Multi-algorithm collaborative external fixation support pose determination method, device and medium

Abstract

The invention relates to a multi-algorithm collaborative external fixation support pose determination method and device and a medium, and the method comprises the steps: obtaining the algorithm starting information of a basic layer and a collaborative layer in a processing system configured by a user, calling a started basic algorithm in a basic layer algorithm module, obtaining a basic pose, and if a basic neural network approximation method is started, obtaining a basic pose; if the basic pose solution is a basic pose solution and a collaborative algorithm in the collaborative layer algorithm module is started, inputting the basic pose solution into the started collaborative algorithm for refining, and taking the refined pose solution as a final candidate pose solution; otherwise, directly taking the basic pose solution as a final candidate pose solution of the corresponding path; theoretical branch chain length vectors corresponding to the final candidate pose solutions are calculated through inverse kinematics, the theoretical branch chain length vectors are compared with actually-measured branch chain length vectors of the current group, and the branch chain with the minimum matching error is selected as a final relative pose result corresponding to the actually-measured branch chain length vectors of the group to be output. Compared with the prior art, the method has the advantages of wide application range, high robustness, high efficiency and the like.

Description

Technical Field

[0001] This invention relates to the field of skeletal deformity correction technology, and in particular to a method, device and medium for determining the pose of an external fixation device using a multi-algorithm collaborative approach. Background Technology

[0002] External fixators play a vital role in orthopedic surgery. Faced with different symptoms, surgeons need to select appropriate fixators based on clinical experience, thus leading to the design of various fixator configurations. Among these, the six-axis external fixator is a typical advanced ring-shaped external fixator, widely used by orthopedic surgeons worldwide. The six-axis external fixator is a modified Stewart mechanism frame consisting of two rings connected at a universal joint by six adjustable telescopic screws. This allows the external fixator, like the Ilizarov fixator, to simultaneously correct complex, multi-dimensional deformities.

[0003] Treatment algorithms plan the treatment process based on external fixation stents, and the algorithm for solving the stent's pose is the core of the treatment algorithm. Doctors can precisely control the correction process with computer assistance. Treatment algorithms can pre-calculate the theoretical screw length for each correction state and adjust the treatment plan according to the actual situation during treatment to predict the treatment effect. Due to the characteristics of parallel mechanisms, inverse kinematics algorithms are simple and efficient for calculating the theoretical screw length, but research on forward kinematics algorithms that assist in designing the treatment process and predicting the actual treatment effect of the stent is limited.

[0004] Existing forward kinematics algorithms have several limitations: When using external fixation stents such as typical Taylor stents to handle special deformities, the significant differences in location and stent link length can easily lead to solution failures; personalized treatment needs require personalized stents, such as asymmetric branch installation methods or different branch support forms, making it difficult for existing forward kinematics algorithms to meet the complex and diverse stent configurations and application scenarios, thus limiting their application scope; inefficient and low-precision forward kinematics algorithms seriously affect the patient's recovery process and final outcome, because the algorithm's results are not accurate, the solution speed is not timely, and in some cases, it may even fail to provide convergence results, posing a potential high risk to treatment; at the same time, the algorithm places certain demands on the performance of local medical equipment and even network servers, and inefficient algorithms increase the equipment burden and affect the real-time performance of computational responses. Therefore, how to provide an efficient, stable, adaptable, and reliable forward kinematics solution method for external fixation stents that can operate in different clinical scenarios is a technical problem that needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a multi-algorithm collaborative method, device and medium for determining the pose of an external fixation device. By establishing a two-layer algorithm architecture of parallel prediction or solving at the basic layer and refinement at the collaborative layer, it achieves directional collaboration between rapid prediction and high-precision optimization. This solves the problems of failure, non-convergence and insufficient accuracy caused by poor initial values ​​in the forward kinematics solution of traditional external fixation devices while ensuring computational efficiency, thus improving the robustness, speed and accuracy of the pose determination process.

[0006] The objective of this invention can be achieved through the following technical solutions: According to one aspect of the present invention, a method for determining the pose of an external fixation device using a multi-algorithm collaborative approach is provided, which operates in a processing system pre-loaded with multiple pose-solving algorithms. The specific steps include: S1. Obtain the structural parameter set and measured branch length of the target external fixation bracket; the structural parameter set includes proximal ring parameters, distal ring parameters, branch parameters, and installation position parameters of each branch on the proximal and distal rings; S2. Obtain the algorithm activation information of the basic layer and the collaborative layer in the processing system configured by the user, call the basic algorithms that have been enabled in the basic layer algorithm module, and perform parallel computation with the structural parameter set and the measured branch length of the current group as common inputs; the basic algorithms include: basic neural network approximation method, basic numerical iterative algorithm and basic intelligent optimization algorithm, wherein at least one of the basic numerical iterative algorithm and basic intelligent optimization algorithm is enabled, and each enabled basic algorithm independently outputs a corresponding basic pose solution; S3. If the basic neural network approximation method is enabled and the cooperative algorithm in the cooperative layer algorithm module is enabled, the basic pose solution output by the basic neural network approximation method is input into the enabled cooperative algorithm for refinement, and the refined pose solution is used as the final candidate pose solution of the current algorithm; otherwise, the basic pose solution of the basic layer is directly used as the final candidate pose solution of the corresponding path. S4. For each final candidate pose solution, calculate the corresponding theoretical branch length vector through inverse kinematics, and compare it with the measured branch length vector of the current group. Select the final candidate pose solution with the smallest matching error as the final relative pose result corresponding to the measured branch length vector of the group.

[0007] Furthermore, the cooperative algorithm includes a cooperative numerical iterative algorithm and a cooperative intelligent optimization algorithm, with at most one algorithm being enabled in the cooperative algorithm; both the basic numerical iterative algorithm and the cooperative numerical iterative algorithm are Newton's iterative method for calculating the Jacobian matrix using the central difference method; both the basic intelligent optimization algorithm and the cooperative intelligent optimization algorithm are particle swarm optimization algorithms or genetic algorithms. When the activated cooperative algorithm is the cooperative numerical iterative algorithm, the basic pose solution output by the basic neural network approximation method is used as the initial solution for iteration of the cooperative numerical iterative algorithm for iterative calculation. When the cooperative algorithm is the cooperative intelligent optimization algorithm, a local search space is constructed with the basic pose solution output by the basic neural network approximation method as the center, and optimization search is performed in the local search space.

[0008] Furthermore, the basic neural network approximation method is implemented based on a pre-trained neural network model, which is obtained through the following steps: Based on the proximal loop parameters, distal loop parameters, and branch parameters in the structural parameter set, the sampling space of the pose parameters is determined; Multiple pose samples are randomly generated within the sampling space, and the theoretical branch length vector corresponding to each pose sample is obtained by inverse kinematics calculation, forming multiple training sample pairs. Each training sample pair includes a pose sample and the corresponding theoretical branch length vector. The initial neural network is trained using the multiple training samples to obtain the pre-trained neural network model.

[0009] Furthermore, the pre-trained neural network model includes a backpropagation neural network and a radial basis function neural network; the backpropagation neural network includes multiple parallel sub-networks, each sub-network being used to predict one dimension of the final relative pose result; the radial basis function neural network uses Gaussian radial basis functions as activation functions.

[0010] Furthermore, the measured branch lengths obtained in S1 are in multiple sets, and corresponding continuous flags indicating whether there is a continuous pose change relationship between the multiple sets of data are obtained. When the continuous flag indicates a continuous pose change relationship, during the execution of S2 for the latter set of measured branch lengths: If the basic numerical iteration algorithm is enabled, the final relative pose result output in S4 for the previous set of data is used as the initial solution for the iteration when solving for the current set of data, or as the value of the corresponding dimension in the initial solution. If the basic intelligent optimization algorithm is enabled, the final relative pose result output in S4 for the previous set of data is used as the center of the search space when solving for the current set of data.

[0011] Furthermore, the structural parameter set also includes the installation order identifiers of the proximal ring and the distal ring; in step S4, the correspondence between the proximal coordinate system and the distal coordinate system on which the inverse kinematics calculation is based is determined according to the installation order identifiers.

[0012] Furthermore, the mounting position parameters of the branches on the proximal and distal rings support asymmetrical arrangement. Specifically, the mounting position parameters are the hole numbers on the rings to which each branch is connected. The ring structure of the external fixing bracket is a full ring, a 2 / 3 ring, or a U-shaped ring. The joint types of the branches include ball joint-sliding joint-ball joint, revolute joint-sliding joint-ball joint, and revolute joint-sliding joint-revolute joint.

[0013] Further, in S4, the matching error is obtained by calculating the Euclidean distance or mean absolute difference between the theoretical branch length vector and the measured branch length vector.

[0014] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0015] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) By integrating the multi-algorithm collaborative architecture of the basic layer parallel computing and the collaborative layer refinement processing, the robustness and success rate of the solution are significantly improved. Through the basic layer, which includes the basic neural network approximation method, the basic numerical iterative algorithm and the basic intelligent optimization algorithm, users can enable at least two different types of algorithms for parallel solution as needed. Even if a single algorithm fails due to poor initial values ​​or an inappropriate search space, other algorithms may still provide effective pose solutions, thus avoiding the problem of solution failure or non-convergence that is common with existing single algorithms. On this basis, a collaborative layer algorithm, namely the collaborative numerical iterative algorithm or the collaborative intelligent optimization algorithm, which is dedicated to receiving the neural network prediction results and performing local refinement, is further used to perform fast and accurate secondary optimization in a very small neighborhood by utilizing the high-quality initial guesses provided by the neural network. This further improves the accuracy and reliability of the final pose results while ensuring the computational speed.

[0017] (2) By supporting a wide range of stent parameter configurations and flexible data processing logic, the applicability and clinical applicability of the method are greatly expanded. The method can not only handle typical six-axis stents with symmetrical installation, but also explicitly support asymmetrical arrangement of branches and various ring structures such as full ring, 2 / 3 ring, and U-shaped ring, as well as various branch joint types such as SPS, RPS, and RPR by defining a complete set of structural parameters including proximal ring parameters, distal ring parameters, branch parameters and installation position numbers. This can meet the complex and diverse stent configuration requirements in personalized treatment. At the same time, the method introduces optimization logic for reusing the results of previous solutions when processing multiple sets of continuously changing branch data, and adapts to the clinical accident of reverse installation of proximal and distal rings by using installation order identifiers. These designs enable the method to flexibly respond to different treatment stages and clinical scenarios, enhancing its adaptability and applicability in real medical environments.

[0018] (3) Through modular and configurable algorithm calling strategies and a unified verification mechanism based on inverse kinematics, the optimal balance between computational efficiency and resource consumption is achieved. Users can freely configure the algorithm combination of the basic layer and the collaborative layer through the algorithm activation information. For example, in scenarios requiring fast response, only the neural network algorithm with fast convergence speed can be enabled, or in scenarios pursuing high accuracy, the complete process of "neural network prediction + collaborative refinement" can be enabled. When computing power is limited, the neural network can be turned off and only the traditional algorithm can be used. This allows the method to flexibly adjust the running logic according to different accuracy requirements and hardware computing power conditions. Finally, all candidate pose solutions generated by the algorithm path need to be verified through unified inverse kinematics calculation. The optimal solution is selected based on the objective standard of minimizing the matching error between the theoretical branch length and the measured length. This closed-loop verification mechanism ensures that even when multiple algorithms are in parallel, a unique and reliable optimal pose result can be output. Thus, a good balance between solution speed, result accuracy and computational resource consumption is achieved overall, enabling the method to run smoothly on different computing power platforms from high-performance workstations to portable medical devices. Attached Figure Description

[0019] Figure 1 Flowchart of a multi-algorithm collaborative method for determining the pose of an external fixation device; Figure 2 Data flow diagram for a multi-algorithm collaborative method for determining the pose of an external fixation device; Figure 3 This is a schematic diagram of the six-axis external fixation bracket in this embodiment; Figure 4 This is a schematic diagram of the coordinate system of the proximal and distal rings of the six-axis external fixation bracket in this embodiment; Figure 5 This diagram illustrates the mounting holes and branch sections for different types of full rings and their branches in the personalized treatment described in this embodiment. Figure 6 A schematic diagram of the principle of Newton's iteration method, which is an approximation of the central difference method; Figure 7 This is a schematic diagram of the particle swarm optimization algorithm. Figure 8 This is a schematic diagram of the genetic algorithm. Figure 9 Schematic diagram of data generation and preprocessing principles; Figure 10 This is a schematic diagram of the backpropagation neural network and a schematic diagram of the Tanh activation function commonly used in backpropagation neural networks; Figure 11 This is a schematic diagram of a radial basis function neural network and a schematic diagram of the Gaussian radial basis function activation function commonly used in radial basis function neural networks. Figure 12 This is a schematic diagram of the forward kinematics solution method for the external fixation bracket using multiple algorithms in Example 1; Figure 13 This is a schematic diagram of the forward kinematics solution method for the external fixation bracket using multiple algorithms in Example 2; Figure 14 This is a schematic diagram of the forward kinematics solution method for the external fixation bracket using multiple algorithms in Example 3; Figure 15 This is a schematic diagram of the forward kinematics solution method for the external fixation bracket using multiple algorithms in Example 4.

[0020] The labels in the diagram are as follows: 1. Proximal bone segment; 2. Proximal ring; 3. Branch; 4. Distal ring; 5. Distal bone segment. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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] Example 1 In the clinical application of skeletal deformity correction, the forward kinematics solution of medical external fixators is a core step in formulating and adjusting treatment plans. Current forward kinematics algorithms have limitations. When faced with special deformity positions, asymmetric support configurations, or significant differences in branch lengths, traditional single algorithms, such as numerical iteration methods or intelligent optimization algorithms, are prone to failure or non-convergence due to poor initial solutions or excessively large search spaces. While relying solely on neural networks for approximation offers speed advantages, their prediction accuracy often falls short of the stringent precision requirements of clinical correction. Furthermore, existing methods struggle to flexibly adapt to complex and personalized scenarios that may arise in clinical practice, such as reversed ring installation order or asymmetric branch arrangements, and are inefficient when processing multiple consecutive sets of correction data. These shortcomings collectively limit the practical application of external fixators in precise, efficient, and personalized treatment.

[0023] The multi-algorithm collaborative method for determining the pose of external fixation stents provided in this embodiment integrates and coordinates the advantages of various algorithms, such as rapid prediction by neural networks, numerical iterative local refinement, and intelligent heuristic global search, by constructing a computational architecture that combines a basic layer and a collaborative layer. This ensures high robustness and high accuracy while improving solution efficiency and is widely applicable to various stent configurations and complex clinical situations.

[0024] like Figure 1 As shown, this embodiment presents a multi-algorithm collaborative method for determining the pose of an external fixation device. This method operates within a processing system that has multiple pre-set pose-solving algorithms. The specific steps include: S1. Obtain the structural parameter set and measured branch length of the target external fixation bracket; the structural parameter set includes proximal ring parameters, distal ring parameters, branch parameters, and installation position parameters of each branch on the proximal and distal rings; S2. Obtain the algorithm activation information of the basic layer and collaborative layer in the user-configured processing system, call the basic algorithms that have been enabled in the basic layer algorithm module, and perform parallel computation with the structural parameter set and the measured branch length of the current group as common inputs; the basic algorithms include: basic neural network approximation method, basic numerical iterative algorithm and basic intelligent optimization algorithm, of which at least one of the basic numerical iterative algorithm and basic intelligent optimization algorithm is enabled, and each enabled basic algorithm independently outputs a corresponding basic pose solution; S3. If the basic neural network approximation method is enabled and the cooperative algorithm in the cooperative layer algorithm module is enabled, then the basic pose solution output by the basic neural network approximation method is input into the enabled cooperative algorithm for refinement, and the refined pose solution is used as the final candidate pose solution of the current algorithm; otherwise, the basic pose solution of the basic layer is directly used as the final candidate pose solution of the corresponding path. S4. For each final candidate pose solution, calculate the corresponding theoretical branch length vector through inverse kinematics, and compare it with the measured branch length vector of the current group. Select the final candidate pose solution with the smallest matching error as the final relative pose result corresponding to the measured branch length vector of the group.

[0025] like Figure 2 The diagram shows the data flow of the method in this embodiment. As can be seen from the diagram, the execution flow of the data flow includes three logical parts: the input part, the solution part, and the output part. The input part receives and integrates the various information required for the solution; the solution part adopts a strategy that coordinates the basic layer algorithm and the cooperative layer algorithm to complete the pose calculation; the output part outputs the optimal pose result obtained from the solution along with relevant process information.

[0026] The six-axis external fixation bracket used in this embodiment is as follows: Figure 3 As shown, it consists of a proximal ring 2, a distal ring 4, and six branches 3 for connecting the proximal ring 2 and the distal ring 4. The proximal ring 2 is used to fix the proximal bone segment 1 of the fractured bone, and the distal ring 4 is used to fix the distal bone segment 5 of the fractured bone. This embodiment uses a six-axis external fixator as an example for illustration, but it is not limited to a six-axis external fixator.

[0027] like Figure 4 As shown, a three-dimensional Cartesian coordinate system is established with the center of the proximal loop as the origin and the plane containing the proximal loop as the XY plane. Another three-dimensional Cartesian coordinate system is established with the center of the distal loop as the origin and the plane containing the distal loop as the XY plane. These two coordinate systems are the global coordinate system and the local coordinate system. The pose relationship between the two loops is described using the positional relationship between the global and local coordinate systems, with each coordinate system being relatively fixed relative to its respective loop.

[0028] like Figure 5 As shown in the figure, a is a symmetrical mounting ring, b is a radially translated asymmetrical mounting ring, and c is a radially radial asymmetrical mounting ring. On one ear of the ring, the solid line outlines the holes for the default symmetrical mounting of the branches, and the dashed line outlines the holes for optional branch mounting during personalized treatment. This invention supports asymmetrical mounting of branches, starting from the core needs of clinical treatment. It provides personalized solutions designed for complex skeletal deformity correction, fracture reduction, or limb lengthening. Its core advantage lies in adapting to differences in human anatomy, accurately addressing asymmetrical lesions, and simultaneously ensuring fixation stability and functional preservation.

[0029] like Figure 5As shown in the diagram, d represents a schematic diagram of some types of branches applicable to this embodiment. The SPS branch, or ball-joint-sliding-ball-joint branch, uses a revolute joint with three axes perpendicular to each other to approximate a ball joint, and a helical joint to approximate a sliding joint. This is the branch type used in six-axis external fixation braces. Other types of external fixation braces also use RPS branches (revolute-sliding-ball-joint branches), RPR branches (revolute-sliding-revolute), and some branch types not listed in the diagram. This invention is applicable not only to six-axis external fixation braces using SPS branches, but also to external fixation braces using RPS and RPR branches.

[0030] Correspondingly, the basic layer algorithm can be selected from numerical iteration method, intelligent heuristic optimization algorithm, and neural network approximation method, while the collaborative layer algorithm can be selected from numerical iteration method and intelligent heuristic optimization algorithm. Figure 2 middle This represents the length of branches 1 through 6, and is one of the inputs for solving the problem. The pose result is used... The table represents the three coordinates of the local coordinate system origin in the global coordinate system and the three ordered rotation angles used to progressively rotate the global coordinate system attitude to the local coordinate system attitude. The process of multi-algorithm collaborative solution is as follows: The information received by the input section includes: The known conditions required for the solution include the set of structural parameters of the target external fixator, such as the parameters of the proximal and distal rings, the parameters of the branches (such as connecting screws), the specific installation position parameters of the branches on the proximal and distal rings, and the measured branch length vector in the current state; information controlling the specific solution process, that is, the instruction information configured by the user to control the algorithm's running logic, such as the enabled or disabled status of various basic and collaborative layer algorithms (i.e., algorithm enable information), the order identifier of the proximal and distal rings, the numbering rules of the branches, etc. This information enables the method to flexibly cope with various complex situations that may arise in clinical practice; internal parameters that need to be adjusted for certain solution algorithms: in order to better adapt the algorithm to specific stent and malformation parameters, users can fine-tune or specify some internal parameters of the algorithm, such as specifying the initial solution or difference step size for the numerical iteration method, setting the search space or solution options for the intelligent optimization algorithm, and setting relevant hyperparameters for the training or structure of the neural network, etc.

[0031] The proximal and distal ring sequence markings enable the method to automatically correct or adapt to ring reversal during surgical installation without requiring physical reinstallation, greatly improving the fault tolerance and convenience of clinical applications.

[0032] The foundational layer algorithms in the solution process independently calculate the corresponding foundational pose solution based on the measured branch lengths of the input. These foundational layer algorithms include three different types: numerical iteration, intelligent heuristic optimization, and neural network approximation. These algorithms run in parallel and independently within the foundational layer, without data transfer between them, and each produces its own foundational pose solution. Specifically, the numerical iteration method uses Newton's iteration method to approximate the derivative using the central difference approach; the intelligent heuristic optimization algorithms include particle swarm optimization and genetic algorithms; and the neural network approximation method refers to a neural network model built based on backpropagation or radial basis function principles. In particular, when using the neural network approximation method, it relies on pre-generated training data, and its output foundational pose solution can be fed into a collaborative layer for further refinement.

[0033] When numerical iterative methods or intelligent heuristic optimization algorithms exist in the base layer, the objective function and constraints are first constructed; when neural network approximation methods exist in the base layer algorithms, training data for training the neural network is first generated based on the scaffold parameters through data generation and preprocessing algorithms. Given a six-dimensional branch length vector, in the base layer, a numerical iterative method iterates continuously based on the initial solution to obtain a temporary pose result close to the true pose solution. An intelligent heuristic optimization algorithm generates a large population of individuals and iteratively searches to obtain a temporary pose result close to the true pose solution. A neural network trained using the training data outputs a predicted value as the temporary pose result. Each base layer algorithm has the possibility of failure, but the probability of finding a solution using multiple algorithms is significantly better than using a single algorithm. The temporary pose result from the neural network approximation method can be further processed in the collaborative layer. Using the temporary pose result from the neural network approximation method as the initial solution for the numerical iterative method results in a very fast iteration process with a high success rate because the initial solution is very close to the true solution. Using the temporary pose result from the neural network approximation method as the center of the initial search space for the intelligent heuristic optimization algorithm allows for searching with a small radius. Because the search space is very small, the search process is very fast and has a high success rate. Therefore, the collaborative layer algorithm can significantly improve the accuracy of the temporary pose result from the neural network approximation method. Without selecting an algorithm in the collaborative layer, the temporary pose result from the neural network approximation method will not be further processed and will be directly output.

[0034] The errors of multiple temporary pose results are compared, that is, the rod length corresponding to each temporary pose result is calculated, the errors of these rod lengths are compared with the input rod length, and the temporary pose result with the smallest error is used as the final output.

[0035] like Figure 6The image shows Newton's iteration method using the central difference in numerical iteration methods. It has a very fast convergence speed. Using the central difference method eliminates the need for derivative analytical expressions, further accelerating its iteration speed on a computer. Its iterative principle is as follows: Determine the variable dimensions and objective function Search space, etc.; Calculate the initial solution objective function value at ;judge Does the convergence condition meet? If it does, output the result directly; otherwise, proceed to the next step. Calculate the objective function value. and , It is the step size of the central difference; Calculate the initial solution : ; Calculate the initial solution objective function value at ; judge Check if the convergence condition is met. If it is, output directly. If not, repeat the above iterative steps.

[0036] like Figure 7 As shown, the iterative principle of the particle swarm optimization algorithm in intelligent heuristic optimization algorithms is as follows: Define the variable dimensions, objective function, search space, etc.; generate according to certain rules. There are 10 particles, each representing a potential solution. Each particle is initialized with an initial position and an initial velocity. The individual optimal position of each particle and the global optimal position of the entire swarm are initialized. The velocity of each particle is updated based on individual and swarm experience, and the position of the particle is updated based on the new velocity. The objective function value of each particle is calculated, and the individual optimal position and the global optimal position of the entire swarm are updated. The iteration is repeated until the convergence condition is met. When the iteration ends, the global optimal position is the final approximate optimum.

[0037] like Figure 8 As shown, the iterative principle of the genetic algorithm is as follows: The problem is transformed into gene coding, represented by strings of 0s and 1s in the diagram, with solid black squares representing 1 and hollow squares representing 0; the fitness function, constraints, etc., are defined; and the code is generated according to certain rules. Each individual is considered as a population; the fitness function value of each individual is calculated; a selection operation is performed to give individuals with higher fitness values ​​a greater probability of being retained in the next generation. Methods such as roulette wheel selection, tournament selection, and elite retention strategies are used to obtain a population size that is still [missing information]. The parent population is used to generate new individuals through crossover (gene recombination) and mutation (gene mutation). Crossover simulates gene exchange in biological reproduction, and mutation simulates random mutation of biological genes. Finally, a offspring population of the same size as the parent population is generated. The process is repeated iteratively until the convergence condition is met. The optimal individual in the population at this point is decoded into the original solution format, which is the approximate optimal solution to the problem.

[0038] like Figure 9 As shown, the data generation and preprocessing algorithm generates input six-dimensional rod length data and output six-dimensional pose data for training the neural network. The specific principle is as follows: The selected stent parameters include the selected proximal ring parameters, the selected distal ring parameters, and the selected 6-branch parameters. Combined with some parameter ranges recommended for external fixation stents, the approximate range of the selected stent pose can be obtained. A relatively conservative estimate is used as the sampling space for possible poses. In spherical coordinates, restrictions , , The range of the sample space is used as the sampling space for the three-dimensional coordinates in the pose. After random and uniform sampling, the initial sample set is transformed into a Cartesian coordinate system to obtain the three-dimensional coordinates. , , The initial sample set is transformed using the following formula: In a three-dimensional Cartesian coordinate system, the range of coordinates is restricted, and the resulting space is used as the sampling space for the three-dimensional pose in the pose. The three-dimensional pose is obtained after random and uniform sampling. , , Initial sample set; The two initial sample sets are combined to form a six-dimensional pose. , , , , , An initial sample set is used, and inverse kinematics is used to calculate the rod length corresponding to each pose, resulting in a six-dimensional rod length. , , , , , Initial sample set; After selecting samples that meet the criteria, the final six-dimensional rod length is obtained. , , , , , and six-dimensional pose , , , , , The sample set that is composed of.

[0039] exist Figure 10 In this context, the backpropagation neural network consists of three parts, specifically: Input layer, the input layer has There are 10 neurons, corresponding to the input. The number of items does not involve any parameter calculations; Hidden layers can contain one or more layers; the diagram shows a single hidden layer. There are 10 neurons, each with a bias value. The same number of neurons in the previous layer but with different weights. Each weight corresponds to a neuron in the upper layer. For a neuron in the hidden layer, the information transmitted by each neuron in the upper layer is multiplied by its corresponding weight, summed, and then added to the bias of this hidden layer neuron before being passed through the activation function. The mapping can then be passed to the next layer; The output layer can also have multiple neurons, the same number as the output quantity. Similar to the hidden layer, each neuron in the output layer has a bias. The same number of neurons in the previous layer but with different weights. Each weight corresponds to a neuron in the upper layer. For a neuron in the output layer, the information transmitted by each neuron in the upper layer is multiplied by its corresponding weight, summed, and then added to the bias of this output layer neuron before being passed through the activation function. The mapping can be output, and in the scenario of predicting the positive solution of a parallel mechanism, the activation function here is a linear activation function.

[0040] The Tanh activation function is a commonly used activation function in backpropagation neural networks. Its output range is (-1, 1), making it suitable for solving the forward problems of parallel mechanisms in the hidden layers of shallow backpropagation neural networks. Different activation functions can be used for the hidden and output layers of the neural network. Other activation functions, such as the Sigmoid function, also perform well in this type of problem. The formula for the Tanh activation function is: like Figure 11 As shown, the radial basis function neural network consists of three parts, specifically: Input layer, the input layer has There are 10 neurons, corresponding to the input. The number of items does not involve any parameter calculations; Hidden layers, typically a single layer, have There are 10 neurons, each containing a central parameter with the same dimension as the input data. and a width parameter (Or scale parameter). For a neuron in the hidden layer, the input vector and the center of the neuron need to be calculated. The distance (usually Euclidean distance) is used as the input to the radial basis function, and the result is the output of the neuron passed to the next layer.

[0041] The output layer can have multiple neurons, the same number as the output quantity; each neuron has the same weight as the neurons in the hidden layer. Each weight corresponds to a neuron in the hidden layer. For a neuron in the output layer, the information transmitted by each neuron in the hidden layer is multiplied by its corresponding weight and then summed (some structures will add a bias term) to directly obtain the final output result without going through an activation function mapping.

[0042] Radial basis function neural networks commonly use Gaussian radial basis functions as activation functions, with an output range of (0, 1] and different center parameters. and width parameter The Gaussian radial basis functions have different forms, and the expression for the activation function is: .

[0043] The collaborative layer algorithm in the solution section is used to improve the accuracy of the computation results for a specific path in the base layer. Specifically, the collaborative layer algorithm receives and processes the base pose solution output by the neural network approximation method in the base layer. Collaborative layer algorithms include numerical iterative methods (such as the central difference Newton method) and intelligent heuristic optimization algorithms (such as particle swarm optimization and genetic algorithms). Their key difference from similar algorithms in the base layer lies in their initial conditions: the numerical iterative methods in the collaborative layer use the base pose solution output by the neural network as their initial solution for iteration; the intelligent optimization algorithms in the collaborative layer construct a local neighborhood with a small radius centered on this base pose solution as their search space. This allows the collaborative layer algorithm to perform fast and high-precision local optimization in regions very close to the true solution, thus forming an efficient collaborative relationship of "prediction-refinement".

[0044] Cooperative algorithms include cooperative numerical iterative algorithms and cooperative intelligent optimization algorithms. At most one algorithm can be enabled in a cooperative algorithm. Both the basic numerical iterative algorithm and the cooperative numerical iterative algorithm are Newton's iteration method that uses the central difference method to calculate the Jacobian matrix. Both the basic intelligent optimization algorithm and the cooperative intelligent optimization algorithm are particle swarm optimization algorithms or genetic algorithms. When the cooperative algorithm is the cooperative numerical iterative algorithm, the basic pose solution output by the basic neural network approximation method is used as the initial solution for the cooperative numerical iterative algorithm for iterative calculation. When the cooperative algorithm is the cooperative intelligent optimization algorithm, a local search space is constructed with the basic pose solution output by the basic neural network approximation method as the center, and optimization search is performed within the local search space.

[0045] The basic neural network approximation method is based on a pre-trained neural network model, which is obtained through the following steps: The sampling space for pose parameters is determined based on the proximal loop parameters, distal loop parameters, and branch parameters in the structural parameter set. Multiple pose samples are randomly generated in the sampling space, and the theoretical branch length vector corresponding to each pose sample is obtained by inverse kinematics calculation, forming multiple training sample pairs. Each training sample pair includes a pose sample and the corresponding theoretical branch length vector. The initial neural network is trained using multiple training samples to obtain a pre-trained neural network model. The pre-trained neural network model includes a backpropagation neural network and a radial basis function neural network; the backpropagation neural network includes multiple parallel sub-networks, each of which is used to predict one dimension of the final relative pose result; the radial basis function neural network uses Gaussian radial basis functions as activation functions.

[0046] The measured branch lengths obtained in S1 are in multiple sets, and corresponding continuous flags indicating whether there is a continuous pose change relationship between the multiple sets of data are obtained. When the continuous flag indicates a continuous pose change relationship, when performing S2 for the latter set of measured branch lengths: If the basic numerical iterative algorithm is enabled, the final relative pose result output in S4 for the previous set of data is used as the initial solution for the current set of data, or as the value of the corresponding dimension in the initial solution. If the basic intelligent optimization algorithm is enabled, the final relative pose result output in S4 for the previous set of data is used as the center of the search space when solving for the current set of data.

[0047] The structural parameter set also includes the installation sequence identifiers of the proximal and distal rings; in S4, the correspondence between the proximal coordinate system and the distal coordinate system used for inverse kinematics calculations is determined based on the installation sequence identifiers.

[0048] In S4, the matching error is obtained by calculating the Euclidean distance or mean absolute difference between the theoretical branch length vector and the measured branch length vector.

[0049] The final results provided in the output section include: Final relative pose result: This describes the relative position and orientation (i.e., relative pose) between the proximal and distal loops. The final relative pose result is a vector with a maximum dimension of six, including three translations and three rotations. This result is selected from all the final candidate pose solutions output by the solution part, and after inverse kinematics verification, it has the smallest matching error with the measured branch length.

[0050] Record of the solution process: In addition to the final relative pose result, the output section also includes a process record, which records all the information of the input section, the parameters used by each algorithm in the solution section, and the important intermediate variables generated during the calculation process, which is convenient for traceability and analysis.

[0051] The mounting position parameters of the branches on the proximal and distal rings support asymmetrical arrangement. The specific mounting position parameters are the hole numbers on the rings to which each branch is connected. The ring structure of the external fixing bracket is a full ring, a 2 / 3 ring, or a U-shaped ring. The joint types of the branches include ball joint-sliding joint-ball joint, revolute joint-sliding joint-ball joint, and revolute joint-sliding joint-revolute joint.

[0052] Users can combine different algorithms in the base layer and the cooperating layer by enabling and disabling the built-in algorithm information. For example, they can disable the neural network path to reduce the computational load, or disable the cooperating layer and use only the results of the base layer.

[0053] In this embodiment, the enabled base layer algorithms include: numerical iteration method, intelligent heuristic optimization algorithm, and neural network approximation method. Furthermore, the collaborative layer algorithm is enabled. The input consists of scaffold parameters and a set of 6 branch lengths, including the selected proximal loop parameters, the selected distal loop parameters, the selected 6 branch parameters, and the six-dimensional branch length. , , , , , All other inputs are default values, such as the default symmetric installation for the 6-branch installation and the default solution precision.

[0054] The data processing steps of the method in this embodiment are as follows: The system will use the acquired set of structural parameters of the target external fixation bracket and the measured branch length vector of the current group as common inputs. Based on the activation information, the basic layer algorithm module simultaneously invokes the basic neural network approximation method, the basic numerical iterative algorithm, and the basic intelligent optimization algorithm for parallel computation, specifically including: The basic neural network approximation method feeds the input data into its pre-trained neural network model and outputs the first basic pose solution; The basic numerical iterative algorithm uses, for example, Newton's iteration method with central difference. It starts with the user-defined initial solution, constructs the objective function using the common input during the iteration process, performs iterative calculations, and outputs the second basic pose solution. Basic intelligent optimization algorithms, such as particle swarm optimization, initialize a population within a preset global search space and perform iterative search based on the fitness function defined by the common input to output a third basic pose solution. Since both the basic neural network approximation method and the collaborative layer algorithm are enabled, the output of the basic neural network approximation method, specifically the first basic pose solution, undergoes refinement. This first basic pose solution is then used as a high-quality initial guess and input into the enabled collaborative algorithm. Depending on the type of collaborative algorithm, the processing method is as follows: If the cooperative algorithm used is a cooperative numerical iterative algorithm, the first basic pose solution is set as the initial solution of the algorithm for iteration, and fast and high-precision local iteration is performed to output the refined pose solution. If the cooperative algorithm used is the cooperative intelligent optimization algorithm, then a local neighborhood much smaller than the global search space is constructed as the center of the first basic pose solution as the new search space. Efficient search is performed in this space to output the refined pose solution. The final candidate pose solution set is formed, resulting in three final candidate pose solutions: the refined pose solution, the second basic pose solution, and the second basic pose solution. For each final candidate pose solution set, the precise inverse kinematics formula is used to calculate the corresponding theoretical branch length vector. The matching error between each theoretical branch length vector and the measured branch length vector is calculated. The final candidate pose solution with the smallest error value is selected as the final relative pose result corresponding to the measured branch length vector of this set.

[0055] After calculating and outputting the final relative pose result, i.e., the target pose of the distal loop relative to the proximal loop, the final relative pose result is compared with a pre-planned, desired target corrective pose. Inverse kinematics calculations determine the theoretical length adjustment required for each adjustable branch on the external fixator to move from the current state (reflected by the measured branch length) to the target corrective pose. This theoretical length adjustment is converted into a specific digital drive signal that can be recognized by the external fixator's actuator. This drive signal can be wirelessly transmitted via a wireless communication module integrated into the device, such as Wi-Fi or Bluetooth, to a miniature actuator or controller connected to each branch of the external fixator. The actuator receiving the drive signal will precisely drive the branches to extend and retract according to the instructions, thereby actually changing the geometry of the external fixator and propelling the bone fragments gradually and precisely towards the preset corrective pose.

[0056] like Figure 12As shown, the input consists of the stent parameters and a set of 6 link lengths, including the selected proximal loop parameters, the selected distal loop parameters, the selected 6 link parameters, and the six-dimensional link length. , , , , , All other inputs are default values, i.e., the 6-branch installation method is the default symmetrical installation, and the solution accuracy is the default accuracy, etc. The support parameters are the specifications of the selected proximal ring, selected branches, and selected distal ring, such as... Figure 3 As shown in the image, 2 represents the selected proximal ring, 3 represents the selected branch, and 4 represents the selected distal ring. The default installation is symmetrical, as shown below. Figure 4 As shown, on one ear, the area highlighted by the solid line represents the holes for branch installation during the default symmetrical installation. Two branches are installed on one ear, requiring a total of three ears. These three ears are not adjacent on the loop. From the patient's perspective, the branches are always numbered counterclockwise on the loop, from 1 to 6, and branches numbered 1 and 2 are always on the same ear of the loop. The output of the solution method is a set of pose results, namely the six-dimensional pose of the selected distal loop x relative to the selected proximal loop. , , , , , .

[0057] The pose, such as Figure 4 As shown, a three-dimensional Cartesian coordinate system is established with the center of the proximal loop as the origin and the plane containing the proximal loop as the XY plane. Another three-dimensional Cartesian coordinate system is established with the center of the distal loop as the origin and the plane containing the distal loop as the XY plane. These two coordinate systems are the global coordinate system and the local coordinate system. The pose relationship between the two loops is described using the positional relationship between the global and local coordinate systems. , respectively representing the three coordinate information of the origin of the local coordinate system in the global coordinate system and the three ordered rotation angles that gradually rotate the attitude of the global coordinate system to the attitude of the local coordinate system. Each coordinate system is relatively fixed relative to the ring it is in.

[0058] The multi-algorithm collaborative algorithm for solving the positive kinematics of the external fixation brace in this embodiment is as follows: Figure 2 As shown, after calculations by the base layer and the collaborative layer, multiple final candidate pose solutions are obtained. The pole lengths corresponding to the final candidate pose solutions are calculated by inverse kinematics, and the error between the pole lengths and the input pole lengths is obtained to obtain the optimal pose result for output.

[0059] The numerical iteration method enabled in this embodiment, such as Figure 6As shown, the Newton-Raphson iteration method, approximated by the central difference method, calculates the derivative of the current solution with respect to the objective function using the central difference method, thus obtaining a result closer to the true solution and achieving faster convergence. This embodiment enables an intelligent heuristic optimization algorithm, such as... Figure 7 As shown, a particle swarm optimization algorithm is used to generate multiple particles in the search space. These particles update their speed based on individual and collective experience, continuously searching to find the globally optimal position. For example... Figure 8 As shown, the genetic algorithm simulates gene exchange in biological reproduction by performing operations such as selection, crossover, and mutation on individuals, iteratively obtaining the globally optimal individual. Furthermore, the neural network approximation method implemented in this embodiment requires pre-generated training data for training the neural network, such as... Figure 9 As shown, the data generation and preprocessing algorithm generates input six-dimensional rod length data and output six-dimensional pose data for training the neural network. Random uniform sampling yields an initial sample set of six-dimensional pose data, and inverse kinematics calculations yield the corresponding initial sample set of six-dimensional rod length data. By filtering samples that meet requirements for rod length, joint angles, etc., a final sample set consisting of six-dimensional rod length and six-dimensional pose data is obtained for neural network training. The initiated neural network approximation method, such as... Figure 10 As shown, backpropagation neural networks are a mature type of neural network. Shallow backpropagation neural networks are suitable for forward posing problems and offer ideal training speed and prediction accuracy. A backpropagation neural network consists of an input layer, hidden layers, and an output layer. The number of neurons in the input and output layers corresponds to the number of input and output values, respectively. The hidden and output layers contain trainable weights and biases, as well as activation functions used to introduce non-linearity. The Tanh activation function is commonly used in the hidden layers, while a linear activation function is commonly used in the output layer. For forward posing problems, six sub-neural networks are constructed. Each sub-neural network takes a six-dimensional rod length as input and outputs a value from the six-dimensional pose. This approach has a simple structure and smaller error, making it suitable for forward posing problems. The neural network approximation method used is as follows: Figure 11 As shown, radial basis function (RBF) neural networks are a mature type of neural network commonly used for function fitting. RBF neural networks consist of an input layer, hidden layers, and an output layer. The number of neurons in the input and output layers corresponds to the number of input and output values, respectively. The number of neurons in the hidden layer depends on the amount of training data. The hidden layer often uses the Gaussian radial basis function as the activation function. The activation function's center parameter *c* and width parameter... σ It is trainable; the weights and biases of the output layer are trainable. For forward solving problems, a six-input, six-output Gaussian radial basis function neural network can be constructed using 1000-2000 training samples to meet the requirements.

[0060] Example 2 like Figure 13As shown, this is the solution of this embodiment. The difference between the method in this embodiment and the solution in Embodiment 1 lies in the algorithm activation information. The enabled basic layer algorithms include: numerical iteration method and intelligent heuristic optimization algorithm. Furthermore, the collaborative layer algorithm is disabled. That is, the data generation and preprocessing algorithm, the neural network approximation method, and the collaborative layer algorithm are all in a disabled state. The method in this embodiment runs under low load.

[0061] The data processing steps in this embodiment are as follows: The system uses the acquired set of structural parameters of the target external fixator and the measured branch length vector of the current group as common inputs; Based on the activation information, the basic layer algorithm module calls the enabled basic numerical iterative algorithm and basic intelligent optimization algorithm for parallel computation, and the basic neural network approximation method path is skipped; Basic numerical iterative algorithms, such as the Newton iteration method with central difference, start with a default or user-defined initial solution, construct the objective function using the common input during the iteration process, perform iterative calculations, and output the fourth basic pose solution. Basic intelligent optimization algorithms, such as particle swarm optimization, initialize the population within a preset global search space and perform iterative search according to the fitness function defined by the common input, outputting the fifth basic pose solution; For the obtained fourth and fifth basic pose solutions, i.e., the final candidate pose solutions, the inverse kinematics formula is used to calculate their respective theoretical branch length vectors; the matching error between each theoretical branch length vector and the measured branch length vector is calculated; the two matching errors are compared, and the final candidate pose solution with the smallest error value is selected as the final relative pose result corresponding to the measured branch length vector of this group, and the output is completed.

[0062] By disabling the neural network prediction and collaborative refinement paths, the algorithm relies solely on basic numerical iterative algorithms and basic intelligent optimization algorithms for parallel solution and competitive selection. This configuration simplifies the computation process, reduces dependence on pre-trained models and additional computational resources, i.e., the collaborative layer, and is suitable for simple scenarios with limited computing power or where the applicability of neural networks is questionable, or as a reliable backup solution. Although it may sacrifice some of the speed and accuracy advantages brought by the prediction-refinement collaboration, it still maintains robustness superior to a single algorithm through the parallel and verification mechanisms of two traditional algorithms.

[0063] Similarly, you can choose to disable numerical iteration or intelligent heuristic optimization algorithms, allowing the program to run with two basic layer algorithms; you can also disable both algorithms at the basic layer, allowing the program to run with only one algorithm; or disable the collaborative layer algorithm, allowing the program to run with three basic layer algorithms.

[0064] Example 3 The algorithm activation information in this embodiment is consistent with that in Embodiment 1. The enabled basic layer algorithms include: numerical iteration method, intelligent heuristic optimization algorithm, and neural network approximation method. Furthermore, the collaborative layer algorithm is enabled. The difference between this embodiment and Embodiment 1 is that the input part consists of support parameters, multiple sets of 6-link lengths, and whether they are continuous; the output part consists of multiple sets of pose results, i.e., solving multiple forward positron problems for the same support.

[0065] For multiple sets of 6-link lengths with no obvious relationship, the numerical iteration method needs to repeatedly iterate from the original custom 6-dimensional pose vector, while the intelligent heuristic optimization algorithm needs to repeatedly search in a large search space, and the neural network approximation method does not need to retrain the neural network. For multiple sets of 6-link lengths with obvious relationship, that is, the poses corresponding to these link lengths are approximately an arithmetic sequence or very close in some dimensions, the numerical iteration method uses the corresponding dimension result from the final iteration result of the previous set to replace the corresponding dimension data of the initial solution as the initial solution for iteration, while the intelligent heuristic optimization algorithm searches in a space centered on the corresponding dimension result from the final iteration result of the previous set with an appropriate radius, and the neural network approximation method does not need to retrain the neural network.

[0066] Example 4 The input portion and algorithm activation information in this embodiment are consistent with those in Embodiment 1. The difference lies in the fact that the parameters in the algorithm can be modified to meet a wide range of scenarios, such as personalized clinical treatment. The input portion of the solution method in this embodiment includes: stent parameters, one or more sets of 6-branch lengths, branch installation parameters, forward and reverse installation parameters, solution accuracy, etc. One or more sets of 6-branch lengths indicate that the solution method can handle one or more forward problems; branch installation parameters indicate that the solution method supports custom and asymmetric branch installation; forward and reverse installation parameters are used to adjust the installation order of the proximal and distal loops in the algorithm; solution accuracy is used to adjust the convergence accuracy of the algorithm. There are also custom functions such as custom internal solution algorithms. The output portion of the solution method includes: pose results and solution process information. The solution process information accurately records the information in the input portion, the magnitude of each parameter in the solution portion, the magnitude of intermediate variables in the solution, the number of iterations, and other information.

[0067] Therefore, this invention is applicable to external fixation devices of various specifications and installation methods, supporting personalized treatment scenarios, such as branched asymmetric installation for unilateral deformities and eccentric loads, and correction of foot and ankle deformities with complex anatomical structures and difficult forward solution methods, thus having a wide range of applications. It uses multiple algorithms for collaborative computation, balancing global search and local optimization, avoiding the shortcomings of single algorithms, improving the robustness of the computation process, reducing the probability of non-convergence, and ensuring the accuracy and reliability of the results. Furthermore, by using efficient and fast solution algorithms, computation time is shortened; the algorithm's operating logic can be flexibly set according to different needs, meeting the requirements for both coarse and accurate solutions, and can run smoothly on different computing devices.

[0068] Although embodiments are described and presented herein, those skilled in the art will readily conceive of various other means and / or structures for performing functions and / or obtaining results and / or one or more of the advantages described herein, and each of such variations and / or modifications is considered to be within the scope of the embodiments described herein. More generally, those skilled in the art will readily understand that all parameters, dimensions, materials, and configurations described herein are exemplary, and actual parameters, dimensions, materials, and / or configurations will depend on the specific application of the teachings of the invention for one or more specific applications therein. Those skilled in the art will recognize or be able to determine many equivalent forms of the particular embodiments described herein using only conventional experimentation. Therefore, it should be understood that the foregoing embodiments are given by way of example only, and that embodiments may be practiced in ways other than those specifically described and protected by the claims within the scope of the appended claims and their equivalents. For example, in other embodiments of the present invention, mature intelligent heuristic optimization algorithms such as ant colony optimization, bee colony optimization, and gray wolf optimization are used, along with other mature numerical iterative methods and neural network approximation methods, and combinations thereof; for example, in other embodiments of the present invention, common stent configurations for surgical fixation are used, as well as common types of rings and branches, and stents with three rings and two sets of branches are used; for example, in other embodiments of the present invention, external fixation stents are used for fixation in different locations such as the forearm, lower leg, and near joints; for example, in other embodiments of the present invention, the present invention is extended to parallel mechanisms in other fields, such as parallel mechanisms used in manufacturing, robotics, aerospace, and other fields, many of which are very similar to medical external fixation stents, and the solutions to forward kinetic problems are also very similar or even identical; for example, in other embodiments of the present invention, when selecting the optimal pose result from temporary pose results, not only the theoretical rod length corresponding to the pose result is considered, but also the possible deformations corresponding to the pose result are considered, and the optimal pose result is selected comprehensively. Therefore, each individual feature, system, article, material, kit, and / or method involved in the embodiments disclosed in the present invention can be combined. Furthermore, any combination of two or more such features, systems, articles, materials, kits, and / or methods is also included within the scope of this disclosure, provided that these features, systems, articles, materials, kits, and / or methods are not contradictory. All means or steps in the appended claims, plus corresponding structures, materials, operations, and equivalents of the functional elements, if any, are intended to include any structure, material, or operation for performing the function in combination with other claimed elements as specifically claimed. The invention has been described for purposes of illustration and description, but this description is not exhaustive or intended to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the invention.The embodiments were selected and described in order to best explain the principles and practical application of one or more aspects of the invention and to enable others of ordinary skill in the art to understand one or more aspects of the invention for various embodiments with various modifications suitable for the particular intended use.

[0069] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0070] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0071] Multiple components in the device are connected to an I / O interface, including: input units such as a keyboard, mouse, etc.; output units such as various types of displays, speakers, etc.; storage units such as disks, optical disks, etc.; and communication units such as network interface cards, modems, wireless transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks. The processing unit performs the various methods and processes described above, such as the method of the present invention. For example, in some embodiments, the method of the present invention may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or the communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the method of the present invention described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute the method of the present invention by any other suitable means (e.g., by means of firmware).

[0072] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0073] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0074] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0075] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for determining the pose of an external fixation device using a multi-algorithm collaborative approach, characterized in that, Running in a processing system with pre-installed multiple pose solving algorithms, the specific steps include: S1. Obtain the structural parameter set and measured branch length of the target external fixation bracket; the structural parameter set includes proximal ring parameters, distal ring parameters, branch parameters, and installation position parameters of each branch on the proximal and distal rings; S2. Obtain the algorithm activation information of the basic layer and the collaborative layer in the processing system configured by the user, call the basic algorithms that have been enabled in the basic layer algorithm module, and perform parallel computation with the structural parameter set and the measured branch length of the current group as common inputs; the basic algorithms include: basic neural network approximation method, basic numerical iterative algorithm and basic intelligent optimization algorithm, wherein at least one of the basic numerical iterative algorithm and basic intelligent optimization algorithm is enabled, and each enabled basic algorithm independently outputs a corresponding basic pose solution; S3. If the basic neural network approximation method is enabled and the cooperative algorithm in the cooperative layer algorithm module is enabled, the basic pose solution output by the basic neural network approximation method is input into the enabled cooperative algorithm for refinement, and the refined pose solution is used as the final candidate pose solution of the current algorithm; otherwise, the basic pose solution of the basic layer is directly used as the final candidate pose solution of the corresponding path. S4. For each final candidate pose solution, calculate the corresponding theoretical branch length vector through inverse kinematics, and compare it with the measured branch length vector of the current group. Select the final candidate pose solution with the smallest matching error as the final relative pose result corresponding to the measured branch length vector of the current group.

2. The method for determining the pose of an external fixation device using a multi-algorithm collaborative approach according to claim 1, characterized in that, The cooperative algorithm includes a cooperative numerical iterative algorithm and a cooperative intelligent optimization algorithm, with at most one algorithm enabled in the cooperative algorithm; both the basic numerical iterative algorithm and the cooperative numerical iterative algorithm are Newton's iterative method for calculating the Jacobian matrix using the central difference method; both the basic intelligent optimization algorithm and the cooperative intelligent optimization algorithm are particle swarm optimization algorithms or genetic algorithms; When the activated cooperative algorithm is the cooperative numerical iterative algorithm, the basic pose solution output by the basic neural network approximation method is used as the initial solution for iteration of the cooperative numerical iterative algorithm for iterative calculation. When the cooperative algorithm is the cooperative intelligent optimization algorithm, a local search space is constructed with the basic pose solution output by the basic neural network approximation method as the center, and optimization search is performed in the local search space.

3. The method for determining the pose of an external fixation device using a multi-algorithm collaborative approach according to claim 1, characterized in that, The basic neural network approximation method is implemented based on a pre-trained neural network model, which is obtained through the following steps: Based on the proximal loop parameters, distal loop parameters, and branch parameters in the structural parameter set, the sampling space of the pose parameters is determined; Multiple pose samples are randomly generated within the sampling space, and the theoretical branch length vector corresponding to each pose sample is obtained by inverse kinematics calculation, forming multiple training sample pairs. Each training sample pair includes a pose sample and the corresponding theoretical branch length vector. The initial neural network is trained using the multiple training samples to obtain the pre-trained neural network model.

4. The method for determining the pose of an external fixation device using a multi-algorithm collaborative approach according to claim 3, characterized in that, The pre-trained neural network model includes a backpropagation neural network and a radial basis function neural network; the backpropagation neural network includes multiple parallel sub-networks, each of which is used to predict one dimension of the final relative pose result; the radial basis function neural network uses Gaussian radial basis functions as activation functions.

5. The method for determining the pose of an external fixation device using a multi-algorithm collaborative approach according to claim 1, characterized in that, The measured branch lengths obtained in S1 are in multiple sets, and corresponding continuous flags indicating whether there is a continuous pose change relationship between the multiple sets of data are obtained. When the continuous flag indicates a continuous pose change relationship, during the execution of S2 for the latter set of measured branch lengths: If the basic numerical iteration algorithm is enabled, the final relative pose result output in S4 for the previous set of data is used as the initial solution for the iteration when solving for the current set of data, or as the value of the corresponding dimension in the initial solution. If the basic intelligent optimization algorithm is enabled, the final relative pose result output in S4 for the previous set of data is used as the center of the search space when solving for the current set of data.

6. The method for determining the pose of an external fixation device using a multi-algorithm collaborative approach according to claim 1, characterized in that, The structural parameter set also includes the installation order identifiers of the proximal and distal rings; in step S4, the correspondence between the proximal coordinate system and the distal coordinate system used for inverse kinematics calculation is determined based on the installation order identifiers.

7. The method for determining the pose of an external fixation device using a multi-algorithm collaborative approach according to claim 1, characterized in that, The mounting position parameters of the branches on the proximal and distal rings support asymmetrical arrangement. Specifically, the mounting position parameters are the hole numbers on the rings to which each branch is connected. The ring structure of the external fixing bracket is a full ring, a 2 / 3 ring, or a U-shaped ring. The joint types of the branches include ball joint-sliding joint-ball joint, revolute joint-sliding joint-ball joint, and revolute joint-sliding joint-revolute joint.

8. The method for determining the pose of an external fixation device using a multi-algorithm collaborative approach according to claim 1, characterized in that, In S4, the matching error is obtained by calculating the Euclidean distance or mean absolute difference between the theoretical branch length vector and the measured branch length vector.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images