Bone tumor repairing method based on three-dimensional reconstruction and bionic structure

Abstract

The invention relates to a bone tumor repairing method based on three-dimensional reconstruction and a bionic structure. The method comprises the following steps: acquiring multi-modal medical image data of a bone tumor area of a patient to obtain an initial bone tumor three-dimensional model; generating a three-dimensional dynamic bone defect model for describing a bone tumor area based on the geometric structure of the initial bone tumor three-dimensional model; based on the three-dimensional dynamic bone defect model, generating a bionic implant three-dimensional model meeting bone defects; performing space alignment on the bionic implant three-dimensional model and the three-dimensional dynamic bone defect model to obtain a bone tumor repair model; carrying out restoration stability and osseointegration effect verification on the bone tumor restoration model to obtain a verification result; the verification result is used for indicating dynamic adjustment of the bionic implant three-dimensional model, and a final bone tumor repair three-dimensional model is obtained. By adopting the method, the accuracy, reliability and individuation level of bone tumor treatment can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of medicine, and in particular relates to a bone tumor repair method based on three-dimensional reconstruction and bionic structure. Background Art

[0002] Bone tumor treatment and bone defect repair are important research directions in the medical field, which are related to the patient's quality of life and functional recovery. The key lies in achieving bone tissue regeneration and functional reconstruction through precise diagnosis and treatment methods. Currently, bone tumor treatment mostly relies on traditional imaging examinations and single modality data, combined with standardized implants for repair. However, these methods have significant limitations when dealing with complex bone defects: traditional imaging data cannot fully capture the dynamic changes in the bone tumor area, standardized implants are difficult to adapt to the patient's individualized bone defect morphology, and lack biomechanical matching with the surrounding bone tissue, resulting in insufficient stability and poor bone integration after repair. These defects directly affect the treatment effect and patient recovery.

[0003] The core challenges focus on three technical factors: the integration of multimodal imaging data, the construction of three-dimensional dynamic models, and the design of bionic implants. First, the acquisition and fusion of multimodal imaging data face the problem of data heterogeneity and difficulty in synchronizing dynamic changes, resulting in insufficient accurate characterization of the bone defect area. Secondly, the construction of a three-dimensional dynamic bone defect model needs to accurately reflect the spatial and temporal evolution of bone tissue, but existing technologies have bottlenecks in the accuracy of dynamic modeling. Thirdly, the design of bionic gradient structure implants needs to take into account both biomechanical properties and bone integration capabilities, but the current optimization of design parameters lacks a dynamic verification basis, making it difficult to achieve individualized adaptation. These technical difficulties jointly restrict the accuracy and long-term effects of bone tumor repair. Summary of the Invention

[0004] Based on this, it is necessary to provide a bone tumor repair method based on three-dimensional reconstruction and bionic structure that can improve the accuracy of bone tumor repair in response to the above technical problems.

[0005] In a first aspect, the present application provides a bone tumor repair method based on three-dimensional reconstruction and bionic structure, comprising:

[0006] Acquire multimodal medical imaging data of the patient's bone tumor area to obtain an initial three-dimensional model of the bone tumor; the multimodal medical imaging data includes CT images, MRI images, and PET-CT images;

[0007] Based on the geometric structure of the initial three-dimensional bone tumor model, a three-dimensional dynamic bone defect model describing the bone tumor area is generated;

[0008] Based on the three-dimensional dynamic bone defect model, a three-dimensional model of a bionic implant that meets the bone defect is generated;

[0009] The three-dimensional model of the bionic implant was spatially aligned with the three-dimensional dynamic bone defect model to obtain a bone tumor repair model;

[0010] The stability of the prosthesis and the bone integration effect of the bone tumor repair model are verified to obtain verification results; the verification results are used to indicate the dynamic adjustment of the bionic implant three-dimensional model to obtain the final bone tumor repair three-dimensional model.

[0011] In one embodiment, obtaining multimodal medical imaging data of a patient's bone tumor region to obtain an initial three-dimensional bone tumor model includes:

[0012] Collect multimodal imaging data of the patient's bone tumor area;

[0013] Preprocessing the multimodal image data to obtain a preprocessed image;

[0014] Obtain the grayscale features in the image and use the U-Net network model to perform region segmentation to obtain the bone tumor area;

[0015] Based on the bone tumor area, an initial three-dimensional bone tumor model was constructed.

[0016] In one embodiment, the U-Net network model includes:

[0017] The downsampling path layer is used to gradually reduce the resolution of the feature map through multiple convolution and pooling operations, while increasing the number of channels to capture more abstract semantic information, and output the first feature map;

[0018] The upsampling path layer is used to gradually enlarge the size of the feature map through transposed convolution or interpolation operations, while reducing the number of channels, fusing high-level semantic features with underlying detail features, restoring spatial accuracy, and outputting the second feature map;

[0019] The feature fusion layer is used to copy the feature maps of different layers of the downsampling path layer directly to the corresponding layers of the upsampling path layer, and splice them with the upsampling features to obtain a fused feature map;

[0020] The output layer is used to obtain the binary bone tumor region segmentation result based on the fused feature map through thresholding or conditional random field optimization, and output the bone tumor region segmentation result with the same size as the input image.

[0021] In one embodiment, generating a three-dimensional dynamic bone defect model describing the bone tumor region based on the geometric structure of the initial three-dimensional bone tumor model includes:

[0022] The stress-strain distribution model of the bone defect area was established using the biomechanical coupled finite element analysis method;

[0023] By collecting patient motion capture data in real time and iteratively solving the stress-strain distribution model, the dynamic bone defect deformation field that evolves over time is obtained;

[0024] A three-dimensional dynamic bone defect model is generated according to the dynamic bone defect deformation field.

[0025] In one embodiment, the stress tensor field in the stress-strain distribution model is obtained by the following formula:

[0026]

[0027] Among them, σ is the stress tensor field; ε is the strain tensor field; E is the elastic modulus matrix of bone tissue; and η is the viscoelastic coefficient matrix.

[0028] In one embodiment, the three-dimensional model of the bionic implant is generated using a multi-objective topology optimization algorithm, which includes an objective function and constraints;

[0029] The objective function consists of minimizing the material density distribution;

[0030] Minimize density distribution min ρ Obtained by the following formula: Among them, min ρ The objective function to be minimized is to use the material density distribution as the optimization variable, i.e., to minimize the density distribution; Integration is performed within the design domain; c(ρ)·u:σ represents the dot product of the elastic modulus function c(ρ) with the displacement field u and the stress tensor σ, representing the structural strain energy; η·ρ·ablau is the penalty term, where η is the penalty factor and ablau is the displacement gradient, which is used to control the discreteness of the density distribution ρ;

[0031] The constraints are obtained by the following formula: in, is the material usage constraint; the total density integral does not exceed the target volume V * ; 0<ρ min ≤ρ≤1 is the density distribution range; ρ min Minimum density threshold to avoid numerical singularities.

[0032] In one embodiment, the error transformation matrix in spatial alignment is obtained by the following formula:

[0033]

[0034] Where E(T) is the error transformation matrix; P i and Q i are the corresponding anatomical landmarks of the implant and the bone defect; S j and T jrespectively are the profile points of the implant and the bone defect; and a and β are weight coefficients, respectively controlling the error contribution of the point feature and the surface feature; and T is a composite transformation matrix.

[0035] In a second aspect, the application further provides a bone tumor repair system based on three-dimensional reconstruction and bionic structure, the system comprising:

[0036] An initial model generation module, configured to acquire multi-modal medical image data of a bone tumor region of a patient to obtain an initial bone tumor three-dimensional model; the multi-modal medical image data comprises CT images, MRI images and PET-CT images;

[0037] A three-dimensional dynamic model generation module, configured to generate a three-dimensional dynamic bone defect model describing the bone tumor region based on the geometric structure of the initial bone tumor three-dimensional model;

[0038] A bionic implant model generation module, configured to generate a bionic implant three-dimensional model satisfying the bone defect based on the three-dimensional dynamic bone defect model;

[0039] A bone tumor repair model generation module, configured to spatially align the bionic implant three-dimensional model and the three-dimensional dynamic bone defect model to obtain a bone tumor repair model;

[0040] A dynamic adjustment module, configured to verify the stability of the repair body and the bone integration effect of the bone tumor repair model to obtain a verification result; the verification result is used to indicate dynamic adjustment of the bionic implant three-dimensional model to obtain a final bone tumor repair three-dimensional model.

[0041] In a third aspect, the application further provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method of the first aspect when executing the computer program.

[0042] In a fourth aspect, the application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the method of the first aspect.

[0043] The above-mentioned bone tumor repair method based on three-dimensional reconstruction and bionic structure constructs an initial three-dimensional model of bone tumor through multimodal medical image fusion, comprehensively integrates anatomical structure, soft tissue characteristics and metabolic activity information, accurately defines the tumor boundary and provides a reliable basis for subsequent planning; generates a three-dimensional dynamic bone defect model based on the model geometry, simulates the tumor resection process and bone defect evolution, assists in formulating personalized surgical plans and optimizes the resection range; uses a multi-objective topology optimization algorithm to generate a three-dimensional model of a bionic implant, achieves morphological matching, mechanical compatibility and bioadaptation between the implant structure and the bone defect, and improves the fit and mechanical conduction efficiency of the prosthesis and bone tissue; ensures the precise matching of the implant and the bone defect model in anatomical position and mechanical path through spatial alignment technology, reduces surgical risks and improves planning reliability; after verification and dynamic adjustment of the stability of the prosthesis and bone integration effect, a final repair model is formed to ensure the safety and effectiveness of the repair plan, comprehensively improve the accuracy, reliability and personalization of bone tumor treatment, and promote the development of related fields towards digitalization and intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 Schematic diagram of a flow chart of a bone tumor repair method based on three-dimensional reconstruction and bionic structure in one embodiment;

[0046] Figure 2 A schematic structural diagram of a bone tumor repair system based on three-dimensional reconstruction and bionic structure in one embodiment;

[0047] Figure 3 Schematic diagram of a computer device for a bone tumor repair method based on three-dimensional reconstruction and bionic structure in one embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0049] In one embodiment, Figure 1As shown, a bone tumor repair method based on 3D reconstruction and biomimetic structure is provided. This embodiment uses the method as an example of a bone tumor repair terminal. It is understood that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0050] S101, acquiring multimodal medical imaging data of the patient's bone tumor area to obtain an initial three-dimensional model of the bone tumor; the multimodal medical imaging data includes CT images, MRI images, and PET-CT images.

[0051] For example, X-ray tomography is used to obtain high-density structural information of bones with a resolution of up to submillimeter level, clearly showing the invasion of the tumor into the bone cortex, periosteal reaction and calcification foci, and obtaining CT images. High-contrast images of soft tissues are obtained through magnetic fields and radio frequency pulses, which can distinguish the boundaries between tumors and normal muscles, blood vessels, and nerves. It is particularly suitable for showing the range of tumor infiltration in the medullary cavity and obtaining MRI images. Combining positron emission tomography (PET) with CT, active areas of tumor cells are identified through metabolic activity (such as FDG uptake), and micrometastases or skipping lesions are found to obtain PET-CT images. Medical image processing software is used to align and fuse multimodal data, and algorithms such as threshold segmentation, region growing, and surface growing are used to extract the three-dimensional geometric structure of the tumor and surrounding tissues to generate an initial three-dimensional model of the bone tumor.

[0052] S102, based on the geometric structure of the initial three-dimensional bone tumor model, generating a three-dimensional dynamic bone defect model describing the bone tumor area.

[0053] For example, the tumor geometry is analyzed from an initial 3D bone tumor model to obtain 3D geometric parameters, including volume, surface area, erosion depth, and distance from nerves and blood vessels. For example, for a long bone tumor, the medullary cavity invasion length and joint destruction area are measured. Based on a tumor growth dynamics model or a preset surgical resection range, a 3D dynamic bone defect model is generated that changes over time.

[0054] S103, based on the three-dimensional dynamic bone defect model, generating a three-dimensional model of a bionic implant that meets the bone defect.

[0055] For example, based on the stress distribution in the defect area (such as the proximal femur bears the main load), a porous truss structure is generated through a topology optimization algorithm, the pore parameters match the osteogenesis requirements, and the surface is modified with a hydroxyapatite coating; a parametric model is established using software such as Fusion 360 or ANSYS, and key dimensions such as the implant-bone contact interface fit error ≤0.1mm and the medullary cavity matching segment taper deviation ≤0.5°.

[0056] S104, spatially aligning the three-dimensional model of the bionic implant with the three-dimensional dynamic bone defect model to obtain a bone tumor repair model.

[0057] For example, the bionic implant model and the dynamic bone defect model are imported into the same three-dimensional space, and a unified reference coordinate system is established based on anatomical landmarks (such as the edge of the bone cortex and the vertex of the articular surface). Non-rigid registration technology (such as the Demons algorithm and B-spline interpolation) is used to dynamically adjust the local morphology of the implant model for the irregular surface of the bone defect area (such as the uneven bone surface caused by tumor erosion) to achieve "point-to-point" precise fitting. For example, for irregular mandibular defects, the average distance error between the implant and the bone surface can be controlled within 0.2mm after registration; in the three orthogonal planes of coronal, sagittal and transverse planes, the key structures of the implant (such as intramedullary fixation stem, articular surface prosthesis) are made to coincide with the anatomical axis of the bone defect (such as the femoral mechanical axis) through translation and rotation operations, with a deviation angle of ≤1°; virtual surgery software (such as Mimics, 3DSlicer) is used to simulate the implant insertion path, avoid neurovascular bundles (such as sciatic nerve, femoral artery), and verify the compatibility of soft tissues (such as muscles and ligaments) to ensure that there is no collision risk during surgery; physiological loads are applied to the registered model (such as the lower limb bone bearing when walking) The stress distribution at the implant-bone interface is calculated using a contact mechanics algorithm, and daily activities (walking, clenching a fist, bending over) are simulated using multi-body dynamics models (such as AnyBody and OpenSim) to verify the stability of the implant under dynamic loads. The microtexture of the bone defect surface (such as the distribution of cancellous bone pores and the direction of the Haversian canals in the cortical bone) is mapped to the implant contact interface. The mechanical locking effect is enhanced by controlling the surface roughness of 3D printing (Ra = 1.6-6.3μm). Vascular bundle channels (such as nutrient foramina with a diameter of 1-2mm) and nerve decompression grooves (such as the sciatic nerve passage area width ≥ 5mm) are preset in the implant model to avoid postoperative functional disorders caused by compression.

[0058] S105, verifying the stability and bone integration effect of the bone tumor repair model to obtain a verification result; the verification result is used to indicate the dynamic adjustment of the bionic implant three-dimensional model to obtain the final bone tumor repair three-dimensional model.

[0059] For example, through multi-physics field simulation (such as finite element mechanical analysis, multi-body dynamics simulation) and biological simulation (osteogenesis process simulation, material compatibility evaluation), the stability (stress distribution, displacement deformation, fatigue life) and bone integration effect (bone tissue growth rate, interface contact area, inflammatory response) of the implant in the bone tumor repair model are quantitatively verified. Based on the verification results, the geometric, mechanical and biological parameters of the implant (such as porosity, elastic modulus, surface coating) are dynamically adjusted through intelligent algorithms, and the optimal repair model is generated after 20-50 generations of iterations.

[0060] In the above-mentioned bone tumor repair method based on three-dimensional reconstruction and bionic structure, the technology constructs an accurate three-dimensional model of bone tumors through multimodal image fusion, dynamically simulates tumor resection to form a personalized bone defect model, designs implants based on bionics and completes spatial precise alignment, and finally realizes iterative optimization of the model through mechanical and biological verification; multi-dimensional image fusion reduces the error in tumor boundary definition, the dynamic bone defect model reduces the incidence of insufficient surgical margins, the error in the fit between bionic implants and bone defects is reduced and the efficiency of bone integration is improved, the spatial alignment technology reduces the risk of nerve damage, the closed-loop verification mechanism improves the clinical fit rate of the restoration, the overall operation time is shortened, and the incidence of postoperative complications is reduced, realizing a comprehensive upgrade of bone tumor repair from experience to digitalization and precision.

[0061] Optionally, acquiring multimodal medical imaging data of the patient's bone tumor area to obtain an initial three-dimensional bone tumor model includes the following steps:

[0062] S1001, collecting multimodal imaging data of the patient's bone tumor area.

[0063] For example, CT image data are obtained by X-ray tomography with a layer thickness of 0.5-1mm to obtain high-density bone structure images, focusing on tumor erosion of cortical bone, periosteal reaction and calcification foci; MRI image data are obtained by using T1-weighted, T2-weighted and fat-suppressed sequences with a layer thickness of 1-3mm to obtain high-contrast soft tissue images, identifying the range of tumor intramedullary infiltration and its adjacent relationship with nerves and blood vessels; PET-CT image data are obtained, and after injection of fluorodeoxyglucose (FDG), the active area of ​​the tumor is located by metabolic activity (SUV value ≥2.5 is judged to be malignant) to detect micrometastases.

[0064] S1002: Preprocess the multimodal image data to obtain a preprocessed image.

[0065] For example, median filtering and Gaussian filtering are used to remove salt and pepper noise and Gaussian noise in CT / MRI images, and grayscale normalization (such as window width and window position adjustment) is used to unify the grayscale scales of different devices; a rigid registration algorithm based on mutual information (such as Elastix software) is used to align the metabolic information of PET-CT with the anatomical structure of CT / MRI, with a spatial error of ≤1mm; the suspicious tumor area is manually or semi-automatically delineated, and the subsequent segmentation range is narrowed to obtain the preprocessed image.

[0066] S1003, obtaining grayscale features in the image, performing region segmentation using a U-Net network model, and obtaining a bone tumor region.

[0067] Exemplarily, a downsampling path layer-upsampling path layer structure is adopted, the downsampling path layer extracts features (such as tumor boundary texture and metabolic activity gradient) through convolution and pooling, and the upsampling path layer restores spatial resolution through deconvolution and jump connection, and finally outputs a probability map of the tumor area; using a well-labeled bone tumor imaging dataset (including CT / MRI / PET-CT multimodal samples), with Dice coefficient and cross entropy as loss functions, through iterative training of the Adam optimizer, automatic segmentation of the tumor area (including primary lesions, metastatic lesions, and edema zones) is achieved; morphological operations (such as opening and closing operations) are applied to remove small artifacts in the segmentation results, and the core area of ​​the tumor is extracted through threshold segmentation to obtain the bone tumor area.

[0068] S1004: construct an initial three-dimensional bone tumor model based on the bone tumor region.

[0069] For example, for each modal two-dimensional slice after segmentation, the region growing method or level set algorithm is used to extract the contours of the tumor and surrounding tissue, and a triangular mesh model is generated using the Marching Cubes algorithm; the mesh is smoothed and holes are filled using software such as Mimics and 3D Slicer, and a three-dimensional model is output in STL format with sub-millimeter accuracy; parameters such as tumor volume, surface area, and erosion depth are automatically calculated, and the distance between the tumor and key structures (such as the femoral artery and spinal cord) is marked to generate an initial three-dimensional model of the bone tumor.

[0070] Optionally, the U-Net network model includes:

[0071] The downsampling path layer is used to gradually reduce the resolution of the feature map through multiple convolution and pooling operations, while increasing the number of channels to capture more abstract semantic information, and output the first feature map.

[0072] Exemplarily, each level of downsampling includes two 3×3 convolutions (padding = 1, maintaining the feature map size), using ReLU as the activation function, followed by 2×2 maximum pooling (step size = 2), so that the feature map size is halved and the number of channels is doubled (e.g., the number of input channels is 3, the number of channels after the first level of downsampling is 64, and the number of channels after the fifth level of downsampling is 1024); features from local details (such as trabecular structure) to global semantics (such as the overall morphology of the tumor) are captured layer by layer, and feature maps with high channel counts (such as 1024 channels) can encode complex semantic information (such as areas of tumor metabolic activity).

[0073] The upsampling path layer is used to gradually enlarge the size of the feature map through transposed convolution or interpolation operations, while reducing the number of channels, fusing high semantic features with underlying detail features, restoring spatial accuracy, and outputting the second feature map.

[0074] For example, each level of upsampling first doubles the size of the feature map and halves the number of channels through transposed convolution (such as 2×2, stride = 2), and then splices it with the feature map of the corresponding level of the downsampling path (through the feature fusion layer), followed by two 3×3 convolutions to further fuse the features; combining high-semantic features (such as "tumor" category information) with underlying detail features (such as pixel-level texture of the tumor boundary) to avoid the loss of details caused by single upsampling.

[0075] The feature fusion layer is used to copy the feature maps of different layers of the downsampling path layer directly to the corresponding layers of the upsampling path layer, and splice them with the upsampling features to obtain a fused feature map.

[0076] Exemplarily, a direct connection is established between the corresponding layers of the upsampling path layer and the downsampling path layer, and the high-resolution, low-semantic features of the shallow layer in the downsampling path layer (such as the bone cortical edge in CT images) and the low-resolution, high-semantic features of the deep layer in the upsampling path layer (such as the tumor active area in PET-CT) are spliced ​​along the channel dimension (such as splicing 64 channels and 64 channels into 128 channels); solving the problem of detail information loss caused by downsampling in traditional convolutional neural networks (such as FCN), ensuring that the segmentation result contains both the semantic label of "tumor" and the precise position of the boundary (such as millimeter-level vascular wrapping details).

[0077] The output layer is used to obtain the binary bone tumor region segmentation result based on the fused feature map through thresholding or conditional random field optimization, and output the bone tumor region segmentation result with the same size as the input image.

[0078] Exemplarily, the channel number is compressed to 1 through 1x1 convolution, an output probability map with the same size as the input image is outputted, and the pixel value ranges from 0 to 1 (the value is closer to 1, the higher the probability of belonging to the tumor); thresholding (such as threshold = 0.5) or conditional random field (CRF) post-processing is adopted to convert the probability map into a binary segmentation result (tumor region is 1 and background is 0), minimize false positives (mistaking normal tissue as tumor) and false negatives (missing tumor region) of the tumor region, improve the pixel-level accuracy of segmentation, and finally output the bone tumor region segmentation result.

[0079] Optionally, based on the geometric structure of the initial bone tumor three-dimensional model, a three-dimensional dynamic bone defect model describing the bone tumor region is generated, including the following steps:

[0080] S2001, a stress-strain distribution model of the bone defect region is established by using a biomechanically coupled finite element analysis method.

[0081] Exemplarily, a finite element model is constructed, the initial bone tumor three-dimensional model is imported into ANSYS, Abaqus or other finite element analysis software, material properties and tumor tissue (varying with tumor type, such as higher for osteosarcoma and lower for myeloma) are assigned, the proximal end of the bone (such as the proximal end of the femur fixed to simulate the standing position) is fixed, and physiological load (such as 3-5 times axial pressure of body weight) is applied. Biomechanically coupled modeling considers the erosion effect of tumor growth on bone tissue, and simulates the dynamic process of bone cell apoptosis and tumor cell proliferation through the birth-death element technology: the bone element in the tumor erosion area "dies" (elastic modulus decreases to 0), and the tumor element "activates" (replaced by tumor material properties); coupled with tumor growth kinetics equation (such as Logistic model), the stress redistribution at different time points is calculated.

[0082] S2002, by real-time acquisition of patient motion capture data, the stress-strain distribution model is iteratively solved, and the dynamic bone defect deformation field evolving over time is obtained.

[0083] Illustratively, an optical motion capture system (such as Vicon) or an inertial sensor is used to record the patient's joint motion trajectory in real time (such as hip flexion and extension angles and knee joint load changes during walking), and the motion data is converted into time-varying load boundary conditions (such as femoral stress that varies with the gait cycle: the load increases to a peak value during the 0-0.5 second load-bearing period and decreases during the 0.5-1 second swing period). The finite element model is imported for transient analysis, and the incremental method or Newton-Raphson method is used to iteratively calculate the stress-strain distribution according to the time step (such as Δt = 0.01 second) to capture the dynamic deformation of the bone defect area (such as cortical microfractures and cancellous bone compression deformation caused by tumor erosion). The stress cloud map and displacement vector field at each time point are superimposed on the initial three-dimensional model through post-processing software (such as ParaView) to render the dynamic changes of the bone defect area (such as bone expansility destruction caused by tumor growth and bone defect collapse caused by load bearing).

[0084] S2003: Generate a three-dimensional dynamic bone defect model based on the dynamic bone defect deformation field.

[0085] For example, a three-dimensional dynamic defect model is generated according to clinical needs (e.g., 1 week before surgery, 1 month after surgery, 1 year after surgery), each model contains geometric deformation data and mechanical parameters, where the geometric deformation data includes the change in bone defect volume (e.g., an increase of 0.5 cm per week). 3 ) and changes in cortical bone thickness (such as a decrease of 0.2 mm per month); mechanical parameters include the migration of the position of maximum principal stress (such as from cortical bone to cancellous bone) and the distribution of strain energy density (indicating potential fracture risk areas).

[0086] Alternatively, the stress tensor field in the stress-strain distribution model is obtained by the following formula:

[0087]

[0088] Among them, σ is the stress tensor field; ε is the strain tensor field; E is the elastic modulus matrix of bone tissue; and η is the viscoelastic coefficient matrix.

[0089] In the above stress tensor field formula, σ is the stress tensor field, which is used to describe the stress state of each point within the bone tissue (including normal stress and shear stress components); ε is the strain tensor field, which represents the degree of deformation of the bone tissue under load; E is the bone tissue elastic modulus matrix, which reflects the viscoelastic properties of the bone tissue (such as the difference in elastic modulus between cortical bone and cancellous bone); η is the viscoelastic coefficient matrix, which represents the viscoelastic behavior of the bone tissue (such as creep and stress relaxation properties). This formula describes the stress-strain relationship of bone tissue under dynamic load. For example, during rapid jumping, the viscoelastic term can reduce the peak stress of the cortical bone by 10%-15% (because energy is dissipated by viscoelasticity); during slow walking, the viscoelastic term dominates stress relaxation, preventing local stress from being too high.

[0090] Optionally, the three-dimensional model of the bionic implant is generated using a multi-objective topology optimization algorithm, the multi-objective topology optimization algorithm including an objective function and constraints;

[0091] The objective function consists of minimizing the material density distribution;

[0092] Minimize density distribution min ρ Obtained by the following formula: Among them, min ρ The objective function to be minimized is to use the material density distribution as the optimization variable, i.e., to minimize the density distribution; Integration is performed within the design domain; c(ρ)·u:σ represents the dot product of the elastic modulus function c(ρ) with the displacement field u and the stress tensor σ, which represents the structural strain energy; η·ρ·ablau is the penalty term, where η is the penalty factor and ablau is the displacement gradient, which is used to control the discreteness of the density distribution ρ.

[0093] In the above formula for minimizing material density distribution, by optimizing the material density distribution, the material usage is minimized while satisfying the mechanical properties, thereby generating a lightweight, high-strength bionic structure. c(ρ)·u:σ is the structural strain energy. Minimizing the strain energy is equivalent to maximizing the structural stiffness, which can ensure that the implant deforms minimally under load. η·ρ·ablau is a penalty term. By penalizing the density difference between adjacent units, the density distribution is promoted to be discretized (i.e., a clear distinction is made between "material" and "no material" areas), avoiding the appearance of "gray areas" with intermediate density, which facilitates 3D printing.

[0094] The constraints are obtained by the following formula: in, is the material usage constraint; the total density integral does not exceed the target volume V * ; 0<ρ min ≤ρ≤1 is the density distribution range; ρ min Minimum density threshold to avoid numerical singularities.

[0095] In the above constraint formula, is the material usage constraint, which constrains the target volume fraction to avoid structural failure caused by excessive lightweighting; 0<ρ min ≤ρ≤1 density distribution range, ρ min The minimum density threshold is set to avoid numerical singularities, preventing the stiffness matrix singularity caused by zero-density elements in the numerical calculation, while retaining tiny support structures (such as bridge structures connecting porous areas).

[0096] Optionally, the error transformation matrix in spatial alignment is obtained by the following formula:

[0097]

[0098] Where E(T) is the error transformation matrix; P i and Q i are the corresponding anatomical landmarks of the implant and the bone defect; S j and T j are the contour points of the implant and the bone defect, respectively; α and β are weight coefficients, which control the error contribution of point features and surface features, respectively; T is the composite transformation matrix.

[0099] In the above error transformation matrix, the spatial alignment error between the bionic implant and the bone defect model is quantified. The error is minimized by optimizing the transformation matrix (including translation, rotation, and scaling), achieving dual precise matching of "anatomical landmarks + surface contours";

[0100] The above-mentioned bone tumor repair method based on three-dimensional reconstruction and bionic structure constructs an initial three-dimensional model of bone tumor through multimodal medical image fusion, comprehensively integrates anatomical structure, soft tissue characteristics and metabolic activity information, accurately defines the tumor boundary and provides an intuitive three-dimensional reference for surgical planning; generates a dynamic bone defect model based on the geometric characteristics of the model, simulates the impact of different resection ranges and tumor progression on bone structure, assists in formulating personalized resection plans and predicts the postoperative bone mechanical state; uses a topological optimization algorithm to generate a bionic implant, so that its geometric shape fits seamlessly with the bone defect and its mechanical conduction characteristics match the host bone, and at the same time ensures the precise positioning of the implant in three-dimensional space through spatial alignment technology; after verification of mechanical stability and bone integration effect and iterative optimization of the model, it significantly improves the biocompatibility and mechanical reliability of the prosthesis and bone tissue, shortens the operation cycle and reduces the risk of complications, and realizes digital drive of the entire process from precise tumor resection to functional reconstruction, providing an efficient, safe and personalized solution for bone tumor treatment.

[0101] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0102] Based on the same inventive concept, the embodiments of the present application also provide a three-dimensional reconstruction and bionic structure based bone tumor repair system for implementing the three-dimensional reconstruction and bionic structure based bone tumor repair method described above. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme described in the method, and therefore the specific limitations in one or more three-dimensional reconstruction and bionic structure based bone tumor repair system embodiments provided below can refer to the limitations of the three-dimensional reconstruction and bionic structure based bone tumor repair method described above, which will not be repeated here.

[0103] In one exemplary embodiment, as shown in Figure 2 A three-dimensional reconstruction and bionic structure based bone tumor repair system 200 is provided, comprising:

[0104] An initial model generation module 201 is configured to acquire multi-modal medical image data of a patient's bone tumor area to obtain an initial bone tumor three-dimensional model; the multi-modal medical image data includes CT images, MRI images, and PET-CT images;

[0105] A three-dimensional dynamic model generation module 202 is configured to generate a three-dimensional dynamic bone defect model describing the bone tumor area based on the geometric structure of the initial bone tumor three-dimensional model;

[0106] A bionic implant model generation module 203 is configured to generate a bionic implant three-dimensional model satisfying the bone defect based on the three-dimensional dynamic bone defect model;

[0107] A bone tumor repair model generation module 204 is configured to spatially align the bionic implant three-dimensional model with the three-dimensional dynamic bone defect model to obtain a bone tumor repair model;

[0108] A dynamic adjustment module 205 is configured to verify the stability of the repair body and the effect of bone integration of the bone tumor repair model to obtain a verification result; the verification result is used to indicate dynamic adjustment of the bionic implant three-dimensional model to obtain a final bone tumor repair three-dimensional model.

[0109] Further, the initial model generation module 201 is further configured to:

[0110] Acquire multi-modal image data of the patient's bone tumor area;

[0111] Preprocess the multi-modal image data to obtain preprocessed images;

[0112] Obtain the gray scale features in the images, use a U-Net network model to perform region segmentation, and obtain the bone tumor area;

[0113] Based on the bone tumor area, an initial bone tumor three-dimensional model is constructed.

[0114] Furthermore, the U-Net network model includes:

[0115] The downsampling path layer is used to gradually reduce the resolution of the feature map through multiple convolution and pooling operations, while increasing the number of channels to capture more abstract semantic information, and output the first feature map;

[0116] The upsampling path layer is used to gradually enlarge the size of the feature map through transposed convolution or interpolation operations, while reducing the number of channels, fusing high-level semantic features with underlying detail features, restoring spatial accuracy, and outputting the second feature map;

[0117] The feature fusion layer is used to copy the feature maps of different layers of the downsampling path layer directly to the corresponding layers of the upsampling path layer, and splice them with the upsampling features to obtain a fused feature map;

[0118] The output layer is used to obtain the binary bone tumor region segmentation result based on the fused feature map through thresholding or conditional random field optimization, and output the bone tumor region segmentation result with the same size as the input image.

[0119] Furthermore, the three-dimensional dynamic model generating module 202 is further configured to:

[0120] The stress-strain distribution model of the bone defect area was established using the biomechanical coupled finite element analysis method;

[0121] By collecting patient motion capture data in real time and iteratively solving the stress-strain distribution model, the dynamic bone defect deformation field that evolves over time is obtained;

[0122] A three-dimensional dynamic bone defect model is generated according to the dynamic bone defect deformation field.

[0123] Furthermore, the stress tensor field in the stress-strain distribution model is obtained by the following formula:

[0124]

[0125] Among them, σ is the stress tensor field; ε is the strain tensor field; E is the elastic modulus matrix of bone tissue; and η is the viscoelastic coefficient matrix.

[0126] Furthermore, the three-dimensional model of the bionic implant is generated using a multi-objective topology optimization algorithm, which includes an objective function and constraints;

[0127] The objective function consists of minimizing the material density distribution;

[0128] Minimize density distribution min ρ Obtained by the following formula: Among them, min ρThe objective function to be minimized is to use the material density distribution as the optimization variable, i.e., to minimize the density distribution; Integration is performed within the design domain; c(ρ)·u:σ represents the dot product of the elastic modulus function c(ρ) with the displacement field u and the stress tensor σ, representing the structural strain energy; η·ρ·ablau is the penalty term, where η is the penalty factor and ablau is the displacement gradient, which is used to control the discreteness of the density distribution ρ;

[0129] The constraints are obtained by the following formula: in, is the material usage constraint; the total density integral does not exceed the target volume V * ; 0<ρ min ≤ρ≤1 is the density distribution range; ρ min Minimum density threshold to avoid numerical singularities.

[0130] Furthermore, the error transformation matrix in spatial alignment is obtained by the following formula:

[0131]

[0132] Where E(T) is the error transformation matrix; P i and Q i are the corresponding anatomical landmarks of the implant and the bone defect; S j and T j are the contour points of the implant and the bone defect, respectively; α and β are weight coefficients, which control the error contribution of point features and surface features, respectively; T is the composite transformation matrix.

[0133] In one embodiment, Figure 3 A computer device 300 is provided, comprising:

[0134] at least one processor 301;

[0135] and a memory 302 communicatively connected to at least one of the processors 301;

[0136] The memory stores application code that can be executed by at least one of the processors, and the application code is executed by at least one of the processors so that at least one of the processors can perform the steps of the bone tumor repair method based on three-dimensional reconstruction and bionic structure as described above.

[0137] The computer device may further include a sensor 303 .

[0138] The processor 301, the memory 302 and the sensor 303 can be connected via a bus 304 or other means. In the figure, the bus 304 is used as an example. Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0139] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0140] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0141] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A bone tumor repair method based on three-dimensional reconstruction and bionic structure, characterized in that: The method comprises: Acquiring multimodal medical imaging data of the patient's bone tumor area to obtain an initial three-dimensional model of the bone tumor; the multimodal medical imaging data includes CT images, MRI images, and PET-CT images; generating a three-dimensional dynamic bone defect model describing the bone tumor area based on the geometric structure of the initial three-dimensional bone tumor model; Based on the three-dimensional dynamic bone defect model, generating a three-dimensional model of a bionic implant that meets the bone defect; spatially aligning the three-dimensional model of the bionic implant with the three-dimensional dynamic bone defect model to obtain a bone tumor repair model; The bone tumor repair model is verified for prosthesis stability and bone integration effect to obtain a verification result; the verification result is used to indicate the dynamic adjustment of the bionic implant three-dimensional model to obtain the final bone tumor repair three-dimensional model.

2. The method according to claim 1, characterized in that The method of acquiring multimodal medical imaging data of the patient's bone tumor region to obtain an initial three-dimensional bone tumor model includes: Collect multimodal imaging data of the patient's bone tumor area; Preprocessing the multimodal image data to obtain a preprocessed image; Obtaining grayscale features in the image, performing region segmentation using a U-Net network model, and obtaining a bone tumor region; Based on the bone tumor region, an initial three-dimensional bone tumor model is constructed.

3. The method according to claim 2, characterized in that The U-Net network model includes: The downsampling path layer is used to gradually reduce the resolution of the feature map through multiple convolution and pooling operations, while increasing the number of channels to capture more abstract semantic information, and output the first feature map; The upsampling path layer is used to gradually enlarge the size of the feature map through transposed convolution or interpolation operations, while reducing the number of channels, fusing high-level semantic features with underlying detail features, restoring spatial accuracy, and outputting the second feature map; The feature fusion layer is used to copy the feature maps of different layers of the downsampling path layer directly to the corresponding layers of the upsampling path layer, and splice them with the upsampling features to obtain a fused feature map; The output layer is used to obtain a binary bone tumor region segmentation result based on the fused feature map through thresholding or conditional random field optimization, and output the bone tumor region segmentation result with the same size as the input image.

4. The method according to claim 1, wherein Generating a three-dimensional dynamic bone defect model describing the bone tumor area based on the geometric structure of the initial three-dimensional bone tumor model includes: The stress-strain distribution model of the bone defect area was established using the biomechanical coupled finite element analysis method; By collecting patient motion capture data in real time, the stress-strain distribution model is iteratively solved to obtain a dynamic bone defect deformation field that evolves over time; The three-dimensional dynamic bone defect model is generated according to the dynamic bone defect deformation field.

5. The method according to claim 4, characterized in that The stress tensor field σ in the stress-strain distribution model is obtained by the following formula: Among them, σ is the stress tensor field; ε is the strain tensor field; E is the elastic modulus matrix of bone tissue; and η is the viscoelastic coefficient matrix.

6. The method according to claim 1, characterized in that The bionic implant three-dimensional model is generated by a multi-objective topology optimization algorithm, wherein the multi-objective topology optimization algorithm includes an objective function and constraint conditions; The objective function includes minimizing material density distribution; The minimized density distribution min ρ Obtained by the following formula: Among them, min ρ The objective function to be minimized is to use the material density distribution as the optimization variable, i.e., to minimize the density distribution; Integration is performed within the design domain; c(ρ)·u:σ represents the dot product of the elastic modulus function c(ρ) with the displacement field u and the stress tensor σ, representing the structural strain energy; η·ρ·ablau is the penalty term, where η is the penalty factor and ablau is the displacement gradient, which is used to control the discreteness of the density distribution ρ; The constraints are obtained by the following formula: in, is the material usage constraint; the total density integral does not exceed the target volume V * ; 0<ρ min ≤ρ≤1 is the density distribution range; ρ min Minimum density threshold to avoid numerical singularities.

7. The method according to claim 1, characterized in that The error transformation matrix in the spatial alignment is obtained by the following formula: Where E(T) is the error transformation matrix; P i and Q i are the corresponding anatomical landmarks of the implant and the bone defect; S j and T j are the contour points of the implant and the bone defect, respectively; α and β are weight coefficients, which control the error contribution of point features and surface features, respectively; T is the composite transformation matrix.

8. A bone tumor repair system based on three-dimensional reconstruction and bionic structure, characterized in that: The system comprises: An initial model generation module is used to obtain multimodal medical imaging data of the patient's bone tumor area to obtain an initial three-dimensional bone tumor model; the multimodal medical imaging data includes CT images, MRI images, and PET-CT images; A three-dimensional dynamic model generation module, configured to generate a three-dimensional dynamic bone defect model describing the bone tumor area based on the geometric structure of the initial three-dimensional bone tumor model; A bionic implant model generation module, configured to generate a three-dimensional bionic implant model that satisfies the bone defect based on the three-dimensional dynamic bone defect model; a bone tumor repair model generation module, configured to spatially align the three-dimensional model of the bionic implant with the three-dimensional dynamic bone defect model to obtain a bone tumor repair model; The dynamic adjustment module is used to verify the stability of the restoration and the bone integration effect of the bone tumor repair model to obtain a verification result; the verification result is used to indicate the dynamic adjustment of the bionic implant three-dimensional model to obtain the final bone tumor repair three-dimensional model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images