Negative Poisson's ratio stent structure for heart and cerebral vessels and optimization method of negative Poisson's ratio stent structure

Abstract

The invention discloses a negative Poisson's ratio stent structure for heart and cerebral vessels and an optimization method of the negative Poisson's ratio stent structure, belongs to the technical field of medical instruments, and solves the problems of axial shortening, contradiction between radial supporting force and flexibility and low calculation efficiency in an optimization process of a traditional vascular stent due to a positive Poisson's ratio effect. Wherein a bracket structure of a curved-edge concave hexagonal geometric unit is adopted, and the bracket synchronously extends in the axial direction to present a negative Poisson's ratio effect when being subjected to an expansion load in the radial direction; degradable macromolecules or degradable metals are selected as the material of the stent; according to the optimization method, candidate design is generated through parametric modeling, performance indexes are obtained through finite element simulation, a neural network agent model is trained to predict multi-target performance, a Pareto optimal solution set is obtained through optimization by adopting a multi-target optimization algorithm, and manufacturing parameters are determined through finite element review. According to the invention, the implantation precision and the positioning accuracy of the stent are improved, the mechanical property and the biocompatibility are optimized, and the design efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the field of medical device technology, and in particular relates to a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular diseases and its optimization method. Background Technology

[0002] Cardiovascular and cerebrovascular diseases, especially coronary heart disease and ischemic stroke, have become one of the leading causes of death worldwide. Their pathological basis is atherosclerosis, which leads to narrowing or occlusion of blood vessels, thereby affecting blood supply to distal tissues. With the continuous development of medical technology, vascular stents, as an important means of treating atherosclerosis, have been widely used in clinical practice. However, traditional vascular stents (such as drug-eluting stents and metallic stents) still have insurmountable limitations in their application, significantly affecting treatment outcomes. Traditional stents often exhibit a positive Poisson's ratio effect, meaning they contract laterally during axial stretching and expand laterally during axial compression. This results in axial shortening during stent expansion, directly affecting the accuracy of stent positioning and preventing complete coverage of the lesion area, thus impacting treatment effectiveness. Furthermore, there is an inherent contradiction between the radial support force and flexibility of traditional stents. To improve radial support force, stents are usually designed to be relatively large, resulting in poor flexibility in tortuous blood vessels, making it difficult to perfectly conform to the vessel wall. This may exert excessive stress on the vessel wall, leading to intimal hyperplasia and ultimately restenosis.

[0003] To address these issues, researchers have proposed a negative Poisson's ratio stent structure design. The negative Poisson's ratio effect refers to the phenomenon where a material, under stress, elongates in one direction while simultaneously contracting in the perpendicular direction. By designing a negative Poisson's ratio structure, the stent can achieve simultaneous axial expansion during radial expansion, effectively avoiding the axial shortening problem of traditional stents, improving implantation precision and positioning accuracy, improving stress distribution on the vessel wall, and reducing the risk of intimal hyperplasia. However, despite the significant potential of negative Poisson's ratio materials in stent design, structural optimization still faces numerous challenges. Currently, stent design optimization primarily relies on traditional finite element analysis (FEM), simulating different design parameters to obtain various performance indicators. However, traditional FEM often requires substantial computation time, especially when dealing with complex structures, making it difficult to efficiently obtain optimized design parameters. Therefore, combining advanced machine learning and artificial intelligence technologies, particularly neural networks as powerful nonlinear fitting tools, can learn the complex relationships between design parameters and performance indicators from large amounts of simulation data, providing a novel solution for stent design optimization. By establishing a rapid surrogate model between design parameters and performance indicators, and combining it with multi-objective optimization algorithms, the negative Poisson's ratio stent can maximize key performance aspects such as radial support force, compliance, and positioning accuracy. Furthermore, with the development of materials science, the application of biodegradable materials has gradually become a research hotspot. These biodegradable materials can gradually degrade and be absorbed by the body after fulfilling their support function, avoiding the long-term foreign body reaction and restenosis problems that may occur with traditional metal stents. Therefore, negative Poisson's ratio stents using biodegradable materials not only provide ideal mechanical properties but also achieve better biocompatibility and long-term efficacy. In summary, the design optimization of negative Poisson's ratio stents, by combining novel materials, advanced mechanical models, and modern computational techniques, provides a more precise and efficient solution for the treatment of cardiovascular and cerebrovascular diseases. However, existing stent designs still have shortcomings in optimization, and there is an urgent need to further improve the overall performance of stents through more efficient and intelligent methods to meet the needs of clinical treatment. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular systems and its optimization method, which can improve the implantation precision and positioning accuracy of the stent.

[0005] To achieve the above objectives, the present invention provides a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular systems, comprising a deployable mesh tubular body composed of multiple structural unit cells, wherein the structural unit cells adopt curved concave hexagonal geometric units; when the stent is subjected to radial expansion load, the curved concave hexagonal geometric units undergo lateral stretching, resulting in axial synchronous elongation, and the whole exhibits a negative Poisson's ratio effect.

[0006] Optionally, the geometric parameters of the structural unit cell include arm length, concavity angle, arm width, and axial length; the concavity angle is greater than 0 degrees and not greater than 70 degrees; the arm width is not less than 0.05 times the arm length and not greater than 0.3 times the arm length; and adjacent units in the circumferential and axial directions are connected through an integrated design.

[0007] Optionally, the deployable mesh tubular body is formed by curling a single-layer two-dimensional negative Poisson's ratio structure and connecting it at the unit nodes; the support material is selected from at least one of biodegradable polymers and biodegradable metals; the biodegradable polymer is L-polylactic acid or its copolymer; the biodegradable metal is magnesium alloy or high-purity WE43 magnesium alloy.

[0008] On the other hand, to achieve the above objectives, the present invention also provides an optimization method for a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular systems, comprising: Candidate designs are generated by using parameterized model design parameters as independent variables; For each candidate design, finite element simulation is performed to obtain performance indicators; A neural network agent model is trained based on performance metrics, and multi-objective prediction values ​​are output. The neural network surrogate model is optimized using a multi-objective optimization algorithm to obtain the Pareto optimal solution set; The solutions in the Pareto optimal solution set are verified by finite element analysis to determine the manufacturing parameters.

[0009] Optionally, the process of generating candidate designs includes: The parametric model uses arm length, concave angle, arm width, and axial length as independent variables; The optimal Latin hypercube sampling method is used to generate a candidate design set. The arm length, concave angle, arm width, and axial length are the geometric parameters of the negative Poisson's ratio scaffold structure unit cell, which is a curved concave hexagonal geometric unit.

[0010] Optionally, the process of training a neural network agent model includes: A multi-output deep feedforward neural network is used as a surrogate model; The input to the multi-output deep feedforward neural network are design parameters that have been dimensionless and standardized, and the outputs are predicted values ​​and uncertainty estimates of negative Poisson's ratio, radial support force, bending compliance, surface area coverage, and support retention rate during degradation. The multi-output deep feedforward neural network is integrated from multiple sub-models, each sub-model including two to four hidden layers, each hidden layer including sixty-four to two hundred and fifty-six neurons, using the ReLU activation function and batch normalization or layer normalization; The loss function is a weighted multi-objective loss function, which includes weighted error terms and heteroscedasticity uncertainty terms for each performance index.

[0011] Optionally, the process of integrating multiple sub-models includes: The number of the multiple sub-models is five to nine, and they are constructed through different random initializations and bootstrap sampling. The ensemble output is the mean of the outputs of each sub-model as the final prediction, and the variance is used as the uncertainty measure. The uncertainty estimation includes random uncertainty and cognitive uncertainty. Random uncertainty is estimated by predicting the noise level of each target through the variance head of the neural network, and cognitive uncertainty is estimated by estimating the dispersion of the mean of the same input and output by each sub-model.

[0012] Optionally, the optimization process using a multi-objective optimization algorithm includes: A multi-objective adaptive guided differential evolution algorithm is used for optimization. The comprehensive optimization objectives include maximizing the negative Poisson's ratio, maximizing radial support force, maximizing bending compliance, minimizing surface area coverage, and maximizing the smoothness and matching degree of the support retention curve during the degradation period. Constraints are imposed, including the upper limit of the outer diameter of the delivery system, the lower limit of the inner diameter of the expansion system, and the lower limit of the high-cycle fatigue life under cardiac cycle load.

[0013] Technical Effects of this Invention: This invention discloses a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular systems and its optimization method. It employs a stent structure with a negative Poisson's ratio effect and biodegradable materials, improving the stent's radial support force, flexibility, and biocompatibility. This avoids the axial shortening problem of traditional stents during expansion, thereby enhancing implantation precision and positioning accuracy, effectively ensuring precise coverage of the lesion area. By combining a neural network surrogate model with a multi-objective optimization algorithm, the relationship between design parameters and performance indicators is quickly established, and efficient multi-objective optimization is performed based on this. This method significantly reduces the computation time required for traditional finite element simulation analysis, improves the efficiency of the design process, and simultaneously maximizes the optimization of key stent performance aspects such as radial support force, flexibility, and positioning accuracy, providing a more precise and intelligent solution for stent design. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating an optimization method for a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular diseases according to an embodiment of the present invention. Figure 2This is a design process diagram of a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular applications according to an embodiment of the present invention. Detailed Implementation

[0015] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0016] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0017] like Figure 2 As shown, this embodiment provides a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular applications, which is composed of deployable mesh tubular bodies. The structural unit cell adopts a curved concave hexagonal geometric unit, and the unit arm length is defined. l Concave angle θ arm width t axial length h When the stent is subjected to radial expansion load, the units undergo lateral stretching, resulting in synchronous axial elongation. This exhibits a negative Poisson's ratio effect, thereby reducing axial shortening during expansion and improving fit and support uniformity in tortuous vessels. The geometric parameters of the concave hexagonal units satisfy… 0° < θ ≤ 70° , 0.05l≤t≤0.3l Furthermore, adjacent circumferential and axial units are connected through an integrated design to reduce stress concentration and improve fatigue life. The tubular body is formed by curling a single-layer two-dimensional negative Poisson's ratio structure and connecting it at the unit nodes to enhance radial support and suppress torsional instability. The scaffold material is selected from at least one of biodegradable polymers and biodegradable metals; the biodegradable polymer is polylactic acid (PLLA) or its copolymers, and the biodegradable metal is magnesium alloy or high-purity WE43 magnesium alloy, which can be loaded with anti-proliferation drugs to form a biodegradable drug-eluting scaffold. The scaffold is prepared by 3D printing, additive manufacturing, or micro-laser processing. Additive manufacturing preferably uses selective laser melting (SLM) or selective laser sintering, followed by support removal, heat treatment, and surface finishing to obtain the target roughness and dimensional accuracy; micro-laser processing preferably involves patterned cutting on thin-walled tubing followed by shape-setting annealing. When balloon dilatation reaches the target inner diameter range, the stent axial elongation rate is 0% to +8%, preferably +1% to +5%; the stent radial recoil rate is no greater than 6%, and the outer and inner arc support distribution is maintained uniformly in a tortuous vessel model with a radius of curvature of 3 to 10 mm. Balloon dilatation is an intravascular procedure performed through a catheter with a balloon to expand and open narrowed or blocked vessels, ensuring that the stent can expand in the appropriate position and support the vessel.

[0018] like Figure 1 As shown, this embodiment also provides an optimization method for negative Poisson's ratio stent structures used in cardiovascular and cerebrovascular systems, including: utilizing a parameterized model to [ l, θ, t, h Candidate designs are generated by optimal Latin hypercube sampling with [ ] as independent variables; finite element simulations are performed on each candidate design to obtain performance indicators such as negative Poisson's ratio, radial support force, bending compliance, surface area coverage, and support retention rate during degradation; based on the data, a neural network surrogate model is trained to output multi-objective predicted values ​​to replace high-cost simulations.

[0019] The preferred proxy model is a multi-output deep feedforward neural network (MDNN), which is suitable for achieving high-precision, low-overhead multi-objective performance prediction based on finite element offline datasets. The specific steps are as follows: (1) The input is the design parameters after dimensionless and standardized processing [ l, θ, t, h ], output the predicted values ​​of negative Poisson ratio, radial support force, bending compliance, surface area coverage and support retention rate during degradation and their uncertainty estimates; (2) Each sub-model is a multi-layer feedforward network, containing 2 to 4 hidden layers, each layer with 64 to 256 neurons, using the activation function ReLU and batch normalization / layer normalization, the loss function is weighted multi-objective loss, including the weighted error term and heteroscedasticity uncertainty term of each index; (3) The number of ensemble sub-models is M (preferably M = 5 to 9), constructed through different random initialization and bootstrap sampling, the ensemble output takes the mean of the output of each sub-model as the final prediction, and the variance is used as the uncertainty measure; (4) The training data is divided into training set, validation set and test set in 8:1:1, early stopping and L2 regularization are used to suppress overfitting, and the determination coefficient of the validation set and test set is used. R ² Not less than 0.95 and 0.92 respectively, and the physical quantities such as radial support force and surface area coverage are restricted to non-negative physical feasibility constraints and within the preset range during the training process; (5) Under the condition of meeting the above accuracy and physical feasibility requirements, the surrogate model realizes the explicit calculation of the sensitivity of the design parameters by outputting gradients, so as to guide the search direction and weight adaptation of the subsequent evolutionary algorithm.

[0020] Multi-output deep feedforward neural networks do not require explicit solution of complex elasticity and degradation coupling equations, exhibiting high training stability, strong generalization ability to variations in geometric and load boundary conditions, and fast computation speed. Their mathematical expression is as follows: (1) Wherein, input vector The output is K Each performance indicator is denoted as... ; For the first Hidden representation vectors of layers; For the first Layer weight matrix and bias vector; L The number of hidden layers; It is a non-linear activation function; The predicted mean for each objective; Let be the weight matrix, connecting the first... L Output of hidden layer Output the mean / variance vector; This is the bias vector, which is used to shift the output mean / variance vector. For the prediction noise variance (random uncertainty) of each target, use softplus Ensure the value is positive.

[0021] Uncertainty estimation addresses random uncertainties arising from noise / irreducible errors in the data itself (such as simulation numerical truncation errors, grid scale differences, measurement noise, etc.). The variance head of a neural network is used to directly predict the noise level for each target. Cognitive uncertainty stems from insufficient model cognition (incomplete data coverage, parameter uncertainty), and is estimated using ensemble training (training multiple randomly initialized / bootsampling sub-models): the dispersion of the mean of each sub-model with respect to the same input and output is the cognitive uncertainty. The specific calculation formula is as follows: (2) in, M The number of integrated sub-models; k For the first k Index of output targets / metrics; x For input vectors ; For the first m The sub-model for the first k The predicted mean of each indicator; For the integrated first k Predicted average of each indicator; For the first m The sub-model for the first k The prediction noise variance of each indicator; The total variance is composed of two parts: the average random uncertainty and the cognitive uncertainty of the aggregate variance.

[0022] A multi-objective loss function is used, applying multi-objective weighting and heteroscedastic Gaussian negative log-likelihood to a single sub-model, further superimposed with physical feasibility constraints and... L2 Regular expressions, in their specific mathematical form, are as follows: (3) in, N Total sample size; i For sample index; No. k The weight of each task; For the sample i The k The true value of the indicator; For the model to sample i The k The predicted mean and the predicted noise variance of the indicator; c For constrained indexes; For the first c The penalty coefficient for each constraint; For the model to sample The predicted value; For the first c A constraint function is applied to the model's predicted values. ; This is the weight decay coefficient; For the first Layer weight matrix; This represents the total loss.

[0023] The multi-objective adaptive guided differential evolution algorithm (MOAGDE) was used for optimization. The comprehensive optimization objectives included (1) maximizing the negative Poisson ratio, (2) maximizing the radial support force, (3) maximizing the bending compliance or minimizing the bending stiffness, (4) minimizing the surface area coverage, and (5) maximizing the smoothness and matching degree of the support retention curve during the degradation period. Constraints were applied, including the upper limit of the outer diameter of the transport vessel, the lower limit of the inner diameter of the expansion vessel, and the lower limit of the high-cycle fatigue life under cardiac cycle load. The Pareto optimal solution set was obtained by coupling the MDNN surrogate model and the MOAGDE algorithm. The selected solutions were then verified by high-precision finite element analysis to finally determine the manufacturing parameters and use them for structural fabrication and experimental verification. The specific optimization steps of the multi-objective adaptive guided differential evolution algorithm are as follows: (1) Initialize the population: randomly generate initial population individuals, each individual containing multiple objective function values; (2) Fitness evaluation: evaluate the multi-objective fitness of each individual; (3) Adaptive parameter adjustment: dynamically adjust the parameters of the algorithm, such as mutation factor, crossover probability, etc., to adapt to the current search requirements; (4) Mutation and crossover: use differential mutation operation to generate new candidate solutions, and generate new individuals through crossover operation; (5) Pareto front update: select individuals with better fitness through Pareto dominance criterion and update the Pareto front solution set; (6) Select the optimal solution: select the best performing individual in each generation for retention; (7) Termination condition check: determine whether the maximum number of iterations or the accuracy requirement has been met, if so, end the algorithm; otherwise, return to step 2.

[0024] This invention discloses an optimization method for negative Poisson's ratio stent structures. It employs a stent structure with a negative Poisson's ratio effect and biodegradable materials, improving the stent's radial support force, flexibility, and biocompatibility. This avoids the axial shortening problem of traditional stents during expansion, thereby enhancing implantation precision and positioning accuracy, effectively ensuring precise coverage of the lesion area. By combining a neural network surrogate model with a multi-objective optimization algorithm, the relationship between design parameters and performance indicators is quickly established, and efficient multi-objective optimization is performed based on this. This method significantly reduces the computation time required for traditional finite element simulation analysis, improves the efficiency of the design process, and simultaneously maximizes the optimization of key stent performance aspects such as radial support force, flexibility, and positioning accuracy, providing a more precise and intelligent solution for stent design.

[0025] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A negative Poisson's ratio stent structure for cardiovascular and cerebrovascular, characterized in that, The negative Poisson's ratio stent structure for cardiovascular system comprises an expandable grid tube, which is composed of a plurality of structural unit cells adopting curved-side concave hexagonal geometric units; when the stent is subjected to an expansion load in the radial direction, the curved-side concave hexagonal geometric units are stretched transversely to cause synchronous axial elongation, and the whole exhibits a negative Poisson's ratio effect.

2. The negative Poisson's ratio stent structure for cardiovascular system according to claim 1, wherein the geometric parameters of the structural unit cell include an arm length, a concave angle, an arm width and an axial length; the concave angle is greater than 0 degrees and not greater than 70 degrees; the arm width is not less than 0.05 times the arm length and not greater than 0.3 times the arm length; and the circumferential and axial adjacent units are connected through integrated design.

3. The negative Poisson's ratio stent structure for cardiovascular system according to claim 1, wherein the expandable grid tube is formed by rolling a single-layer two-dimensional negative Poisson's ratio structure and connecting at the unit nodes; the stent material is selected from at least one of a degradable polymer and a degradable metal; the degradable polymer is poly-L-lactic acid or a copolymer thereof; and the degradable metal is a magnesium alloy or high-purity WE43 magnesium alloy. The optimization method for implementing the negative Poisson's ratio stent structure for cardiovascular system according to any one of claims 1-3 comprises: generating candidate designs by taking the design parameters of the parametric model as independent variables; performing finite element simulation on each candidate design to obtain performance indicators; 4. An optimization method for a cardiovascular negative Poisson's ratio stent structure, characterized in that, training a neural network surrogate model based on the performance indicators to output multi-objective prediction values; optimizing the neural network surrogate model using a multi-objective optimization algorithm to obtain a Pareto optimal solution set; performing finite element rechecking on the solutions in the Pareto optimal solution set to determine manufacturing parameters.

5. The optimization method for the negative Poisson's ratio stent structure for cardiovascular system according to claim 4, wherein the process of generating candidate designs comprises: the parametric model takes the arm length, the concave angle, the arm width and the axial length as independent variables; an optimal Latin hypercube sampling method is used to generate a candidate design set; wherein the arm length, the concave angle, the arm width and the axial length are geometric parameters of the negative Poisson's ratio stent structure unit cell, and the structural unit cell is a curved-side concave hexagonal geometric unit.

6. The optimization method for the negative Poisson's ratio stent structure for cardiovascular system according to claim 4, wherein the process of training the neural network surrogate model comprises: a multi-output deep feedforward neural network is used as the surrogate model; the multi-output deep feedforward neural network takes dimensionless and standardized design parameters as input and outputs prediction values and uncertainty estimates of the negative Poisson's ratio, the radial support force, the bending flexibility, the surface area coverage rate and the support retention rate during the degradation period; the multi-output deep feedforward neural network is integrated by a plurality of sub-models, each sub-model includes two to four hidden layers, each hidden layer includes sixty-four to two hundred and fifty-six neurons, and uses a ReLU activation function and batch normalization or layer normalization; the loss function is a weighted multi-objective loss function, which includes weighted error terms and heteroscedastic uncertainty terms of each performance indicator. ​ ​ ​ ​ ​ 7. The method for optimizing a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular diseases as described in claim 6, characterized in that, The process of integrating multiple sub-models includes: The number of the multiple sub-models is five to nine, and they are constructed through different random initializations and bootstrap sampling. The ensemble output is the mean of the outputs of each sub-model as the final prediction, and the variance is used as the uncertainty measure. The uncertainty estimation includes random uncertainty and cognitive uncertainty. Random uncertainty is estimated by predicting the noise level of each target through the variance head of the neural network, and cognitive uncertainty is estimated by estimating the dispersion of the mean of the same input and output by each sub-model.

8. The method for optimizing a negative Poisson's ratio stent structure for cardiovascular and cerebrovascular diseases as described in claim 4, characterized in that, The process of finding the optimal solution using a multi-objective optimization algorithm includes: A multi-objective adaptive guided differential evolution algorithm is used for optimization. The comprehensive optimization objectives include maximizing the negative Poisson's ratio, maximizing radial support force, maximizing bending compliance, minimizing surface area coverage, and maximizing the smoothness and matching degree of the support retention curve during the degradation period. Constraints are imposed, including the upper limit of the outer diameter of the delivery system, the lower limit of the inner diameter of the expansion system, and the lower limit of the high-cycle fatigue life under cardiac cycle load.

Images