Implicit organizational model construction method, model and application based on digital human registration

Through the implicit organization model construction method, the shortcomings of the three-dimensional digital human model in internal organization classification and positioning are solved, and a higher precision and more accurate model is achieved. It is suitable for different posture research in the medical field, reducing radiation risks.

CN118761341BActive Publication Date: 2024-11-26XIAOZHI FUTURE (CHENGDU) TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411256127.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-09
Publication Date
2024-11-26
Estimated Expiration
2044-09-09

AI Technical Summary

Technical Problem

The existing three-dimensional digital human model has shortcomings in internal tissue classification and precise positioning, and cannot be effectively applied in the medical field, especially because it is impossible to accurately classify and track tissues in computer-built 3D digital human body.

Method used

Using the implicit organization model construction method based on registered digital people, deforming mapping, depose mapping and deforming mapping are performed through the SMPL model, combining pseudo-inverse matrix and iterative optimization algorithm to realize the organization classification and classification of the 3D digital people model and Positioning, using CT and DR image data to build a precise implicit organization model.

Benefits of technology

The precision and positioning accuracy of the 3D digital human model are improved, and the tissue position and morphology of the human body under different postures can be simulated in the medical field, reducing the radiation dose when determining the lesion position through imaging technology, and expanding the application scenario.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761341B_ABST
    Figure CN118761341B_ABST
Patent Text Reader

Abstract

The present application relates to the field of image data processing technology, and discloses a method, model and application of implicit tissue model construction based on registered digital human, including steps of de-deformation mapping #imgabs0#, de-gesture mapping #imgabs1# and de-morphology mapping #imgabs2# of the registered digital human model established through SMPL based on image data including CT and DR, and steps of tissue classification of the model. The present invention provides a method for constructing an implicit tissue model, and the constructed implicit tissue model is more accurate in terms of body surface and bones than the existing standard digital human model, and has the advantages of higher model precision and more accurate positioning compared to the construction of the display tissue model. At the same time, the implicit tissue model can realize the simulation of the human body model in different postures, and the application scenarios are more extensive.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, in particular to the technical field of 3D modeling using computer graphics, and specifically to a method for constructing an implicit tissue model based on registering digital humans, a model and an application thereof. Background Art

[0002] 3D digital human generation technology refers to the process of creating a virtual human image (i.e. 3D digital human or virtual human) using computer technology and algorithms. This technology usually involves the following steps:

[0003] Data collection: Data on human images is obtained through various means, including 2D images, videos, depth images, laser scanning, etc. The collected data is converted into a three-dimensional model using computer graphics and machine learning algorithms. This process may include the generation of facial features, body structure, texture mapping, etc. After obtaining the three-dimensional model, it is given dynamic expression through animation technology, enabling it to show various expressions, movements and interactions. Combining data from different modalities, the generated three-dimensional digital human is optimized to make it more visually realistic and natural.

[0004] The application range of 3D digital human generation technology is wide, and it can play an important role in many fields: For example, virtual characters: in video games, animated films and virtual reality experiences, high-quality virtual characters are generated to enhance user experience. Users can create personalized virtual images according to their preferences. Users can use virtual human images as their own representatives on social platforms to interact with others. At the same time, virtual human images can be used in social applications for facial replacement and special effects enhancement. In the medical and educational fields, 3D digital human models can be used for anatomical teaching and simulated surgical practice, etc.

[0005] In the existing technology, 3D models can be generated from single or multiple 2D images. Common methods include deep learning, convolutional neural networks (CNN), etc. Of course, video sequences are also used to extract motion information and expression changes to generate more dynamic 3D models, which are mainly used in animation production, virtual reality (VR) and augmented reality (AR). Although the existing technology can generate 3D digital human models from 2D images and videos, the digital humans generated based on different data have more or less some shortcomings and one-sidedness. In terms of comprehensiveness, realism, especially the generation of structures including bones inside the digital human, the accuracy needs to be improved. More importantly, the existing three-dimensional digital humans can hardly be actually applied in the field of medical technology. Since it is impossible to accurately classify the tissues in the body of the 3D digital human created by the computer, it is impossible to qualitatively judge and track the target or area of ​​interest. Summary of the invention

[0006] In order to solve the problem of internal organization classification of 3D digital human, this application provides a method, model and application for constructing an implicit organization model based on the registered digital human, which can solve at least one of the following technical problems:

[0007] 1. It solves the problem of implicit organizational model construction. Compared with explicit organizational model construction, it has the advantages of higher model precision and more accurate positioning.

[0008] 2. It solves the tissue classification problem of 3D digital human models. It can take a certain point in the model as the research object, and conduct tissue characterization and positioning of any point, so as to better simulate different states of the human body.

[0009] 3. The implicit tissue model established by the present invention can change the posture and shape of the 3D digital human by changing the SMPL model parameters, simulating different states of the real human body, so that it can provide a model basis for the study of the position and morphology of the body tissue under different postures of the human body in the medical field, and solve the problem that the existing technology uses the human body repeatedly through imaging technologies such as CT to determine the position and shape of the lesion tissue, which will lead to excessive cumulative radiation dose and affect health.

[0010] In order to achieve the above purpose, the technical solution adopted in this application is:

[0011] The implicit tissue model construction method based on the registered digital human includes de-deformation mapping of the registered digital human model established by SMPL based on image data including CT and DR. , to pose mapping and demorphological mapping Steps;

[0012] Step STP100, De-deformation mapping , any point on the body surface mesh or skeletal mesh on the deformed SMPL template Using the pseudo-inverse matrix Dedistortion Mapping , get the corresponding points after de-deformation ,

[0013]

[0014] in, is any point on the body surface mesh or skeletal mesh on the deformed SMPL template, is the affine transformation matrix from the point on the SMPL standard template to the corresponding point on the deformed SMPL template;

[0015] Step STP200, go to posture mapping , through the linear blend skinning function Establish the in-body points of the SMPL template with pose obtained in step STP100 Mapping to points in the SMPL template with morphology The mapping relationship is expressed as , and then through the iterative optimization algorithm through the SMPL template with morphology Get points in the SMPL template with pose , the iterative optimization process is as follows:

[0016] Step STP210, Initialize to The corresponding point of the nearest SMPL template body surface point with posture on the SMPL template with shape - that is, the point with the same serial number, initialize for The skin weight of

[0017] Step STP220, for each iteration and ,renew ;

[0018] Step STP230, prediction ,but , ,renew ;

[0019] Step STP240, repeat steps STP220-STP230 until and The difference is small enough, that is When , the iteration is completed;

[0020] Among them, the network is the same point in the predicted morphological space Displacement networks in different SMPL templates, networks It is a point in the predicted SMPL standard template space The skin weight of

[0021] Step STP300, remove morphological mapping , through the network In-body points in SMPL space with morphology obtained in prediction step STP200 In vivo points with SMPL standard template Displacement , expressed as , then the SMPL standard template body point is obtained Expressed as

[0022]

[0023] in, The morphological parameters of the currently registered digital human model; The posture parameters of the currently registered digital human model; is the weight parameter;

[0024] Step STP400, tissue classification, will complete the de-deformation mapping through steps STP100-STP300 , to pose mapping and demorphological mapping The SMPL standard template obtained was used for organizational classification, and the network predict The organizational categories are:

[0025]

[0026] in Represents non-solid tissue, such as lung cavities and cavity cavities; Represents bones; represents subcutaneous adipose tissue; Represents lean tissue and completes the implicit tissue model construction.

[0027] The present invention also includes the network in step STP200 , and the network in step STP400 The steps for loss training are as follows:

[0028] Step X100, training network ,set up is a vertex on the SMPL template with shape, is the corresponding vertex on the SMPL standard template, the network The training loss function is defined as:

[0029]

[0030] set up is a vertex of the SMPL standard template, the network The training loss function is defined as:

[0031]

[0032] network The loss function is defined as the cross entropy between the predicted tissue classification and the training data table labels:

[0033]

[0034] in, Indicates a point in the body The type of prediction, expressed as a probability, with a value between 0 and 1. Indicates a point in the body The actual type of

[0035] Step X200, the network in step X100 Replace with Conduct training;

[0036] Step X300, the network in step X100 Replace with Conduct training;

[0037] Among them, the training data uses the HIT dataset.

[0038] Preferably, the method for establishing the HIT data set includes:

[0039] Step J100, collecting samples from the whole body MRI scan data of the subject;

[0040] Step J200, using a segmentation algorithm to divide the MRI tomographic image into cavities, bones, subcutaneous fat tissue, and lean tissue;

[0041] Step J300, extracting the outer contour of the body surface, generating a point cloud from the extracted outer contour and fitting the obtained point cloud with the SMPL model, thereby generating a corresponding HIT data set of the SMPL body surface and the internal tissue.

[0042] The present invention also provides an implicit tissue model, which is specifically constructed by using the above-mentioned implicit tissue model construction method based on registering digital human.

[0043] The present invention also provides an application of an implicit tissue model in locating a static region of interest, wherein any point in the region of interest is located The spatial coordinate method is as follows:

[0044] Step Y100, define the target point and define the shooting Second, in the Find the target point in the image data in the region of interest under different shooting angles , the spatial coordinates of the target point on the image data ,have

[0045]

[0046] in, Radiation source to The distance of the radiographic detector is SID , the width and height of the detector imaging area are , , the detector imaging pixel width and height are and , the pixel coordinates of the imaging point of the ray center on the detector are ;

[0047] Step Y200, spatial transformation of the target point, based on the ray source With the target point The geometric relationship on the same straight line, then the transformation relationship between the current ray coordinate system and the world coordinate system is established as follows:

[0048]

[0049] Establish Straight line The equation of the straight line is as follows:

[0050]

[0051] in, is the transformation matrix between the ray coordinate system and the world coordinate system;

[0052] Step Y300, solve the target point Corresponding points in the world coordinate system The spatial coordinates of , calculate the straight line Direction vector , calculate the vector from a point on the line to the target point , so there is

[0053]

[0054] The least squares method can be used to fit the points coordinates to achieve static positioning.

[0055] The present invention also provides an application of an implicit tissue model in dynamic region of interest positioning. Based on the positioning principle of the implicit tissue model in static region of interest positioning, the target point is positioned at Mapped into the SMPL standard template, let yes At the corresponding point in the SMPL standard template, and save When the object's body changes, the camera parameter module is used to re-acquire the current body's morphological parameters. and attitude parameters Input into the SMPL standard template, and then through the in vivo mapping relationship get Repositioning in current pose.

[0056] The present invention also provides an application of an implicit tissue model in estimating body fat percentage, which can be specifically achieved by the following steps: traversing each target point inside the current human body grid by uniformly sampling at equal intervals , using the in vivo mapping relationship Get the target point Corresponding points in the body of the SMPL standard template , and point Organize and classify, count points Number of times classified as SAT , then the body fat percentage is:

[0057]

[0058] in, is a collection of subcutaneous fat tissue points. is the set of all points, Is body fat percentage.

[0059] Beneficial effects:

[0060] 1. The present invention provides a method for constructing an implicit tissue model. The implicit tissue model constructed is more accurate in terms of body surface and bones than the existing standard digital human model, and has the advantages of higher model precision and more accurate positioning than the construction of the displayed tissue model. At the same time, the implicit tissue model can realize the simulation of the human body model in different postures, and has a wider range of application scenarios.

[0061] 2. The 3D digital human model constructed by the present invention can realize accurate classification and statistics of tissues, can take a certain point in the model as a research object, and perform tissue characterization and positioning at any point, so as to better simulate different states of the human body.

[0062] 3. The implicit tissue model established by the present invention can change the posture and shape of the 3D digital human by changing the SMPL model parameters, simulating different states of the real human body, so that it can provide a model basis for the study of the position and morphology of the body tissue under different postures of the human body in the medical field, and solve the problem that the existing technology uses the human body repeatedly through imaging technologies such as CT to determine the position and shape of the lesion tissue, which will lead to excessive cumulative radiation dose and affect health. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described in combination with the embodiments of the present application below. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments.

[0064] Embodiment 1:

[0065] The implicit tissue model construction method based on the registered digital human includes de-deformation mapping of the registered digital human model established by SMPL based on image data including CT and DR. , to pose mapping and demorphological mapping Steps;

[0066] In the de-deformation mapping , to pose mapping and demorphological mapping Before introducing the steps, in order to better illustrate the background of the technical solution provided by the present invention and its role, the following contents are first explained so that those skilled in the art can understand more accurately through the description of this embodiment.

[0067] Image data refers to all image data that can be used to revise the standardized SMPL model, such as the most commonly used X-ray images such as CT, MRI images, single or continuous DR or CBCT, etc. Different image data types will have different registration methods or revision methods to revise the standardized SMPL in a personalized way, that is, to integrate the actual real image parameters into the standard idealized SMPL model, so that the model becomes very similar or identical to an actual human body.

[0068] The registered digital human model refers to the personalized SMPL model after revision or fusion. The reason why the SMPL model can be called personalized at this time is that it integrates the parameters of a real human body. However, since the personalized SMPL model has the characteristics of a real human body, it does not have the separable types of subsequent tissues, and its personalized part must be removed, that is, the composite mapping from the idealized SMPL model to the personalized SMPL model is removed; it is worth noting that removing these mappings is not equivalent to the idealized SMPL model, because when establishing the registered digital human, a large number of parameters of the real human body are integrated into the rationalized SMPL model, therefore, other surface or internal tissues or bones of the real human body are also included; removing the converted composite mapping is to convert the personalized SMPL model into a standardized SMPL model system or spatial coordinates, rather than clearing all the personalized information of the personalized SMPL model. For example, if there is a tumor in the registered digital human model, then after removing the converted composite mapping, the tumor also exists in the SMPL model after removing the composite mapping. The registered digital human model is the object to be processed in this embodiment and is also the object basis of the input of the present invention. The above description is intended to clearly explain the premise of the processing object of the present invention. The models in the group obtained by registering digital human models in different fields and under different technologies are different. For example, the three-dimensional digital human model of Zhang San and the three-dimensional digital human model of Li Si are definitely different after registration, but they can be standardized through the construction method provided in this embodiment to facilitate the subsequent classification and application of the organization. The specific construction process includes:

[0069] Step STP100, De-deformation mapping , any point on the body surface mesh or skeletal mesh on the deformed SMPL template Using pseudo-inverse matrix Dedistortion Mapping , get the corresponding points after de-deformation ,

[0070]

[0071] in, is any point on the body surface mesh or skeletal mesh on the deformed SMPL template, is the affine transformation matrix from the point on the SMPL standard template to the corresponding point of the deformed SMPL template; if the deformed SMPL template is obtained by non-rigid registration, then for ease of understanding, It can be considered as the affine transformation matrix from the points on the SMPL standard template grid to the grid points after non-rigid registration;

[0072] Step STP200, go to posture mapping , through the linear blend skinning function Establish the in-body points of the SMPL template with pose obtained in step STP100 Mapping to points in the SMPL template with morphology The mapping relationship is expressed as , and then through the iterative optimization algorithm through the SMPL template with morphology Get points in the SMPL template with pose , the iterative optimization process is as follows:

[0073] Step STP210, Initialize to The corresponding point of the nearest SMPL template body surface point with posture on the SMPL template with shape - that is, the point with the same serial number, initialize for The skin weight of

[0074] Step STP220, for each iteration and ,renew ;

[0075] Step STP230, prediction ,but , ,renew ;

[0076] Step STP240, repeat steps STP220-STP230 until and The difference is small enough, that is When , the iteration is completed;

[0077] Among them, the network is the same point in the predicted morphological space Displacement networks in different SMPL templates, networks It is a point in the predicted SMPL standard template space The skin weight of

[0078] Step STP300, remove morphological mapping , through the network In-body points in SMPL space with morphology obtained in prediction step STP200 In vivo points with SMPL standard template Displacement , expressed as , then the SMPL standard template body point is obtained Expressed as

[0079]

[0080] in, The morphological parameters of the currently registered digital human model; The posture parameters of the currently registered digital human model; is the weight parameter;

[0081] Step STP400, tissue classification, will complete the de-deformation mapping through steps STP100-STP300 , to pose mapping and demorphological mapping The SMPL standard template obtained was used for organizational classification, and the network predict The organizational categories are:

[0082]

[0083] in Represents non-solid tissue, such as lung cavities and cavity cavities; Represents bones; represents subcutaneous adipose tissue; Represents lean tissue and completes the implicit tissue model construction.

[0084] Embodiment 2:

[0085] This embodiment further includes the following steps based on the embodiment 1: , and the network in step STP400 The steps for loss training are as follows:

[0086] Step X100, training network ,set up is a vertex on the SMPL template with shape, is the corresponding vertex on the SMPL standard template, the network The training loss function is defined as:

[0087]

[0088] set up is a vertex of the SMPL standard template, the network The training loss function is defined as:

[0089]

[0090] network The loss function is defined as the cross entropy between the predicted tissue classification and the training data table labels:

[0091]

[0092] in, Indicates a point in the body The type of prediction, expressed as a probability, with a value between 0 and 1. Indicates a point in the body The actual type of

[0093] Step X200, the network in step X100 Replace with Conduct training;

[0094] Step X300, the network in step X100 Replace with Conduct training;

[0095] Among them, the training data uses the HIT data set. The existing extensive HIT data set is used for training in this embodiment; of course, the method of this embodiment can also be trained through a specific data set for application to special human groups. For example, if the subsequent application population is mainly for disabled or handicapped people, data samples of the application group can be collected separately for training, and only the data set samples are changed, which does not affect the implementation of the method provided in this embodiment.

[0096] Embodiment 3:

[0097] This embodiment provides a method for establishing a data set for establishing a general, special or specific field data set. Specifically, the method for establishing a data set includes:

[0098] Step J100, collecting samples from the whole body MRI scan data of the subject;

[0099] Step J200, using a segmentation algorithm to divide the MRI tomographic image into cavities, bones, subcutaneous fat tissue, and lean tissue;

[0100] Step J300, extracting the outer contour of the body surface, generating a point cloud from the extracted outer contour and fitting the obtained point cloud with the SMPL model, thereby generating a corresponding HIT data set of the SMPL body surface and the internal tissue.

[0101] Embodiment 4:

[0102] The present invention also provides an implicit tissue model, which is specifically constructed by using the implicit tissue model construction method based on the registration of digital human according to any of the above embodiments.

[0103] This embodiment also provides an application of an implicit tissue model in locating a static region of interest, wherein any point in the region of interest is located. The spatial coordinate method is as follows:

[0104] Step Y100, define the target point and define the shooting Second, in the Find the target point in the image data in the region of interest under different shooting angles , the spatial coordinates of the target point on the image data ,have

[0105]

[0106] in, Radiation source to The distance of the radiographic detector is SID , the width and height of the detector imaging area are , , the detector imaging pixel width and height are and , the pixel coordinates of the imaging point of the ray center on the detector are ;

[0107] Step Y200, spatial transformation of the target point, based on the ray source With target point The geometric relationship on the same straight line, then the transformation relationship between the current ray coordinate system and the world coordinate system is established as follows:

[0108]

[0109] Establish Straight line The equation of the straight line is as follows:

[0110]

[0111] in, is the transformation matrix between the ray coordinate system and the world coordinate system;

[0112] Step Y300, solve the target point Corresponding points in the world coordinate system The spatial coordinates of , calculate the straight line Direction vector . Calculate the vector from a point on the line to the target point , so there is

[0113]

[0114] The least squares method can be used to fit the points coordinates to achieve static positioning.

[0115] This embodiment also provides an application of an implicit tissue model in dynamic ROI positioning. Based on the positioning principle of the implicit tissue model in the above embodiment in static ROI positioning, the target point is positioned at Mapped into the SMPL standard template, let yes At the corresponding point in the SMPL standard template, and save When the object's body changes, the camera parameter module is used to re-acquire the current body's morphological parameters. and attitude parameters Input into the SMPL standard template, and then through the in vivo mapping relationship get Repositioning in current pose.

[0116] The present invention also provides an application of an implicit tissue model in estimating body fat percentage, which can be specifically achieved by the following steps: traversing each target point inside the current human body grid by uniformly sampling at equal intervals , using the in vivo mapping relationship Get the target point Corresponding points in the body of the SMPL standard template , and point Organize and classify, count points Number of times classified as SAT , then the body fat percentage is:

[0117]

[0118] in, is a collection of subcutaneous fat tissue points. is the set of all points, Is body fat percentage.

[0119] Embodiment 5:

[0120] This embodiment also provides a method for establishing a registered digital human model as the basis for constructing an implicit tissue model of the present invention, specifically a human body mesh non-rigid registration method based on X-ray fluoroscopic images, for registering a human body mesh M including a body surface mesh and a skeleton mesh. The human body mesh model (including the idealized standard SMPL model or the personalized SMPL model) is non-rigidly registered to obtain an accurate human body mesh model (registered digital human model), including the following steps:

[0121] Step STP100, extracting contours based on actual X-ray image data, using contour detection algorithms to extract the human body surface contours on each X-ray image data , and use U-Net or Mask-R-CNN algorithm to segment the skeleton contour in the body ;

[0122] Step STP200, when the X-ray image data is a CT or MRI image, the human body surface contour obtained in step STP100 is converted into and bone contours Uniform downsampling is performed to obtain the sampling points on all images, and spatial stacking is performed according to the actual pixel size to form a surface point cloud. and bone point cloud ; Among them, the surface point cloud middle Represents the number of points on the body surface point cloud, is the first homogeneous coordinate on the surface point cloud Points, bone point cloud middle Represents the number of points in the bone point cloud, is the first homogeneous coordinate on the bone point cloud points;

[0123] Step STP300, using a diffusion pruning method based on local geometric similarity and inconsistent counterparts to calculate the vertices of the body surface mesh M in the human body mesh to be registered and the body surface point cloud The mapping between and the skeleton mesh in the human body mesh to be registered Vertex and bone point cloud The mapping between ;

[0124] Step STP400, build in mapping and Under the premise of obtaining the body surface grid M, transfer it to the target body surface grid Non-rigid registration , is the set or matrix of non-rigid registration, Indicate point The transformation matrix of the skeleton mesh Go to Target Skeletal Mesh Non-rigid registration , is the set or matrix of non-rigid registration, Indicate point The transformation matrix of

[0125] Step STP500, establish non-rigid registration energy function ,

[0126]

[0127] in, Represents the weight, when the surface point cloud and bone point cloud When present ,otherwise ; , and They are point cloud loss, smoothness constraint and orthogonality constraint; get the minimum value of That is, the best non-rigid registration is completed.

[0128] Preferably, when the X-ray image data is any one or more of CT, MRI, single or continuous DR or CBCT in step STP200, a step based on contour non-rigid registration is further included, specifically including:

[0129] Step STP210, any point on the body surface grid M Convert the standard world coordinate system to the tube coordinate system of the X-ray image data shooting state to obtain the corresponding point ,

[0130]

[0131] in, is the transformation matrix from the world coordinate system to the spherical tube coordinate system, n is the number of shots;

[0132] Step STP220, obtaining pixel coordinates of all points of the human body mesh model according to the size of the detector imaging area and the placement of the tube and the detector

[0133]

[0134] in, , Represent the actual length of a pixel in the horizontal and vertical directions respectively. The pixel coordinates of the imaging point of the ray center on the detector are ;

[0135]

[0136] in, , are the width and height of the detector imaging area, respectively. and are the detector imaging pixel width and height, respectively;

[0137] Step STP230, extract the set of all points on the outermost edge to obtain the body surface contour ;

[0138] Step STP240, according to steps STP210-STP230, any point on the skeleton mesh Convert to get the bone outline ;

[0139] Step STP250, the body surface contour obtained in step STP230 Surface profile extracted with step STP100 Perform registration and place the bone contour obtained in step STP240 The bone contour extracted in step STP100 Perform registration and establish a non-rigid registration energy function ,

[0140]

[0141] in, Represents the weight, when the surface point cloud and bone point cloud When present ,otherwise ; When the body surface contour and bone contours When present , otherwise ,in , , and They are point cloud loss, contour loss, smoothness constraint and orthogonality constraint.

[0142] In this embodiment, the point cloud loss Obtained by the following algorithm

[0143]

[0144] in, represents the error weight of body surface points, Represents the error weight of the bone point, if There are corresponding points on the target shape, then the weight is 1, otherwise the weight is 0; if There are corresponding points on the target shape, then the weight is 1, otherwise the weight is 0.

[0145] In this embodiment, the smoothness constraint Obtained by the following algorithm

[0146]

[0147] in, and are adjacent body surface vertices, and are adjacent bone vertices.

[0148] In this embodiment, the orthogonality constraint Obtained by the following algorithm

[0149]

[0150] in, , yes Rotation matrix, , It is extracted , The constant of the rotation component matrix, and make sure and is the rotation matrix, Represents completing the above summation under the condition that...

[0151] In this embodiment, the contour loss Obtained by the following algorithm

[0152]

[0153] in, It is 0 or a positive integer.

[0154] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. Implicit organizational model construction method based on digital human registration, It is characterized in that Image data including CT and DR are de-deformed by the registered digital human model established by SMPL , to pose mapping and demorphological mapping Steps; Step STP100, De-deformation mapping , any point on the body surface mesh or skeletal mesh on the deformed SMPL template Using pseudo-inverse matrix Dedistortion Mapping , get the corresponding points after de-deformation , in, is any point on the body surface mesh or skeletal mesh on the deformed SMPL template, is the affine transformation matrix from the point on the SMPL standard template to the corresponding point on the deformed SMPL template; Step STP200, go to posture mapping , through the linear blend skinning function Establish the in-body points of the SMPL template with pose obtained in step STP100 Mapping to points in the SMPL template with morphology The mapping relationship is expressed as , and then through the iterative optimization algorithm through the SMPL template with morphology Get points in the SMPL template with pose , the iterative optimization process is as follows: Step STP210, Initialize to The corresponding point of the nearest SMPL template body surface point with posture on the SMPL template with shape - that is, the point with the same serial number, initialize for The skin weight of Step STP220, for each iteration and ,renew ; Step STP230, prediction ,but , ,renew ; Step STP240, repeat steps STP220-STP230 until and The difference is small enough, that is When , the iteration is completed; Among them, the network is the same point in the predicted morphological space Displacement networks in different SMPL templates, networks is a point in the predicted SMPL standard template space The skin weight of Step STP300, remove morphological mapping , through the network In-body points in SMPL space with morphology obtained in prediction step STP200 In vivo points with SMPL standard template Displacement , expressed as , then the SMPL standard template body point is obtained Expressed as in, The morphological parameters of the currently registered digital human model; The posture parameters of the currently registered digital human model; is the weight parameter; Step STP400, tissue classification, will complete the de-deformation mapping through steps STP100-STP300 , to pose mapping and demorphological mapping The SMPL standard template obtained was used for organizational classification, and the network predict The organizational categories are: in Represents non-solid tissue, such as lung cavities and cavity cavities; Represents bones; represents subcutaneous adipose tissue; Represents lean tissue and completes the implicit tissue model construction.

2. The implicit tissue model construction method based on the registration digital human according to claim 1, It is characterized in that It also includes the network in step STP200 , and the network in step STP400 The steps for loss training are as follows: Step X100, training network ,set up is a vertex on the SMPL template with shape, are the corresponding vertices on the SMPL standard template, the network The training loss function is defined as: set up is a vertex of the SMPL standard template, the network The training loss function is defined as: network The loss function is defined as the cross entropy between the predicted tissue classification and the training data table labels: in, Indicates a point in the body The type of prediction, expressed as a probability, with a value between 0 and 1. Indicates a point in the body The actual type of Step X200, the network in step X100 Replace with Conduct training; Step X300, the network in step X100 Replace with Conduct training; Among them, the training data uses the HIT dataset.

3. The implicit tissue model construction method based on digital human registration according to claim 2, It is characterized in that The method for establishing the HIT data set includes: Step J100, collecting samples from the whole body MRI scan data of the subject; Step J200, using a segmentation algorithm to divide the MRI tomographic image into cavities, bones, subcutaneous fat tissue, and lean tissue; Step J300, extracting the outer contour of the body surface, generating a point cloud from the extracted outer contour and fitting the obtained point cloud with the SMPL model, thereby generating a corresponding HIT data set of the SMPL body surface and the internal tissue.

4. An implicit organizational model, It is characterized in that The implicit tissue model is constructed by using the method for constructing an implicit tissue model based on registering a digital human as described in any one of claims 1 to 3.

5. A method for applying an implicit tissue model in static region of interest positioning, based on the implicit tissue model of claim 4, Features: Locate any point in the area of ​​interest The spatial coordinate method is as follows: Step Y100, define the target point and define the shooting Second, in the Find the target point in the image data in the region of interest under different shooting angles , the spatial coordinates of the target point on the image data ,have in, Radiation source to The distance of the radiographic detector is , the width and height of the detector imaging area are , , the detector imaging pixel width and height are and , the pixel coordinates of the imaging point of the ray center on the detector are ; Step Y200, spatial transformation of the target point, based on the ray source With the target point The geometric relationship on the same straight line, then the transformation relationship between the current ray coordinate system and the world coordinate system is established as follows: Establish Straight line The equation of the straight line is as follows: in, is the transformation matrix between the ray coordinate system and the world coordinate system; , , They are the three-dimensional coordinates corresponding to the conversion of the current ray coordinate system to the world coordinate system; Step Y300, solve the target point Corresponding points in the world coordinate system The spatial coordinates of , calculate the straight line Direction vector ; Calculate the vector from a point on the line to the target point , so there is The least squares method can be used to fit the points coordinates to achieve static positioning, where N is the number of shots.

6. An application method of implicit tissue model in dynamic region of interest localization, Features: Based on the positioning principle of the method for applying the implicit tissue model in static region of interest positioning described in claim 5, the target point Mapped into the SMPL standard template, let yes At the corresponding point in the SMPL standard template, and save When the object's body changes, the camera parameter module is used to re-acquire the current body's morphological parameters. and attitude parameters Input into the SMPL standard template, and then through the in vivo mapping relationship get Relocalization in the current pose, where Represents sequential operation function ,symbol Represents the sequence operator.

7. A method for applying an implicit tissue model to body fat percentage estimation, based on the implicit tissue model as claimed in claim 4, Features: This is achieved by the following steps: traversing each target point inside the current human body mesh by uniformly sampling at equal intervals , using the in vivo mapping relationship Get the target point Corresponding points in the body of the SMPL standard template , and point Organize and classify, count points Number of times classified as SAT , then the body fat percentage is: in, is a collection of subcutaneous fat tissue points. is the set of all points, is the body fat percentage, Represents sequential operation function ,symbol Represents the sequence operator.