A virtual patient modeling method, electronic device, storage medium and program product

Abstract

This application provides a virtual patient modeling method, electronic device, storage medium, and program product. The method includes: acquiring surface morphology data, internal tomographic image data, and dynamic functional state data of the object to be modeled; generating three-dimensional surface point cloud data based on the surface morphology data, generating voxel model data based on the internal tomographic image data, and generating motion trajectory data based on the dynamic functional state data; performing spatial registration and fusion processing on the three-dimensional surface point cloud data, the voxel model data, and the motion trajectory data using a fusion reconstruction algorithm to generate a three-dimensional mesh model including pathological state characterization information; and generating a virtual patient model based on the three-dimensional mesh model. This solves the problem of lacking pathological features in modeling from a single data source and improves the realism of the model.

Description

Technical Field

[0001] This application relates to the field of medical technology, and more specifically, to a virtual patient modeling method, electronic device, storage medium, and program product. Background Technology

[0002] Virtual patient modeling is an auxiliary technology for medical simulation teaching, surgical planning, and diagnosis and treatment training. Constructing high-fidelity three-dimensional digital models can provide medical students and clinicians with an interactive environment for observing anatomical structures and simulating pathological states, thereby improving the intuitiveness of medical teaching and the realism of diagnosis and treatment training.

[0003] However, related technologies lack a systematic characterization of pathological states, making it difficult to intuitively present the morphological changes, distribution patterns, and spatial relationships of diseased tissues with surrounding tissues in the model. Furthermore, traditional models often rely on a single data source, making it difficult to integrate multi-dimensional information such as surface morphology, internal structure, and dynamic functional states. The generated models are mostly static anatomical structures, failing to provide a foundation for subsequent dynamic pathological simulations. These shortcomings result in the generation of highly realistic virtual patients failing to meet the needs of clinical teaching and training. Summary of the Invention

[0004] The purpose of this application is to provide a virtual patient modeling method, electronic device, storage medium, and program product to achieve the technical effect of improving the realism of virtual patient models.

[0005] A first aspect of this application provides a virtual patient modeling method, the method comprising: Acquire surface morphology data, internal tomographic image data, and dynamic functional status data of the object to be modeled; Three-dimensional surface point cloud data is generated based on the surface morphology data, voxel model data is generated based on the internal tomographic image data, and motion trajectory data is generated based on the dynamic functional state data. The three-dimensional surface point cloud data, the voxel model data and the motion trajectory data are spatially registered and fused using a fusion reconstruction algorithm to generate a three-dimensional mesh model that includes pathological state characterization information. A virtual patient model is generated based on the aforementioned three-dimensional mesh model.

[0006] In the above implementation process, by acquiring and fusing multi-source data, a three-dimensional mesh model containing pathological state characterization information is generated, which provides a foundation for constructing a high-fidelity virtual patient, solves the problem of lack of pathological features in modeling from a single data source, and improves the realism of the model.

[0007] Further, the step of spatially registering and fusing the three-dimensional surface point cloud data, the voxel model data, and the motion trajectory data using a fusion reconstruction algorithm to generate a three-dimensional mesh model including pathological state characterization information includes: Spatial registration is performed between the three-dimensional surface point cloud data and the voxel model data to obtain spatially registered data. Dynamic constraints are constructed based on the motion trajectory data; The dynamic constraints are applied to the spatially registered data, and fusion weights are assigned according to the organization type for differentiated fusion to simulate the response characteristics of local organizations under functional states, thereby generating the three-dimensional mesh model.

[0008] In the above implementation process, spatial registration ensures accurate alignment of multi-source data, dynamic constraints are used to simulate the response characteristics of tissues under functional states, and tissue type differentiation is combined to enable the generated mesh model to realistically reflect the deformation and stress distribution under physiological movement, thereby enhancing the dynamic realism of the model.

[0009] Furthermore, the differentiated fusion based on the allocation of fusion weights according to organization type includes: The corresponding fusion weights are assigned according to the tissue type, and pathological feature identifiers are embedded in the three-dimensional mesh model.

[0010] In the above implementation process, fusion weights are assigned according to tissue type to ensure accurate reconstruction of the internal structure of hard tissue and the surface details of soft tissue. At the same time, pathological feature identifiers are embedded to achieve visualization and quantification of the lesion area, which facilitates intuitive observation and measurement analysis in clinical teaching.

[0011] Furthermore, before spatially registering the three-dimensional surface point cloud data with the voxel model data, the method further includes: Adaptive threshold surface reconstruction is performed on the voxel model data to obtain voxel surface data; The voxel surface data and the three-dimensional surface point cloud data are pre-registered.

[0012] In the above implementation process, high-quality voxel surface data is obtained through adaptive threshold surface reconstruction before fine registration, and pre-registration is performed to provide a good initial alignment state for subsequent spatial registration and avoid the algorithm from getting trapped in local optima.

[0013] Furthermore, generating a virtual patient model based on the three-dimensional mesh model includes: By using a correlation model between systemic diseases and local tissue lesions, systemic disease parameters are mapped onto the three-dimensional mesh model to obtain the virtual patient model.

[0014] In the above implementation process, the correlation model between systemic diseases and local tissue lesions is used to dynamically map systemic disease parameters into a three-dimensional mesh model, so that the virtual patient model can reflect the impact of the overall health status on local pathology, and enhance the clinical relevance and dynamic simulation capability of the model.

[0015] Furthermore, the association model is constructed through the following steps: Acquire clinical sample data, which includes systemic disease parameters and corresponding local tissue pathological manifestations; Based on the clinical sample data, the mapping relationship between systemic disease parameters and local tissue pathological manifestations is trained to generate the association model.

[0016] In the above implementation process, a correlation model is generated based on clinical sample data to ensure that the mapping relationship between systemic disease parameters and local pathological manifestations is scientific and accurate, and the model can be updated as data accumulates, providing a reliable basis for dynamic pathological simulation.

[0017] Furthermore, the method also includes: Based on a pre-set pathological feature library, a personalized virtual patient model that conforms to the target pathological state is generated through parameterized configuration; wherein, the pathological feature library includes multiple disease types and corresponding pathological parameter dimensions.

[0018] In the above implementation process, personalized virtual patient models that conform to different target pathological states can be quickly generated by pre-setting a pathological feature library and parameterized configuration, thereby improving the efficiency of case generation and meeting the needs of clinical teaching and training for diverse and highly realistic cases.

[0019] A second aspect of this application provides an electronic device, the electronic device comprising: processor; Memory used to store processor-executable instructions; Wherein, when the processor invokes the executable instructions, it implements any of the methods described in the first aspect.

[0020] A third aspect of this application provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of any of the methods described in the first aspect.

[0021] A fourth aspect of this application provides a computer program product, the computer program product including a computer program, which, when executed by a processor, implements any of the methods described in the first aspect. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a virtual patient modeling method provided in an embodiment of this application; Figure 2 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Taking the field of oral medicine as an entry point for analysis, the demand for highly realistic virtual patients is particularly prominent in clinical teaching and multidisciplinary collaborative diagnosis and treatment training. In particular, the diagnosis and treatment of periodontal disease involves multiple dimensions such as soft and hard tissue assessment, systemic disease correlation analysis, and functional state simulation, which puts forward higher requirements for the realism and dynamism of the model.

[0027] However, related technologies for virtual patient models in the oral cavity have significant limitations: general models often focus on reconstructing the anatomical morphology of teeth, lacking a systematic depiction of periodontal tissue-specific pathological features such as periodontal pocket depth distribution and alveolar bone resorption gradient; traditional models lack the correlation mechanism between systemic diseases and local lesions, making it difficult to simulate the impact of systemic diseases such as diabetes on periodontal status; furthermore, case generation efficiency is low, making it difficult to quickly cover clinical scenarios with different degrees of lesions. The root causes of these problems include: related technologies mainly rely on single-modal medical imaging data, generating static three-dimensional anatomical models through image segmentation and surface reconstruction, without fully considering the specificity of periodontal tissues, lacking the ability to fuse multi-source data and dynamically simulate pathology, and failing to meet the needs of standardized oral training and clinical practice for highly realistic cases.

[0028] To address this, this application proposes to acquire three-dimensional morphological data from intraoral scanning, CBCT tomographic images, and mandibular movement trajectory data. After preprocessing, these data are input into an optimized triangular mesh reconstruction algorithm to achieve precise mapping of periodontal soft and hard tissues and enhance the visualization of periodontal pathological features. Furthermore, by calling a pre-defined associated model through a systemic disease association interface, parameters of systemic diseases such as diabetes are embedded into the three-dimensional model to form a dynamic pathological state. Finally, a parameterized model is established based on a desensitized case database to rapidly generate virtual cases of periodontal disease with varying degrees of severity, solving the problem of lack of specificity and dynamism in general oral models and providing a highly realistic platform for standardized oral training.

[0029] Based on this, embodiments of this application provide a virtual patient modeling method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a virtual patient modeling method provided in an embodiment of this application.

[0030] In this embodiment, the method includes: Step S10: Obtain the surface morphology data, internal tomographic image data, and dynamic functional status data of the object to be modeled; It should be noted that the method is applicable to various medical scenarios, such as orthopedics, neurology, cardiology, dentistry, etc., and is used to construct high-fidelity three-dimensional virtual patient models that include pathological features.

[0031] Surface morphology data refers to the three-dimensional geometric information of the surface of an object or organ. It can be obtained through devices such as optical scanners, laser scanners, and structured light scanners. For example, high-precision point cloud data can be obtained by scanning the surface of a patient, or surface contours can be obtained through stereo photography.

[0032] Internal tomographic imaging data: A sequence of tomographic images of the internal structures of an object. These can be acquired using equipment such as computed tomography (CT), cone-beam computed tomography (CBCT), magnetic resonance imaging (MRI), and ultrasound. For example, acquiring CT scan images of a patient's lesion site, which contain grayscale information about bones and soft tissues, can reflect internal anatomical structures and pathological changes.

[0033] Dynamic functional status data: Temporal data of an object during movement or functional activities. This data can be acquired using devices such as motion capture systems, dynamic MRI, gait analyzers, and mandibular motion tracking systems. Examples include recording the trajectory of a patient's joint movements and the dynamic deformation sequence of organs.

[0034] Optionally, in a dental setting, the method can be used to model virtual patients with periodontal disease. Regarding the acquisition of surface morphology data: an intraoral scanner can be used to acquire three-dimensional surface morphology data of the patient's crowns and gingiva, clearly recording the dentition morphology, gingival contour, and soft tissue surface details.

[0035] Regarding the acquisition of internal tomographic images: CBCT equipment is used to acquire tomographic images of the patient's jawbone, alveolar bone, periodontal ligament, and tooth roots. The slice thickness can be set to 0.125mm, which can display internal structural information such as the boundary of alveolar bone resorption and the location of the mandibular nerve canal.

[0036] For the acquisition of dynamic functional status data: The mandibular motion trajectory of patients such as chewing and opening and closing mouth is recorded using a mandibular motion trajectory device, which can reflect the functional status of the temporomandibular joint and the deviation in occlusal movement.

[0037] Step S20: Generate three-dimensional surface point cloud data based on the surface morphology data, generate voxel model data based on the internal tomographic image data, and generate motion trajectory data based on the dynamic functional state data; It should be noted that 3D surface point cloud data is generated from surface morphology data after preprocessing such as denoising, filtering, and simplification, and is used to describe the external surface geometry of the object to be modeled. For example, the raw point cloud obtained from an optical scan of a patient's body surface generally contains noise and redundant data. Algorithms such as Gaussian filtering and bilateral filtering can be used to remove high-frequency noise while preserving surface features; subsequently, methods such as uniform sampling, curvature adaptive sampling, or voxel mesh filtering are used to simplify the point cloud, reducing the data volume while retaining geometric details, resulting in clean and lightweight surface point cloud data. For example, in a dental medicine scenario, the following processing flow can be used for crown and gingival surface data obtained from intraoral scans: First, Gaussian filtering is applied to remove noise introduced by saliva reflection or the scanning environment; then, the Poisson surface reconstruction algorithm is used to transform the scattered point cloud into a smooth, continuous triangular mesh surface. This algorithm can effectively repair micro-holes caused by scanning blind spots, generating a complete crown-gingival surface model; finally, as needed, uniform sampling can be used to simplify the mesh vertices to balance model accuracy and subsequent processing efficiency.

[0038] Voxel model data: Generated from internal tomographic image data through image segmentation and 3D reconstruction. Each voxel contains a grayscale value or tissue type label to characterize the spatial distribution of internal structures. For example, threshold segmentation or deep learning-based semantic segmentation can be performed on CT images to extract target tissue regions (such as bones and organs). Then, a 3D voxel model is generated through voxelization. The grayscale value of each voxel can reflect tissue density, or different tissue types can be distinguished by label values. For example, in the context of oral medicine, an improved U-Net network can be used to segment CBCT data to identify structures such as alveolar bone, periodontal ligament, tooth structure, and mandibular nerve canal. During segmentation, a grayscale threshold range (e.g., 226-1900 HU for bone tissue) can be used to improve segmentation accuracy. After segmentation, a 3D voxel model is constructed through voxelization, and pathological information such as periodontal pocket depth and bone resorption degree are labeled at the voxel level, providing a data foundation for the visualization of pathological features.

[0039] Motion trajectory data: Generated from dynamic functional state data through temporal alignment, filtering, interpolation, etc., it describes the motion patterns of an object during functional activities. For example, Kalman filtering is applied to the point trajectories recorded by motion capture to remove outliers, and time normalization is performed according to the motion cycle to obtain a smooth motion trajectory curve. For example, in a dental setting, for chewing and opening / closing motion data recorded by a mandibular motion tracker, a Kalman filtering algorithm is used to remove outliers caused by poor patient cooperation or equipment vibration; then, temporal segmentation and alignment are performed according to the chewing cycle, and motion trajectory curves with equal time intervals are generated through interpolation.

[0040] Step S30: Spatial registration and fusion processing of the three-dimensional surface point cloud data, the voxel model data and the motion trajectory data are performed using a fusion reconstruction algorithm to generate a three-dimensional mesh model including pathological state characterization information; It should be noted that spatial registration involves first registering the 3D surface point cloud data and the voxel model data to the same spatial coordinate system. This can be achieved using the Iterative Closest Point (ICP) algorithm or a feature-point-based registration method, resulting in high-precision alignment between surface and internal data. For example, for orthopedic models, the femoral surface point cloud can be registered with the bone contour in the CT voxel model, with the error controlled to the sub-millimeter level.

[0041] Dynamic constraint construction: Based on motion trajectory data, the motion patterns of the object under functional states are extracted, and dynamic constraint equations are constructed. For example, for the knee joint, the joint angle change is calculated based on the gait trajectory and used as a constraint condition for the subsequent fusion process, so that the generated model can simulate soft tissue deformation during movement.

[0042] Fusion Processing: Based on the registered data, fusion weights are assigned according to tissue type for differentiated fusion. For example, voxel data is given higher weights for skeletal regions to ensure geometric accuracy; surface data is given higher weights for soft tissue regions to preserve surface details. Simultaneously, dynamic constraints are applied to the fusion process to generate a 3D mesh model that reflects functional state response.

[0043] The 3D mesh model includes pathological state representation information: during the fusion process, pathological markers are added to the corresponding areas of the mesh model based on prior clinical knowledge or pathological features automatically identified from images (such as tumor boundaries, fracture lines, and osteoporotic areas). These markers can be vertex attributes (such as color and labels), texture coordinates, or additional annotation layers, thereby enabling the generated model to contain quantifiable pathological information, such as the volume of the lesion area and the grayscale threshold range.

[0044] Step S40: Generate a virtual patient model based on the three-dimensional mesh model.

[0045] It should be noted that a 3D mesh model containing pathological state representation information is used as a foundation, and further processing is performed to generate a virtual patient model that can be directly used for teaching, training, or clinical simulation. This may include operations such as texture mapping, material assignment, and motion driving. For example, textures are added to the skeletal model, mechanical properties are added to the soft tissue model, and motion trajectory data is bound to the model to enable it to simulate physiological movement. The final generated virtual patient model can be interactively observed and manipulated in a virtual reality environment.

[0046] In this embodiment, by acquiring and fusing multi-source data, a three-dimensional mesh model containing pathological state characterization information is generated, providing a foundation for constructing a high-fidelity virtual patient. This solves the problem of lacking pathological features in modeling from a single data source and improves the realism of the model.

[0047] Based on any of the above embodiments, the step of spatially registering and fusing the three-dimensional surface point cloud data, the voxel model data, and the motion trajectory data using a fusion reconstruction algorithm to generate a three-dimensional mesh model including pathological state characterization information includes: Spatial registration is performed between the three-dimensional surface point cloud data and the voxel model data to obtain spatially registered data. Dynamic constraints are constructed based on the motion trajectory data; The dynamic constraints are applied to the spatially registered data, and fusion weights are assigned according to the organization type for differentiated fusion to simulate the response characteristics of local organizations under functional states, thereby generating the three-dimensional mesh model.

[0048] It should be noted that the purpose of spatial registration is to align 3D surface point cloud data and voxel model data from different devices and coordinate systems to the same spatial coordinate system, laying the foundation for subsequent fusion.

[0049] Specifically, the coordinate transformation relationship between surface point cloud data and voxel model data is first established. For surface point clouds, the coordinate system is typically defined by the scanning device (e.g., the local coordinate system of an intraoral scanner); for voxel models, the coordinate system is defined by the imaging device (e.g., the global coordinate system of CBCT or the DICOM coordinate system). Initial registration can be performed using marker points or feature points. Marker point method: Before scanning and image acquisition, non-transmissive marker points (such as small balls) are placed on the surface of the object to be modeled. These marker points are visible in both surface scanning and tomographic images. Initial alignment is achieved by extracting the corresponding positions of the marker points in the two coordinate systems and calculating the rigid body transformation matrix (rotation matrix R and translation vector T).

[0050] Feature point method: If no manual markers are available, anatomical feature points (such as bony prominences, tooth cusps, organ boundaries, etc.) can be used for matching. The initial transformation matrix is ​​solved by manually or automatically identifying at least three pairs of non-collinear corresponding points.

[0051] After initial registration, fine registration is performed using the Iterative Closest Point (ICP) algorithm or its variants. The ICP algorithm iteratively finds the closest point pairs and minimizes the mean square error, continuously optimizing the transformation matrix until convergence. In the context of oral medicine, the crown-gingival surface point cloud acquired from intraoral scanning is ICP-registered with the alveolar bone and tooth surface reconstructed from the CBCT voxel model. Due to the high precision of intraoral scan data (0.01 mm) and the high resolution of CBCT (0.125 mm slice thickness), the registration error can be less than 0.02 mm, ensuring precise alignment between the crown surface and the internal structures of the tooth root and alveolar bone. After registration, the 3D surface point cloud data is transformed to the coordinate system of the voxel model using a transformation matrix to obtain spatially registered data. At this point, the surface point cloud and the voxel model are strictly corresponding in space, allowing for subsequent fusion.

[0052] It should be understood that dynamic constraints are deformation patterns in functional states extracted from motion trajectory data, used to guide the fusion process so that the generated mesh model can simulate the response characteristics of tissues in motion.

[0053] Optionally, the motion trajectory data is usually time-series displacement or angle data, which can be smoothed and periodically extracted first: Filtering and noise reduction: Kalman filtering or low-pass filtering is used to remove high-frequency noise and abnormal jump points while preserving the true motion components; Periodic segmentation: Long sequences are segmented into individual periods based on motion characteristics (such as gait period and chewing period) to facilitate the extraction of typical motion patterns; Parametric representation: Fitting trajectory data to a time function, such as a Fourier series, yields a continuous equation of motion.

[0054] Specifically, based on the preprocessed motion trajectory, dynamic constraint equations are constructed to describe the kinematic or dynamic behavior of the object to be modeled in a functional state. Constraint types include: Displacement constraints: defining the displacement changes of specific locations (such as the mandible and articular surfaces) over time to drive model motion; Deformation constraints: establishing deformation field functions based on the deformation laws of soft tissues during motion (such as muscle contraction and skin stretching) to constrain the displacement and strain relationships of mesh nodes; Contact constraints: defining the constraint relationships between contact surfaces for scenarios involving occlusion or joint contact to prevent penetration and simulate interaction forces. In the oral medicine scenario, dynamic constraint equations are constructed based on the opening and closing mouth and chewing motion data recorded by the mandibular motion trajectory analyzer. The trajectory is fitted as a curve of time function to define the displacement and rotation laws of the mandible relative to the maxilla, and further derives the stress distribution pattern of periodontal tissues during occlusion.

[0055] The spatially registered data is combined with dynamic constraints, and the final 3D mesh model is generated through differential fusion. The process is as follows: First, based on the grayscale range or segmentation labels of the voxel model data, different tissue types are identified. Common tissue types include hard tissues (such as bone and teeth), soft tissues (such as muscle, gingiva, and skin), and cavities or fluid regions. For each tissue type, a preset fusion weight coefficient is used to balance the contributions of surface point cloud data and voxel model data in the fusion process. For hard tissue regions: voxel model data can accurately reflect internal structure and density distribution, so it is given a higher weight (e.g., 0.8); surface point cloud data mainly supplements surface details, so it is given a lower weight (e.g., 0.2). For soft tissue regions: surface point cloud data can better capture fine surface morphology (such as gingival contours), so it is given a higher weight (e.g., 0.7); voxel model data provides internal structural reference, so it is given a lower weight (e.g., 0.3). For transitional regions (such as the boundary between bone and soft tissue): a weight gradient strategy is adopted, using a distance field or fuzzy membership function to achieve a smooth transition of weights, avoiding unnatural seams after fusion.

[0056] Next, the constructed dynamic constraint equations are applied to the spatially registered data, allowing the fusion process to proceed within the framework of the functional state. The application of dynamic constraints includes: driving the displacement of key nodes based on motion trajectory data, and transferring the displacement to the entire mesh region through interpolation algorithms (such as radial basis function interpolation or finite element shape function interpolation) to generate a dynamic deformation field; if it is necessary to simulate the mechanical response of the tissue in the functional state, stress distribution can be calculated using finite element analysis or a mass-spring model in conjunction with material properties (such as elastic modulus and Poisson's ratio). The application of dynamic constraints ensures that the subsequently fused model can realistically reflect the deformation and stress characteristics of the tissue during motion. Under the guidance of dynamic constraints, the spatially registered data is fused differentially, specifically in two aspects: geometric fusion and attribute fusion. Geometric fusion: For each mesh vertex, based on the weight coefficient of its tissue type, the coordinates of the corresponding points in the surface point cloud and the corresponding points in the voxel model contour are weighted and averaged to obtain the fused vertex position. Attribute fusion: The grayscale values ​​of the voxel model, tissue labels, and information such as the color and texture of the surface point cloud are mapped to the mesh vertices. For example, the grayscale values ​​at corresponding positions in the voxel model are normalized and stored as vertex attributes for subsequent density analysis; tissue type labels obtained from segmentation are assigned to vertices for easy classification rendering; if the surface point cloud contains color information (such as the gingival color in intraoral scans), it is mapped to the mesh vertices to form a realistic appearance. For pathological feature embedding: during the fusion process, pathological feature identifiers are added to the corresponding mesh positions based on the pathological regions segmented from the voxel model (such as periodontal pockets, bone resorption areas, and tumor boundaries). Identification methods include: adding special labels to the vertices of pathological regions for easy subsequent visualization rendering (e.g., red indicates a periodontal pocket depth ≥3mm); storing pathological quantitative indicators (such as bone resorption percentage and periodontal pocket depth values) as vertex attributes; and setting a transparency coefficient for bone resorption areas to make the internal structure visible.

[0057] In the context of oral medicine, the differential fusion process is as follows: The alveolar bone region is reconstructed using a high-density mesh (appearance density 1.5 times that of the tooth body region) to precisely represent the trabecular bone structure, while CBCT grayscale values ​​are mapped to pseudo-color bone density. Based on the segmentation results, regions with periodontal pocket depth ≥3mm are marked with red vertices; regions with bone resorption ≥50% are set to a vertex transparency of 0.6 for semi-transparent display. The occlusal stress distribution values ​​obtained from dynamic constraint calculations are stored as vertex attributes to generate stress cloud maps. After fusion, the generated 3D mesh model undergoes mesh smoothing (e.g., Laplacian smoothing), cavity repair, adaptive simplification, and quality checks (e.g., checking mesh closure and normal consistency). The final output 3D mesh model possesses high geometric accuracy (oral model error ≤0.03mm), rich pathological information, and dynamic response capabilities.

[0058] In this embodiment, spatial registration ensures accurate alignment of multi-source data, dynamic constraints are used to simulate the response characteristics of tissues in functional states, and tissue type differentiation is combined to enable the generated mesh model to realistically reflect the deformation and stress distribution under physiological movement, thereby enhancing the dynamic realism of the model.

[0059] Based on any of the above embodiments, the step of assigning fusion weights according to organization type for differentiated fusion includes: The corresponding fusion weights are assigned according to the tissue type, and pathological feature identifiers are embedded in the three-dimensional mesh model.

[0060] It should be noted that different tissue types have varying degrees of dependence on surface point cloud data and voxel model data during the fusion and reconstruction process. For example, hard tissues (such as bones and teeth) rely on voxel model data to accurately reflect their internal structure and density distribution, while soft tissues (such as gingiva and muscle) rely on surface point cloud data to preserve detailed surface morphology and color information. Therefore, differentiated fusion weights need to be assigned according to tissue type: hard tissue regions are given higher weights to voxel model data (e.g., 0.8) and lower weights to surface point cloud data (e.g., 0.2); soft tissue regions are given the opposite: higher weights to surface point cloud data (e.g., 0.7) and lower weights to voxel model data (e.g., 0.3). For transitional regions (such as the bone-soft tissue interface), a weight gradient strategy (e.g., distance field gradient) is used to achieve smooth fusion and avoid seam artifacts.

[0061] Pathological feature identifiers refer to additional information embedded in a 3D mesh model, used to mark, quantify, or visualize disease-related regions, parameters, or states. Specifically, pathological feature identifiers can exist in various forms: storing pathological parameters as vertex attributes (such as periodontal pocket depth, percentage of bone resorption, and inflammation index); highlighting lesion areas with color labels (such as marking areas with periodontal pocket depth ≥3mm in red); setting transparency attributes to display internal structures (such as displaying areas with bone resorption ≥50% semi-transparently); or adding additional geometric elements as markers (such as lesion boundary outlines and indicator arrows).

[0062] For example, the following steps are used to assign fusion weights based on tissue type and embed pathological feature identifiers into the 3D mesh model: First, tissue type is identified based on the grayscale threshold or segmentation results of the voxel model data. For example, regions with CT values ​​of 226-1900 HU are labeled as hard tissue (bone, teeth), while other soft tissue regions (gingiva, muscle) are obtained through deep learning segmentation. For hard tissue regions, voxel model data is assigned a higher weight (e.g., 0.8) and surface point cloud data a lower weight (e.g., 0.2) to prioritize the accuracy of internal structures; for soft tissue regions, surface point cloud data is assigned a higher weight (e.g., 0.7) and voxel data a lower weight (e.g., 0.3) to preserve surface details; a distance field gradient weight is used at the bone-soft tissue interface to achieve a smooth transition. Subsequently, pathological feature identifiers are embedded in the fused mesh model: pathological parameters obtained from image segmentation or clinical measurement (such as periodontal pocket depth and percentage of bone resorption) are stored as vertex attributes, and the visualization effect is controlled according to these attribute values. For example, areas with periodontal pocket depth ≥3mm are rendered in red, and areas with bone resorption ≥50% are set to semi-transparent display, thereby forming a virtual patient model that includes both anatomical structure and quantitative pathological information.

[0063] In this embodiment, fusion weights are assigned according to tissue type to ensure accurate reconstruction of the internal structure of hard tissue and the surface details of soft tissue. At the same time, pathological feature identifiers are embedded to achieve visualization and quantification of the lesion area, which facilitates intuitive observation and measurement analysis in clinical teaching.

[0064] Based on any of the above embodiments, before spatially registering the three-dimensional surface point cloud data and the voxel model data, the method further includes: Adaptive threshold surface reconstruction is performed on the voxel model data to obtain voxel surface data; The voxel surface data and the three-dimensional surface point cloud data are pre-registered.

[0065] It should be noted that adaptive thresholding is a segmentation threshold that is dynamically adjusted based on local image features (such as grayscale distribution, gradient, and texture) to improve the accuracy of surface reconstruction in different regions.

[0066] Surface reconstruction: The process of extracting isosurfaces from voxel data to generate a three-dimensional mesh surface.

[0067] Voxel surface data: Geometric surface information extracted from voxel model data through surface reconstruction algorithms, represented in the form of triangular meshes.

[0068] Pre-registration: A coarse alignment process performed before fine registration, with the aim of providing a good initial state for subsequent high-precision registration.

[0069] Understandably, before performing fine spatial registration between 3D surface point cloud data and voxel model data, it is necessary to first extract reliable surface geometric information, i.e., voxel surface data, from the voxel model data.

[0070] Traditional global thresholding surface reconstruction methods use a single grayscale threshold to extract isosurfaces. However, in medical images, the grayscale distribution may differ between different tissues and regions, making it difficult for a single threshold to simultaneously meet the reconstruction accuracy requirements of all areas. The adaptive thresholding method shown in this embodiment dynamically adjusts the threshold based on local image features to improve the accuracy of surface reconstruction.

[0071] Alternatively, the adaptive threshold can be determined based on the following strategy: Local grayscale statistical analysis: Within a local window of the voxel model, the statistical characteristics such as the mean, variance, and gradient of grayscale values ​​are calculated, and a threshold is dynamically set based on these characteristics. For example, in the bone tissue region, due to the large fluctuations in grayscale caused by the trabecular structure, 80% of the local mean can be used as the threshold; in the cortical bone region, the grayscale values ​​are high and stable, and 90% of the local mean can be used as the threshold.

[0072] Edge intensity detection: The gradient magnitude of the voxel data is calculated, and regions with larger gradients typically correspond to tissue boundaries. Near the boundaries, the threshold is adaptively adjusted according to the gradient direction to make the extracted isosurfaces more closely resemble the actual anatomical boundaries.

[0073] Machine learning prediction: Based on labeled training data, a regression model or classifier is trained to predict the optimal segmentation threshold according to local image features. For example, a random forest model can be used, taking local gray-level statistical features and texture features as input, and outputting the best threshold for that region.

[0074] Multiscale analysis: Voxel data is analyzed at different scales. The coarse scale is used to identify tissue types and general boundaries, while the fine scale is used to accurately locate surface positions, thereby adaptively adjusting the threshold at different scales.

[0075] Specifically, in the context of oral medicine, an adaptive voxel threshold is set for the alveolar bone trabecular region in CBCT data: in areas rich in trabecular bone, due to lower bone density and looser structure, a lower threshold (e.g., 226 HU) is used to preserve fine structures; in dense cortical bone regions, a higher threshold (e.g., 500 HU) is used to accurately extract the bone surface. This adaptive strategy can simultaneously preserve the fine structure of the trabecular bone and the clear boundaries of the cortical bone.

[0076] Optionally, based on a determined adaptive threshold, a surface reconstruction algorithm is used to extract isosurfaces from the voxel model to generate voxel surface data. Surface reconstruction algorithms include, but are not limited to: The Moving Cubes (MC) algorithm traverses voxel cells and determines the topological structure of isosurfaces within a cell by comparing the gray values ​​of cell vertices with a threshold, generating triangular patches. Based on the standard MC algorithm, an improvement is made for the alveolar bone trabecular region by setting an adaptive voxel threshold. This dynamically adjusts the threshold for isosurface extraction according to the gray-level characteristics of different regions, improving the reconstruction accuracy of the trabecular bone structure.

[0077] Dual contour algorithm: It can preserve sharp features while extracting isosurfaces, and is suitable for scenarios that require high-precision boundaries (such as tooth cusps and bone ridges).

[0078] Deep learning-based reconstruction method: Use neural networks to directly predict surface positions from voxel data to generate smooth and complete surface meshes.

[0079] For example, in a dental setting, an improved MC algorithm is used to triangulate CBCT tomographic images. For the alveolar bone trabecular region, an adaptive voxel threshold is set through local grayscale statistical analysis to make the reconstructed trabecular structure clearer and more complete, while avoiding artifacts caused by noise. The reconstruction result is voxel surface data, represented in the form of a triangular mesh, containing surface geometric information of hard tissues such as alveolar bone and tooth structure.

[0080] Optionally, the reconstructed voxel surface data may contain noise, voids, or non-manifold structures, which can be optimized. Mesh smoothing: Laplacian smoothing or Taubin smoothing algorithms are used to remove staircase artifacts without shrinking features; Hole repair: Holes are filled by triangulation based on surrounding geometry to ensure continuous surface closure.

[0081] It should be understood that the purpose of the pre-registration includes: Narrowing the search space: Adjusting the relative poses of the two datasets to be roughly aligned to avoid the fine registration algorithm getting stuck in local optima.

[0082] Improve registration efficiency: Reduce the number of iterations required for fine registration and speed up the overall processing speed.

[0083] Enhanced registration robustness: For data with large-angle rotation or translation differences, direct ICP registration may fail, while pre-registration can provide a reliable initial transformation.

[0084] For example, depending on the data characteristics and application scenario, the following pre-registration methods can be selected: Feature-point-based pre-registration: Automatically or manually identify corresponding anatomical feature points (such as bony prominences, tooth cusps, organ boundaries, etc.) in voxel surface data and surface point cloud data. Based on at least three pairs of non-collinear corresponding points, calculate the rigid body transformation matrix (rotation matrix R and translation vector T) to achieve initial alignment. Feature point recognition can employ automatic detection algorithms based on curvature, normal variation, or local shape descriptors (such as point feature histogram (PFH) and fast point feature histogram (FPFH)).

[0085] Pre-registration based on Principal Component Analysis (PCA): The principal axis directions of the voxel surface data and surface point cloud data are calculated separately (three principal component vectors are obtained through PCA). The principal axes of the two datasets are aligned to achieve rotation correction. Translation vectors are calculated based on the data center points to achieve position alignment.

[0086] Based on global registration algorithms: This method employs algorithms such as Sample Consistency Initial Registration (SAC-IA) to quickly estimate the initial transformation matrix through random sampling and feature matching. It is insensitive to initial pose and is suitable for situations without prior information.

[0087] Pre-registration based on markers: If artificial markers (such as non-transparent spheres) are placed during the data acquisition stage, the positions of the markers in the two datasets can be directly extracted, and the transformation matrix can be calculated.

[0088] In practical implementation, in the context of oral medicine, pre-registration is performed as follows: Easily identifiable anatomical features such as cusps, incisal edges, and molar ridges are selected, and 5-8 pairs of corresponding points are manually or automatically marked on the surface of the intraoral scan point cloud and the voxel reconstructed by CBCT. Based on the corresponding point pairs, singular value decomposition (SVD) is used to solve the rigid body transformation matrix, minimizing the mean square error between point pairs. The transformation matrix is ​​applied to the voxel surface data to roughly align it with the intraoral scan point cloud, with rotation errors controlled within 5° and translation errors controlled within 1mm.

[0089] After pre-registration, voxel surface data with initial transformation is obtained. At this point, the voxel surface data and the 3D surface point cloud data are roughly aligned and can be directly used as input for the spatial registration step for subsequent fine ICP registration.

[0090] In this embodiment, high-quality voxel surface data is obtained through adaptive threshold surface reconstruction before fine registration, and pre-registration is performed to provide a good initial alignment state for subsequent spatial registration and avoid the algorithm from getting trapped in local optima.

[0091] Based on any of the above embodiments, generating a virtual patient model based on the three-dimensional mesh model includes: By using a correlation model between systemic diseases and local tissue lesions, systemic disease parameters are mapped onto the three-dimensional mesh model to obtain the virtual patient model.

[0092] It should be noted that the correlation model is a mathematical model that describes the quantitative relationship between systemic disease parameters and local tissue lesions. It can be constructed based on clinical sample data (clinical sample data can come from hospital medical record databases, clinical research projects, or public medical datasets).

[0093] Systemic disease parameters include disease type (such as diabetes, hypertension, rheumatoid arthritis), disease course (such as years of illness), and laboratory indicators (such as blood glucose level, glycated hemoglobin, blood pressure, and inflammatory factor levels).

[0094] Local tissue lesions refer to local lesion characteristics related to systemic diseases, such as periodontal pocket depth, alveolar bone resorption, and gingival inflammation index in periodontitis; bone mineral density value in osteoporosis; and joint space width and cartilage thickness in arthritis.

[0095] Understandably, this embodiment, based on the already generated three-dimensional mesh model containing pathological state characterization information, maps systemic disease parameters into the model through the association model, thereby enabling dynamic adjustment of the pathological state.

[0096] Specifically, systemic disease parameters of the object to be modeled can be received through the systemic disease association interface. For example, for a virtual patient with diabetes and periodontal disease, the following parameters can be input: 10-year duration of diabetes, average fasting blood glucose of 8.5 mmol / L in the past 3 months, and glycated hemoglobin of 7.5%. The input systemic disease parameters are then substituted into a preset association model to calculate the corresponding local tissue pathological changes or adjustment coefficients. Taking the association between diabetes and periodontal disease as an example, the association model can define the following mapping rules: calculate the periodontal inflammation index adjustment factor based on blood glucose levels, calculate the alveolar bone resorption rate based on disease duration, and calculate the periodontal pocket depth change based on glycated hemoglobin values. These mapping rules are then applied to the 3D mesh model, such as modifying the numerical attributes of pathological feature identifiers in the mesh model, like adjusting the percentage of bone resorption and periodontal pocket depth in specific areas; fine-tuning the mesh vertex positions based on the calculation results to simulate changes in tissue morphology (such as changes in gingival swelling); and updating the model's color mapping table to reflect the new pathological parameter distribution in pseudo-color rendering. After mapping is complete, the 3D mesh model is updated to a dynamic pathological state model that includes the effects of systemic diseases. The model has the following characteristics: the pathological manifestations of the model are dynamically adjusted according to the input systemic disease parameters, which can reflect the impact of disease control level on local tissues; in interactive teaching or training scenarios, students can adjust systemic disease parameters (such as setting different blood glucose levels), and the model updates and displays the corresponding periodontal status changes in real time; the model supports displaying the impact of systemic diseases in a variety of ways, such as displaying the distribution of inflammation degree through pseudo-color, displaying the progress of bone resorption through changes in transparency, and simulating stress changes under functional conditions through animation.

[0097] For example, taking an oral medicine scenario, a virtual patient model is created for a 55-year-old male with moderate chronic periodontitis and type 2 diabetes: Input systemic disease parameters: 10-year history of diabetes, mean fasting blood glucose level of 8.5 mmol / L in the past 3 months, and glycated hemoglobin of 7.5%.

[0098] Association model mapping calculation: Based on the preset diabetes-periodontal disease association rule, the periodontal inflammation enhancement factor under poor blood glucose control is calculated to be 1.3; the alveolar bone resorption rate is calculated according to the disease course and adjusted to 1.5 times the baseline value; the periodontal pocket depth in the posterior tooth region is calculated to be increased by 0.8 mm from the baseline value.

[0099] Model update: The above calculation results were mapped to the 3D mesh model of periodontal disease: In the periodontal pocket depth annotation layer, the depth of the posterior tooth area was updated from 5mm to 5.8mm; in the alveolar bone resorption gradient layer, the degree of bone resorption in the molar area was updated from 1 / 2 to nearly 2 / 3; the pseudo-color rendering of the gingival area was updated from light red to dark red, indicating the aggravation of inflammation.

[0100] A virtual patient model is generated: a virtual model of periodontal disease is obtained that reflects the patient's current blood glucose control level. The trainee can adjust the blood glucose parameter to 6.0 mmol / L through the interface. The model automatically updates, showing reduced gingival redness and swelling, decreased periodontal pocket depth, and slowed bone resorption rate, intuitively demonstrating the improvement effect of blood glucose-lowering treatment on periodontal condition.

[0101] In this embodiment, a correlation model between systemic diseases and local tissue lesions is used to dynamically map systemic disease parameters into a three-dimensional mesh model, enabling the virtual patient model to reflect the impact of overall health status on local pathology and enhancing the model's clinical relevance and dynamic simulation capabilities.

[0102] Based on any of the above embodiments, the association model is constructed through the following steps: Acquire clinical sample data, which includes systemic disease parameters and corresponding local tissue pathological manifestations; Based on the clinical sample data, the mapping relationship between systemic disease parameters and local tissue pathological manifestations is trained to generate the association model.

[0103] It should be noted that local histopathological data refers to local lesion characteristics related to systemic diseases, which can be obtained through imaging examinations, clinical measurements, pathological biopsies, etc. In the field of oral medicine, these include periodontal pocket depth, clinical attachment level, alveolar bone resorption, gingival bleeding index, and tooth mobility.

[0104] Optionally, in a dental setting, clinical data from more than 50 patients with diabetes and periodontal disease are obtained, with the following fields recorded for each sample: Systemic disease parameters: type of diabetes, duration of disease (years), fasting blood glucose (mmol / L), glycated hemoglobin (%). Local pathological data: average periodontal pocket depth (mm), deepest periodontal pocket depth (mm), percentage of alveolar bone resorption at the most severe site (%), and gingival bleeding index (0-3 grade).

[0105] Specifically, clinical sample data is used as input, and machine learning or statistical analysis methods are employed to train a model that establishes a mapping relationship between systemic disease parameters and local pathological manifestations, generating a callable association model. Optionally, the sample data is preprocessed before model training: Feature selection: Systemic parameters strongly correlated with local pathological manifestations are selected as input features through correlation analysis, principal component analysis, or domain knowledge screening. For example, in the association between diabetes and periodontal disease, glycated hemoglobin and disease duration can be selected as key features. Data standardization: Parameters of different dimensions are normalized or standardized to eliminate the impact of scale differences on model training. Dataset partitioning: The sample data is divided into a training set (e.g., 70%), a validation set (e.g., 15%), and a test set (e.g., 15%) for model training, parameter tuning, and performance evaluation. Next, based on the data type and complexity of the relationships, an appropriate modeling method is selected. For example, for complex scenarios with multivariate nonlinear interactions, the following machine learning algorithms can be used for training: ① Neural Networks: Construct a Multilayer Perceptron (MLP) network. The input layer nodes correspond to the filtered systemic disease parameters (such as glycated hemoglobin, disease course), the hidden layers perform nonlinear transformations on the input through activation functions (such as ReLU, Sigmoid), and the output layer nodes correspond to the target local pathological manifestations (such as periodontal pocket depth, percentage of bone resorption). The network iteratively adjusts the weights through the backpropagation algorithm to minimize the loss function (such as mean squared error) between the predicted and true values. For more complex spatial structure data (such as radiomics features), convolutional neural networks (CNNs) can also be used to extract deep features. ② Support Vector Machines: Suitable for small sample sizes, nonlinear regression, or classification problems. The input data is mapped to a high-dimensional feature space through a kernel function (such as radial basis function kernel RBF). An optimal hyperplane is constructed in the high-dimensional space to make the sample points as close to the hyperplane as possible (regression) or maximize the class margin (classification). The penalty coefficient and kernel function parameters need to be determined through cross-validation in the model to prevent overfitting. Regardless of the method used, the model's performance can be evaluated on validation and test sets after training to ensure that its generalization ability meets the requirements of clinical applications.

[0106] In this embodiment, a correlation model is trained and generated based on clinical sample data to ensure that the mapping relationship between systemic disease parameters and local pathological manifestations is scientific and accurate, and the model can be updated as data accumulates, providing a reliable basis for dynamic pathological simulation.

[0107] Based on any of the above embodiments, the method further includes: Based on a pre-set pathological feature library, a personalized virtual patient model that conforms to the target pathological state is generated through parameterized configuration; wherein, the pathological feature library includes multiple disease types and corresponding pathological parameter dimensions.

[0108] It should be noted that the pathological feature library is a structured database that stores pathological feature parameters of various diseases, used to support parameterized configuration and case generation.

[0109] For example, the data sources of the pathological feature database include clinical desensitized cases (collecting clinical data of real patients, including disease diagnosis, imaging features, pathological measurement indicators, etc. For example, in the field of oral medicine, CBCT images, intraoral scan data, periodontal examination records, etc. of more than 50 patients with periodontal disease can be collected), literature and expert knowledge (extracting the pathological parameter range of typical cases from medical literature and clinical guidelines, or defining the pathological features of standard cases by experts in the field).

[0110] The pathology feature database is stored in a structured manner, with each record corresponding to a disease type or a specific case, and containing the following fields: Types of diseases: such as chronic periodontitis, aggressive periodontitis, osteoporosis, osteoarthritis, etc.

[0111] Pathological parameters: Quantifiable indicators describing the severity and characteristics of the disease, such as: Periodontal disease: periodontal pocket depth, clinical attachment level, alveolar bone resorption, bleeding on probing index, tooth mobility, etc. Orthopedic diseases: bone mineral density T-score, fracture risk score, joint space width, cartilage thickness, etc. Cardiovascular diseases: plaque burden, vascular stenosis rate, calcification score, etc.

[0112] Anatomical location: The specific location where the disease occurs, such as the entire mouth, molar region, anterior region, proximal femur, lumbar spine, etc.

[0113] Typical case data: Optional complete 3D model data or key feature point coordinates, used as templates for generating new cases.

[0114] The pathological feature database can be stored in a relational database (such as MySQL) or a non-relational database (such as MongoDB), and supports retrieval by disease type, severity, anatomical location, and other conditions.

[0115] In the context of oral medicine, the pathological feature database includes: Disease types: chronic periodontitis, aggressive periodontitis, gingivitis, etc.

[0116] Pathological parameters include: average periodontal pocket depth, deepest periodontal pocket depth, percentage of alveolar bone resorption at the most severe site, percentage of positive sites for probing bleeding, and tooth mobility grading.

[0117] Typical cases: Stores the above parameters and corresponding three-dimensional model features of more than 50 desensitized cases.

[0118] Specifically, the parameterized configuration interface serves as the user's window for interacting with the system. It is used to select the disease type, set pathological parameters, and preview the generated virtual patient model. After the user completes the parameter configuration and triggers the generation command, the system quickly instantiates a personalized virtual patient model that conforms to the target pathological state based on the pathological feature library and the generated 3D mesh model.

[0119] In the context of oral medicine, to address the need for periodontal disease cases of varying severity in standardized dental training, the implementation process is as follows: First, a pathological feature database was constructed: based on data from 50 desensitized patients with diabetes and periodontal disease, a pathological feature database was established. Each record includes: periodontal disease stage, average periodontal pocket depth across the entire mouth, deepest pocket depth, percentage of the site with the most severe alveolar bone resorption, positive rate of bleeding on probing, duration of diabetes, and glycated hemoglobin.

[0120] Secondly, case settings are configured through a parametric configuration interface: Students or teachers select the disease type "chronic periodontitis" on the configuration interface, and then set the following parameters using the slider: Periodontal disease stage: Stage II (moderate); Average periodontal pocket depth: 4.5mm; Bone resorption degree: 35%; Optional: Input systemic disease parameters (such as 8-year duration of diabetes, 7.8% glycated hemoglobin) to call the associated model to adjust the inflammatory manifestations.

[0121] Subsequently, the system retrieves the closest case template from the pathological feature database based on the user-configured parameters and maps the template's pathological features onto the basic oral cavity 3D mesh model: the system retrieves the closest case template from the feature database based on "Stage II, average pocket depth 4.5mm, bone resorption 35%". The pathological features of the template are then mapped onto the basic oral cavity model: a 5mm deep periodontal pocket is generated in the molar region, and a 3mm deep pocket is generated in the anterior region; based on the 35% bone resorption, semi-transparent rendering is applied to the corresponding root bifurcation area. If systemic disease parameters are input, the associated model calculates the inflammation index, making the gingival area appear a deeper red (simulating redness and swelling caused by poor blood sugar control).

[0122] Finally, after clicking the "Generate" button, the system outputs a virtual oral patient model that perfectly matches the set parameters within approximately 10 seconds. The generated model supports interactive operation: users can measure the periodontal pocket depth at any location, rotate to observe the three-dimensional morphology of alveolar bone resorption, switch stress distribution cloud maps to view occlusal forces, and try adjusting blood glucose parameters (such as adjusting glycated hemoglobin from 7.8% to 6.5%) to observe the dynamic changes in gingival redness and swelling and bone resorption progression, thereby intuitively understanding the impact of systemic diseases on periodontal status.

[0123] In this embodiment, by using a preset pathological feature library and parameterized configuration, personalized virtual patient models that conform to different target pathological states can be quickly generated, improving the efficiency of case generation and meeting the needs of diverse and highly realistic cases in clinical teaching and training.

[0124] Based on the methods described in any of the above embodiments, this application also provides, as follows: Figure 2 The diagram shows the structure of an electronic device. Figure 2 At the hardware level, the electronic device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to implement the methods described in any of the above embodiments.

[0125] Based on the methods described in any of the above embodiments, this application also provides a computer storage medium storing a computer program, which, when executed by a processor, can be used to perform the methods described in any of the above embodiments.

[0126] Based on the methods described in any of the above embodiments, this application also provides a computer program product, which includes one or more computer programs or instructions. The computer program or instructions may be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. When executed by a processor, the computer program implements the methods described in any of the above embodiments.

[0127] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0128] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0129] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0131] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0132] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A virtual patient modeling method, characterized in that, The method includes: Acquire surface morphology data, internal tomographic image data, and dynamic functional status data of the object to be modeled; Three-dimensional surface point cloud data is generated based on the surface morphology data, voxel model data is generated based on the internal tomographic image data, and motion trajectory data is generated based on the dynamic functional state data. The three-dimensional surface point cloud data, the voxel model data and the motion trajectory data are spatially registered and fused using a fusion reconstruction algorithm to generate a three-dimensional mesh model that includes pathological state characterization information. A virtual patient model is generated based on the aforementioned three-dimensional mesh model.

2. The method according to claim 1, characterized in that, The process of spatially registering and fusing the 3D surface point cloud data, the voxel model data, and the motion trajectory data using a fusion reconstruction algorithm to generate a 3D mesh model including pathological state characterization information includes: Spatial registration is performed between the three-dimensional surface point cloud data and the voxel model data to obtain spatially registered data. Dynamic constraints are constructed based on the motion trajectory data; The dynamic constraints are applied to the spatially registered data, and fusion weights are assigned according to the organization type for differentiated fusion to simulate the response characteristics of local organizations under functional states, thereby generating the three-dimensional mesh model.

3. The method according to claim 2, characterized in that, The differentiated fusion based on the allocation of fusion weights according to organization type includes: The corresponding fusion weights are assigned according to the tissue type, and pathological feature identifiers are embedded in the three-dimensional mesh model.

4. The method according to claim 2, characterized in that, Before spatially registering the 3D surface point cloud data with the voxel model data, the method further includes: Adaptive threshold surface reconstruction is performed on the voxel model data to obtain voxel surface data; The voxel surface data and the three-dimensional surface point cloud data are pre-registered.

5. The method according to claim 1, characterized in that, The process of generating a virtual patient model based on the three-dimensional mesh model includes: By using a correlation model between systemic diseases and local tissue lesions, systemic disease parameters are mapped onto the three-dimensional mesh model to obtain the virtual patient model.

6. The method according to claim 5, characterized in that, The association model is constructed through the following steps: Acquire clinical sample data, which includes systemic disease parameters and corresponding local tissue pathological manifestations; Based on the clinical sample data, the mapping relationship between systemic disease parameters and local tissue pathological manifestations is trained to generate the association model.

7. The method according to claim 1, characterized in that, The method further includes: Based on a pre-set pathological feature library, a personalized virtual patient model that conforms to the target pathological state is generated through parameterized configuration; wherein, the pathological feature library includes multiple disease types and corresponding pathological parameter dimensions.

8. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store processor-executable instructions; Wherein, when the processor invokes the executable instructions, it implements the method described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It stores computer instructions that, when executed by a processor, implement the steps of any of the methods described in claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images