Facial plastic surgery effect prediction method and system based on three-dimensional reconstruction

Abstract

The application discloses a face plastic surgery effect prediction method and system based on three-dimensional reconstruction, and relates to the technical field of image recognition. The method comprises the following steps: obtaining face original three-dimensional point cloud data of a surgery receiver, surgery adjustment parameters and individual physical parameters; constructing initial coupling proportion field data; inputting the surgery adjustment parameters into a directional stress release model, combining skin line direction data to calculate asymmetric stress vector data, and performing spatial orientation correction on the initial coupling proportion field to obtain asymmetric deformation trend data, which is input into a time rebound prediction model to generate time evolution prediction data in combination with individual physical parameters and tissue elastic characteristics; limiting local volume increment in the prediction data according to a scar tendency weight, generating target postoperative deformation prediction data, and performing component decomposition to output multi-dimensional interpretable deformation decomposition results. The application realizes high-precision and interpretable prediction of the soft tissue dynamic deformation process of the surgery effect.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, specifically to a method and system for predicting the effects of facial plastic surgery based on three-dimensional reconstruction. Background Technology

[0002] Facial plastic surgery prediction technology is an important research direction at the intersection of medical aesthetics and digital medicine. It aims to use computer simulations to predict the postoperative facial morphology, providing doctors with precise surgical planning references and alleviating patients' uncertainty about postoperative results. In recent years, with advancements in high-precision 3D scanning equipment and biomechanical modeling algorithms, facial simulation solutions based on 3D point clouds have gradually been applied clinically, realizing the transition from simple image overlay to 3D morphological evolution.

[0003] Current techniques primarily employ three-dimensional facial data combined with simplified linear elastic models to linearly simulate soft tissue deformation under stress, predicting immediate postoperative morphological changes. These methods provide basic visual references for routine tissue repositioning and can calculate the follow-up displacement of soft tissues based on changes in the skeletal framework, exhibiting a degree of individualized adaptability. However, for plastic surgeries involving deep tissue dissection or complex incision designs, the biomechanical environment of the patient's face is highly complex. Because the surface of human facial skin features directional skin lines, the direction of these lines determines the anisotropic characteristics of internal soft tissue tension.

[0004] During surgery, different dissection paths and incision designs can disrupt the original tension balance, triggering an asymmetric release of internal soft tissue stress. This stress release causes the soft tissue to deviate from its predetermined trajectory during healing and repositioning, resulting in an asymmetric deformation trend. Simultaneously, as a typical viscoelastic material, soft tissue's postoperative morphology is not immediately stable but rather exhibits continuous rebound during the swelling reduction period and due to tissue memory effects. Current techniques fail to spatially correct for the tension imbalance caused by individual skin lines and lack coupled modeling to counteract scar hyperplasia and long-term rebound patterns, easily leading to significant morphological deviations between the predicted model and the actual postoperative stable state. This deviation not only weakens the reliability of the prediction results but may also lead to incorrect surgical decisions, increasing the risk of unnatural facial contours or asymmetry postoperatively. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for predicting the effects of facial plastic surgery based on three-dimensional reconstruction.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention discloses a method for predicting the effects of facial plastic surgery based on three-dimensional reconstruction, comprising the following steps:

[0008] The original three-dimensional point cloud data of the patient's face, surgical adjustment parameters, and individual physical parameters are obtained; wherein, the individual physical parameters include at least skin line direction data and scar tendency weights.

[0009] Based on the original three-dimensional point cloud data of the face, bone and soft tissue hierarchical segmentation is performed to construct initial coupled proportional field data;

[0010] The surgical adjustment parameters are input into the pre-trained directional stress release model, and the anisotropic tension difference generated by the soft tissue under different dissection paths is determined by combining the skin texture direction data. This is then converted into asymmetric stress vector data of the soft tissue. The asymmetric stress vector data is used to spatially correct the initial coupling ratio field data to obtain asymmetric deformation trend data.

[0011] The asymmetric deformation trend data is input into a pre-trained time rebound prediction model. Combined with the individual physical parameters, the tissue elastic modulus and thickness distribution characteristics are analyzed from the original three-dimensional point cloud data of the face. The follow-up rebound amount within the preset observation period is calculated to generate time evolution prediction data.

[0012] Based on the scar tendency weight, the local volume increment represented by the time evolution prediction data is restricted to generate target postoperative deformation prediction data;

[0013] The target postoperative deformation prediction data is deconstructed into components, and multidimensional interpretable deformation decomposition results are output.

[0014] Secondly, this invention discloses a method and system for predicting the effects of facial plastic surgery based on three-dimensional reconstruction. Using the aforementioned method for predicting the effects of facial plastic surgery based on three-dimensional reconstruction, the method includes:

[0015] The data acquisition module is used to acquire the patient's original three-dimensional point cloud data of the face, surgical adjustment parameters, and individual physical parameters; wherein, the individual physical parameters include at least skin line direction data and scar tendency weights;

[0016] The coupling construction module is used to perform bone and soft tissue hierarchical segmentation based on the original three-dimensional point cloud data of the face, and to construct the initial coupling scale field data;

[0017] The spatial correction module is used to input the surgical adjustment parameters into the pre-trained directional stress release model, combine the skin texture direction data to determine the anisotropic tension difference generated by the soft tissue under different dissection paths, and convert it into asymmetric stress vector data of the soft tissue; the asymmetric stress vector data is used to perform spatial orientation correction on the initial coupling ratio field data to obtain asymmetric deformation trend data.

[0018] The time evolution module is used to input the asymmetric deformation trend data into a pre-trained time rebound prediction model, combine it with the individual physical parameters, analyze the tissue elastic modulus and thickness distribution characteristics from the original three-dimensional point cloud data of the face, calculate the follow-up rebound amount within a preset observation period, and generate time evolution prediction data.

[0019] The prediction module is used to limit the local volume increment represented by the time evolution prediction data according to the scar tendency weight, and generate target postoperative deformation prediction data.

[0020] The interpretation and analysis module is used to deconstruct the target postoperative deformation prediction data into components and output multidimensional interpretable deformation decomposition results.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] 1. This invention introduces a directional stress release model based on the initial coupled proportional field, transforming the deformation processing of the facial 3D model from traditional isotropic linear simulation to asymmetric spatial correction considering skin texture direction. By extracting the anisotropic tension difference determined by the skin texture direction data, the stress vector of the soft tissue point cloud under different dissection paths can be quantitatively mapped, correcting the spatial displacement orientation of soft tissue under complex surgical interference. This avoids the geometric simulation deviation caused by neglecting the internal initial tension distribution in conventional graphic deformation algorithms, resulting in a deeper fit between the predicted asymmetric deformation trend and the actual physiological and mechanical environment of the individual patient.

[0023] 2. This invention constructs a time-rebound prediction model and analyzes the elasticity characteristics of individual tissues, enabling the system to simulate the temporal rebound evolution of soft tissues during the postoperative recovery period, driven by biological memory effects and tissue properties, based on asymmetric deformation trends. This allows the prediction process to capture dynamic morphological changes from the immediate postoperative period to the stable period, rather than providing a static, instantaneous preview of the surgical outcome. This allows surgeons to anticipate the intermediate and final morphologies during swelling reduction and tissue remodeling, providing a basis for developing more rational postoperative management and communication strategies.

[0024] 3. This invention, through multi-dimensional deconstruction output, provides a traceable physical basis for the complex graphic evolution logic, offering doctors an intuitive quantitative reference for the degree of influence of different technical variables on deformation results. This assists them in optimizing and adjusting the plan based on accurate deformation source analysis, thereby enhancing the clinical decision support value of the plastic surgery prediction system. Attached Figure Description

[0025] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0026] Figure 1 This is a flowchart of the steps of the present invention;

[0027] Figure 2 This is a schematic diagram illustrating the working principle of the present invention;

[0028] Figure 3 This is a diagram illustrating the steps for generating asymmetric deformation trend data according to the present invention.

[0029] Figure 4 This is a system module connection diagram of the present invention. Detailed Implementation

[0030] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0031] In existing technologies, the prediction of facial plastic surgery outcomes largely relies on a single three-dimensional morphological superposition or a general biomechanical model, making it difficult to simultaneously consider the personalized and dynamic temporal aspects of the prediction. Traditional methods, when simulating long-term postoperative morphology, fail to comprehensively consider the anisotropic tension distribution caused by the surgical dissection path and individual biological healing characteristics, leading to deviations between the predicted results and the final stable morphology. Existing three-dimensional simulation technologies cannot effectively integrate key dynamic factors such as intraoperative mechanical disturbances, postoperative tissue rebound, and individual scar hyperplasia tendencies. Especially when extensive dissection or high-tension suturing is involved, the soft tissue distribution predicted by static models is prone to systematic errors, making it difficult to meet the needs of personalized and precise preoperative planning and doctor-patient communication.

[0032] To address these issues, the study discovered a coupling relationship between the asymmetric stress generated during surgical dissection and the temporal rebound of soft tissue. Dynamic deformation simulation was achieved by constructing a multi-stage, multi-physics-field-connected prediction model. Further investigation revealed that the initial bone-soft tissue coupling relationship plays a dominant role in morphological trends, but individual skinline orientation and scarring tendencies can lead to spatial orientation corrections and final volume constraints. Therefore, a sequential prediction approach was proposed, progressing from "anatomical static coupling" to "intraoperative mechanical correction," then to "temporal biological response," and finally to "individualized modulation." Furthermore, through technological integration, the mapping relationship between asymmetric stress vector data, the temporal rebound prediction model, and individual physical parameters was incorporated into the deformation data generation process, forming a full-cycle prediction system covering the entire process from surgical operation to long-term stability.

[0033] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0034] Example 1:

[0035] like Figure 1 As shown, the method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction includes the following steps:

[0036] The original three-dimensional point cloud data of the patient's face, surgical adjustment parameters, and individual physical parameters are obtained. Among them, the individual physical parameters include at least skin line direction data and scar tendency weight. Skin line direction data is obtained by gradient feature extraction of facial texture images, and scar tendency weight is quantified based on the patient's past medical history and dermoscopic examination data.

[0037] Based on the original 3D point cloud data of the face, bone and soft tissue hierarchical segmentation is performed to construct initial coupled scale field data;

[0038] The surgical adjustment parameters are input into the pre-trained directional stress release model. Combined with the skin texture data, the anisotropic tension difference generated by the soft tissue under different dissection paths is determined and converted into asymmetric stress vector data of the soft tissue. The asymmetric stress vector data is used to spatially correct the initial coupled proportional field data to obtain asymmetric deformation trend data.

[0039] The asymmetric deformation trend data is input into the pre-trained time rebound prediction model. Combined with individual physical parameters, the tissue elastic modulus and thickness distribution characteristics are analyzed from the original three-dimensional point cloud data of the face. The follow-up rebound amount within the preset observation period is calculated to generate time evolution prediction data.

[0040] Based on the scar tendency weight, the local volume increment represented by the time evolution prediction data is restricted to generate the target postoperative deformation prediction data; the scar tendency weight refers to the quantitative parameter reflecting the individual's postoperative scar hyperplasia potential, with a value range of [0,1].

[0041] The target postoperative deformation prediction data is deconstructed into components, and the multidimensional interpretable deformation decomposition results are output.

[0042] like Figure 2 The diagram shown illustrates the working principle of this application. The working principle of this application is achieved through deep collaboration between hardware execution units and software algorithm logic. During the data acquisition phase, the system uses a three-dimensional optical scanning device (such as a structured light scanner or a multi-camera array) to collect the spatial geometric features of the patient's face in real time, generating raw three-dimensional point cloud data of the face containing high-density three-dimensional coordinates. Simultaneously, surgical adjustment parameters (including osteotomy line position, soft tissue dissection range, etc.) and individual physical parameters imported from the clinical testing terminal are acquired through an interactive workstation.

[0043] In the initial coupling construction phase, the system invokes segmentation algorithm logic, utilizing the normal vectors and curvature features of the point cloud to achieve hierarchical segmentation of facial skin, subcutaneous tissue, and the skeletal framework. The system then uses a spatial topological association algorithm to calculate the spatial correlation degree of each soft tissue node relative to the bone anchor points, thereby constructing initial coupling scale field data. This initial coupling scale field data establishes a primary mapping relationship for the transmission of bone displacement to soft tissue.

[0044] The system uses a pre-trained directional stress release model to simulate the physical damage caused by surgical dissection. Combined with anisotropic initial tension distribution defined by skin grain direction data, it calculates the stress release of soft tissue along different dissection paths. The hardware system utilizes tensor arithmetic to transform the tension imbalance in the dissection area into asymmetric stress vector data. This vector data is then injected into a spatial orientation correction operator to directionally weight the initial coupling scale field, transforming static geometric deformation into asymmetric deformation trend data that takes into account mechanical imbalances.

[0045] The directional stress relief model aims to simulate the disruption of mechanical equilibrium caused by surgical incision and dissection. Its core is a hybrid architecture based on graph convolutional neural networks (GCN) and finite element analysis (FEA). An anisotropic viscoelastic tensor field is used as its foundation. Specifically, the directional stress relief model defines a graph structure consisting of node coordinates and edge tensions, where edge weights are influenced by dermal grain direction data. Constraints represent the orientation field data of facial skin texture, used to constrain the weights of edges in the constraint graph structure, and affecting the equivalent release force vector. The direction and magnitude are determined by the edge weight matrix in the GCN. In the local coordinate system, the stress release vector... Discrete form of the following equilibrium equations:

[0046] ;

[0047] in, For local stress tensor;

[0048] The equivalent release force is mapped by the incision location and the stripping depth.

[0049] The above formula describes the balance between the stress state and the release of external force within local soft tissue under surgical incision and dissection. GCN is used to construct the graph structure of soft tissue point cloud and output the initial stress state and boundary conditions of each node; FEA solves the equilibrium equations based on this output to obtain asymmetric stress vector data.

[0050] The pre-training of the directional stress release model was based on a large-scale, de-identified biomechanical database. This database integrated over 5000 ex vivo skin stretching experimental data (including stress-strain curves, anisotropic parameters, etc.) from publicly available biomechanical research literature, as well as intraoperative digital strain measurement data from 1200 anonymized facial surgery patients collected after approval from the ethics committees of collaborating hospitals. All clinical data were used with informed consent from patients and underwent strict anonymization to ensure the compliance and ethical nature of the data sources. The initial learning rate was 0.0001, with a weight decay of 1×10⁻⁶. −5 The AdamW optimizer was used. The batch size was set to 64, and the training cycle was 200 epochs.

[0051] The pre-trained temporal rebound prediction model receives asymmetric deformation trend data as initial input. The system analyzes the tissue elastic modulus and thickness distribution characteristics of different anatomical regions of the face by comparing and mapping the local density of the original point cloud with an anatomical template. Combined with built-in viscoelastic mechanics formulas, it calculates the soft tissue deformation attenuation and follow-up rebound within preset observation periods such as the immediate period, swelling reduction period, and stabilization period. Finally, it generates temporal evolution prediction data with time-step characteristics, solving the problem of long-term stability loss in plastic surgery prediction.

[0052] The time-rebound prediction model is used to predict the morphological evolution of tissues over a recovery period of up to one year. Its core is based on a combination of Long Short-Term Memory (LSTM) networks and a generalized Maxwell's rheological model. The time-rebound prediction model simulates tissue rebound as a multi-stage relaxation process. Its rebound displacement vector... Fitting using a multinomial Proni series:

[0053] ;

[0054] Where t is the time variable calculated from the end of the surgery;

[0055] i is the stage index, with values ​​1, 2, and 3, corresponding to the three postoperative stages: i=1: immediate postoperative period (0-48 hours); i=2: tissue swelling reduction period (3 days to 3 months); i=3: collagen remodeling stabilization period (3 months to 1 year).

[0056] Let be the weight coefficient for the i-th stage;

[0057] This is the time constant corresponding to the immediate phase, the swelling reduction phase, and the stable phase.

[0058] Let be the reference displacement vector for stage i, representing the theoretical maximum rebound displacement direction and magnitude under the assumption of no attenuation in this stage.

[0059] Parameters at each stage , , Dynamic predictions are made using a time-rebound prediction model (trained based on LSTM and clinical follow-up data), and parameters are adjusted by combining individual tissue elastic modulus and thickness distribution characteristics.

[0060] The training data for the time rebound prediction model came from a standardized follow-up database established by a multi-center clinical research project. This database contained long-term clinical follow-up data (3D scan point clouds and associated clinical assessment information at 1 day, 7 days, 1 month, 3 months, 6 months, and 1 year post-surgery) of 1200 patients undergoing facial plastic surgery, which underwent strict anonymization. The research project has been approved by the relevant ethical review committee. The hidden layer dimension of the model was set to 256 layers, and the packet loss ratio was 0.2 to prevent overfitting. Cosine annealing was used to adjust the learning rate. The model was continuously run for approximately 72 hours on a high-performance computing node (256GB RAM) until the temporal trajectory correlation coefficients on the validation set were obtained. .

[0061] The system automatically detects extreme points of normal displacement in the time evolution prediction data and identifies regions of volume expansion. Based on the scar tendency weight in individual physical parameters, operator constraints are applied to the volume increment of these regions to simulate the final tissue morphology after restricted scar proliferation. Finally, the system uses a differential decomposition algorithm to deconstruct the target postoperative deformation prediction data, projecting the complex deformation field to multiple physical feature dimensions and outputting multidimensional interpretable deformation decomposition results.

[0062] By introducing anisotropic stress correction and time-dimension-based springback compensation, this application effectively solves the prediction bias problem caused by neglecting the internal tension distribution and dynamic self-healing characteristics of soft tissue in existing technologies. By using individual physical parameters to restrictively modulate deformation prediction, the final output three-dimensional morphology more closely resembles the actual physical evolution of biological tissues. This not only improves the geometric fidelity of the prediction results but also reveals the physical essence of the deformation source, providing quantitative technical support for scientific decision-making and safety assessment of surgical procedures.

[0063] This application further proposes that the initial coupled scale field data is first constructed by acquiring the original three-dimensional point cloud data of the face through the system. The original three-dimensional point cloud data of the face comes from one of two methods: 1) medical image three-dimensional reconstruction point cloud containing bone and soft tissue information obtained by CT or CBCT scan; or 2) fused point cloud data formed by multimodal registration of epidermal point cloud obtained by external three-dimensional optical scanner and bone point cloud obtained by medical image.

[0064] Specifically, the system utilizes sampling consistency algorithms or curvature distribution-based clustering operators in the Point Cloud Library (PCL) to perform feature extraction and topological analysis on the input raw 3D point cloud data. This identifies skeletal surface feature points with stable spatial geometric characteristics (such as the orbital rim, mandibular angle, and zygomatic arch) and their corresponding soft tissue point cloud clusters. For fused point cloud data, the system directly separates skeletal and soft tissue point cloud clusters based on differences in intensity or color features from different data sources. For medical image reconstruction point clouds from a single source, the system calls a pre-trained deep learning segmentation model (such as a 3D segmentation network based on the U-Net architecture). This model is trained on tens of thousands of labeled samples (such as craniofacial CT datasets labeled by radiologists) and can automatically identify and segment skeletal surfaces with stable spatial geometric characteristics (such as the orbital rim, mandibular angle, and zygomatic arch) and their corresponding soft tissue regions.

[0065] During this process, the system establishes a spatial neighborhood retrieval mechanism, which typically uses the KD tree algorithm to search for neighboring bone reference points within a preset radius (e.g., 5mm to 15mm) for each soft tissue point cloud point, laying the geometric skeleton foundation for subsequent hierarchical association.

[0066] The system calculates the displacement transfer function of each soft tissue point cloud relative to adjacent bone surface feature points using a pre-defined mapping algorithm. This displacement transfer function essentially describes the attenuation logic of energy diffusion to the surface viscoelastic soft tissue when the bone, acting as a supporting rigid body, undergoes displacement. The system generates a displacement transfer matrix M describing the mapping relationship between bone momentum and soft tissue changes through least squares fitting or radial basis function (RBF) interpolation, serving as the initial coupled scale field data.

[0067] To further refine the displacement transfer matrix M, the system can call a pre-trained displacement sensitivity model. This model is trained on thousands of clinical medical image datasets (such as CT and MRI registration data). The training environment typically employs a high-performance computing cluster (such as a hardware environment with single-precision floating-point operation capability of at least 100 TFLOPS), with a learning rate set between 0.001 and 0.01. The model converges after at least 500 iterations using the Adam optimizer, ensuring that the matrix parameters accurately reflect the anatomical tissue connections. The mapping relationship of the displacement transfer function can be expressed as:

[0068] ;

[0069] in, Represents the target space coordinate vector of a soft tissue point cloud;

[0070] Represents the displacement vector of skeletal feature points;

[0071] The generated displacement transfer matrix;

[0072] This is the soft tissue deformation bias term calculated based on a regional anatomical statistical model (such as a Gaussian process regression model), which is trained on historical surgical data and used to correct the linear transfer matrix. Uncaptureable nonlinear anatomical constraints.

[0073] The weight coefficients contained in the displacement transfer matrix M are usually distributed in the interval [0,1]. Their values ​​are determined by the Euclidean distance between the soft tissue point and the bone surface, as well as the damping characteristics of the tissue anatomical layers (such as muscle layer, fat layer, and dermis).

[0074] By constructing the aforementioned displacement transfer matrix, this application achieves a high-dimensional mapping between facial skeletal framework changes and surface tissue deformation, transforming discrete point cloud geometric information into a coupled proportional field with anatomical logic. This processing method can quantify the contribution of skeletal support to facial morphology. The generated displacement transfer matrix not only realistically simulates the associated deformations guided by hard tissue changes, but also avoids model tearing or geometric distortion caused by traditional linear stretching algorithms through multi-dimensional weight distribution, thereby improving the geometric fidelity and anatomical consistency of the simulated morphology immediately after surgery.

[0075] like Figure 3The diagram illustrates the steps involved in generating asymmetric deformation trend data according to this application. This application further proposes a step where asymmetric stress vector data is used to spatially correct the initial coupled proportional field data to obtain asymmetric deformation trend data. In this step, the system first obtains precise surgical adjustment parameters through the surgical planning interface, and then uses a spatial coordinate analysis algorithm to extract incision location information (including the starting coordinates, curvature, and length of the incision path) and tissue dissection depth information (defining the vertical depth of the damaged soft tissue layer). These parameters determine the boundary conditions under which the originally continuous and balanced stress field of the face is physically disrupted.

[0076] The system invokes a pre-trained directional stress release model, injecting the extracted incision and peeling parameters into the mechanical simulation environment. This model, based on the fracture energy release principle of materials mechanics, calculates the tension imbalance caused by uneven tension release on both sides of the fracture surface. During this process, the model automatically matches the individual patient's skin texture direction and analyzes the angle between the incision vector and the skin texture tension direction. When the incision direction is not parallel to the skin texture direction, the system identifies a sharp increase in shear stress, thereby quantifying the tension bias values ​​of each component of the surgical site in three-dimensional space.

[0077] To transform this mechanical imbalance into geometric deformation correction, the system generates a corresponding spatial correction coefficient based on the degree of tension imbalance. This spatial correction factor It is not globally uniform, but rather exhibits a nonlinear attenuation distribution based on distance from the cut. The system reconstructs the coordinate mapping weights of the initial coupled scale field data, and adjusts the spatial correction coefficients accordingly. This is superimposed on the original displacement transfer weights. Specifically, for a specific soft tissue point i, the final corrected mapping relationship can be expressed as:

[0078] ;

[0079] in, The corrected spatial location;

[0080] These are the initial coupling weights;

[0081] This is the spatial correction factor;

[0082] This is asymmetric stress vector data;

[0083] This is the displacement vector of the skeletal feature points.

[0084] This superposition process causes the soft tissue point cloud to have an offset in line with the direction of stress release as it moves with the bones, ultimately generating asymmetric deformation trend data.

[0085] By applying spatial correction coefficients to the initial coordinate mapping weights, this scheme effectively simulates the "centripetal reduction" or "directional shift" phenomenon caused by soft tissue dissection during clinical surgery. This correction mechanism overcomes the limitation of traditional simulation techniques, where soft tissue can only move perpendicular to the bone surface, allowing the deformation trend to reflect the asymmetric characteristics constrained by skin tension. Therefore, the final deformation prediction is more biologically realistic in terms of geometric details, especially in stress-sensitive areas such as the corners of the mouth and eyelids, accurately capturing minute offset deviations and reducing directional errors in surgical outcome prediction.

[0086] This application further proposes that, after obtaining the asymmetric deformation trend data, in order to ensure the extremely high accuracy of the stress simulation results, the system initiates the dynamic self-calibration logic of the directional stress release model.

[0087] The system first retrieves skin texture data from the individual's physical parameters and, combined with a pre-stored biomechanical distribution matrix describing the physiological tension attributes of different areas of the human face, establishes a local coordinate system within the tangential plane surrounding the incision. Within this local coordinate system, the processor constructs the tension distribution matrix through tensor product operations. This matrix accurately depicts the initial prestress distribution characteristics of the skin surface and superficial fascia layer in a resting state, providing a personalized mechanical reference for model calibration.

[0088] The system utilizes this tension distribution matrix to anisotropically correct the asymmetric stress vector data generated in the preceding steps. This process simulates the constraint or guiding effect of the local tension field on tissue displacement after skin perturbation by calculating the product of the tension matrix and the initial stress vector. The calibration unit calculates the prediction residual value of the asymmetric stress vector data before and after correction by comparing the corrected vector field with the anatomical constraints of the clinical gold standard or historical prediction deviation data. The formula is as follows:

[0089] ;

[0090] Where k represents the kth sampling point, and its value range is 1≤k≤n;

[0091] n is the total number of point cloud sampling points within the local feature region;

[0092] The stress vector at each point; This represents the asymmetric stress vector initially calculated using the directional stress release model, without correction for dermal tension. This indicates the use of dermal grain direction data. The stress vector after correction of the tension distribution matrix T.

[0093] If the calculated predicted residual value Exceeding the preset convergence threshold (For example, the preset convergence threshold) The range was set between 0.05 mm and 0.2 mm. This range was obtained through statistical analysis of the deviation between postoperative 3D scan data and predicted data from 100 facial surgery patients. When the difference between the predicted and actual morphology is ≤0.2mm (the difference is indistinguishable to the human eye), the system determines that the stress release model in the current direction fails to fully reflect the biomechanical characteristics of the individual. At this point, the system automatically calculates the distribution gradient of the predicted residual value in the point cloud space. The weight coefficients (such as tangential damping coefficient and peel strain energy coefficient) of each peeling path in the directional stress release model are adjusted using the backpropagation algorithm or gradient descent method. This iterative process is achieved through the parallel computing capabilities of the graphics processor, and usually after no less than 10 fine-tuning iterations, the residual value falls back to within the convergence threshold.

[0094] By introducing a dynamic self-calibration process, this application achieves a precise transition of the prediction model from general simulation to personalized feature adaptation. This self-calibration mechanism can effectively identify and compensate for nonlinear stress disturbances caused by dermal differences, making the generated anisotropic corrections more consistent with the patient's actual tissue mechanical response. This closed-loop optimization based on prediction residual feedback not only narrows the fluctuation range of prediction bias but also ensures the reliability of asymmetric deformation trend data at the microscopic feature level, thereby providing a high-precision mechanical initial state for the subsequent time evolution prediction stage.

[0095] This application further proposes that after completing the asymmetric stress correction, the system enters the deep simulation stage in the time dimension, and simulates the long-term stability of facial morphology by constructing dynamic time-effect evolution laws.

[0096] The system first analyzes the thickness distribution characteristics and then calls upon the time-rebound prediction model based on these characteristics. This model, fundamentally based on linear viscoelasticity (LVE) theory and biological tissue rheology, divides the postoperative soft tissue recovery process into three key stages: the immediate postoperative period (immediate to 48 hours post-surgery), the swelling reduction period (3 days to 3 months post-surgery), and the stabilization period (3 months to 1 year post-surgery). For thickness differences in different anatomical regions, the model automatically matches the corresponding relaxation modulus, thereby quantitatively simulating the shape evolution of soft tissue under internal tension and external constraints within each stage.

[0097] During the simulation, the system focuses on calculating the rebound displacement vector of soft tissue due to the tissue memory effect. The tissue memory effect originates from the tendency of dermal collagen fibers and subcutaneous fascia to recover from their initial state after deformation. The system uses a decay function to simulate this recovery dynamic that decreases over time, and the calculation process involves exponential integration of the initial strain energy. Subsequently, the system executes a nonlinear spatial vector superposition command to superimpose the calculated rebound displacement vector. Nonlinear spatial vector superposition with asymmetric deformation trend data generates time evolution prediction data.

[0098] Time evolution prediction data The superposition process employs a nonlinear weighting operator to ensure that the displacement synthesis conforms to the strain constraints of biological tissue. Its core computational logic can be expressed as follows:

[0099] ;

[0100] Where t represents the postoperative time variable;

[0101] This represents the time evolution prediction data at time t;

[0102] This is the initial position before surgery;

[0103] This data represents asymmetric deformation trends.

[0104] It is an attention-based temporal fusion network or a parameterized relaxation-recovery equation integrator whose parameters are dynamically modulated by tissue thickness and elastic modulus.

[0105] During this process, the system sets the damping weight according to the thickness distribution characteristics. For example, in the cheek where the subcutaneous fat is thicker, the attenuation period of the rebound displacement vector is set in the ratio range of [0.6, 0.9], while in the forehead where the tissue is thinner, it is set between [0.3, 0.5].

[0106] By simulating the three-stage shape evolution pattern and introducing the nonlinear superposition of tissue memory effects, this application can accurately predict the contour rebound or position retraction that may occur in patients after the swelling subsides. This effectively avoids the drawback of traditional simulations that only focus on immediate shape and ignore the instability of long-term effects. This evolution prediction based on the physical time constant gives the final prediction data not only geometric accuracy but also biological predictive depth, providing an objective scientific basis for clinicians to develop personalized post-operative maintenance plans.

[0107] This application further proposes that, in the refined processing flow of generating time evolution prediction data, the system introduces an accuracy verification logic based on three-stage time series fitting, which aims to correct parameter drift in biomechanical simulation through a closed-loop feedback mechanism.

[0108] The system executes a time-series decomposition algorithm to break down the initially generated full-cycle time evolution prediction data into prediction subsets corresponding to three independent time periods: immediate elastic recoil, tissue swelling fluid dynamics evolution, and collagen remodeling stabilization. Specifically, these prediction subsets include: the "immediate elastic recoil" subset, corresponding to the release of physical tension in the very short time following surgery; the "tissue swelling fluid dynamics evolution" subset, corresponding to the lymphatic and blood circulation metabolism in the weeks following surgery; and the "collagen remodeling stabilization" subset, corresponding to the fibroblast activity-driven process in the months to one year following surgery.

[0109] The system accesses a pre-stored clinical statistical database via a high-speed data interface. This database integrates the real evolution trajectories of a large-scale sample under similar surgical and physical parameters, forming a series of standardized benchmark reference streams. The system calculates the instantaneous deviation rate between each predicted subset and the corresponding benchmark reference stream data under similar parameters in the clinical statistical database by comparing each predicted subset with the corresponding benchmark reference stream data. The instantaneous deviation rate This reflects the degree of deviation of the current mechanical model from the predictions at a specific physiological stage. Specifically, the system uses the Dynamic Time Warping (DTW) algorithm to register the predicted trajectory with the baseline trajectory and calculates the normalized distance in Euclidean space. :

[0110] ;

[0111] in, To predict the displacement vector of the subset at time t;

[0112] This is the reference displacement vector in the database.

[0113] If normalized distance Exceeding the preset tolerance threshold (e.g., preset tolerance threshold) If the swelling reduction period is set between 0.08 and 0.12, the system will determine that the model's time response characteristics need to be corrected.

[0114] when When =0, define =0 indicates that the predicted displacement and the reference displacement are both zero at the current moment, with no deviation.

[0115] The system uses the instantaneous deviation rate to determine the time constant of the time rebound prediction model. Perform reverse compensation. Time constant. The relaxation rate of tissue viscoelasticity is determined by a compensation operator, which allows the system to automatically increase or decrease the rate based on the distribution gradient of instantaneous deviations. This value is used to accelerate or slow down the deformation evolution during the simulation process. This compensation logic is usually executed in parallel in the background, and through multiple iterations of fitting, the matching degree between the corrected prediction subset and the clinical baseline flow data is improved to a preset confidence level.

[0116] Through three-stage temporal fitting and inverse compensation, this application enhances the dynamic accuracy of temporal evolution prediction, effectively overcoming the limitations of purely theoretical mechanical models in simulating complex physiological evolution in the human body (such as differences in fluid metabolism rates and collagen deposition rates). By independently decomposing and correcting different time periods, the system can adaptively match the tissue recovery characteristics of different patients, ensuring that the final output temporal evolution data is not only physically and logically consistent but also possesses extremely high empirical reliability in clinical statistics.

[0117] This application further proposes specific steps for generating target postoperative deformation prediction data by restricting local volume increments in time evolution prediction data based on scar tendency weights. The system, by introducing biological constraint logic, corrects simple physical rebound simulations to a final stable form that conforms to the characteristics of human healing.

[0118] The system first performs spatial geometric detection on the temporal evolution prediction data generated in the preceding steps, extracting the displacement changes of each point cloud node in the normal dimension. By executing a region search command, it identifies areas in the temporal evolution prediction data where the normal displacement exceeds a preset displacement threshold. (For example, the preset displacement threshold is set between 0.5mm and 1.5mm, the preset displacement threshold) Based on the normal physiological bulge range of facial soft tissue during postoperative healing (exceeding this range is judged as "abnormal volume increment" and requires scar restraint treatment), the set of point cloud clusters defines the volume expansion area as local volume increment, and identifies local bulges in the surgical area that may be caused by tissue stacking, hydraulic fluctuations or healing reactions.

[0119] The system calls upon the scar tendency weight from the individual's physical parameters. Based on this weight, it nonlinearly reduces the normal displacement of the local volume increment to simulate the final stable state of tissue after scar hyperplasia is restricted. The scar tendency weight, as a modulating operator, reflects the potential probability of excessive proliferation of fibroblasts in the patient. In the specific implementation, the system establishes a displacement reduction model to reduce the normal displacement... Weighting of scarring tendency Mapping processing is performed. When the patient has a low scarring tendency (scarring tendency weighting)... When the level is low, the system applies only slight restriction to preserve natural tissue fullness; while when the patient has a high tendency to scar, the system increases the reduction of normal displacement to simulate the tissue stability state after scar hyperplasia is physically restricted under pressure therapy or drug intervention.

[0120] This reduction logic is typically calculated using a logistic regression function or a pre-defined attenuation operator, and its formula can be expressed as:

[0121] ;

[0122] in, This represents the final stable displacement after reduction.

[0123] This is a regional sensitivity coefficient, quantified based on the results of retrospective clinical studies on the incidence of postoperative complications in that region (e.g., areas with high tension concentration, such as the nasal alae and mandibular margin, have higher coefficients). The general range is 0.5~1.0; the range for highly sensitive areas is 1.2~2.0.

[0124] The weighting for scarring tendency is 0-0.3 for low risk and 0.6-1.0 for high risk.

[0125] This represents the initial normal displacement (usually positive, indicating tissue elevation) in the time evolution prediction data without scar constraint.

[0126] Typical value examples are shown in Table 1 below:

[0127] Table 1 Typical values ​​for hypertrophic scars

[0128] Anatomical sites Regional sensitivity coefficient κ Scarring tendency weight w Scar inhibitory factor Constraint Effect Description Nasal wings (highly sensitive) 1.8 Low risk 0.2 ≈0.70 Slight restraint, preserving the natural bulge Jawline edge (highly sensitive) 1.8 High risk 0.8 ≈0.24 Strong constraints significantly inhibit hyperplasia Cheek (general) 0.8 Medium risk 0.5 ≈0.67 Medium constraints Forehead (general) 0.8 Low risk 0.2 ≈0.85 Weak constraints

[0129] By correcting the normal displacement at each incremental point, the system regenerates the constrained point cloud spatial coordinates and finally synthesizes the target postoperative deformation prediction data. This operator, through the aforementioned exponential decay logic, simulates the displacement saturation effect of the skin layer after injury due to restricted fibroblast activity. When the scar tendency weight... When the activation threshold is exceeded, the operator automatically increases the reduction force of the normal displacement to simulate the volume reduction of individuals with a high tendency to scar during the tissue remodeling stabilization period. Through this nonlinear modulation based on thickness and body weight, the system achieves secondary correction of the predicted point cloud coordinates, ensuring that the final output of the target postoperative deformation prediction data not only conforms to the physical rebound law, but also conforms to the individual biological healing logic of the patient.

[0130] By introducing scar tendency weights to limit local volume increments, this application achieves a scientific regression from "ideal evolution" to "clinical reality." This modulation mechanism can effectively predict and simulate differences in tissue bulges that may occur after healing in individuals with different constitutions, avoiding the problem of overly smooth or idealized predicted morphology caused by neglecting scar hyperplasia patterns in traditional predictions. This ensures that the generated deformation data not only has mechanical support but also biomedical predictive accuracy, resulting in a stable, clinically deliverable output.

[0131] This application further proposes a step for outputting multidimensional interpretable deformation decomposition results.

[0132] The system first executes a global spatial comparison command to calculate the total displacement field of the postoperative deformation prediction data of the target relative to the original three-dimensional point cloud data of the face. The total displacement field It contains the three-dimensional vector changes of each point cloud node from its initial state to its final stable state, and is the cumulative result of the combined effects of all biomechanical variables.

[0133] To decompose the contribution of each technical factor to the final deformation, the system uses a differential decomposition algorithm to project the total displacement field onto a matrix composed of initial coupled proportional field data (representing the bony basis), asymmetric stress vector data (representing surgical intervention), follow-up rebound (representing tissue characteristics), and scar tendency weights. In the orthogonal dimensional space (representing healing constraints), it serves as a multidimensional interpretable deformation decomposition result.

[0134] Since these biomechanical characteristics often exhibit linear correlation in their original state, the system employs Schmidt orthogonalization or principal component transformation (PCA) logic to decouple each feature vector.

[0135] In mathematical terms, the system constructs a set of basis vectors. These correspond to the four dimensions mentioned above. Using the difference operator, the system calculates the projection components of each dimension into the total displacement field. The percentage contribution of each component to the postoperative deformation prediction data is then calculated. The weighted projection formula can be defined as follows:

[0136] ;

[0137] in, represents the percentage contribution of the k-th dimension, and represents the proportion of the contribution of the k-th deformation causal dimension to the total deformation field;

[0138] This represents the projection vector of the total displacement field onto the k-th characteristic dimension;

[0139] k,j represent dimension indices, with values ​​ranging from 1 to 4, corresponding to the four deformation causal dimensions respectively;

[0140] Let represent the k-th orthogonal basis vector, and let represent the orthogonal basis vector after processing by Schmidt orthogonalization or principal component analysis (PCA), corresponding to different deformation causal dimensions;

[0141] Specifically, Extract rigid linkage deformation caused by changes in the skeletal framework. Extracting the anisotropic shift caused by the release of the incision. Extracting displacement compensation caused by viscoelastic recoil, and The amount of normal displacement reduction caused by the limited extraction volume is then calculated.

[0142] Example: Suppose the magnitudes of the total displacement field of a patient after projection are shown in Table 2 below:

[0143] Table 2. Magnitudes of the Total Displacement Field of the Patient after Projection

[0144] Dimension Projection module length (mm) Contribution percentage ck 3.5 3.5 / 10.0×100%=35% 2.5 25% 2.0 20% 2.0 20% total 10.0 100%

[0145] Table 2 can be interpreted as follows: In this patient's postoperative deformation, 35% was due to bone changes, 25% was due to stress release, 20% was due to tissue rebound, and 20% was affected by scar constraints.

[0146] Ultimately, the system outputs multidimensional interpretable deformation decomposition results in the form of radar charts or contribution ratio tables. Through this deconstruction method, this application elevates the technical level from "predicting what it will look like" to "why it will look this way." This processing logic clearly informs doctors how much of the postoperative morphological change is due to bone resection and how much is limited by the patient's constitution leading to rebound or hyperplasia. This high level of interpretability enhances the patient's trust and provides doctors with quantitative evidence for reverse optimization of surgical parameters, fundamentally avoiding errors in outcomes caused by blind experience-based judgments, and embodying the characteristics of data-driven precision medicine.

[0147] This application further proposes that after outputting multidimensional interpretable deformation decomposition results, the method also includes security verification, which prevents the simulated deformation generated by the algorithm from exceeding the physiological tolerance range of the human body or causing anatomical distortions by introducing "hard constraints" of medical anatomical structures.

[0148] The system first calls the preset anatomical constraint database, which stores a massive spatial probability distribution map of standard anatomical morphology, covering key constraint information such as the skin thinning limit, the forbidden zone for the distribution of neurovascular bundles, and the displacement threshold of muscle attachment points.

[0149] The system executes spatial matching instructions to calculate the overlap between the target postoperative deformation prediction data and the anatomical constraint database. The degree of overlap This is achieved by calculating the overlap volume or boundary distance between the deformed model point cloud and the anatomical restricted area spatial surface.

[0150] Specifically, the system uses Minkowski and / or collision detection algorithms to analyze whether the predicted soft tissue location has invaded a pre-defined dangerous anatomical area (such as the superficial layer where facial nerve branches run). Its mathematical evaluation model can be expressed as follows:

[0151] ;

[0152] in, The overlapping volume between the target postoperative deformation prediction data and the anatomically constrained no-go zone;

[0153] The total volume of all tissue points that are collided with the anatomical constraint database in the target postoperative deformation prediction data.

[0154] The closer to 1, the higher the anatomical compatibility.

[0155] The system determines the degree of overlap. Is it below the preset safety threshold? (This was obtained through statistical analysis of deformation simulation data from 100 safe surgical cases, such as the preset safety threshold.) (Set between 0.95 and 0.98).

[0156] If the calculated overlap is lower than a preset safety threshold, the system determines that adjusting the current surgical parameters may lead to risks such as local tissue tension overload, blood supply obstruction, or nerve pathway exposure, and immediately triggers feedback logic to generate a preoperative parameter adjustment instruction. This instruction includes specific risk point coordinates and suggested adjustment amounts, such as "It is recommended to reduce the mandibular angle dissection depth by 2mm" or "It is recommended to adjust the incision position to avoid nerve distribution areas."

[0157] By introducing a safety verification based on an anatomical constraint database, this application examines complex prediction results within a realistic anatomical framework, effectively mitigating potential extreme biases in biomechanical simulations. Through automatically generated parameter adjustment instructions, the system extends its functionality from "effect prediction" to "risk warning and scheme optimization," ensuring that the output surgical prediction scheme is not only visually natural and physically consistent, but also feasible and safe from an anatomical and medical perspective.

[0158] The following is a specific example of a method for predicting the effects of facial plastic surgery based on three-dimensional reconstruction:

[0159] A female patient underwent mandibular angle osteotomy due to a wide jawline. Preoperatively, a high-precision structured light scanner (0.1mm accuracy) was used to acquire raw 3D point cloud data of the face, obtaining full-face morphological information containing over 1.2 million spatial coordinate points. Surgical parameters were set as follows: 8mm width of mandibular angle outer plate resection, osteotomy line 15mm from the posterior border of the mandibular ramus, incision located in the buccal vestibule intraorally, and tissue dissection depth reaching 2mm subperiosteally. Individual skin texture direction data was obtained through gradient feature extraction of skin texture images, showing that the patient's mandibular border skin texture was nearly horizontal (angled at 15° with the long axis of the mandible). Combined with her history of chest hyperplasia and dermoscopic collagen density data, a quantified scar tendency weight was assigned. (Low-risk range is 0-0.3, high-risk range is 0.6-1.0).

[0160] The system first performs bone and soft tissue hierarchical segmentation on the original point cloud, identifying 17 skeletal surface feature points such as the mandibular angle and zygomatic arch. It then calculates the displacement transfer function of soft tissue point cloud clusters, including the masseter muscle region and buccal fat pad region, relative to adjacent skeletal feature points, generating initial coupled proportional field data. Specifically, the soft tissue displacement transfer matrix for the mandibular angle region shows that for every 1mm movement of the bone, the surface skin displacement is 0.62mm, and the masseter muscle attachment area displacement is 0.89mm.

[0161] Surgical adjustment parameters were input into a pre-trained directional stress release model, and anisotropic tension differences were calculated using skin texture data. Simulation results showed that due to the 25° angle between the incision direction and the skin texture direction, a shear stress increment of 0.12 MPa was generated in the incision area, which was converted into asymmetric stress vector data pointing backward and upward (X-axis: 0.23 mm, Y-axis: -0.18 mm, Z-axis: 0.31 mm). This vector data was used to spatially correct the initial coupled proportional field, obtaining asymmetric deformation trend data. After correction, the predicted soft tissue displacement in the mandibular angle region shifted by an average of 12% compared to the initial model, with the point cloud shift in the lateral cheek reaching 0.45 mm.

[0162] Asymmetric deformation trend data were input into the time-elasticity prediction model, and the tissue elastic modulus and thickness distribution characteristics were analyzed from the original point cloud: the patient's mandibular border skin thickness was 2.1 mm, with an elastic modulus of 12 MPa; the cheek subcutaneous fat thickness was 8.3 mm, with an elastic modulus of 4.2 MPa. Based on this, the follow-up rebound amount within a 6-month observation period after surgery was calculated. The simulation showed that the soft tissue rebound rate was 18% in the immediate postoperative period (0-48 hours), 35% in the swelling reduction period (3 days-3 months), and 8% in the stable period (3 months-6 months). The follow-up rebound vector and the asymmetric deformation trend data were superimposed nonlinearly to generate time evolution prediction data, in which the final stable displacement of the mandibular angle region shrank by 0.92 mm compared with the immediate prediction value.

[0163] Based on a scar tendency weight ω=0.75, the local volume increment in the time evolution prediction data is restricted. The system identifies volume expansion regions with normal displacement exceeding 1.2mm within a 5mm radius around the incision. A nonlinear reduction operator is used to reduce the normal displacement of this region by 40%, simulating the final stable state of tissue after restricted scar proliferation, and generating target postoperative deformation prediction data.

[0164] The target postoperative deformation prediction data were deconstructed by component analysis. The projection components of the total displacement field in four orthogonal dimensions—initial coupling proportional field, asymmetric stress vector, follow-up rebound amount, and scar tendency weight—were calculated to obtain multidimensional interpretable deformation decomposition results: 72% contribution from bone scaffold variation, 15% contribution from surgical stress release, 10% contribution from tissue follow-up rebound, and 3% contribution from scar hyperplasia restriction.

[0165] Finally, the anatomical constraint database was used for safety verification, and the overlap between the target postoperative deformation prediction data and the course area of ​​the mandibular branch of the facial nerve was calculated. (Safety threshold) This procedure complies with anatomical safety standards. The system simultaneously generates a visual prediction report, displaying an animation of facial morphological evolution from pre-operative to 6 months post-operatively, and annotating key morphological parameters at each stage, providing quantitative evidence for doctor-patient communication and surgical plan optimization.

[0166] This embodiment achieves accurate prediction throughout the entire cycle from surgical operation to long-term stability through the synergistic effect of multi-dimensional biomechanical modeling, individual physical parameter modulation, and anatomical safety verification, effectively improving the personalized design level of facial plastic surgery and the efficiency of doctor-patient communication.

[0167] Example 2:

[0168] like Figure 4 As shown, the facial plastic surgery effect prediction system based on three-dimensional reconstruction uses the aforementioned facial plastic surgery effect prediction method based on three-dimensional reconstruction, including:

[0169] The data acquisition module is used to acquire the patient's original three-dimensional point cloud data of the face, surgical adjustment parameters, and individual physical parameters; among which, the individual physical parameters include at least skin line direction data and scar tendency weights;

[0170] The coupling construction module is used to perform bone and soft tissue hierarchical segmentation based on the original 3D point cloud data of the face and to construct the initial coupling scale field data.

[0171] The spatial correction module is used to input surgical adjustment parameters into the pre-trained directional stress release model, combine skin texture data to determine the anisotropic tension difference generated by soft tissue under different dissection paths, and convert it into asymmetric stress vector data of soft tissue; the asymmetric stress vector data is used to spatially correct the initial coupled proportional field data to obtain asymmetric deformation trend data.

[0172] The time evolution module is used to input asymmetric deformation trend data into a pre-trained time rebound prediction model, combine individual physical parameters, analyze the tissue elastic modulus and thickness distribution characteristics from the original three-dimensional point cloud data of the face, calculate the follow-up rebound amount within the preset observation period, and generate time evolution prediction data.

[0173] The prediction module is used to limit the local volume increment represented by the time evolution prediction data according to the scar tendency weight, and generate the target postoperative deformation prediction data.

[0174] The interpretation and analysis module is used to deconstruct the target postoperative deformation prediction data into components and output multidimensional interpretable deformation decomposition results.

[0175] The system of this application mainly consists of a high-performance computing platform, a three-dimensional spatial acquisition array, and a clinical parameter input terminal at the physical level. The modules are coordinated at the instruction level through a high-speed data bus.

[0176] The data acquisition module integrates a high-precision structured light scanner or multi-view camera array as its hardware core, used to non-invasively acquire raw three-dimensional point cloud data of the patient's face. The module also interconnects with the electronic medical record system through a clinical parameter interface to acquire surgical adjustment parameters (such as osteotomy path and depth) and individual physical parameters (such as skin line direction and scar weight).

[0177] The coupling construction module relies on a high-performance graphics processing unit (GPU) and parallel computing units. Its hardware logic automatically separates the skeleton and soft tissue point cloud in a three-dimensional coordinate system by executing a hierarchical segmentation algorithm, and calculates the displacement transfer matrix through dedicated logic circuits to construct the initial coupling scale field, realizing hardware-accelerated mapping from discrete points to the anatomically linked skeleton.

[0178] The spatial correction module consists of a dedicated spatial correction processor. This processor receives simulation instructions from the directional stress release model and performs anisotropic tension difference calculations on the peeling path and grain direction in the tensor calculation unit. By generating an asymmetric stress vector and applying it to the coupled proportional field weight address in memory, spatial orientation correction is achieved, and finally, asymmetric deformation trend data is output.

[0179] The time evolution module utilizes a time-series simulation chip to model tissue viscoelasticity. The hardware unit maps the tissue elastic modulus by analyzing the local curvature and density of the point cloud and iteratively calculates the follow-up rebound within a preset time series step. This module synthesizes rebound vectors in real time through an instruction pipeline, generating time evolution prediction data covering the entire surgical cycle.

[0180] The prediction module is equipped with a local constraint operation unit. This unit extracts local volume increments from the prediction data, combines them with scar tendency weights stored in the cache, and applies constraints to the normal displacement of the point cloud through a nonlinear reduction operator, thereby outputting the target postoperative deformation prediction data.

[0181] This nonlinear reduction operator, by transforming scar tendency weights into nonlinear damping of spatial displacement, can accurately capture local tissue collapse or proliferation limitations caused by individual differences, and control the prediction error within a preset convergence threshold. Internally, nonlinear modulation based on biological characteristics enables the predicted postoperative deformation data to possess interpretable physiological constraints in terms of geometric morphology.

[0182] The interpretation and analysis module employs a differential decomposition processor. By executing the Schmidt orthogonalization projection instruction, it deconstructs the complex deformation field into a multidimensional contribution feature space, ultimately driving the display terminal to output multidimensional interpretable results containing the contribution rates of bony, stress, and rebound.

[0183] In a scenario predicting a mandibular angle osteotomy, the system first acquires the patient's point cloud using a 3D scanner. The coupling construction module calculates the initial relevance weight of bone reduction to the masseter muscle region based on the mandibular osteotomy line position. The spatial correction module identifies a 30° angle between the facial incision and skin lines, generating a posterior-upward stress vector to offset the initial displacement. The temporal evolution module calculates the natural tissue rebound during the 3-month postoperative swelling reduction period based on the patient's subcutaneous fat thickness, compensating for a 1.5mm displacement inward. The prediction module identifies the patient as having a high tendency to scar, automatically reducing the prediction increment around the wound by 20%. The interpretation and analysis module ultimately shows that in the postoperative contour change, bone resection contributes 70%, stress release contributes 15%, and late-stage rebound correction contributes 15%, providing the surgeon with precise quantitative data.

[0184] This application addresses the limitations of traditional prediction systems' singular and static morphological simulations through real-time analysis and algorithmic compensation of facial biomechanical features using hardware units, ensuring dual accuracy in both three-dimensional space and time. It reveals the physical causes of deformation, providing biomechanically sound and safe technical support for scientific decision-making in surgical planning.

[0185] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for predicting the effects of facial plastic surgery based on three-dimensional reconstruction, characterized in that, Includes the following steps: The original three-dimensional point cloud data of the patient's face, surgical adjustment parameters, and individual physical parameters are obtained; wherein, the individual physical parameters include at least skin line direction data and scar tendency weights. Based on the original three-dimensional point cloud data of the face, bone and soft tissue hierarchical segmentation is performed to construct initial coupled proportional field data; The surgical adjustment parameters are input into the pre-trained directional stress release model, and the anisotropic tension difference generated by the soft tissue under different dissection paths is determined by combining the skin texture direction data. This is then converted into asymmetric stress vector data of the soft tissue. The asymmetric stress vector data is used to spatially correct the initial coupling ratio field data to obtain asymmetric deformation trend data. The asymmetric deformation trend data is input into a pre-trained time rebound prediction model. Combined with the individual physical parameters, the tissue elastic modulus and thickness distribution characteristics are analyzed from the original three-dimensional point cloud data of the face. The follow-up rebound amount within the preset observation period is calculated to generate time evolution prediction data. Based on the scar tendency weight, the local volume increment represented by the time evolution prediction data is constrained to generate target postoperative deformation prediction data; wherein, the constraint is achieved through the following nonlinear reduction formula: ； In the formula, For the final stable displacement after reduction, κ represents the initial normal displacement without scar constraint in the time evolution prediction data, ω is the region sensitivity coefficient, and ω is the scar tendency weight. The target postoperative deformation prediction data is deconstructed into components, and multidimensional interpretable deformation decomposition results are output. The specific steps for outputting multidimensional interpretable deformation decomposition results include: Calculate the total displacement field of the target postoperative deformation prediction data relative to the original three-dimensional point cloud data of the face; Using a differential decomposition algorithm, the total displacement field is projected into an orthogonal dimensional space composed of the initial coupled proportional field data, the asymmetric stress vector data, the follow-up rebound amount, and the scar tendency weight. The percentage contribution of each component to the target postoperative deformation prediction data is calculated as the result of the multidimensional interpretable deformation decomposition.

2. The method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps for constructing the initial coupled scale field data include: Feature extraction is performed on the original three-dimensional point cloud data of the face to identify feature points on the bone surface and soft tissue point cloud clusters; Calculate the displacement transfer function of each soft tissue point cloud relative to the feature points on the adjacent bone surface, and generate a displacement transfer matrix describing the mapping relationship between bone momentum and soft tissue change, which serves as the initial coupled proportional field data.

3. The method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps for spatially correcting the initial coupled proportional field data using the asymmetric stress vector data to obtain asymmetric deformation trend data include: Extract the incision location information and tissue dissection depth information from the surgical adjustment parameters; The tension imbalance caused by the incision location information and the tissue dissection depth information is calculated using the directional stress release model. Based on the tension imbalance, a corresponding spatial correction coefficient is generated, and the spatial correction coefficient is superimposed on the coordinate mapping weight of the initial coupled proportional field data to obtain the asymmetric deformation trend data.

4. The method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction according to claim 3, characterized in that, After obtaining the asymmetric deformation trend data, the method further includes a dynamic self-calibration process for the directional stress release model, specifically comprising the following steps: Based on the skin grain direction data and its corresponding biomechanical distribution matrix, a tension distribution matrix in a local coordinate system is established. The anisotropic correction of the asymmetric stress vector data is performed using the tension distribution matrix, and the prediction residuals of the asymmetric stress vector data before and after the correction are calculated. If the predicted residual value exceeds the preset convergence threshold, the weight coefficients of each stripping path in the directional stress release model are adjusted according to the distribution gradient of the predicted residual value.

5. The method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps for generating time evolution prediction data include: Based on the thickness distribution characteristics, the time rebound prediction model is invoked to simulate the shape evolution of the soft tissue in the three stages of the immediate surgical period, the swelling reduction period, and the stabilization period. The rebound displacement vector generated by the soft tissue due to the tissue memory effect is calculated, and the rebound displacement vector is superimposed with the asymmetric deformation trend data in a nonlinear spatial vector to generate the time evolution prediction data.

6. The method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction according to claim 5, characterized in that, The generation of time evolution prediction data also includes accuracy verification logic based on three-stage time series fitting, the specific steps of which include: The time evolution prediction data is decomposed into prediction subsets corresponding to three independent time periods: immediate elastic recoil, tissue swelling fluid dynamics evolution, and collagen remodeling stabilization. Call the pre-stored clinical statistics database to calculate the instantaneous deviation rate between the predicted subset and the benchmark reference stream data with the same parameters in the clinical statistics database; The time constant of the time rebound prediction model is inversely compensated based on the instantaneous deviation rate.

7. The method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps for generating target postoperative deformation prediction data by limiting the local volume increment in the time evolution prediction data based on the scar tendency weight include: The volume expansion region in the time evolution prediction data where the normal displacement is greater than a preset displacement threshold is extracted as the local volume increment. Based on the scar tendency weight, the normal displacement of the local volume increment is reduced to simulate the final stable state of the tissue after scar hyperplasia is restricted, and the target postoperative deformation prediction data is generated.

8. The method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction according to claim 1, characterized in that, After outputting the multidimensional interpretable deformation decomposition results, the method also includes security verification, the specific steps of which include: Call the preset anatomical constraint database and calculate the overlap between the target postoperative deformation prediction data and the anatomical constraint database; Determine whether the overlap is lower than a preset safety threshold; If the overlap is lower than the preset safety threshold, a preoperative parameter adjustment instruction is generated.

9. A facial plastic surgery effect prediction system based on three-dimensional reconstruction, characterized in that, The method for predicting the effect of facial plastic surgery based on three-dimensional reconstruction as described in any one of claims 1 to 8 includes: The data acquisition module is used to acquire the patient's original three-dimensional point cloud data of the face, surgical adjustment parameters, and individual physical parameters; wherein, the individual physical parameters include at least skin line direction data and scar tendency weights; The coupling construction module is used to perform bone and soft tissue hierarchical segmentation based on the original three-dimensional point cloud data of the face, and to construct the initial coupling scale field data; The spatial correction module is used to input the surgical adjustment parameters into the pre-trained directional stress release model, combine the skin texture direction data to determine the anisotropic tension difference generated by the soft tissue under different dissection paths, and convert it into asymmetric stress vector data of the soft tissue; the asymmetric stress vector data is used to perform spatial orientation correction on the initial coupling ratio field data to obtain asymmetric deformation trend data. The time evolution module is used to input the asymmetric deformation trend data into a pre-trained time rebound prediction model, combine it with the individual physical parameters, analyze the tissue elastic modulus and thickness distribution characteristics from the original three-dimensional point cloud data of the face, calculate the follow-up rebound amount within a preset observation period, and generate time evolution prediction data. The prediction module is used to constrain the local volume increment represented by the time evolution prediction data according to the scar tendency weight, and generate target postoperative deformation prediction data; wherein, the constraint is achieved by the following nonlinear reduction formula: ； In the formula, For the final stable displacement after reduction, κ represents the initial normal displacement without scar constraint in the time evolution prediction data, ω is the region sensitivity coefficient, and ω is the scar tendency weight. The interpretation and analysis module is used to deconstruct the target postoperative deformation prediction data into components and output multidimensional interpretable deformation decomposition results. The specific steps for outputting multidimensional interpretable deformation decomposition results include: Calculate the total displacement field of the target postoperative deformation prediction data relative to the original three-dimensional point cloud data of the face; Using a differential decomposition algorithm, the total displacement field is projected into an orthogonal dimensional space composed of the initial coupled proportional field data, the asymmetric stress vector data, the follow-up rebound amount, and the scar tendency weight. The percentage contribution of each component to the target postoperative deformation prediction data is calculated as the result of the multidimensional interpretable deformation decomposition.

Images