Robot-assisted fracture reduction path planning method and system

Abstract

The invention relates to the field of biomechanical modeling, in particular to a robot-assisted fracture reduction path planning method and system, and the method comprises the steps: predicting the displacement field and maximum principal strain of peripelvic muscles through employing a physical information neural network; a strain threshold constraint and a strain sensing cost function are embedded into an RRT * path planning process, and soft tissue injury risks are actively avoided in a tree extension and path optimization stage; according to the robot-assisted fracture reduction path planning method and system provided by the embodiment of the invention, a physical information neural network (Muscle-PINNs) is adopted to quickly predict the displacement field and the maximum principal strain of the peripelvic muscle, and a strain threshold constraint and a strain sensing cost function are embedded into an RRT * path planning process; the soft tissue injury risk is actively avoided in the tree expansion and path optimization stage, and meanwhile millisecond-level reasoning is achieved to meet the requirement for real-time performance in the operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of biomechanical modeling, and in particular, to a robot-assisted fracture reduction path planning method and system. BACKGROUND

[0002] In the pelvic fracture reduction surgery, the mechanical response of the surrounding pelvic muscles and other soft tissues is a key factor to ensure the accuracy of reduction and the safety of the tissues. The reaction force and strain generated by excessive traction not only affect the accuracy of reduction, but also may cause muscle injury, so the mechanical effect of soft tissues during reduction needs to be systematically evaluated. However, the existing methods focus on bone structures and oversimplify muscles and ligaments, and although finite element-based simulation can accurately reproduce muscle response, the modeling and single-scenario calculation are time-consuming and difficult to meet the real-time requirements of reduction path planning.

[0003] Lei et al. reconstructed a pelvic-muscle finite element model based on patient CT using finite element analysis software, and planned a spatial reduction path from the initial to the aligned state using a three-dimensional A* algorithm. The model was fixed on the healthy side and loaded on the affected side according to the path, and boundary / contact conditions such as muscle-bone bonding and muscle-muscle non-frictional contact were set for dynamic simulation to simulate the stress, strain and displacement distribution of the main muscle groups, and then identify the high-risk muscle groups during reduction and provide basis for robot parameters and safety thresholds.

[0004] Pan et al. used Opensim software to simulate the stretching of the pelvic muscle group, combined with an enhanced A* path planning algorithm. The muscle force was calculated at this node through the OpenSim muscle-bone model, and the muscle force was integrated as a heuristic influence factor to make the path meet the collision avoidance and minimum muscle resistance, and obtain the lowest muscle resistance throughout the path under the condition of limited path cost increase, thereby reducing the traction load and improving the safety during the operation.

[0005] However, the existing methods have the following shortcomings: a) A* algorithm has low planning efficiency because it needs to traverse three-dimensional space points; b) The pelvic-muscle simulation modeling based on finite elements and the muscle-bone dynamics calculation based on OpenSim are costly, and the finite element solution of a single displacement condition takes about 2-10 minutes, which is difficult to meet the real-time linkage planning requirements in pelvic fracture reduction surgery; C) The finite element muscle modeling process is heavy and strongly dependent on the grid, while the Opensim muscle modeling uses Hill system muscle tendon unit system, which is not consistent with muscle biomechanics. SUMMARY

[0006] The embodiment of the application provides a robot-assisted fracture reduction path planning method and system, which significantly reduces strain prediction deviation and improves efficiency while keeping displacement error controllable.

[0007] According to an embodiment of the application, a robot-assisted fracture reduction path planning method is provided, comprising the following steps: The physical information neural network is used to predict the displacement field and maximum principal strain of the surrounding pelvic muscles. The strain threshold constraint and the strain-aware cost function are embedded in the RRT * The path planning process actively avoids the risk of soft tissue injury in the tree expansion and path optimization stages.

[0008] Further, before the physical information neural network is used to predict the displacement field and maximum principal strain of the surrounding pelvic muscles, the method further comprises: A dual-branch space-load decoupled physical information neural network is constructed.

[0009] Further, the construction of the dual-branch space-load decoupled physical information neural network comprises: The geometric topology and load conditions are respectively encoded and fused in the feature layer, and the data, PDE, boundary and strain loss are trained.

[0010] Further, before the dual-branch space-load decoupled physical information neural network is constructed, the method further comprises: Based on the CT reconstructed pelvic musculoskeletal tissue of the pelvic fracture patient, the muscle and bone are separated according to the different gray values to generate a 3D model using segmentation software.

[0011] Further, the physical information neural network comprises a dual-branch fully connected network and a physical calculator. The dual-branch fully connected network is composed of a coordinate branch, a load branch and a merging subnetwork: the coordinate branch receives spatial coordinates, and also has two hidden layers; after the two-branch features are fused, they are input into the merging subnetwork, and a normalized displacement vector is output; The physical calculator is configured to: for each batch sample, the physical gradient is calculated by PyTorch automatic differentiation and component-by-component calculation u / X, and the physical gradient is corrected according to the normalized scale; based on the hyperelastic strain energy, the equivalent second invariant I1, I2 correction term is calculated, and the polar part σ iso is constructed, and the volume compression part σ vol is added; and the Green strain tensor S is converted into a 6-dimensional vector for strain fitting loss.

[0012] Further, the strain threshold constraint and the strain-aware cost function are embedded in the RRT * The path planning process comprises: Predicting embeddings in RRT * For each candidate pose (R,T), the displacement of each muscle attachment point is calculated. The maximum principal stress E_max is obtained through the model. High-risk branches are screened out with a threshold, and the final path is optimized with path length + strain cost.

[0013] Furthermore, the prediction will be embedded in RRT * For each candidate pose (R,T), the displacement of each muscle attachment point is calculated. The maximum principal stress E_max is obtained through the model. High-risk branches are filtered out using a threshold, and the final path is optimized by combining path length and strain cost. In RRT * Within the algorithm's framework, the path planning process begins at the starting position and gradually expands towards the target position. The algorithm randomly samples a point in the search space and finds the node closest to that point in the current path tree. By calculating the connection paths between neighboring nodes in the path tree and the sampled point, a new path node is generated. Collision detection checks whether the newly generated path node collides with an obstacle. If a collision occurs, the node is discarded.

[0014] Furthermore, during the path tree expansion process, for each newly generated path node, the Muscle-PINNs model is invoked to predict muscle strain; the displacement of the adhesion points between the muscle and the affected bone is calculated through rigid body transformation.

[0015] Furthermore, the global maximum principal strain is statistically analyzed, the maximum value is recorded, and the result is returned to the RRT. * Path planning module; Muscle strain limit check determines whether the maximum strain of the muscle at the path node exceeds a set threshold. The threshold is the maximum principal strain at the target point. If the strain of the node exceeds the threshold, the path node is discarded.

[0016] According to another embodiment of the present invention, a robot-assisted fracture reduction path planning system is provided, comprising: The prediction unit is used to predict the displacement field and maximum principal strain of the peripelvic muscles using a physical information neural network. Path planning unit, used to embed strain threshold constraints and strain-aware cost function into RRT * The path planning process proactively avoids the risk of soft tissue damage during the tree expansion and path optimization stages.

[0017] A storage medium storing program files capable of implementing any of the above-described robot-assisted fracture reduction path planning methods.

[0018] A processor for running a program, wherein the program executes any of the above-mentioned robot-assisted fracture reduction path planning methods during runtime.

[0019] The robot-assisted fracture reduction path planning method and system in this embodiment of the invention employs Muscle-PINNs to rapidly predict the displacement field and maximum principal strain of the peripelvic muscles, and embeds strain threshold constraints and strain-aware cost functions into the RRT (Reduction and Reduction Therapy). * During the path planning process, the risk of soft tissue damage is actively avoided in the tree expansion and path optimization stages, while achieving millisecond-level inference to meet the real-time requirements during surgery. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart illustrating a robot-assisted fracture reduction path planning method; Figure 2 This is a flowchart illustrating the preoperative planning process for robot-assisted pelvic fracture reduction according to the present invention. Figure 3 This is a comparison diagram of the simulation results of the gluteus maximus and iliacus muscles in this invention; Figure 4 This is a comparison chart of the path planning results based on the PINNs muscle simulation model of this invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] Example 1 According to an embodiment of the present invention, a robot-assisted fracture reduction path planning method is provided, comprising the following steps: A physical information neural network was used to predict the displacement field and maximum principal strain of the peripelvic muscles. Embedding strain threshold constraints and strain-sensing cost functions into RRT * The path planning process proactively avoids the risk of soft tissue damage during the tree expansion and path optimization stages.

[0024] The robot-assisted fracture reduction path planning method in this embodiment of the invention uses Muscle-PINNs to quickly predict the displacement field and maximum principal strain of the peripelvic muscles, and embeds strain threshold constraints and strain-aware cost functions into the RRT. * During the path planning process, the risk of soft tissue damage is actively avoided in the tree expansion and path optimization stages, while achieving millisecond-level inference to meet the real-time requirements during surgery.

[0025] The robot-assisted fracture reduction path planning method of the present invention will be described in detail below with specific embodiments: To overcome the shortcomings of existing technologies, this invention proposes an integrated soft tissue biomechanical prediction-path planning method for pelvic repositioning: It employs Muscle-PINNs to rapidly predict the displacement field and maximum principal strain of the peripelvic muscles, and embeds strain threshold constraints and a strain-aware cost function into the RRT (Responsive Tracing and Path Planning) mechanism. * During the path planning process, the risk of soft tissue damage is actively avoided in the tree expansion and path optimization stages, while achieving millisecond-level inference to meet the real-time requirements during surgery.

[0026] This invention proposes a real-time muscle strain-constrained path planning algorithm based on a neural network model of muscle physical information for real-time planning of reduction paths in pelvic fracture reduction surgery. It uses physical priors and grid-independent reasoning of data to provide displacement / strain fields in milliseconds, and directly embeds strain thresholds into the search and optimization, taking into account both efficiency and soft tissue safety boundaries.

[0027] The basic content of the technical solution of this invention is as follows: This invention proposes a Muscle-PINNs+RRT method for closed reduction of pelvic fractures. * A strain-constrained path planning method is proposed. First, a dual-branch, space-load decoupled physical information neural network is constructed: geometric topology and load conditions are encoded separately and fused at the feature layer. The network is trained using data, PDE, boundary, and strain losses to predict the muscle displacement field and maximum principal strain independently of the grid. Second, the predictions are embedded into the RRT (Restricted Path Therapy). *The displacement of each muscle attachment point is calculated for each candidate pose (R,T). The maximum principal stress E_max is obtained through the model. High-risk branches are screened out using a threshold, and the final path is optimized by "path length + strain cost" to achieve near real-time intraoperative planning. This scheme significantly reduces strain prediction bias and significantly improves efficiency while keeping displacement error controllable.

[0028] Please see Figure 1 This is a flowchart illustrating a robot-assisted fracture reduction path planning method; the technical solution of this invention is described in detail below: (1) Biomechanical modeling of pelvic muscle tissue First, see Figure 1 This invention reconstructs pelvic muscle and skeletal tissue based on CT scans of patients with pelvic fractures. Segmentation software is used to separate muscles and bones according to different grayscale values ​​to generate a 3D model. The invention employs a second-order polynomial hyperelastic model, where the strain energy density function W consists of a deviation component and a volumetric component.

[0029] , The deviation component Wdel is represented by the simplified invariants I1 and I2 of the right Cauchy-Green deformation tensor C.

[0030] , Where C 10 C 01 Describing stiffness under low strain, C 20 C 11 C 02 It is a parameter that captures nonlinear behavior and complex deformations.

[0031] Volume term W vol Consider near-incompressibility.

[0032] , Where D1 and D2 are incompressibility parameters, Let F be the determinant of the deformation gradient.

[0033] Biomechanical modeling of muscle tissue is based on the continuum mechanics framework, assuming quasi-static conditions and no body loads, where the system satisfies the momentum balance equations. Starting with deformation mapping, partial differential equations (PDEs) are derived step-by-step. Deformation mapping is defined. :X x = X + u(X), where X is the reference configuration coordinates, x is the current configuration coordinates, and u is the displacement vector. The deformation gradient tensor F is obtained by partial differentiation of x.

[0034] , in X represents the reference configuration gradient operator. The right Cauchy-Green tensor C is C=F. T F, calculate the invariants I1 and I2.

[0035] , For hyperelastic materials, the strain energy density W depends on the modified invariants I1, I2, and J, and the second Piola-Kirchhoff stress tensor S is derived from the strain energy density W.

[0036] , S can be further converted into the Cauchy stress tensor σ=J-1FSFT

[20] , and under quasi-static conditions, the momentum balance equation is always zero.

[0037] , (2) Muscle biomechanical physical information neural network model This invention proposes a two-branch physical information neural network (Muscle-PINNs). It mainly comprises a two-branch fully connected network and a physical calculator.

[0038] Two-branch fully connected network: The network consists of a coordinate branch, a load branch, and a merged subnet. The coordinate branch receives spatial coordinates (Lx, Ly, Lz) and also has two hidden layers. The features from the two branches are fused and input into the merged subnet, outputting a normalized displacement vector u=[ux,uy,uz]. T .

[0039] , Here, f represents a multilayer perceptron (MLP), with each layer undergoing a linear transformation followed by a Tanh activation function. This dual-branch design allows the model to capture geometric and loading dependencies separately, improving its adaptability to various operating conditions. Material parameters are fixed and embedded in the network to ensure physical consistency.

[0040] Physics calculator: For each batch of samples, automatic differentiation and component-by-component calculations are performed using PyTorch. u / X is obtained by correcting for normalized scaling and obtaining the physical gradient; based on hyperelastic strain energy, the equivalent second-order invariants I1 and I2 correction terms are calculated, and the biased part σ is constructed. iso And add volume compression part σ vol The Green strain tensor S is then converted into a 6-dimensional vector for strain fitting loss.

[0041] The total loss in Muscle-PINNs includes data fitting, boundary conditions, balance constraints, and strain loss. It dynamically balances these components through adaptive weighting, combining inverse and proportional weighting with EMA smoothing, weight range constraints, warmup, and stagnation fallback strategies to dynamically adjust the relative contribution of each component's loss. The total loss function is defined as a weighted sum of five terms: , in, For data fitting loss term, For fixed boundary loss terms, To load boundary loss, This is the loss term in the balance equation. This represents the strain fitting loss term. Adam is used for rapid convergence in the early stages of the loss calculation; after reaching the preset epoch, L-BFGS is switched to obtain more refined optimization results.

[0042] (3) Pelvic fracture reduction pathway planning This invention is based on RRT * The algorithm proposes a Muscle-PINNs+RRT method. * Strain-constrained path planning method. The specific process is as follows: In RRT... * Within the algorithm's framework, the path planning process begins at the starting position and gradually expands towards the target position. The algorithm randomly samples a point in the search space and finds the node in the current path tree that is closest to that point. New path nodes are generated by calculating the connecting paths between neighboring nodes in the path tree and the sampled point. Collision detection checks whether the newly generated path node collides with any obstacles. If a collision occurs, the node is discarded.

[0043] During the path tree expansion process, for each newly generated path node, the Muscle-PINNs model is invoked to predict muscle strain. The displacement of the adhesion points between the muscle and the affected bone is calculated through rigid body transformation. , in, R is the position of the adhesion point in the initial configuration, R is the rotation matrix, and T is the translation vector.

[0044] After the Muscle-PINNs model prediction is completed, the deformation gradient is calculated based on the displacement field predicted by the model, and the Cauchy strain, which is symmetric to the displacement gradient, is obtained: , The maximum principal strain at each point is extracted by eigenvalue decomposition.

[0045] , Statistical analysis of the global maximum principal strain is performed, the maximum value is recorded, and the result is returned to the RRT. * Path planning module. Muscle strain limit check determines whether the maximum strain of the muscle at the path node exceeds a set threshold, which is the maximum principal strain at the target point. If the strain of the node exceeds the threshold, the path node is discarded.

[0046] The key points and areas to be protected in this invention are: (1) A preoperative planning framework for robot-assisted pelvic fracture reduction based on muscle biomechanical modeling is proposed; (2) A muscle biomechanics simulation algorithm based on physical information neural network is proposed; (3) Propose the physical residual equation for polynomial second-order hyperelasticity; (4) A path planning algorithm that considers the maximum principal strain of muscles in pelvic fracture planning is proposed.

[0047] Compared with the prior art, the advantages of the present invention are: (1) See Figure 4 The use of Physical Information Neural Network (PINN) to construct muscle biomechanical modeling greatly improves the efficiency of muscle stretching simulation and provides support for real-time planning of pelvic fracture reduction paths. (2) The influence of muscle force was considered in the reduction path of pelvic fracture, which ensured the safety threshold of soft tissue during the reduction of pelvic fracture.

[0048] Example 2 According to another embodiment of the present invention, a robot-assisted fracture reduction path planning system is provided, comprising: The prediction unit is used to predict the displacement field and maximum principal strain of the peripelvic muscles using a physical information neural network. Path planning unit, used to embed strain threshold constraints and strain-aware cost function into RRT * The path planning process proactively avoids the risk of soft tissue damage during the tree expansion and path optimization stages.

[0049] The robot-assisted fracture reduction path planning system in this embodiment of the invention uses a physical information neural network (Muscle-PINNs) to quickly predict the displacement field and maximum principal strain of the peripelvic muscles, and embeds strain threshold constraints and strain-aware cost functions into the RRT. * During the path planning process, the risk of soft tissue damage is actively avoided in the tree expansion and path optimization stages, while achieving millisecond-level inference to meet the real-time requirements during surgery.

[0050] The robot-assisted fracture reduction path planning system of the present invention will be described in detail below with reference to specific embodiments: To overcome the shortcomings of existing technologies, this invention proposes an integrated soft tissue biomechanical prediction-path planning system for pelvic repositioning: It employs Muscle-PINNs to rapidly predict the displacement field and maximum principal strain of the peripelvic muscles, and embeds strain threshold constraints and a strain-aware cost function into the RRT (Responsive Tissue Repositioning Therapy). * During the path planning process, the risk of soft tissue damage is actively avoided in the tree expansion and path optimization stages, while achieving millisecond-level inference to meet the real-time requirements during surgery.

[0051] This invention proposes a real-time muscle strain-constrained path planning algorithm based on a neural network model of muscle physical information for real-time planning of reduction paths in pelvic fracture reduction surgery. It uses physical priors and grid-independent reasoning of data to provide displacement / strain fields in milliseconds, and directly embeds strain thresholds into the search and optimization, taking into account both efficiency and soft tissue safety boundaries.

[0052] The basic content of the technical solution of this invention is as follows: This invention proposes a Muscle-PINNs+RRT method for closed reduction of pelvic fractures. * A strain-constrained path planning system is proposed. First, a dual-branch, space-load decoupled physical information neural network is constructed: geometric topology and load conditions are encoded separately and fused at the feature layer. Training is performed using data, PDE, boundary, and strain losses to predict the muscle displacement field and maximum principal strain independently of the grid. Second, the predictions are embedded into the RRT (Restricted Path Theory). * The displacement of each muscle attachment point is calculated for each candidate pose (R,T). The maximum principal stress E_max is obtained through the model. High-risk branches are screened out using a threshold, and the final path is optimized by "path length + strain cost" to achieve near real-time intraoperative planning. This scheme significantly reduces strain prediction bias and significantly improves efficiency while keeping displacement error controllable.

[0053] The technical solution of the present invention is described in detail below: (1) Biomechanical modeling of pelvic muscle tissue First, based on CT scans of patients with pelvic fractures, the pelvic muscle and skeletal tissue is reconstructed. Segmentation software is then used to separate the muscles and bones according to their grayscale values, generating a 3D model. This invention employs a second-order polynomial hyperelastic model, where the strain energy density function W consists of a deviation component and a volumetric component.

[0054] , The deviation component Wdel is represented by the simplified invariants I1 and I2 of the right Cauchy-Green deformation tensor C.

[0055] , Where C 10 C 01Describing stiffness under low strain, C 20 C 11 C 02 It is a parameter that captures nonlinear behavior and complex deformations.

[0056] Volume term W vol Consider near-incompressibility.

[0057] , Where D1 and D2 are incompressibility parameters, Let F be the determinant of the deformation gradient.

[0058] Biomechanical modeling of muscle tissue is based on the continuum mechanics framework, assuming quasi-static conditions and no body loads, where the system satisfies the momentum balance equations. Starting with deformation mapping, partial differential equations (PDEs) are derived step-by-step. Deformation mapping is defined. :X x = X + u(X), where X is the reference configuration coordinates, x is the current configuration coordinates, and u is the displacement vector. The deformation gradient tensor F is obtained by partial differentiation of x.

[0059] , in X represents the reference configuration gradient operator. The right Cauchy-Green tensor C is C=F. T F, calculate the invariants I1 and I2.

[0060] , For hyperelastic materials, the strain energy density W depends on the modified invariants I1, I2, and J, and the second Piola-Kirchhoff stress tensor S is derived from the strain energy density W.

[0061] , S can be further converted into the Cauchy stress tensor σ=J-1FSFT

[20] , and under quasi-static conditions, the momentum balance equation is always zero.

[0062] , (2) Muscle biomechanical physical information neural network model This invention proposes a dual-branch physical information neural network (Muscle-PINNs). The Muscle-PINNs network structure is as follows: Figure 2 As shown, it mainly includes a two-branch fully connected network and a physical calculator.

[0063] Two-branch fully connected network: The network consists of a coordinate branch, a load branch, and a merged subnet. The coordinate branch receives spatial coordinates (Lx, Ly, Lz) and also has two hidden layers. The features from the two branches are fused and input into the merged subnet, outputting a normalized displacement vector u=[ux,uy,uz]. T .

[0064] , Here, f represents a multilayer perceptron (MLP), with each layer undergoing a linear transformation followed by a Tanh activation function. This dual-branch design allows the model to capture geometric and loading dependencies separately, improving its adaptability to various operating conditions. Material parameters are fixed and embedded in the network to ensure physical consistency.

[0065] Physics calculator: For each batch of samples, automatic differentiation and component-by-component calculations are performed using PyTorch. u / X is obtained by correcting for normalized scaling and obtaining the physical gradient; based on hyperelastic strain energy, the equivalent second-order invariants I1 and I2 correction terms are calculated, and the biased part σ is constructed. iso And add volume compression part σ vol The Green strain tensor S is then converted into a 6-dimensional vector for strain fitting loss.

[0066] The total loss in Muscle-PINNs includes data fitting, boundary conditions, balance constraints, and strain loss. It dynamically balances these components through adaptive weighting, combining inverse and proportional weighting with EMA smoothing, weight range constraints, warmup, and stagnation fallback strategies to dynamically adjust the relative contribution of each component's loss. The total loss function is defined as a weighted sum of five terms: , Where L data L is the data fitting loss term. fixed For the fixed boundary loss term, L stretch For loading boundary loss, L PDE L is the loss term in the balance equation. strain This represents the strain fitting loss term. Adam is used for rapid convergence in the early stages of the loss calculation; after reaching the preset epoch, L-BFGS is switched to obtain more refined optimization results.

[0067] (3) Pelvic fracture reduction pathway planning This invention is based on RRT * The algorithm proposes a Muscle-PINNs+RRT method. * Strain-constrained path planning method. The specific process is as follows: In RRT... *Within the algorithm's framework, the path planning process begins at the starting position and gradually expands towards the target position. The algorithm randomly samples a point in the search space and finds the node in the current path tree that is closest to that point. New path nodes are generated by calculating the connecting paths between neighboring nodes in the path tree and the sampled point. Collision detection checks whether the newly generated path node collides with any obstacles. If a collision occurs, the node is discarded.

[0068] During the path tree expansion process, for each newly generated path node, the Muscle-PINNs model is invoked to predict muscle strain. The displacement of the adhesion points between the muscle and the affected bone is calculated through rigid body transformation. , in, R is the position of the adhesion point in the initial configuration, R is the rotation matrix, and T is the translation vector.

[0069] After the Muscle-PINNs model prediction is completed, the deformation gradient is calculated based on the displacement field predicted by the model, and the Cauchy strain, which is symmetric to the displacement gradient, is obtained: , The maximum principal strain at each point is extracted by eigenvalue decomposition.

[0070] , Statistical analysis of the global maximum principal strain is performed, the maximum value is recorded, and the result is returned to the RRT. * Path planning module. Muscle strain limit check determines whether the maximum strain of the muscle at the path node exceeds a set threshold, which is the maximum principal strain at the target point. If the strain of the node exceeds the threshold, the path node is discarded.

[0071] The key points and areas to be protected in this invention are: (1) A preoperative planning framework for robot-assisted pelvic fracture reduction based on muscle biomechanical modeling is proposed; (2) A muscle biomechanics simulation algorithm based on physical information neural network is proposed; (3) Propose the physical residual equation for polynomial second-order hyperelasticity; (4) A path planning algorithm that considers the maximum principal strain of muscles in pelvic fracture planning is proposed.

[0072] Compared with the prior art, the advantages of the present invention are: (1) See Figure 3 The use of Physical Information Neural Network (PINN) to construct muscle biomechanical modeling greatly improves the efficiency of muscle stretching simulation and provides support for real-time planning of pelvic fracture reduction paths. (2) The influence of muscle force was considered in the reduction path of pelvic fracture, which ensured the safety threshold of soft tissue during the reduction of pelvic fracture.

[0073] Example 3 A storage medium storing program files capable of implementing any of the above-described robot-assisted fracture reduction path planning methods.

[0074] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements the steps of robot-assisted fracture reduction path planning. The computer storage medium can be any available medium or data storage device accessible by a computer, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0075] Example 4 A processor for running a program, wherein the program executes any of the above-mentioned robot-assisted fracture reduction path planning methods during runtime.

[0076] In an exemplary embodiment, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of robot-assisted fracture reduction path planning. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0077] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0078] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0079] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The system embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0080] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0082] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0083] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A robot-assisted fracture reduction path planning method, characterized in that, Includes the following steps: A physical information neural network was used to predict the displacement field and maximum principal strain of the peripelvic muscles. Embedding strain threshold constraints and strain-sensing cost functions into RRT * The path planning process proactively avoids the risk of soft tissue damage during the tree expansion and path optimization stages.

2. The robot-assisted fracture reduction path planning method according to claim 1, characterized in that, Before using a physical information neural network to predict the displacement field and maximum principal strain of the peripelvic muscles, the method further includes: Construct a physical information neural network with dual-branch space-load decoupling.

3. The robot-assisted fracture reduction path planning method according to claim 2, characterized in that, The construction of the dual-branch space-load decoupled physical information neural network includes: Geometric topology and load conditions are encoded separately and fused at the feature layer, and trained using joint data, PDE, boundary and strain loss.

4. The robot-assisted fracture reduction path planning method according to claim 2, characterized in that, Before constructing the dual-branch space-load decoupled physical information neural network, the method further includes: Based on CT reconstruction of pelvic muscle and skeletal tissue in patients with pelvic fractures, segmentation software was used to separate muscles and bones according to different gray values ​​to generate a 3D model.

5. The robot-assisted fracture reduction path planning method according to claim 1, characterized in that, The physical information neural network includes: a two-branch fully connected network and a physical calculator; The dual-branch fully connected network consists of a coordinate branch, a load branch, and a merged subnet: the coordinate branch receives spatial coordinates and also has two hidden layers; the features of the two branches are fused and input into the merged subnet, and the output is a normalized displacement vector. The physical calculator is configured to perform component-wise calculations using automatic differentiation in PyTorch for each batch of samples. u / X is obtained by correcting for normalized scaling and obtaining the physical gradient; based on hyperelastic strain energy, the equivalent second-order invariants I1 and I2 correction terms are calculated, and the biased part σ is constructed. iso And add volume compression part σ vol The Green strain tensor S is then converted into a 6-dimensional vector for strain fitting loss.

6. The robot-assisted fracture reduction path planning method according to claim 1, characterized in that, The method of embedding strain threshold constraints and strain sensing cost functions into RRT * The path planning process includes: Predicting embeddings in RRT * For each candidate pose (R,T), the displacement of each muscle attachment point is calculated. The maximum principal stress E_max is obtained through the model. High-risk branches are screened out with a threshold, and the final path is optimized with path length + strain cost.

7. The robot-assisted fracture reduction path planning method according to claim 6, characterized in that, The prediction embedded in RRT * For each candidate pose (R,T), the displacement of each muscle attachment point is calculated. The maximum principal stress E_max is obtained through the model. High-risk branches are filtered out using a threshold, and the final path is optimized by combining path length and strain cost. In RRT * Within the algorithm's framework, the path planning process begins at the starting position and gradually expands towards the target position. The algorithm randomly samples a point in the search space and finds the node closest to that point in the current path tree. By calculating the connection paths between neighboring nodes in the path tree and the sampled point, a new path node is generated. Collision detection checks whether the newly generated path node collides with an obstacle. If a collision occurs, the node is discarded.

8. The robot-assisted fracture reduction path planning method according to claim 7, characterized in that, During the path tree expansion process, for each newly generated path node, the Muscle-PINNs model is invoked to predict muscle strain; the displacement of the adhesion point between the muscle and the affected bone is calculated through rigid body transformation.

9. The robot-assisted fracture reduction path planning method according to claim 8, characterized in that, Statistical analysis of the global maximum principal strain is performed, the maximum value is recorded, and the result is returned to the RRT. * Path planning module; Muscle strain limit check determines whether the maximum strain of the muscle at the path node exceeds a set threshold. The threshold is the maximum principal strain at the target point. If the strain of the node exceeds the threshold, the path node is discarded.

10. A robot-assisted fracture reduction path planning system, characterized in that, include: The prediction unit is used to predict the displacement field and maximum principal strain of the peripelvic muscles using a physical information neural network. Path planning unit, used to embed strain threshold constraints and strain-aware cost function into RRT * The path planning process proactively avoids the risk of soft tissue damage during the tree expansion and path optimization stages.

Images