Surface registration-based target feature extraction method, device, equipment and medium

Abstract

This application relates to a target feature extraction method, apparatus, device, and medium based on surface registration. The method includes: acquiring three-dimensional surface mesh data corresponding to a target to be processed; acquiring pre-generated three-dimensional template data corresponding to the target to be processed; performing surface registration between the surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed to establish a correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed; reading three-dimensional template feature annotation information from the three-dimensional template data; and mapping the three-dimensional template feature annotation information to the three-dimensional surface mesh data to be processed according to the correspondence to extract target features from the three-dimensional surface mesh data to be processed. This method can improve processing efficiency.

Description

[0001] This invention patent application is a divisional application of Chinese invention patent application filed on March 2, 2021, with application number 2021102291660 and titled "Target Feature Extraction Method, Apparatus, Device and Medium Based on Surface Registration". Technical Field

[0002] This application relates to the field of image processing technology, and in particular to a method, apparatus, device and medium for target feature extraction based on surface registration. Background Technology

[0003] With the development of image processing technology, the automatic feature extraction of three-dimensional target organs in CT or MR images can be widely applied in medical image-based assisted diagnosis and treatment. The extracted feature points, linear structures, or regions help doctors diagnose diseases, plan surgeries, or serve as input for subsequent intelligent computer processing, acting as an indispensable key step in fully automated algorithms.

[0004] Traditional techniques include feature detection methods based on parameter fitting, feature detection methods based on local feature analysis, and feature point detection methods based on machine learning. However, these methods require the target to have local specific features; otherwise, the target cannot be located. Alternatively, different algorithms need to be designed for feature points, feature lines, or feature regions. There is no way to design a single algorithm to simultaneously detect multiple types of feature structures.

[0005] Therefore, in order to improve the above method, a feature point, line or region extraction method based on image registration is introduced, which can overcome the above problems.

[0006] However, feature point, line, or region extraction methods based on image registration use a three-dimensional matrix as the target storage carrier, resulting in very low efficiency throughout the entire operation. Furthermore, the registration process focuses on the whole, leading to a very large processing load and further reducing efficiency. Summary of the Invention

[0007] Therefore, it is necessary to provide a surface registration-based target feature extraction method, apparatus, device, and medium that can improve processing efficiency in response to the above-mentioned technical problems.

[0008] A target feature extraction method based on surface registration, the method comprising:

[0009] Obtain the 3D surface mesh data of the target to be processed;

[0010] Obtain pre-generated 3D template data corresponding to the target to be processed;

[0011] The surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed are surface registered to establish the correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed.

[0012] Read the 3D template feature annotation information from the 3D template data;

[0013] Based on the correspondence, the three-dimensional template feature annotation information is mapped to the three-dimensional surface mesh data to be processed, so as to extract the target features in the three-dimensional surface mesh data to be processed.

[0014] In one embodiment, the method for generating the three-dimensional template data includes:

[0015] Obtain the three-dimensional surface mesh data of the sample;

[0016] The surface template mesh data is obtained by surface registration of the three-dimensional surface mesh data of the sample.

[0017] The three-dimensional template data is obtained by annotating the features on the surface template mesh data.

[0018] In one embodiment, the step of annotating the features on the surface template mesh data to obtain the three-dimensional template data includes:

[0019] Record the index or point set index group of the points corresponding to the target feature structure on the three-dimensional template surface mesh data;

[0020] The three-dimensional template data is obtained based on the index of the points or the index group of the point set and the surface template mesh data.

[0021] In one embodiment, the step of performing surface registration on the sample three-dimensional surface mesh data to obtain surface template mesh data includes:

[0022] Obtain the corresponding points in the three-dimensional surface mesh data of each sample;

[0023] After processing the corresponding points according to preset rules, surface template mesh data is obtained.

[0024] In one embodiment, before acquiring the corresponding points in the three-dimensional surface mesh data of each of the samples, the method further includes:

[0025] Align all sample 3D mesh data to the same coordinate space.

[0026] In one embodiment, aligning all sample 3D mesh data to the same coordinate space includes:

[0027] All sample 3D mesh data are aligned to the same coordinate space using affine transformation techniques.

[0028] In one embodiment, the step of mapping the three-dimensional template feature annotation information to the three-dimensional surface mesh data to be processed according to the correspondence relationship, so as to extract the target features in the three-dimensional surface mesh data to be processed, includes:

[0029] Obtain the closest point in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and use the closest point as the feature point corresponding to the target to be processed; or

[0030] Obtain the set of points in the three-dimensional surface mesh data to be processed that are closest to the template feature annotation information, and use the set of closest points as the feature line or feature region corresponding to the target to be processed.

[0031] In one embodiment, the step of performing surface registration between the surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed includes:

[0032] The surface template mesh data in the three-dimensional template data is used as floating mesh data, and the three-dimensional surface mesh data to be processed is used as target mesh data, so as to perform surface registration of the floating mesh data onto the target mesh data; the surface registration includes the mapping of the floating mesh data to points in the target mesh data and the processing of the mapped points.

[0033] A target feature extraction device based on surface registration, the device comprising:

[0034] The data acquisition module is used to acquire the three-dimensional surface mesh data of the target to be processed.

[0035] The template data acquisition module is used to acquire pre-generated three-dimensional template data corresponding to the target to be processed;

[0036] The first surface registration module is used to perform surface registration between the surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed, so as to establish the correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed.

[0037] The mapping module is used to read the three-dimensional template feature annotation information from the three-dimensional template data; and map the three-dimensional template feature annotation information to the three-dimensional surface mesh data to be processed according to the correspondence, so as to extract the target features in the three-dimensional surface mesh data to be processed.

[0038] In one embodiment, the device further includes:

[0039] The sample data acquisition module is used to acquire the three-dimensional surface mesh data of the sample;

[0040] The second surface registration module is used to perform surface registration on the three-dimensional surface mesh data of the sample to obtain surface template mesh data.

[0041] The template generation module is used to annotate the features on the surface template mesh data to obtain three-dimensional template data.

[0042] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method in any of the above embodiments.

[0043] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method in any of the above embodiments.

[0044] The aforementioned target feature extraction method, apparatus, device, and medium based on surface registration process three-dimensional surface mesh data rather than three-dimensional matrix data, thus improving processing efficiency. Secondly, since the three-dimensional template data stores three-dimensional template feature annotation information, after surface registration, it can directly extract various types of target features from the three-dimensional surface mesh data to be processed in one go through mapping, further improving processing efficiency. Finally, since it focuses on feature-rich surface regions, i.e., performing surface registration rather than image registration, it further improves processing efficiency. Attached Figure Description

[0045] Figure 1 This is an application environment diagram of a surface registration-based target feature extraction method in one embodiment;

[0046] Figure 2 This is a flowchart illustrating a target feature extraction method based on surface registration in one embodiment;

[0047] Figure 3 This is a schematic diagram of the hip joint segmentation effect in one embodiment;

[0048] Figure 4 This is a schematic diagram of the three-dimensional reconstruction effect of the hip joint in one embodiment;

[0049] Figure 5 This is a schematic diagram illustrating the mapping between three-dimensional template data and three-dimensional surface mesh data to be processed in one embodiment;

[0050] Figure 6 A flowchart illustrating how three-dimensional template data is generated in one embodiment;

[0051] Figure 7 This is a schematic diagram illustrating the initial state of the three-dimensional template data and the three-dimensional surface mesh data to be processed in one embodiment.

[0052] Figure 8 This is a schematic diagram of coarse mesh alignment in one embodiment;

[0053] Figure 9 This is a schematic diagram of rigid mesh registration in one embodiment;

[0054] Figure 10 This is a schematic diagram of mesh elastic registration in one embodiment;

[0055] Figure 11 This is a flowchart illustrating a target feature extraction method based on surface registration in another embodiment;

[0056] Figure 12 This is a structural block diagram of a target feature extraction device based on surface registration in one embodiment;

[0057] Figure 13 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0058] 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.

[0059] The target feature extraction method based on surface registration 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 can receive 3D images scanned by medical imaging device 104 and stored as a 3D matrix, or retrieve 3D images scanned by medical imaging device 104 and stored as a 3D matrix from a database, etc. It then performs 3D reconstruction on these images to obtain 3D surface mesh data, and subsequently obtains pre-generated 3D template data corresponding to the target to be processed. The surface template mesh data in the 3D template data and the 3D surface mesh data to be processed are then registered to establish a correspondence between the two. Finally, 3D template feature annotation information is read from the 3D template data. Based on the correspondence, the 3D template feature annotation information is mapped to the 3D surface mesh data to be processed to extract target features from the 3D surface mesh data. In this way, it processes 3D surface mesh data rather than 3D matrix data, thus improving processing efficiency. Secondly, since the 3D template data stores 3D template feature annotation information, after surface registration, it can directly extract various types of target features from the 3D surface mesh data to be processed in one go through mapping, further improving processing efficiency. Finally, since it focuses on feature-rich surface areas, i.e., performing surface registration rather than image registration, it further improves processing efficiency.

[0060] The terminal 102 may be, but is not limited to, various personal computers, laptops, smartphones, tablets, portable wearable devices, and functional modules and dedicated circuits of the medical imaging equipment itself. In this embodiment, the terminal 102 may include the patient's mobile terminal device and / or the medical operator's mobile terminal device. The medical imaging equipment 104 includes, but is not limited to, various imaging devices, such as CT imaging equipment (CT: Computed Tomography, which uses a precisely collimated X-ray beam and a highly sensitive detector 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. through CT scans), magnetic resonance imaging equipment (which is a type of tomographic imaging that uses the magnetic resonance phenomenon to obtain electromagnetic signals from the human body and reconstruct human body information images), positron emission tomography (PET) equipment, positron emission tomography / magnetic resonance imaging (PET / MR) systems, etc.

[0061] In one embodiment, such as Figure 2 As shown, a target feature extraction method based on surface registration is provided, which is then applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps:

[0062] S202: Obtain the 3D surface mesh data of the target to be processed.

[0063] Specifically, the three-dimensional surface mesh data to be processed refers to 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.

[0064] In the field of medical imaging, 3D scanned CT or MR medical image data are generally 3D images, that is, medical image data stored in the form of a 3D matrix. This 3D image includes the target to be processed, such as the target organ or tissue from which features are to be extracted. In other embodiments, the 3D surface mesh data to be processed can also be obtained by 3D reconstruction of the human body surface for facial feature extraction and feature region segmentation, used for functions such as facial expression recognition, face recognition, and facial feature analysis. Specifically, when recognizing faces, images can be acquired using a 3D camera for feature extraction. The images acquired by the 3D camera, such as facial point cloud data obtained from scanning a face, can be used. The terminal can extract facial features from this data. The difference between this case and the above examples is that the acquired images are no longer 3D matrix data like CT or MR, but point cloud data; therefore, the only difference lies in the surface reconstruction algorithm. For point cloud data, algorithms such as Poisson surface reconstruction can be used to reconstruct 3D mesh data. Subsequently, the same method can be used to perform 3D mesh data registration to complete feature extraction. For example, it can extract facial features, facial symmetry lines, and different areas of the face (forehead, cheeks, etc.), and apply them to different scenarios such as face recognition and facial feature analysis.

[0065] Specifically, 3D reconstruction may 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 this mask data. 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. Figure 3 This is a schematic diagram of the hip joint segmentation effect in one embodiment. Methods for 3D reconstruction of 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, etc. Specifically, Figure 4 This is a schematic diagram of the three-dimensional reconstruction effect of the hip joint in one embodiment.

[0066] The application of hip joint surface feature extraction in hip replacement surgery is used as an example. The terminal can segment the hip bone in CT data using image segmentation technology based on a deep learning fully convolutional network to obtain segmentation mask data for one hip bone. Then, the segmentation mask data of the hip bone is used to perform three-dimensional reconstruction of the bone surface. Specifically, the above methods will be explained in detail:

[0067] The Marching Cube algorithm includes: reconstructing the contour of the hip bone segmentation mask data using the Marching Cube algorithm. The set threshold for contour reconstruction can be any value between the background pixel value and the target structure pixel value in the segmentation mask data, such as the value between the hip bone and the background pixels. In this example, background pixels in the segmentation mask data are represented by 0, and target structure pixels are represented by 1; therefore, any value between 0 and 1 can be selected. In this example, the threshold chosen is 0.5.

[0068] Interpolation reconstruction using the Marching Cube algorithm near the contour based on a surface threshold involves: specifically, the terminal takes the original CT data, combines it with the contour of the segmentation mask data, and then uses the Marching Cube algorithm to interpolate and reconstruct the CT data near the contour based on a bone surface threshold. The threshold is selected as any grayscale value between the edge of the target structure and the background. In this example, the target structure is the hip bone, and any pixel CT value between the surrounding tissue and the hip bone can be selected. In this example, the threshold chosen is 150 HU. The reconstruction range of the Marching Cube is limited to 1-5 pixels from the contour line of the segmentation mask data. In this example, the Marching Cube reconstruction is performed within a range of 3 pixels from the contour line.

[0069] The Poisson surface reconstruction algorithm includes: performing three-dimensional reconstruction of the contour edge points of segmented mask data based on the Poisson surface reconstruction algorithm.

[0070] S204: Obtain the pre-generated 3D template data corresponding to the target to be processed.

[0071] Specifically, the 3D template data includes 3D template surface mesh data and 3D template feature annotation information. It is a pre-generated template, that is, a template corresponding to the 3D surface mesh data to be processed. This template is generated based on the sample 3D surface mesh data. For example, the surface mesh template is obtained after registering the sample 3D surface mesh data. Then, the target features of the target to be processed in the surface mesh template are extracted manually or semi-automatically. Semi-automatic extraction is mainly for points that are inconvenient to select manually. It requires the use of some algorithms to extract target features. For example, for the center point of the acetabulum, some points on the acetabulum need to be selected manually, then a sphere is fitted to the selected points, and then the center of the sphere is calculated.

[0072] Optionally, the 3D template data can be stored according to organs. This way, after obtaining the 3D surface mesh data to be processed, the corresponding 3D template data that has been stored can be selected according to the organ corresponding to the 3D surface mesh data to be processed.

[0073] S206: Perform surface registration between the surface template mesh data in the 3D template data and the 3D surface mesh data to be processed, so as to establish the correspondence between the surface template mesh data and the 3D surface mesh data to be processed.

[0074] Specifically, surface registration refers to unifying the surface template mesh data and the 3D surface mesh data to be processed into the same coordinate system. Registration establishes a one-to-one correspondence between the positions on the surface template mesh data and the positions on the 3D surface mesh data to be processed, thus laying the foundation for subsequent feature mapping.

[0075] Surface registration can include linear registration and elastic registration. In some special cases, linear registration includes mesh global registration, affine registration, or rigid body registration. Elastic registration relies on the mesh having already completed linear registration. The surface registration methods in this embodiment include, but are not limited to, the Coherent Point Drift (CPD) algorithm, which can achieve various types of linear and elastic registration; the ICP algorithm in rigid registration algorithms, which can achieve affine registration with scale transformation parameters; NDT (Normal-Distributions Transform); and phase correlation algorithms. Global registration includes, but is not limited to, principal axis alignment algorithms based on PCA; feature matching methods based on the RANSAC framework; brute-force search (through traversal search in angle and orientation); 4PCS; and Super4PCS.

[0076] S208: Read the 3D template feature annotation information from the 3D template data; according to the correspondence, map the 3D template feature annotation information to the 3D surface mesh data to be processed, so as to extract the target features in the 3D surface mesh data to be processed.

[0077] Specifically, the mapping refers to mapping the positions of the surface template mesh data and the three-dimensional surface mesh data to be processed, thereby forming a correspondence between the various positions. This allows the reading of the three-dimensional template feature annotation information, thus determining the target features corresponding to the three-dimensional template feature annotation information in the three-dimensional surface mesh data to be processed.

[0078] Specifically, in combination Figure 5 As shown, Figure 5This is a schematic diagram illustrating the mapping between 3D template data and 3D surface mesh data to be processed in one embodiment. Specifically, the 3D template feature annotation information, including feature points, feature lines, and feature regions, from the 3D template data is mapped onto the 3D surface mesh data to be processed, thereby enabling the extraction of target features from the 3D surface mesh data.

[0079] The target feature extraction method based on surface registration described above processes three-dimensional surface mesh data rather than three-dimensional matrix data, thus improving processing efficiency. Secondly, since the three-dimensional template data stores three-dimensional template feature annotation information, after surface registration, it can directly extract various types of target features from the three-dimensional surface mesh data to be processed in one go through mapping, further improving processing efficiency. Finally, since it focuses on feature-rich surface regions, i.e., performing surface registration rather than image registration, it further improves processing efficiency.

[0080] In one embodiment, see Figure 6 As shown, Figure 6 This is a flowchart illustrating a method for generating 3D template data in one embodiment. The method for generating 3D template data may include:

[0081] S602: Obtain the three-dimensional surface mesh data of the sample.

[0082] Specifically, the sample 3D surface mesh data can be obtained by 3D reconstruction based on different 3D images. The specific 3D reconstruction methods can be found above. Taking the hip as an example again, the terminal first collects a large amount of hip medical image data from different patients as a training set. Then, the medical image data in the training set is segmented and reconstructed using the aforementioned 3D reconstruction method to obtain the sample 3D surface mesh data. If the medical image data in the training set is already 3D surface mesh data, then 3D reconstruction is unnecessary.

[0083] One point that should also be noted is that when there is only one set of sample 3D surface mesh data, it is directly used as the surface template mesh data of the 3D template data. If there are at least two sets of sample 3D surface mesh data, the sample 3D surface mesh data is surface registered to obtain surface template mesh data, and after feature annotation, 3D template data is generated. For details, please refer to the following text.

[0084] S604: Perform surface registration on the sample's 3D surface mesh data to obtain surface template mesh data.

[0085] Specifically, the surface registration here is the same as the surface registration method mentioned above, except that the object is transformed into sample three-dimensional surface mesh data, so it will not be described again.

[0086] Specifically, when the positions of the corresponding points are basically consistent after registration, the position of each corresponding point is directly obtained as the surface mesh data. In other embodiments, all sample mesh data can be aligned to a space, and then the positions of the corresponding points of all mesh data can be averaged to obtain the average surface mesh data. Specifically, the position averaging here refers to the process where, after registering the sample 3D surface mesh data in the training set, the terminal can obtain the position of the corresponding point, and then the average position of the position is obtained. Finally, the average position of all corresponding points is obtained to obtain the average 3D surface mesh data.

[0087] S606: The three-dimensional template data is obtained by annotating the features on the surface template mesh data.

[0088] Specifically, the features to be extracted are labeled on the average 3D surface mesh data to obtain 3D template data. This labeling can be manual or semi-automatic. Semi-automatic labeling is mainly for points that are inconvenient to select manually. It requires combining some algorithms to extract the target features. For example, to extract the center point of the acetabulum, some points on the acetabulum need to be manually selected, then a sphere is fitted to the selected points, and then the center of the sphere is calculated.

[0089] In the above embodiments, surface mesh data is obtained by registering the sample's three-dimensional surface mesh data, and then the features in the surface mesh data are labeled to generate three-dimensional template data. Since features can be labeled as needed, this method is suitable for various types of feature extraction, including different types of feature structures on the surface and inside the target, such as feature points, feature lines, and feature regions. It is applicable to both locally specific and non-locally specific features. Furthermore, multiple features can be labeled, resulting in high extraction efficiency and suitability for rapid extraction of a large number of features. It can extract all required features in parallel at once, with no upper limit on the number of features to be detected, and the algorithm efficiency does not decrease with an increase in the number of features to be extracted. Finally, it is not limited to features with volume such as points, lines, and surfaces; it uses the surface of the target structure as a carrier for template matching and feature extraction, which has a natural advantage for feature extraction on surface structures, especially for features without volume such as points, lines, and surfaces.

[0090] In one embodiment, three-dimensional template data is obtained by annotating the features on the surface template mesh data, including: recording the index or point set index group of the points corresponding to the target feature structure on the three-dimensional template surface mesh data; and obtaining the three-dimensional template data based on the index or point set index group of the points and the surface template mesh data.

[0091] During calibration, the terminal can record a list of vertex indices corresponding to each feature structure on the average 3D surface mesh data. If it is a point structure, the indices of the corresponding vertex on the template are recorded; if it is a line or surface structure, an array of indices of all points belonging to that line or surface structure on the template is recorded.

[0092] In the above embodiments, by applying the vertices in the grid data, the position of the target feature can be accurately recorded.

[0093] In one embodiment, surface registration is performed on the sample three-dimensional surface mesh data to obtain surface template mesh data, including: obtaining the corresponding points in the three-dimensional surface mesh data of each sample; and processing the corresponding points according to preset rules to obtain surface template mesh data.

[0094] In one embodiment, before obtaining the corresponding points in the three-dimensional surface mesh data of each sample, the method further includes: aligning all sample three-dimensional mesh data to the same coordinate space.

[0095] In one embodiment, aligning all sample 3D mesh data to the same coordinate space includes: aligning all sample 3D mesh data to the same coordinate space using an affine transformation technique.

[0096] Specifically, if all sample 3D surface mesh data were captured from the same viewpoint, meaning they reside in the same coordinate space, the terminal first determines the corresponding points within each sample's 3D surface mesh data. These corresponding points can be determined by location, for example, by identifying the closest points among different sample 3D surface mesh data. Then, the corresponding points are processed to obtain the surface template mesh data, such as through elastic or rigid registration. If all sample 3D surface mesh data were not captured from the same viewpoint, meaning they reside in different coordinate spaces, then all sample 3D mesh data are preferentially aligned to the same coordinate space. This can be achieved through affine transformation techniques, including translation, rotation, and scaling, to ensure linear alignment of all sample 3D surface mesh data as a whole.

[0097] Specifically, in order to ensure the orderliness of surface registration, in this embodiment, a sample three-dimensional surface mesh data is randomly selected as the reference mesh data, and then the remaining sample three-dimensional surface mesh data in the training set are surface registered to the reference mesh data. That is, the remaining sample three-dimensional surface mesh data are registered to the space corresponding to the reference mesh data through surface registration technology, such as through affine transformation technology. Specifically, all sample three-dimensional mesh data are aligned to the same coordinate space through coarse alignment, and then averaged to obtain the corresponding average three-dimensional surface mesh data.

[0098] The aforementioned surface registration includes at least one of coarse mesh alignment, rigid mesh registration, and elastic mesh registration. The execution of coarse mesh alignment, rigid mesh registration, and elastic mesh registration has a specific order. Generally, coarse mesh alignment, rigid mesh registration, and elastic mesh registration are performed sequentially. When the two mesh data are already well aligned in their initial state, or when the rigid mesh registration algorithm is a global registration algorithm that does not depend on coarse mesh alignment, coarse mesh alignment can be omitted here.

[0099] For specific limitations on coarse mesh alignment, rigid mesh registration, and flexible mesh registration, please refer to the following text.

[0100] In one embodiment, according to the correspondence, the three-dimensional template feature annotation information is mapped to the three-dimensional surface mesh data to be processed to extract the target features in the three-dimensional surface mesh data to be processed, including: obtaining the closest point in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and taking the closest point as the feature point corresponding to the target to be processed; or obtaining the set of the closest points in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and taking the set of the closest points as the feature line or feature region corresponding to the target to be processed.

[0101] Specifically, this embodiment defines the mapping method, which is implemented through nearest neighbor point lookup. This can be further divided into feature point mapping and feature line and feature region mapping. Feature point mapping involves finding the closest point in the 3D surface mesh data to be processed that corresponds to the template feature annotation information, and using this closest point as the feature point corresponding to the target, i.e., the target feature. The mapping of feature lines and feature regions can be broken down into several sets of feature points, thus achieving the mapping through multiple feature point mapping steps.

[0102] In the above embodiments, template matching and feature extraction are performed using the surface of the target structure as a carrier. This method has a natural advantage for feature extraction from surface structures, especially for features such as points, lines, and surfaces that do not have volume. Furthermore, the feature extraction efficiency is high, suitable for the rapid extraction of a large number of features. It can extract all required features in parallel at once, with no upper limit on the number of features to be detected, and the algorithm efficiency does not decrease as the number of features to be extracted increases.

[0103] In one embodiment, surface registration is performed between the surface template mesh data in the 3D template data and the 3D surface mesh data to be processed to establish a correspondence between the surface template mesh data and the 3D surface mesh data to be processed. This includes: using the surface template mesh data in the 3D template data as floating mesh data and the 3D surface mesh data to be processed as target mesh data, and performing surface registration of the floating mesh data onto the target mesh data; wherein surface registration includes the correspondence of the floating mesh data to points in the target mesh data and the processing of the corresponding points.

[0104] Specifically, surface registration technology is implemented through multiple steps, sequentially performing coarse mesh alignment, rigid mesh registration, and elastic mesh registration. Coarse mesh alignment can be omitted in some cases, such as when the two mesh datasets are already well aligned initially, or when the rigid mesh registration algorithm uses a global registration algorithm that does not rely on the coarse mesh alignment.

[0105] Specifically, see Figure 7 As shown, Figure 7 This is a schematic diagram of the initial state of the surface template mesh data and the three-dimensional surface mesh data to be processed in the three-dimensional template data in one embodiment. In this schematic diagram of the initial state, one is the surface template mesh data in the three-dimensional template data and the other is the three-dimensional surface mesh data to be processed. In the registration, the surface template mesh data in the three-dimensional template data is used as floating mesh data and the three-dimensional surface mesh data to be processed is used as target mesh data. The floating mesh data is then surface registered onto the target mesh data.

[0106] Specifically, see Figure 8 , Figure 8 This is a schematic diagram of coarse mesh alignment in one embodiment. In this embodiment, the purpose of the coarse mesh alignment algorithm is to roughly align two structurally similar targets in space and simultaneously perform scale matching. This embodiment employs an alignment method based on PCA principal axis detection. This method only uses point cloud data formed by vertices in the mesh. Assuming the target mesh data is Pt and the floating mesh data is Pf, the following steps are performed:

[0107] First, perform translation alignment: calculate the center of the two grid data, and use translation transformation to align the centers of the two point clouds by moving grid data Pf onto grid data Pt.

[0108] Secondly, the principal axis coordinate system is established: using the PCA algorithm, principal component analysis is performed on the grid data Pt and Pf respectively to obtain 3*3 matrices Rt and Rf formed by the three principal component vectors. This matrix represents the rotation matrix of the point cloud from the current coordinate system to the coordinate system established by its three principal component vectors (hereinafter referred to as the principal axis coordinate system).

[0109] Third, principal axis alignment: Since grid data Pt and grid data Pf are grid data of the same three-dimensional structure from different patients, they are similar in shape and their principal axes are close. After transforming grid data Pt and grid data Pf to the principal axis coordinate system using Rt and Rf respectively, alignment of the two grid data in the direction can be achieved.

[0110] Fourth, scale alignment: Since grid data Pt and Pf may have scale differences, scale correction can be performed on grid data Pf along the three principal axes to ensure that the difference between the maximum and minimum values ​​of grid data Pf and grid data Pt is equal along the three principal axes. The transformation scale of grid data Pf along the three principal axes is calculated using the following formula:

[0111]

[0112] In the formula, max Pt x Scale represents the maximum x-coordinate of the Pt point cloud in the principal axis coordinate system. x Indicates the x-direction in the principal coordinate system

[0113] Fifth, the correction scale of grid data Pf point cloud.

[0114] Sixth, orientation correction: After principal axis alignment, there may still be mismatches in the positive and negative directions of the principal axes. Therefore, the grid data Pf is traversed in both positive and negative directions along the three axes, for a total of 8 directions, and orientation correction is performed for each, resulting in 8 grid data. The nearest grid data Pf to each of these is calculated, and the grid data with the smallest distance is selected as the optimal transformation for grid data Pf. This completes the optimal matching from grid data Pf to grid data Pt.

[0115] Specifically, see Figure 9 , Figure 9 This is a schematic diagram of rigid mesh registration in one embodiment. In this example, rigid mesh registration is achieved using the Iterative Closest Point (ICP) algorithm. The core of the ICP algorithm is to minimize an objective function:

[0116]

[0117] Where P t For the target point cloud, P f Let R be a floating point cloud, and T be the rotation matrix and translation vector to be optimized, respectively. By optimizing and adjusting R and T, f(R,T) is minimized.

[0118] The algorithm is optimized through iteration, specifically including the following steps:

[0119] First, for each point in the two grid data sets, find the nearest corresponding point in the other grid data set. The nearest point pair is solved by SVD decomposition to obtain the transformation matrix, i.e., R and T. Then, perform a rigid transformation and repeat the above steps until the error is less than the set threshold or the number of iterations is reached.

[0120] In addition, point cloud rigid registration algorithms based on deep learning or rigid registration methods based on corresponding feature points can also be used.

[0121] Specifically, see Figure 10 , Figure 10 This is a schematic diagram of mesh elastic registration in one embodiment. In this example, mesh elastic registration is performed iteratively.

[0122] First, for each point in one of the two grid data sets, find its nearest corresponding point in the other grid data set. Then, perform a viscous transformation between the two point pairs. The viscous transformation is defined as displacing each point in the floating grid data set directly in the direction of its corresponding point in the target grid data set. Next, perform an elastic transformation between the point pairs. The elastic transformation is defined as replacing the original coordinate position of each point p in the target grid data set with the weighted average of the coordinate positions of its N nearest neighbors in the same grid data set. This is equivalent to smoothing the position of each point p in the target grid data set. Here, the weight of each neighboring point is determined by its distance from point p; the closer the distance, the greater the weight. In this example, it is defined as the Gaussian radial basis function of the distance to point p. Repeat the above steps until the required number of iterations is met.

[0123] Optionally, the terminal can also incorporate multi-scale concepts for flexible registration to improve operational efficiency. The specific approach is as follows: downsample the grid data to varying degrees, from large to small, to obtain multiple grid data pairs from low to high resolution (which can be 2-6 different resolution grid data pairs); perform flexible registration on the lowest resolution grid data; apply a deformation field to the next level of resolution floating grid data for flexible registration at the next level of resolution; repeat the above steps to perform flexible registration from coarse to fine resolution until the highest resolution (original resolution) grid data registration is completed.

[0124] Alternatively, other elastic registration schemes can be used, such as elastic registration methods based on deep learning for elastic deformation field estimation.

[0125] In one embodiment, see Figure 11 As shown, where Figure 11 The flowchart below shows a target feature extraction method based on surface registration in another embodiment. This embodiment mainly includes two parts: template creation and feature extraction.

[0126] The creation of templates can be combined with Figure 6 As shown, in the template creation process, by utilizing a large amount of existing training set data, the target to be processed in the data is reconstructed into three-dimensional surface mesh data of the sample. After being registered to the same space by three-dimensional mesh surface registration technology, average three-dimensional surface mesh data is formed. Then, the features to be extracted are manually labeled in advance on the average data to form labeled three-dimensional template data.

[0127] During the feature extraction process, which is also the algorithm execution stage, the input 3D image or point cloud data is reconstructed to obtain the 3D mesh data to be processed. Then, the 3D mesh surface registration calculation is used to register the 3D mesh data to be processed with the surface template mesh data in the 3D template data, i.e., spatial alignment. Then, the 3D template feature annotation information in the 3D template data is mapped to the 3D mesh data to be processed, thereby completing the extraction of target features.

[0128] In the above embodiments, since features can be labeled as needed, it is applicable to various types of feature extraction, including various feature structures on the surface and inside of the target, such as feature points, feature lines, feature regions, etc. It is applicable to both features with local specificity and features without local specificity. In addition, multiple features can be labeled during the calibration process, resulting in high extraction efficiency and suitability for the rapid extraction of a large number of features. It can extract all required features in parallel at once, with no upper limit on the number of features to be detected, and the algorithm efficiency will not decrease as the number of features to be extracted increases. Finally, it is not limited to features with volume such as points, lines, and surfaces. It uses the surface of the target structure as a carrier for template matching and feature extraction, which has a natural advantage for feature extraction on surface structures, especially features without volume such as points, lines, and surfaces.

[0129] It should be understood that, although Figure 2 , Figure 6 as well as Figure 11The 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 specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 , Figure 6 as well as Figure 11 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.

[0130] In one embodiment, such as Figure 12 As shown, a target feature extraction device based on surface registration is provided, comprising: a data acquisition module 100, a template data acquisition module 200, a first surface registration module 300, and a mapping module 400, wherein:

[0131] The data acquisition module 100 is used to acquire the three-dimensional surface mesh data of the target to be processed.

[0132] The template data acquisition module 200 is used to acquire pre-generated 3D template data corresponding to the target to be processed.

[0133] The first surface registration module 300 is used to perform surface registration between the surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed, so as to establish the correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed.

[0134] The mapping module 400 is used to read the three-dimensional template feature annotation information from the three-dimensional template data; according to the correspondence, the three-dimensional template feature annotation information is mapped to the three-dimensional surface mesh data to be processed, so as to extract the target features in the three-dimensional surface mesh data to be processed.

[0135] In one embodiment, the above-described surface registration-based target feature extraction device may further include:

[0136] The sample data acquisition module is used to acquire the three-dimensional surface mesh data of the sample.

[0137] The second surface registration module is used to perform surface registration on the sample's three-dimensional surface mesh data to obtain surface template mesh data.

[0138] The template generation module is used to annotate the features on the surface template mesh data to obtain three-dimensional template data.

[0139] In one embodiment, the template generation module described above may include:

[0140] A recording unit is used to record the index or point set index group of the points corresponding to the target feature structure on the three-dimensional template surface mesh data.

[0141] The generation unit is used to obtain three-dimensional template data based on the point index or point set index group and surface template mesh data.

[0142] In one embodiment, the second surface registration module described above includes:

[0143] The corresponding point acquisition unit is used to acquire the corresponding points in the three-dimensional surface mesh data of each sample.

[0144] The data processing unit is used to process the corresponding points according to preset rules to obtain surface template mesh data.

[0145] In one embodiment, the second surface registration module further includes:

[0146] Alignment units are used to align all sample 3D mesh data to the same coordinate space.

[0147] In one embodiment, the alignment unit is used to align all sample 3D mesh data to the same coordinate space using affine transformation techniques.

[0148] In one embodiment, the mapping module 400 is used to obtain the closest point in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and to use the closest point as the feature point corresponding to the target to be processed; or to obtain the set of the closest points in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and to use the set of the closest points as the feature line or feature region corresponding to the target to be processed.

[0149] In one embodiment, the first surface registration module 300 is used to take the surface template mesh data in the three-dimensional template data as floating mesh data and the three-dimensional surface mesh data to be processed as target mesh data, so as to perform surface registration of the floating mesh data onto the target mesh data; wherein the surface registration includes the correspondence of the floating mesh data to points in the target mesh data and the processing of the corresponding points.

[0150] Specific limitations regarding the surface registration-based target feature extraction device can be found in the limitations of the surface registration-based target feature extraction method described above, and will not be repeated here. Each module in the aforementioned surface registration-based target feature 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, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0151] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 13 As 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 surface registration-based target feature 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.

[0152] Those skilled in the art will understand that Figure 13 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.

[0153] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: acquiring three-dimensional surface mesh data corresponding to a target to be processed; acquiring pre-generated three-dimensional template data corresponding to the target to be processed; performing surface registration between the surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed to establish a correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed; reading three-dimensional template feature annotation information from the three-dimensional template data; and mapping the three-dimensional template feature annotation information to the three-dimensional surface mesh data to be processed according to the correspondence to extract target features from the three-dimensional surface mesh data to be processed.

[0154] In one embodiment, the method for generating three-dimensional template data when the processor executes a computer program includes: acquiring sample three-dimensional surface mesh data; performing surface registration on the sample three-dimensional surface mesh data to obtain surface template mesh data; and annotating the features on the surface template mesh data to obtain three-dimensional template data.

[0155] In one embodiment, the process of annotating features on surface template mesh data to obtain three-dimensional template data when the processor executes a computer program includes: recording the index or point set index group of points corresponding to the target feature structure on the three-dimensional template surface mesh data; and obtaining three-dimensional template data based on the index or point set index group of points and the surface template mesh data.

[0156] In one embodiment, the process of performing surface registration on sample three-dimensional surface mesh data to obtain surface template mesh data when the processor executes a computer program includes: acquiring corresponding points in the three-dimensional surface mesh data of each sample; and processing the corresponding points according to a preset rule to obtain surface template mesh data.

[0157] In one embodiment, before the processor executes the computer program to acquire the corresponding points in the three-dimensional surface mesh data of each sample, the method further includes: aligning all the three-dimensional mesh data of the samples to the same coordinate space.

[0158] In one embodiment, the alignment of all sample 3D mesh data to the same coordinate space by the processor executing a computer program includes: aligning all sample 3D mesh data to the same coordinate space using an affine transformation technique.

[0159] In one embodiment, when the processor executes a computer program, it maps 3D template feature annotation information to 3D surface mesh data to be processed according to a correspondence to extract target features from the 3D surface mesh data to be processed. This includes: obtaining the closest point in the 3D surface mesh data to be processed that corresponds to the template feature annotation information, and using the closest point as the feature point corresponding to the target to be processed; or obtaining the set of the closest points in the 3D surface mesh data to be processed that corresponds to the template feature annotation information, and using the set of the closest points as the feature line or feature region corresponding to the target to be processed.

[0160] In one embodiment, the surface registration of the surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed, implemented by the processor executing the computer program to establish the correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed, includes: using the surface template mesh data in the three-dimensional template data as floating mesh data, and using the three-dimensional surface mesh data to be processed as target mesh data, to perform surface registration of the floating mesh data onto the target mesh data; wherein the surface registration includes the correspondence of the floating mesh data to points in the target mesh data and the processing of the corresponding points.

[0161] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it performs the following steps: acquiring three-dimensional surface mesh data corresponding to the target to be processed; acquiring pre-generated three-dimensional template data corresponding to the target to be processed; performing surface registration between the surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed to establish a correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed; reading three-dimensional template feature annotation information from the three-dimensional template data; and mapping the three-dimensional template feature annotation information to the three-dimensional surface mesh data to be processed according to the correspondence to extract target features from the three-dimensional surface mesh data to be processed.

[0162] In one embodiment, the method for generating three-dimensional template data when the computer program is executed by the processor includes: acquiring sample three-dimensional surface mesh data; performing surface registration on the sample three-dimensional surface mesh data to obtain surface template mesh data; and annotating the features on the surface template mesh data to obtain three-dimensional template data.

[0163] In one embodiment, when a computer program is executed by a processor, it generates three-dimensional template data by annotating features on surface template mesh data, including: recording the index or point set index group of points corresponding to the target feature structure on the three-dimensional template surface mesh data; and generating three-dimensional template data based on the index or point set index group of points and the surface template mesh data.

[0164] In one embodiment, when a computer program is executed by a processor, it performs surface registration on sample three-dimensional surface mesh data to obtain surface template mesh data, including: acquiring corresponding points in each sample three-dimensional surface mesh data; and processing the corresponding points according to preset rules to obtain surface template mesh data.

[0165] In one embodiment, before the computer program is executed by the processor to acquire the corresponding points in the three-dimensional surface mesh data of each sample, the method further includes: aligning all the three-dimensional mesh data of the samples to the same coordinate space.

[0166] In one embodiment, the alignment of all sample 3D mesh data to the same coordinate space when the computer program is executed by the processor includes: aligning all sample 3D mesh data to the same coordinate space using an affine transformation technique.

[0167] In one embodiment, the computer program, when executed by a processor, performs a mapping based on positional information in the three-dimensional surface mesh data to be processed to extract target features, including: obtaining the closest point in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and using the closest point as the feature point corresponding to the target to be processed; or obtaining the set of the closest points in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and using the set of the closest points as the feature line or feature region corresponding to the target to be processed.

[0168] In one embodiment, when a computer program is executed by a processor, it performs surface registration between surface template mesh data in three-dimensional template data and processed three-dimensional surface mesh data to establish a correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed. This includes: using the surface template mesh data in the three-dimensional template data as floating mesh data and the three-dimensional surface mesh data to be processed as target mesh data, and performing surface registration between the floating mesh data and the target mesh data; wherein surface registration includes the correspondence between the floating mesh data and points in the target mesh data, as well as the processing of the corresponding points.

[0169] 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.

[0170] 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.

[0171] 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 target feature extraction method based on surface registration, characterized in that, The method includes: Obtain the three-dimensional surface mesh data corresponding to the target to be processed; wherein, processing the three-dimensional surface mesh data is to perform three-dimensional reconstruction on the three-dimensional image obtained by medical imaging equipment scanning in the form of a three-dimensional matrix to obtain the three-dimensional surface mesh data; Obtain pre-generated 3D template data corresponding to the target to be processed; The surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed are surface registered to establish the correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed. Read the three-dimensional template feature annotation information from the three-dimensional template data; according to the correspondence, map the three-dimensional template feature annotation information to the three-dimensional surface mesh data to be processed, so as to extract the target features in the three-dimensional surface mesh data to be processed; wherein, the three-dimensional template feature annotation information includes the annotation information of feature points, feature lines and feature regions; The step of performing surface registration between the surface template mesh data in the 3D template data and the 3D surface mesh data to be processed includes: The surface template mesh data in the 3D template data is used as floating mesh data, and the 3D surface mesh data to be processed is used as target mesh data. Surface registration is performed on the floating mesh data and then onto the target mesh data. The surface registration includes the mapping of points in the floating mesh data to points in the target mesh data and the processing of the mapped points. The surface registration includes sequentially performing mesh coarse alignment, mesh rigid registration, and mesh elastic registration. The mesh elastic registration is performed through an iterative process, which includes: for each floating point in the floating mesh data, obtaining the nearest target point in the target mesh data, shifting the floating point in the direction of the corresponding target point, and performing a weighted average smoothing elastic transformation on the position of the target point based on the neighboring points in the target mesh data. The step of mapping the 3D template feature annotation information to the 3D surface mesh data to be processed according to the correspondence relationship, in order to extract the target features in the 3D surface mesh data to be processed, includes: Obtain the closest point in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and use the closest point as the feature point corresponding to the target to be processed; or Obtain the set of points in the three-dimensional surface mesh data to be processed that are closest to the template feature annotation information, and use the set of closest points as the feature line or feature region corresponding to the target to be processed.

2. The method according to claim 1, characterized in that, The methods for generating the three-dimensional template data include: Obtain the three-dimensional surface mesh data of the sample; The surface template mesh data is obtained by surface registration of the three-dimensional surface mesh data of the sample. The three-dimensional template data is obtained by annotating the features on the surface template mesh data.

3. The method according to claim 2, characterized in that, The process of annotating the features on the surface template mesh data to obtain the three-dimensional template data includes: Record the index or point set index group of the points corresponding to the target feature structure on the surface template mesh data; The three-dimensional template data is obtained based on the index of the points or the index group of the point set and the surface template mesh data.

4. The method according to claim 2, characterized in that, The step of performing surface registration on the sample's three-dimensional surface mesh data to obtain surface template mesh data includes: Obtain the corresponding points in the three-dimensional surface mesh data of each sample; After processing the corresponding points according to preset rules, surface template mesh data is obtained.

5. The method according to claim 4, characterized in that, After processing the corresponding points according to preset rules, surface template mesh data is obtained, including: For each corresponding point, the position of the corresponding point is averaged to obtain the average position of the corresponding point; Based on the average position of each corresponding point, average three-dimensional surface mesh data is obtained, and the average three-dimensional surface mesh data is used as the surface template mesh data.

6. The method according to claim 4, characterized in that, Before obtaining the corresponding points in the three-dimensional surface mesh data of each sample, the method further includes: Align all sample 3D mesh data to the same coordinate space.

7. The method according to claim 6, characterized in that, Aligning all sample 3D mesh data to the same coordinate space includes: All sample 3D mesh data are aligned to the same coordinate space using affine transformation techniques.

8. A target feature extraction device based on surface registration, characterized in that, The device includes: The data acquisition module is used to acquire the three-dimensional surface mesh data corresponding to the target to be processed; wherein, the three-dimensional surface mesh data to be processed is obtained by three-dimensional reconstruction of the three-dimensional image with three-dimensional matrix storage obtained by scanning medical imaging equipment. The template data acquisition module is used to acquire pre-generated three-dimensional template data corresponding to the target to be processed; The first surface registration module is used to perform surface registration between the surface template mesh data in the three-dimensional template data and the three-dimensional surface mesh data to be processed, so as to establish the correspondence between the surface template mesh data and the three-dimensional surface mesh data to be processed. A mapping module is used to read 3D template feature annotation information from the 3D template data; and map the 3D template feature annotation information to the 3D surface mesh data to be processed according to the correspondence, so as to extract target features in the 3D surface mesh data to be processed; wherein, the 3D template feature annotation information includes annotation information of feature points, feature lines and feature regions; The first surface registration module is used to take the surface template mesh data in the three-dimensional template data as floating mesh data and the three-dimensional surface mesh data to be processed as target mesh data, so as to perform surface registration of the floating mesh data onto the target mesh data; the surface registration includes the correspondence of the floating mesh data to points in the target mesh data and the processing of the corresponding points; the surface registration includes sequentially performing mesh coarse alignment, mesh rigid registration and mesh elastic registration; the mesh elastic registration is performed through an iterative process, which includes: for each floating point in the floating mesh data, obtaining the target point in the target mesh data that is closest to the floating point, displacing the floating point in the direction of the corresponding target point, and performing a weighted average smoothing elastic transformation on the position of the target point based on the neighboring points in the target mesh data; The mapping module is used to obtain the closest point in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and use the closest point as the feature point corresponding to the target to be processed; or, to obtain the set of the closest points in the three-dimensional surface mesh data to be processed that corresponds to the template feature annotation information, and use the set of the closest points as the feature line or feature region corresponding to the target to be processed.

9. 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 7.

10. 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 7.

Images