An orthodontic treatment effect prediction method and system based on multi-modal data fusion

Abstract

The present application relates to the field of medical informatics, and provides a kind of orthodontic curative effect prediction method and system based on multi-modal data fusion.The method comprises: the spatial registration fusion of the multi-modal image data of patient craniofacial region is carried out, and multi-modal fusion image is obtained;The voxel attribute mapping in multi-modal fusion image is carried out, and individualized digital twin anatomical model is obtained;The stress distribution after treatment cycle is applied to the finite element analysis calculation of correction force, and time series stress distribution set is obtained;According to the local periodontal membrane hydrostatic pressure in time series stress distribution set, compare and determine with capillary blood vessel filling pressure threshold value, import super-threshold region into pathological evolution logic, update grid node space coordinates according to bone remodeling law in non-super-threshold region, and obtain updated anatomical model;Based on the update anatomical model and corresponding super-threshold determination result, generate curative effect prediction report.The present application improves the biomechanical authenticity of orthodontic simulation, and enhances the credibility of curative effect prediction result.

Description

Technical Field

[0001] This invention relates to the field of medical informatics technology, and in particular to a method and system for predicting the efficacy of orthodontic treatment based on multimodal data fusion. Background Technology

[0002] Orthodontic treatment is a clinical process in which teeth are moved in a controlled manner within the alveolar bone by applying continuous corrective forces. Its core biological mechanism lies in the cascade of bone resorption and deposition triggered by the periodontal ligament under force. With the development of digital medical technology, technologies such as cone-beam computed tomography (CBCT), intraoral 3D scanning, and finite element analysis have been gradually introduced into the field of orthodontics, providing a data foundation for constructing individualized 3D models of the craniofacial region and realizing computer-aided planning of treatment plans.

[0003] Existing digital prediction technologies for orthodontics (such as geometric tooth alignment software like ClinCheck) typically treat alveolar bone as a homogeneous and isotropic ideal material, generating orthodontic animations based solely on the geometric coordinate transformation of the target tooth position. They fail to incorporate the spatial distribution differences in bone density among individual patients into the calculation model, resulting in a significant deviation between the material properties used in finite element stress analysis and the heterogeneous mechanical properties of the actual jawbone. Furthermore, due to the lack of temporal tracking and biological threshold determination mechanisms for local hydrostatic pressure in the periodontal ligament, existing technologies cannot identify the risk of hyaline degeneration and irreversible osteonecrosis caused by local pressure continuously exceeding microvascular filling pressure under orthodontic force. Consequently, they cannot provide early warnings for long-term pathological complications such as root resorption and bone dehiscence during the efficacy prediction stage. Summary of the Invention

[0004] This invention provides a method and system for predicting the efficacy of orthodontic treatment based on multimodal data fusion, in order to overcome the shortcomings of existing technologies.

[0005] This invention provides a method for predicting the efficacy of orthodontic treatment based on multimodal data fusion, comprising: S1: Spatial registration and fusion are performed on the acquired multimodal image data of the patient's craniofacial region to obtain multimodal fused images; S2: Perform voxel attribute mapping processing on the bone grayscale information in the multimodal fused image, convert it into biomechanical attribute parameters of multiple voxel micro-elements, and obtain an individualized digital twin anatomical model. S3: Using the individualized digital twin anatomical model as the calculation basis, the stress distribution after applying corrective force to the treatment cycle divided into multiple time increments is calculated by finite element analysis to obtain the time-series stress distribution set. S4: Based on the local periodontal membrane hydrostatic pressure corresponding to multiple time increments in the time-series stress distribution set, compare and determine with the preset capillary filling pressure threshold, import the over-threshold area into the pathological evolution logic, and update the grid node spatial coordinates of the non-over-threshold area according to the bone remodeling law to obtain the updated anatomical model of multiple time increments. S5: Generate a therapeutic efficacy prediction report based on the updated anatomical model and the corresponding threshold determination results over multiple time increments.

[0006] According to the present invention, a method for predicting the efficacy of orthodontic treatment based on multimodal data fusion is provided, wherein step S1 further includes: S11: Obtain three-dimensional volume data containing bone density grayscale information by cone-beam CT scanning, obtain point cloud data of tooth crown surface by intraoral three-dimensional scanning, and obtain soft tissue surface morphology data by facial three-dimensional scanning to obtain the initial multimodal raw dataset. S12: Extract craniofacial anatomical landmarks from multiple modal data in the initial multimodal raw dataset. Using the cone-beam CT three-dimensional coordinate system as the global reference system, rigidly register the crown surface point cloud data and soft tissue surface morphology data respectively through the iterative nearest point algorithm to obtain a multimodal fused image under a unified coordinate system.

[0007] According to the method for predicting orthodontic treatment efficacy based on multimodal data fusion provided by the present invention, step S2 further includes: S21: Read the HU values ​​of multiple voxels in the multimodal fused image; S22: Classify and label the global voxels based on the HU threshold range of cortical bone and cancellous bone to obtain a bone classification voxel map; S23: Substitute the HU values ​​of multiple voxels in the bone classification voxel map into the empirical formula for elastic modulus density, calculate the corresponding elastic modulus and Poisson's ratio for each voxel, and assign the calculation parameters to multiple finite element mesh elements to obtain an individualized digital twin anatomical model with a heterogeneous distribution of mechanical properties.

[0008] According to the method for predicting the efficacy of orthodontic treatment based on multimodal data fusion provided by the present invention, in step S3, in the finite element analysis calculation of each time increment, the mesh node coordinates and material properties of the updated anatomical model output from the previous time increment are used as the initial boundary conditions. The orthodontic force corresponding to the current time increment is applied to the crown contact surface in the form of a surface load. Static finite element solution is performed, the principal stress tensor components of each periodontal ligament unit node are extracted, and the local hydrostatic pressure value is calculated to form the node hydrostatic pressure distribution field of the corresponding time increment.

[0009] According to the method for predicting orthodontic treatment efficacy based on multimodal data fusion provided by the present invention, step S4 further includes: S41: Compare the hydrostatic pressure distribution field of nodes corresponding to multiple time increments in the time-series stress distribution set with the capillary filling pressure threshold node by node. Mark the nodes with hydrostatic pressure greater than the capillary filling pressure threshold as pathological evolution nodes, and mark the nodes with hydrostatic pressure less than or equal to the capillary filling pressure threshold as normal remodeling nodes to obtain a node state classification map. S42: For pathological evolution nodes, block normal bone remodeling iteration, import the corresponding region into bone resorption morphological degeneration logic, accumulate and record the amount of bone thickness loss, and obtain a pathological cumulative damage distribution map. For normal remodeled nodes in the node state classification diagram, the change in bone density at each node within the corresponding time increment is calculated according to Frost's bone remodeling law. The change in bone density is converted into the displacement of the grid node and the spatial coordinates of the node are updated to obtain an updated anatomical model with multiple time increments.

[0010] According to the method for predicting the efficacy of orthodontic treatment based on multimodal data fusion provided by the present invention, in step S42, the bone resorption morphological degeneration logic of the pathological evolution node is specifically as follows: if the current node is marked as a pathological evolution node within multiple consecutive time increments, it is determined that the corresponding region has entered an irreversible bone loss state, the elastic modulus of the corresponding grid unit is set to 0, and the percentage of bone thickness loss in the current region is marked in the form of a risk heat map to obtain a pathological cumulative damage distribution map.

[0011] According to the method for predicting orthodontic treatment efficacy based on multimodal data fusion provided by the present invention, step S5 further includes: S51: Render the updated anatomical model with multiple time increments into a four-dimensional dynamic evolution animation according to the time frame, and superimpose the hydrostatic pressure distribution of each frame onto the three-dimensional geometric surface of the corresponding frame to form an evolution animation sequence containing a time-series stress heat map. S52: Extract the high-risk anatomical regions recorded in the cumulative pathological damage distribution map in all time increments, and locate the corresponding tooth position number and anatomical location coordinates; S53. Based on the force direction and hydrostatic pressure exceeding the threshold corresponding to the high-risk area, calculate the recommended adjustment amount of the corrective force and output the directional operation command including the force value and torque direction. S54. The directional operation instructions and the evolutionary animation sequence are combined to form a therapeutic effect prediction report.

[0012] A second aspect of the present invention provides an orthodontic treatment efficacy prediction system based on multimodal data fusion, comprising: Fusion module: Used to spatially register and fuse the acquired multimodal image data of the patient's craniofacial region to obtain multimodal fused images; Mapping module: used to perform voxel attribute mapping processing on the bone grayscale information in the multimodal fused image, convert it into biomechanical attribute parameters of multiple voxel micro-elements, and obtain an individualized digital twin anatomical model; Calculation module: Used to perform finite element analysis calculations on the stress distribution after applying corrective force to multiple time increments of the treatment cycle, based on the individualized digital twin anatomical model, to obtain a set of time-series stress distributions; Update module: It is used to compare and determine the local periodontal membrane hydrostatic pressure corresponding to multiple time increments in the time-series stress distribution set with the preset capillary filling pressure threshold, import the over-threshold area into the pathological evolution logic, and update the spatial coordinates of the grid nodes in the non-over-threshold area according to the bone remodeling law to obtain the updated anatomical model of multiple time increments. Output module: Used to update the anatomical model and the corresponding threshold determination results based on multiple time increments, and generate a therapeutic efficacy prediction report.

[0013] A third aspect of the present invention provides an orthodontic treatment efficacy prediction device based on multimodal data fusion, comprising: A memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause a multimodal data fusion-based orthodontic efficacy prediction device to perform a multimodal data fusion-based orthodontic efficacy prediction method as described above.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement a method for predicting the efficacy of orthodontic treatment based on multimodal data fusion as described in any of the preceding claims.

[0015] This invention provides a method and system for predicting orthodontic treatment efficacy based on multimodal data fusion. By converting the grayscale information of bone density from cone-beam computed tomography (CBCT) into a heterogeneous elastic modulus and Poisson's ratio distribution through voxel-attribute mapping, the constructed individualized digital twin anatomical model can realistically reflect the spatial mechanical differences between cortical and cancellous bone. Compared with the existing approach of homogenizing alveolar bone, the consistency between the finite element stress calculation results and the actual biomechanical response of the jawbone is significantly improved, thus giving the stress distribution prediction in subsequent time increments higher individual specificity and physical reliability. Secondly, the introduction of the temporal threshold determination mechanism in this invention enables the system to quantitatively locate high-risk areas where the local hydrostatic pressure of the periodontal ligament continuously exceeds the capillary filling pressure before the formal implementation of the orthodontic plan. This allows for early identification of pathological evolution trends such as hyaline degeneration, root resorption, and bone dehiscence, bringing long-term complications that could only be passively discovered during clinical follow-up into the scope of predictive intervention. This greatly reduces the clinical risk of irreversible bone damage caused by improper orthodontic force settings. In addition, the four-dimensional dynamic evolution animation and directional orthodontic force adjustment instructions output by the system enable doctors to intuitively obtain the full picture of bone morphology evolution over time during the treatment planning stage, and make verifiable quantitative adjustments to the magnitude of the applied force and the direction of the torque accordingly. This not only reduces the uncertainty of the treatment plan caused by relying on experience judgment, but also provides a visual and objective basis for doctor-patient communication. Overall, it promotes the evolution of orthodontic treatment from geometric tooth arrangement to biomechanically driven precision treatment management. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A schematic diagram of the process for predicting the efficacy of orthodontic treatment based on multimodal data fusion provided by the present invention; Figure 2 This is a schematic diagram of the structure of an orthodontic treatment efficacy prediction system based on multimodal data fusion provided by the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0019] The embodiments of the present invention are described below with reference to the figures.

[0020] like Figure 1 As shown, this invention provides a method for predicting the efficacy of orthodontic treatment based on multimodal data fusion, comprising: S1: Spatial registration and fusion of the acquired multimodal image data of the patient's craniofacial region are performed to obtain multimodal fused images.

[0021] Step S1 further includes: S11: Obtain three-dimensional volume data containing bone density grayscale information using cone-beam CT scanning, obtain point cloud data of tooth crown surface using intraoral three-dimensional scanning, and obtain soft tissue surface morphology data using facial three-dimensional scanning to obtain the initial multimodal raw dataset.

[0022] In step S11, the present invention first acquires raw data of the patient's craniofacial region from three independent acquisition channels. The first channel outputs three-dimensional volumetric data through cone-beam CT scanning. This volumetric data consists of several voxels, each carrying a Hounsfield Unit (HU) grayscale value. The HU value reflects the degree of attenuation of X-rays by the tissue at the voxel's location. The higher the bone density, the larger the HU value. The HU value of the cortical bone region is higher than 850, the HU value of the cancellous bone region is lower than this value, and the HU value of the soft tissue and periodontal ligament region is even lower. The second channel outputs point cloud data of the crown surface through intraoral three-dimensional scanning. This point cloud consists of a large number of three-dimensional coordinate points. Each point records the spatial coordinates (x, y, z) of a sampling location on the crown surface. The densely distributed set of coordinate points collectively describes the high-precision geometry of the crown. The third channel outputs soft tissue surface morphology data through facial three-dimensional scanning, also recording the external surface geometry of the facial soft tissue in the form of a three-dimensional point cloud or triangular mesh. The above three sets of data are each established in an independent scanner coordinate system, with different definitions of the coordinate origin and axis, which together constitute the initial multimodal raw dataset.

[0023] S12: Extract craniofacial anatomical landmarks from multiple modal data in the initial multimodal raw dataset. Using the cone-beam CT three-dimensional coordinate system as the global reference system, rigidly register the crown surface point cloud data and soft tissue surface morphology data respectively through the iterative nearest point algorithm to obtain a multimodal fused image under a unified coordinate system.

[0024] Furthermore, since the three sets of data coordinate systems in the initial multimodal raw dataset are different, this invention needs to unify them under the same spatial reference system. This invention uses the cone-beam CT three-dimensional coordinate system as the global reference system because CT volume data covers the entire craniofacial skeletal structure, has the widest spatial range, and contains bone density information, making it suitable as a registration benchmark.

[0025] The registration process is executed in parallel in two paths. First, the present invention manually or automatically identifies a set of craniofacial anatomical landmarks in both cone-beam CT body data and intraoral scan point cloud data. The principle for selecting landmarks is that they are anatomical structures that can be stably identified in both types of data, such as the incisal edge of incisors and the cusps of molars.

[0026] Subsequently, an initial rigid transformation matrix (containing a rotation matrix R and a translation vector t) is calculated using these corresponding marker point pairs. This invention inputs this initial transformation matrix into the Iterative Closest Point (ICP) algorithm for fine registration. Specifically, the processing logic of the ICP algorithm is as follows: the intraoral scan point cloud is transformed to the CT coordinate system using the current transformation matrix; each point in the transformed point cloud is traversed, and the nearest point in the corresponding CT surface data is searched as the corresponding point pair; a new transformation matrix that minimizes the mean square distance error is calculated based on all corresponding point pairs; the current transformation matrix is ​​replaced with the new transformation matrix, and the above process is repeated until the change in the mean square distance error between two adjacent iterations is lower than a preset convergence threshold, and the final rigid transformation matrix is ​​output. Finally, this invention applies this matrix to all coordinate points of the intraoral scan crown surface point cloud data to complete the alignment of the intraoral scan data to the CT coordinate system.

[0027] For soft tissue surface morphology data, the present invention performs the same process, namely, extracting the corresponding soft tissue surface anatomical landmarks from facial scan data and CT data, calculating the initial transformation matrix, inputting it into the ICP algorithm for iterative solution, outputting the final transformation matrix, and applying it to all coordinate points of the soft tissue surface morphology data.

[0028] After the two channels are registered, the cone-beam CT three-dimensional volume data, the transformed crown surface point cloud data, and the transformed soft tissue surface morphology data are all in the same CT coordinate system. The superposition of the three constitutes a multimodal fused image in a unified coordinate system. This fused image simultaneously contains bone density voxel information, high-precision crown geometric information, and soft tissue surface morphology information.

[0029] S2: Perform voxel attribute mapping processing on the bone grayscale information in the multimodal fused image, convert it into biomechanical attribute parameters of multiple voxel micro-elements, and obtain an individualized digital twin anatomical model.

[0030] Step S2 further includes: S21: Read the HU values ​​of multiple voxels in the multimodal fused image.

[0031] In step S21, the present invention reads HU values ​​voxel by voxel from the cone-beam CT volume data portion of the multimodal fusion image and outputs a three-dimensional HU value matrix corresponding to the CT scan resolution. Each element in the matrix corresponds to the HU value of a voxel position in space.

[0032] S22: Classify and label the global voxels based on the HU threshold range of cortical bone and cancellous bone to obtain a bone classification voxel map.

[0033] In step S22, the present invention sets the HU threshold range based on the histological characteristics of the jawbone: voxels with a HU value higher than 850 are classified as high-density cortical bone, voxels with a HU value within the corresponding range of cancellous bone are classified as cancellous bone, and periodontal ligament and soft tissue areas are separately labeled according to their HU range.

[0034] This invention performs threshold comparison on each voxel in the three-dimensional HU value matrix output by S21, writes the classification label back to the corresponding voxel position, and outputs a bone classification voxel map. This map corresponds completely to the original CT volume data in space, and each voxel position carries an additional tissue category label.

[0035] S23: Substitute the HU values ​​of multiple voxels in the bone classification voxel map into the empirical formula for elastic modulus density, calculate the corresponding elastic modulus and Poisson's ratio for each voxel, and assign the calculation parameters to multiple finite element mesh elements to obtain an individualized digital twin anatomical model with a heterogeneous distribution of mechanical properties.

[0036] In step S23, the present invention substitutes the HU value of each voxel in the bone classification voxel map into the elastic modulus-density empirical formula, and calculates the elastic modulus corresponding to each voxel. Compared with Poisson The elastic modulus describes a material's ability to resist deformation under stress; a higher value indicates a harder material. Poisson's ratio describes the ratio of transverse strain to longitudinal strain under uniaxial stress. Together, they determine the linear elastic constitutive relation of the voxel element. Subsequently, this invention generates a finite element mesh based on multimodal fused images. The elastic modulus and Poisson's ratio of each voxel within the coverage area of ​​each finite element mesh are weighted and averaged, and then assigned to that mesh element. Thus, each mesh element carries independent heterogeneous mechanical properties, outputting an individualized digital twin anatomical model with a heterogeneous distribution of mechanical properties.

[0037] S3: Using the individualized digital twin anatomical model as the calculation basis, the stress distribution after applying corrective force to the treatment cycle is divided into multiple time increments, and finite element analysis is performed to obtain the time-series stress distribution set.

[0038] In step S3, in the finite element analysis calculation of each time increment, the mesh node coordinates and material properties of the updated anatomical model output from the previous time increment are used as the initial boundary conditions. The orthodontic force corresponding to the current time increment is applied to the crown contact surface in the form of a surface load. Static finite element solution is performed, the principal stress tensor components of each periodontal ligament unit node are extracted, and the local hydrostatic pressure value is calculated to form the node hydrostatic pressure distribution field of the corresponding time increment.

[0039] In step S3, the present invention divides the total treatment cycle T into several time increments according to a preset time step Δt (ranging from 1 to 4 weeks), and performs finite element static analysis incrementally. Specifically, in the first time increment, the mesh node coordinates and elastic modulus and Poisson's ratio of each element of the individualized digital twin anatomical model output in S23 are used as the initial boundary conditions.

[0040] Subsequently, the present invention applies the orthodontic force (within the range of 10g to 500g) corresponding to the current time increment to the crown contact surface in the form of a surface load. The surface load is that the force is evenly distributed to each finite element node of the contact surface and converted into the equivalent concentrated force of each node.

[0041] Subsequently, a static finite element method is performed. In the solution, this invention establishes a global stiffness matrix, which is assembled from the stiffness matrices of each element according to the nodal degrees of freedom. The element stiffness matrix is ​​calculated from the elastic modulus, Poisson's ratio, and geometry of each element. Using the nodal force vector of the applied surface load as the right-hand side, a system of linear equations is solved to obtain the displacement vectors of all nodes.

[0042] After calculating the displacement vector, this invention calculates the strain tensor of each element based on the displacement vector and the strain-displacement relationship matrix, and then converts the strain tensor into a stress tensor using the constitutive relation matrix. For each finite element node of the periodontal ligament region, this invention extracts the three principal stress components of its stress tensor. , , The local hydrostatic pressure value of the node is calculated. Finally, the hydrostatic pressure values ​​of all periodontal ligament nodes are summed to form the node hydrostatic pressure distribution field at the current time increment.

[0043] In subsequent time increments, the present invention replaces the initial boundary conditions with the updated mesh node coordinates and updated material properties of the dissecting model output from the previous time increment S42, and repeats the above solution process until all time increments are calculated. The nodal hydrostatic pressure distribution fields of all time increments are arranged in time sequence, and the time sequence stress distribution set is output.

[0044] S4: Based on the local periodontal membrane hydrostatic pressure corresponding to multiple time increments in the time-series stress distribution set, compare and determine with the preset capillary filling pressure threshold, import the over-threshold area into the pathological evolution logic, and update the spatial coordinates of the grid nodes in the non-over-threshold area according to the bone remodeling law to obtain the updated anatomical model of multiple time increments.

[0045] Step S4 further includes: S41: The hydrostatic pressure distribution field of nodes corresponding to multiple time increments in the time-series stress distribution set is compared with the capillary filling pressure threshold node by node. Nodes with hydrostatic pressure greater than the capillary filling pressure threshold are marked as pathological evolution nodes, and nodes with hydrostatic pressure less than or equal to the capillary filling pressure threshold are marked as normal remodeling nodes, thus obtaining a node state classification diagram.

[0046] In step S41, the present invention uses the capillary filling pressure threshold (ranging from 20 g / cm² to 30 g / cm²) as the criterion to perform a node-by-node numerical comparison of the hydrostatic pressure distribution field corresponding to each time increment in the time-series stress distribution set. Specifically, for each finite element node in the distribution field, its hydrostatic pressure value is read and compared with a preset threshold: nodes with hydrostatic pressure greater than the threshold are marked as pathological evolution nodes, and nodes with hydrostatic pressure less than or equal to the threshold are marked as normal remodeling nodes. After all nodes are marked, a node state classification map is output. This map corresponds one-to-one with the mesh of the individualized digital twin anatomical model in space. Each node carries one of two state labels, "pathological" or "normal," for use in the subsequent triage processing in S42.

[0047] S42: For pathological evolution nodes, block normal bone remodeling iteration, import the corresponding region into the bone resorption morphological degeneration logic, accumulate and record the amount of bone thickness loss, and obtain the pathological cumulative damage distribution map.

[0048] In step S42, the bone resorption morphological degeneration logic of the pathological evolution node is as follows: if the current node is marked as a pathological evolution node within multiple consecutive time increments, it is determined that the corresponding region has entered an irreversible bone loss state, the elastic modulus of the corresponding grid unit is set to 0, and the percentage of bone thickness loss in the current region is marked in the form of a risk heat map to obtain a pathological cumulative damage distribution map.

[0049] For the normal remodeling nodes in the node state classification diagram, in step S42, the present invention calculates the change in bone density at each node within the corresponding time increment based on Frost's bone remodeling law, converts the change in bone density into the displacement of the grid node and updates the node spatial coordinates to obtain an updated anatomical model with multiple time increments.

[0050] Subsequently, based on the node state classification diagram, the present invention executes two independent sets of data processing logic for pathological evolution nodes and normal reconstruction nodes respectively.

[0051] For normal remodeling nodes, this invention introduces Frost's law of bone remodeling to calculate changes in bone mineral density. The core idea of ​​Frost's law is that bone tissue determines whether to initiate bone deposition or resorption based on local strain stimuli. When the strain value is within the normal physiological window, bone tissue maintains a steady state; above the upper limit, bone deposition accelerates; below the lower limit, bone resorption begins. This invention uses the equivalent strain value of the node within the current time increment Δt as input and calculates the change in bone mineral density at that node within time Δt based on Frost's law. Then The spatial displacement of the node is converted by the linear relationship between bone density and volumetric strain, and the coordinates of the mesh node are updated. Finally, the updated anatomical model at the end of the current time increment is obtained, which serves as the initial geometric boundary condition for the finite element calculation of the next time increment.

[0052] For pathological evolution nodes, this invention blocks the aforementioned bone remodeling iteration process and instead executes the bone resorption morphological degeneration logic. The specific determination rule is as follows: The time increment count of the current node continuously marked as a pathological evolution node on the time axis is counted. When this count reaches a preset threshold M (M is a positive integer not less than 2), the region is determined to have entered an irreversible bone loss state. The elastic modulus value of the corresponding mesh element is forcibly set to 0, meaning that this element will no longer bear any mechanical load transmission in the finite element calculations of all subsequent time increments, simulating substantial necrosis and loss of bone tissue. Simultaneously, this invention accumulates and records the bone thickness loss of each pathological evolution node within each time increment, expressing the loss as a percentage relative to the initial bone thickness, and marking it on the surface of the 3D model in the form of a risk heatmap. The color gradient continuously encodes from green (low loss) to red (high loss), ultimately outputting a pathological cumulative damage distribution map.

[0053] S5: Generate a therapeutic efficacy prediction report based on the updated anatomical model and the corresponding threshold determination results over multiple time increments.

[0054] Step S5 further includes: S51: Render the updated anatomical model with multiple time increments into a four-dimensional dynamic evolution animation according to the time frame, and superimpose the hydrostatic pressure distribution of each frame onto the three-dimensional geometric surface of the corresponding frame to form an evolution animation sequence containing a time-series stress heat map.

[0055] Furthermore, in step S51, the present invention loads the entire time-increment updated anatomical model output in S42 into the rendering engine frame by frame in chronological order, with each frame corresponding to the 3D mesh geometry at the end of a time increment. Subsequently, the numerical values ​​of the hydrostatic pressure distribution field of the corresponding time increment node output in S41 are mapped to the surface vertex color channels of the 3D geometry in that frame, using color coding to encode the magnitude of the hydrostatic pressure, and superimposed onto the surface of the 3D geometry to form the stress heatmap rendering result for that frame. Finally, the rendering frames of all time increments are concatenated in chronological order to output a four-dimensional dynamic evolution animation sequence containing the temporal stress heatmap.

[0056] S52: Extract the high-risk anatomical regions recorded in the cumulative pathological damage distribution map of all time increments, and locate the corresponding tooth position number and anatomical coordinates.

[0057] In step S52, the present invention traverses the pathological cumulative damage distribution map of all time increments, extracts the set of nodes whose elastic modulus has been set to 0 or whose bone thickness loss percentage exceeds the preset risk threshold, and maps these nodes to the standard dentition numbering system (such as FDI two-digit tooth position labeling method) based on their spatial coordinates in the digital twin anatomical model, and outputs the tooth position number list of high-risk anatomical areas and the three-dimensional coordinate set of the corresponding anatomical parts.

[0058] S53. Based on the force direction and hydrostatic pressure exceeding the threshold corresponding to the high-risk area, calculate the recommended adjustment amount of the corrective force and output the directional operation command including the force value and torque direction.

[0059] For each high-risk anatomical region output by S52, this invention reads the cumulative value of the hydrostatic pressure exceeding the threshold in that region over all time increments, and simultaneously reads the direction vector and magnitude of the current orthodontic force at the node in that region. Subsequently, based on the correspondence between the hydrostatic pressure exceeding the threshold and the direction of force application, this invention calculates the force adjustment and torque angle adjustment required to bring the hydrostatic pressure at the node in that region back below the capillary filling pressure threshold. The above calculation results are formatted into directional operation instructions containing specific tooth position numbers, suggested force reduction amounts, and suggested torque adjustment directions and angles.

[0060] S54. The directional operation instructions and the evolutionary animation sequence are combined to form a therapeutic effect prediction report.

[0061] Finally, this invention fuses the directional operation command output by S53 with the evolutionary animation sequence output by S51, and overlays the command content as a text annotation layer onto the time frame when the pathological marker first appears in the corresponding high-risk area in the animation sequence. This allows the corresponding force adjustment suggestions to be obtained synchronously at the time node when the pathological risk appears when the animation is acquired. The two are then encapsulated and output as a therapeutic effect prediction report.

[0062] like Figure 2 As shown, this invention provides a predictive system for orthodontic treatment efficacy based on multimodal data fusion, comprising: Fusion module 100: used to perform spatial registration and fusion of the acquired multimodal image data of the patient's craniofacial region to obtain a multimodal fused image; Mapping module 200: used to perform voxel attribute mapping processing on the bone grayscale information in the multimodal fused image, convert it into biomechanical attribute parameters of multiple voxel micro-elements, and obtain an individualized digital twin anatomical model; Calculation module 300: Used to perform finite element analysis calculations on the stress distribution after applying corrective force to multiple time increments of the treatment cycle, based on the individualized digital twin anatomical model, to obtain a set of time-series stress distributions; Update module 400: It is used to compare and determine the local periodontal membrane hydrostatic pressure corresponding to multiple time increments in the time-series stress distribution set with the preset capillary filling pressure threshold, import the over-threshold area into the pathological evolution logic, and update the grid node spatial coordinates of the non-over-threshold area according to the bone remodeling law to obtain the updated anatomical model of multiple time increments. Output module 500: Used to update the anatomical model and the corresponding threshold determination results based on multiple time increments, and generate a therapeutic effect prediction report.

[0063] The present invention also provides an orthodontic efficacy prediction device based on multimodal data fusion, comprising: a memory and at least one processor, wherein the memory stores instructions; at least one processor invokes the instructions in the memory to cause the orthodontic efficacy prediction device based on multimodal data fusion to perform an orthodontic efficacy prediction method based on multimodal data fusion as described above.

[0064] The present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement a method for predicting the efficacy of orthodontic treatment based on multimodal data fusion as described above.

[0065] This invention introduces the properties of heterogeneous materials, resulting in simulations that more closely resemble real human responses. Secondly, it can identify seemingly perfect geometric tooth arrangements that are biomechanically infeasible (such as those leading to bone fracture). For the output, this invention transforms invisible biomechanical changes into visualized evolutionary animations, facilitating doctor-patient communication. Furthermore, it eliminates the need for manual separation of cortical and cancellous bone, as the algorithm automatically assigns attribute values.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the efficacy of orthodontic treatment based on multimodal data fusion, characterized in that, include: S1: Spatial registration and fusion are performed on the acquired multimodal image data of the patient's craniofacial region to obtain multimodal fused images; S2: Perform voxel attribute mapping processing on the bone grayscale information in the multimodal fused image, convert it into biomechanical attribute parameters of multiple voxel micro-elements, and obtain an individualized digital twin anatomical model. S3: Using the individualized digital twin anatomical model as the calculation basis, the stress distribution after applying corrective force to the treatment cycle divided into multiple time increments is calculated by finite element analysis to obtain the time-series stress distribution set. S4: Based on the local periodontal membrane hydrostatic pressure corresponding to multiple time increments in the time-series stress distribution set, compare and determine with the preset capillary filling pressure threshold, import the over-threshold area into the pathological evolution logic, and update the grid node spatial coordinates of the non-over-threshold area according to the bone remodeling law to obtain the updated anatomical model of multiple time increments. S5: Generate a therapeutic efficacy prediction report based on the updated anatomical model and the corresponding threshold determination results over multiple time increments.

2. The method for predicting orthodontic treatment efficacy based on multimodal data fusion according to claim 1, characterized in that, Step S1 further includes: S11: Obtain three-dimensional volume data containing bone density grayscale information by cone-beam CT scanning, obtain point cloud data of tooth crown surface by intraoral three-dimensional scanning, and obtain soft tissue surface morphology data by facial three-dimensional scanning to obtain the initial multimodal raw dataset. S12: Extract craniofacial anatomical landmarks from multiple modal data in the initial multimodal raw dataset. Using the cone-beam CT three-dimensional coordinate system as the global reference system, rigidly register the crown surface point cloud data and soft tissue surface morphology data respectively through the iterative nearest point algorithm to obtain a multimodal fused image under a unified coordinate system.

3. The method for predicting the efficacy of orthodontic treatment based on multimodal data fusion according to claim 1, characterized in that, Step S2 further includes: S21: Read the HU values ​​of multiple voxels in the multimodal fused image; S22: Classify and label the global voxels based on the HU threshold range of cortical bone and cancellous bone to obtain a bone classification voxel map; S23: Substitute the HU values ​​of multiple voxels in the bone classification voxel map into the empirical formula for elastic modulus density, calculate the corresponding elastic modulus and Poisson's ratio for each voxel, and assign the calculation parameters to multiple finite element mesh elements to obtain an individualized digital twin anatomical model with a heterogeneous distribution of mechanical properties.

4. The method for predicting the efficacy of orthodontic treatment based on multimodal data fusion according to claim 1, characterized in that, In step S3, in the finite element analysis calculation of each time increment, the mesh node coordinates and material properties of the updated anatomical model output from the previous time increment are used as the initial boundary conditions. The orthodontic force corresponding to the current time increment is applied to the crown contact surface in the form of a surface load. Static finite element solution is performed, the principal stress tensor components of each periodontal ligament unit node are extracted, and the local hydrostatic pressure value is calculated to form the node hydrostatic pressure distribution field of the corresponding time increment.

5. The method for predicting the efficacy of orthodontic treatment based on multimodal data fusion according to claim 1, characterized in that, Step S4 further includes: S41: Compare the hydrostatic pressure distribution field of nodes corresponding to multiple time increments in the time-series stress distribution set with the capillary filling pressure threshold node by node. Mark the nodes with hydrostatic pressure greater than the capillary filling pressure threshold as pathological evolution nodes, and mark the nodes with hydrostatic pressure less than or equal to the capillary filling pressure threshold as normal remodeling nodes to obtain a node state classification map. S42: For the pathological evolution nodes in the node state classification diagram, block the normal bone remodeling iteration, import the corresponding region into the bone resorption morphological degeneration logic, accumulate and record the amount of bone thickness loss, and obtain the pathological cumulative damage distribution map. For normal remodeled nodes in the node state classification diagram, the change in bone density at each node within the corresponding time increment is calculated according to Frost's bone remodeling law. The change in bone density is converted into the displacement of the grid node and the spatial coordinates of the node are updated to obtain an updated anatomical model with multiple time increments.

6. The method for predicting the efficacy of orthodontic treatment based on multimodal data fusion according to claim 5, characterized in that, In step S42, the bone resorption morphological degeneration logic of the pathological evolution node is as follows: if the current node is marked as a pathological evolution node in multiple consecutive time increments, it is determined that the corresponding region has entered an irreversible bone loss state, the elastic modulus of the corresponding grid unit is set to 0, and the percentage of bone thickness loss in the current region is marked in the form of a risk heat map to obtain a pathological cumulative damage distribution map.

7. The method for predicting orthodontic treatment efficacy based on multimodal data fusion according to claim 1, characterized in that, Step S5 further includes: S51: Render the updated anatomical model with multiple time increments into a four-dimensional dynamic evolution animation according to the time frame, and superimpose the hydrostatic pressure distribution of each frame onto the three-dimensional geometric surface of the corresponding frame to form an evolution animation sequence containing a time-series stress heat map. S52: Extract the high-risk anatomical regions recorded in the cumulative pathological damage distribution map in all time increments, and locate the corresponding tooth position number and anatomical location coordinates; S53. Based on the force direction and hydrostatic pressure exceeding the threshold corresponding to the high-risk area, calculate the recommended adjustment amount of the corrective force and output the directional operation command including the force value and torque direction. S54. The directional operation instructions and the evolutionary animation sequence are combined to form a therapeutic effect prediction report.

8. A predictive system for orthodontic treatment efficacy based on multimodal data fusion, characterized in that, include: Fusion module: Used to spatially register and fuse the acquired multimodal image data of the patient's craniofacial region to obtain multimodal fused images; Mapping module: used to perform voxel attribute mapping processing on the bone grayscale information in the multimodal fused image, convert it into biomechanical attribute parameters of multiple voxel micro-elements, and obtain an individualized digital twin anatomical model; Calculation module: Used to perform finite element analysis calculations on the stress distribution after applying corrective force to multiple time increments of the treatment cycle, based on the individualized digital twin anatomical model, to obtain a set of time-series stress distributions; Update module: It is used to compare and determine the local periodontal membrane hydrostatic pressure corresponding to multiple time increments in the time-series stress distribution set with the preset capillary filling pressure threshold, import the over-threshold area into the pathological evolution logic, and update the spatial coordinates of the grid nodes in the non-over-threshold area according to the bone remodeling law to obtain the updated anatomical model of multiple time increments. Output module: Used to update the anatomical model and the corresponding threshold determination results based on multiple time increments, and generate a therapeutic efficacy prediction report.

9. A device for predicting the therapeutic effect of orthodontic treatment based on multimodal data fusion, characterized in that, include: A memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause a multimodal data fusion-based orthodontic efficacy prediction device to perform a multimodal data fusion-based orthodontic efficacy prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a processor, implement a method for predicting orthodontic treatment efficacy based on multimodal data fusion as described in any one of claims 1-7.

Images