on the registration relationship between the template and the image to be processed; The third skeletal feature determination unit is used to calculate the target feature points of the image to be processed based on the target points.

14. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 9 or 10 to 12.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9 or 10 to 12.

16. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 9 or 10 to 12.

Images