Absorbable bioprosthetic valve design method based on artificial intelligence and biomechanics

Abstract

The invention provides an absorbable bioprosthetic valve design method based on artificial intelligence and biomechanics. The absorbable bioprosthetic valve design method comprises the following steps: S1, constructing a performance demand system of a valve; s2, constructing a three-dimensional reconstruction model of the original lesion structure of the human body and a parameterized structure model of the valve; s3, establishing a first deep learning model for simulating interventional mechanical behaviors of the valve; s4, constructing a second deep learning model for predicting the relationship between the stress and the material degradation behavior in the long-term use process of the valve; s5, embedding the first deep learning model and the second deep learning model into a reinforcement learning framework; and S6, carrying out simulation and experimental verification based on a design result in the step S5, and evaluating the adaptability and reliability of the design parameters under the function requirements defined in the step S1. According to the method, the biomechanical theory and the artificial intelligence algorithm are fused, the structural parameter interpretability of the absorbable bioprosthetic valve design and the automation of the whole process are realized, and a new path is provided for the intelligent optimization design of the implantable interventional medical device with a complex structure.

Description

Technical Field

[0001] This invention belongs to the field of implantable medical device design technology, specifically relating to a design method for absorbable bio-valve based on artificial intelligence and biomechanics. Background Technology

[0002] As an important type of cardiovascular implantable medical device, absorbable bioprosthetic valves play a significant role in clinical treatment. In the treatment of valvular heart disease, these biodegradable valves, after implantation, can assist circulation and provide necessary mechanical support to promote tissue regeneration. After achieving their function, they are gradually absorbed by the body, thus avoiding the need for secondary surgery.

[0003] Due to the complex structure of absorbable bioprosthetic valves and their highly dynamic physiological effects, including continuous blood flow and cyclical pressure within the body, assessing their long-term lifespan and performance stability presents a significant challenge during the design and development phase. Currently, the design and performance evaluation of such bioprosthetic valves primarily rely on finite element simulation (FEM), which involves establishing a finite element model of the valve structure to simulate its mechanical response under different physiological loads, such as varying blood pressure and blood flow conditions, providing a basis for design optimization. However, FEM simulation is computationally complex and time-consuming, leading to low efficiency in iterative optimization. Furthermore, simulation analysis requires experienced professionals to adjust model parameters, relying on repeated trial and error, making it difficult to integrate into automated design processes and meet the needs for rapid optimization and multi-scheme comparison. In addition, existing absorbable bioprosthetic valve design processes lack sufficient consideration of long-term degradation behavior, thus affecting the accuracy of evaluation results.

[0004] In recent years, the development of artificial intelligence technology, especially deep learning and reinforcement learning, has provided new ideas for the design and optimization of absorbable bioprosthetic valves. Training deep learning models can replace some of the time-consuming and costly simulation calculations. However, how to combine artificial intelligence technology with biomechanical simulation to establish a systematic design method for absorbable bioprosthetic valves, improve model inference speed and design evaluation efficiency while ensuring accuracy, and ultimately form an intelligent design process, remains an unsolved technical challenge in this field.

[0005] Training deep learning models using data accumulated through traditional methods such as finite element simulation can replace some of the high-cost simulation calculations, rapidly predicting valve performance under different design parameters while maintaining a certain level of accuracy, significantly improving model inference speed and design evaluation efficiency. Furthermore, embedding this model into a reinforcement learning framework enables end-to-end joint optimization of valve structure and performance; that is, the AI ​​agent automatically adjusts design parameters to gradually approach the optimal solution, ultimately forming an intelligent design process. However, how to combine artificial intelligence technology with biomechanical simulation to establish a systematic method for designing absorbable bioprosthetic valves, truly leveraging these advantages, remains an unsolved technical challenge in this field. Summary of the Invention

[0006] This application addresses the technical problems of existing absorbable bioprosthetic valve design optimization, such as computational complexity, long iteration cycles, difficulty in efficiently evaluating and optimizing a large number of design schemes, reliance on repeated manual trial and error for parameter adjustment, resulting in low efficiency in the valve design process and a lack of sufficient consideration for long-term degradation behavior. Therefore, it provides an absorbable bioprosthetic valve design method based on artificial intelligence and biomechanics.

[0007] The technical solution adopted by this application to solve the above-mentioned technical problems is as follows: The design method for absorbable bioprosthetic valves based on artificial intelligence and biomechanics includes the following steps: S1: Constructing a performance requirement system for the valve, decomposing the requirements into functional indicators in the functional domain, and mapping them into structural parameters and material parameters in the physical domain; S2: Acquiring the patient's clinical medical imaging data, and using deep learning methods to construct a three-dimensional reconstruction model of the original pathological structure of the human body and a parameterized structural model of the valve, respectively. The parameterized structural model responds to the structural parameters defined in step S1; S3: Establishing a first deep learning model to simulate the interventional mechanical behavior of the valve; the training data of the first deep learning model comes from the finite element simulation results, using the structural model generated in step S2 and the material parameters defined in S1 as inputs, and the core performance indicators extracted from the finite element simulation as outputs, to predict the mechanical response of the valve during intervention; S4: Constructing a pre- A second deep learning model is used to measure the relationship between stress and material degradation behavior during long-term valve use. The training data for the second deep learning model comes from finite element simulation and in vitro experiments to characterize the change in the degradation rate of the material under stress and the resulting evolution of the valve's mechanical properties. It predicts whether the structural and material parameters defined in step S1 meet the requirements. S5: The first and second deep learning models are embedded into a reinforcement learning framework. The reinforcement learning algorithm is used to automatically iteratively optimize the structural and material parameters of the valve, thereby optimizing the interventional performance, mechanical performance, and biodegradability of the valve, achieving end-to-end efficient optimization design from anatomical model input to design parameter output. S6: Based on the design results of S5, simulation and experimental verification are carried out to evaluate the adaptability and reliability of the design parameters under the functional requirements defined in S1.

[0008] The structural parameters of the valve include the outer diameter of the stent after deployment, the stent length, the thickness of the stent support structure, the number and distribution of stent segments, the stent mesh size, the geometric parameters of the stent support ribs and connecting ribs, and the number, thickness, and shape and size parameters of the valve leaflets. The material parameters of the valve include the composition and content of each component of the stent material, the molecular weight distribution of the material, the Young's modulus, Poisson's ratio, tensile strength, and the degradation rate of the material. The performance indicators of the valve include the radial recoil rate, axial shortening rate after interventional implantation, the radial stiffness of the valve under cyclic loading, the stress safety factor, and the fatigue life, as well as the degradation rate, mechanical strength retention rate, and time required for complete degradation of the material during its service life.

[0009] The clinical medical imaging data used in step S2 includes computed tomography (CT) images and / or magnetic resonance imaging (MRI) images; the three-dimensional reconstruction model of the original human lesion structure includes a three-dimensional morphological model of the aorta, aortic valve leaflets, valve annulus, left ventricular outflow tract, and calcified plaque area; the parametric structural model of the valve includes at least three leaflet structures and a support ring structure.

[0010] The parametric structural model also includes a skirt anchoring structure. The parametric structural model is generated based on structural design parameters and spatially fits with the three-dimensional reconstruction model of the original human lesion structure. The valve leaflet structure is defined by a polynomial parametric surface. The support ring structure is generated using rotational geometry and spatially matched with the aortic valve annulus region through Boolean operations or a fitting method.

[0011] In step S3, the predicted mechanical response of the valve during the intervention process includes at least one of the following: valve structural stress safety, anchoring stability, opening and closing functionality, and occlusion coordination.

[0012] The first deep learning model is a regression-type deep neural network model, which is used to approximate the mechanical behavior of valve stents during interventional implantation. The deep learning model is trained on finite element simulation data of valve stent interventional processes under different structural and material parameters to predict the mechanical response index of the stent.

[0013] The mechanical response indicators of the stent include stress-strain distribution during expansion, radial support force, and deformation recovery rate after intervention.

[0014] In step S4, it is predicted whether the structural and material parameters defined in step S1 meet the requirements for material performance evolution and functional continuity.

[0015] The second deep learning model is trained by combining historical simulation and degradation experimental data to simulate changes in valve structure and performance caused by material degradation, in order to evaluate the valve's performance and usage risks throughout its entire life cycle.

[0016] The first deep learning model and the second deep learning model include, but are not limited to, convolutional neural networks, recurrent neural networks, Transformers or other deep neural network structures, and the algorithm type is not limited to a specific architecture.

[0017] The reinforcement learning framework in step S5 adopts an agent-environment interaction mode, in which the agent generates a valve structure design scheme based on the current design parameters.

[0018] The reinforcement learning algorithm is not limited to a specific type; it can be any reinforcement learning algorithm, such as deep Q-network or Actor-Critic, to achieve the optimization search of design parameters.

[0019] In step S6, a valve-heart chamber-vascular coupling simulation model is constructed and simulations of stent deployment, opening and closing cycles, and material degradation are performed; a valve prototype is prepared and structural matching tests, opening and closing function tests, and in vitro degradation experiments are carried out to comprehensively evaluate the adaptability and reliability of the design parameters under the multidimensional functional requirements defined in S1.

[0020] This application uses simulation and experimental verification to systematically evaluate the performance of optimized design parameters under typical operating conditions. The simulation part involves constructing a valve-heart chamber-blood vessel coupling model. The experimental part involves preparing a valve prototype based on the optimized structural and material parameters, verifying structural matching through an implantation anatomical simulation model, conducting opening and closing performance tests using a simulated circulation device, and obtaining the trend of material mechanical parameter changes during the service life through in vitro material degradation experiments. These methods are used to verify the degree to which the design scheme meets the functional requirements and to perform closed-loop calibration of the aforementioned simulation model.

[0021] This application constructs a valve-heart chamber-vascular coupling simulation model and performs simulations of stent deployment, opening and closing cycles, and material degradation. A valve prototype is fabricated and structural matching tests, opening and closing function tests, and in vitro degradation experiments are conducted to comprehensively evaluate the adaptability and reliability of the design parameters under multi-dimensional functional requirements. In step S6, simulation and experimental verification are used to systematically evaluate the performance of the optimized design parameters under typical operating conditions. The simulation part involves constructing a valve-heart chamber coupling model and sequentially conducting multi-field simulations of the valve deployment process, opening and closing cycle process, and material degradation process based on a commercial finite element analysis platform, outputting indicators such as leaflet deformation, opening area, stress distribution, and anchoring stability. The experimental part involves fabricating a valve prototype based on the optimized structural and material parameters, performing structural matching verification through an implantation anatomy simulation model, conducting opening and closing performance tests using a simulated circulation device, and obtaining the trend of material mechanical parameter changes during the service life through in vitro material degradation experiments. This verifies the degree to which the design scheme meets functional requirements and performs closed-loop calibration of the aforementioned simulation model.

[0022] Simultaneously, the embedded deep learning model (obtained in steps S3 and S4) is invoked to rapidly evaluate the mechanical performance of the design scheme. Based on the evaluation results of interventional performance, mechanical performance, biodegradability, and other indicators, a reward / penalty function for reinforcement learning is calculated to update the agent's design strategy. The reinforcement learning algorithm is used to automatically search for and optimize the design scheme, and can employ value iterative algorithms or policy gradient algorithms to achieve multi-objective optimization of valve structural and material parameters.

[0023] This application uses a deep learning model trained by finite element simulation to replace direct finite element calculation for performance evaluation, and combines reinforcement learning algorithms to optimize design parameters, thereby improving the computational efficiency of design iteration and realizing an end-to-end automated and efficient design process based on medical image input to optimized valve design parameter output.

[0024] The geometric modeling in this application includes the establishment of two types of models: one is a three-dimensional reconstruction model of the original human disease structure, used to accurately restore the spatial anatomical morphology of the aorta, aortic valve, and calcified plaques; the other is a parametric structural model of the interventional absorbable bioprosthetic valve, used for design fitting based on the morphology of the target area and clinical functional requirements. These two models together constitute the geometric modeling foundation of this invention, providing anatomical constraints, structural mapping, and morphological response inputs for subsequent mechanical simulation, long-term performance evaluation, and reinforcement learning optimization, supporting the mapping construction between structural design parameters and functional requirements. Specifically, the original human disease structure model uses deep learning methods to automatically segment and three-dimensionally reconstruct clinical medical image data, extracting key tissue regions including the aortic root, the aortic valve tricuspid structure, the valve annulus, the left ventricular outflow tract, and calcified plaques. A multi-organ joint segmentation network model is used to complete structural identification, and an individualized three-dimensional anatomical model is generated through surface reconstruction and parameter extraction. This model accurately reflects the spatial morphology, boundary relationships, and calcification distribution of the diseased tissue, used to evaluate the spatial adaptability and deployment behavior of the interventional structure.

[0025] The interventional valve structure model is parametrically modeled based on structural design parameters, including key design variables such as leaflet thickness, free edge length, support structure pitch, and anchoring boundary width. The leaflet morphology is defined using parametric functions or cubic polynomial surfaces. The support structure is generated through a solid of revolution and automatically adapts to the valve annulus diameter. The calcification avoidance region is determined by Boolean operations intersecting with the anatomical model. The final generated valve structure model is aligned with the human body model, forming a spatial fit, serving as the input basis for subsequent simulation models.

[0026] This application establishes a valve stent interventional mechanical model based on deep learning methods to quickly and accurately predict the mechanical response of the valve stent during interventional surgery, replacing traditional finite element analysis methods. The model's training data comes from numerical simulation calculations obtained under different design parameters, covering a series of key mechanical performance indicators such as stent stress distribution, deformation state, axial shortening rate, and radial recoil rate. The simulation data can be obtained using finite element analysis software such as ABAQUS and ANSYS. The modeling process requires preprocessing of the simulation data, including extracting key mechanical performance features as output indicators of the model, and using structural dimension parameters and material property parameters as inputs, while using specific mechanical response indicators as output features. The data processing ensures that the input and output data formats of the model are uniform and standardized, facilitating rapid model training. The deep learning model can employ methods such as feedforward neural networks (FNN) and graph convolutional networks (GCN), using supervised learning training to optimize predictive performance, aiming to minimize the error between the predicted and simulated results. The model training process uses cross-validation or independent validation set methods to ensure the model's generalization ability.

[0027] This application also establishes a deep learning-based prediction model for the long-term use and material degradation performance of valves, effectively predicting the changes in material properties during long-term valve use. The model's data sources include finite element simulation data under long-term operating conditions and accelerated aging in vitro experimental data. These experiments involve performance change data from various material types, different stress levels, and different degradation environments. The data undergoes preprocessing steps including denoising, interpolation, and feature extraction to form a data format suitable for time-series prediction tasks, extracting key information such as material degradation rate, mechanical property degradation curves, and long-term service life indicators.

[0028] The long-term performance prediction model employs deep learning models such as recurrent neural networks, long short-term memory networks, Transformer models, or temporal convolutional networks, trained using supervised learning methods. The optimization objective is to minimize the error between the model's predictions and actual long-term performance, ensuring that the model's predictive ability meets clinical needs. The predictive ability of the long-term performance model has also been independently validated with experimental data, ensuring its reliable prediction of the long-term mechanical performance and material degradation trends of different valve designs, providing an accurate and efficient predictive tool for assessing the long-term reliability of valve designs.

[0029] This application integrates the aforementioned deep learning model into a deep reinforcement learning optimization framework to automatically achieve the collaborative optimization design of valve structure and material parameters, thereby finding the optimal design scheme that balances short-term performance and long-term reliability. The reinforcement learning process uses a parameterized valve model generated by geometric modeling as the adjustable action parameter space for the reinforcement learning agent. The mechanical model and the long-term degradation model together constitute the simulation environment for reinforcement learning, used to evaluate the effect of each design adjustment. The reinforcement learning optimization design defines a state, an action, and a reward. The state is the current set of valve design parameters, the action is the adjustment strategy for the parameter set, and the reward function comprehensively considers short-term and long-term performance indicators, assigning different weight coefficients to each. The short-term performance indicators include the valve's opening and closing flexibility, instantaneous hemodynamic performance, and stent stability in the initial state. The long-term performance indicators include the valve's structural integrity at the end of its predetermined service life, the proportion of remaining material strength, and the level of functional retention. By setting the weights for the short-term and long-term components (i.e., short-term weights and long-term weights), the total reward value is defined as the comprehensive value obtained by weighted summing of the short-term and long-term indicator scores according to their respective weights. The reward function design enables the reinforcement learning algorithm to simultaneously focus on the initial and endurance performance of the device during optimization, achieving balanced performance optimization. The reinforcement learning process involves continuous parameter optimization through interaction between the agent and the environment. After each action adjustment, the agent selects a set of valve design parameters based on the current strategy, such as valve geometry and stent structure parameters. Using an integrated interventional mechanics and long-term performance prediction model, the corresponding performance evaluation indicators are calculated in real-time to obtain the short-term and long-term performance evaluation of the design scheme, thereby calculating the corresponding reward value. The reinforcement learning process treats each design parameter selection and evaluation process as a training round, allowing the agent to gradually improve the quality of the design scheme through repeated multi-round training.

[0030] The optimization process utilizes methods such as Deep Q-Network (DQN), Actor-Critic method, and Deep Deterministic Policy Gradient (DDPG) to achieve efficient search and optimization. The optimization process determines convergence by monitoring the fluctuation range of the reward value. After the training process reaches a stable state, the optimal combination of valve structure and material parameters is obtained, and the design is verified through finite element simulation or in vitro experiments to ensure its reliability and clinical applicability.

[0031] The simulation verification step, based on the constructed parametric geometric model, mechanical model, long-term performance model, and reinforcement learning optimization model, conducts multi-condition, multi-index digital simulation evaluations of the obtained combination of absorbable bioprosthetic valve design parameters to verify the comprehensive performance of the structural and material design parameters in achieving functional requirements. The simulation verification employs a multiphysics coupling method, combining static structural analysis, dynamic opening and closing fluid-structure interaction simulation, and time-varying material performance modeling to simulate the valve's initial deployment behavior after implantation, its opening and closing behavior during the cardiac cycle after implantation, and its performance evolution during long-term use. Through this simulation process, the stability of the valve stent positioning, leaflet opening and closing function, local stress level, regurgitation flow rate, material modulus evolution, and performance retention during the service life are evaluated item by item, thus providing a theoretical basis for the effectiveness of the design parameters and fundamental guidance for prototype development and experimental verification.

[0032] The experimental verification steps are used to evaluate the in vitro performance of the absorbable bioprosthetic valve prototype designed and optimized using the method of this invention, verifying whether the design parameters meet the functional requirements under actual use conditions. Through structural adaptability testing, opening and closing function testing, and material degradation performance testing, the system evaluates the real-world performance of the valve structure outside the simulation environment, providing experimental support for the engineering implementation and product development of the design scheme. The experimental verification includes, but is not limited to: (a) manufacturing and implantation adaptability testing of the valve structure prototype, verifying the matching between the structural design parameters and the pathological anatomical structure; (b) conducting opening and closing cycle tests under a simulated cardiac platform, measuring indicators such as valve opening area, regurgitation flow, and pressure distribution to verify whether the valve meets functional requirements; and (c) conducting in vitro degradation experiments on the absorbable materials used, observing their performance evolution process in a simulated body fluid environment, and verifying the supporting capabilities of performance retention and degradation controllability in the material design parameters.

[0033] The method of this invention is applicable to different types of valves and various material systems. For example, this method can be used for the design of bioprosthetic valves such as aortic valves, pulmonary valves, and mitral valves, and different valve structures can be optimized using this method. At the same time, the proposed design process is applicable to different material systems (such as natural biological tissue materials, biodegradable polymer materials, biodegradable metals, bio-based composite materials, etc.), thus demonstrating the wide applicability of this method under various valve structures and material combinations.

[0034] Compared with existing valve design methods, this invention has significant advantages: The absorbable bioprosthetic valve design method based on artificial intelligence and biomechanics described in this invention significantly improves design efficiency, enabling multiple design iterations to be completed in a short time. By introducing a deep learning model, the efficiency of design scheme evaluation is significantly improved, allowing for rapid evaluation of a large number of candidate designs. The optimization process simultaneously considers structural performance and long-term durability, achieving synergistic optimization of structural parameters and performance indicators. Furthermore, deep reinforcement learning is used to achieve a high degree of automation in the design process, reducing human intervention and reliance on experience, making design decisions more objective and scientific. Simultaneously, the exploration mechanism of reinforcement learning helps to avoid the local optimum traps easily fallen into by traditional optimization processes, thus increasing the likelihood of obtaining a globally optimized design scheme. In summary, this invention provides a highly efficient, low-cost, comprehensive, and automated absorbable bioprosthetic valve design method, offering a scientific and efficient tool for the development of valve implantation devices, effectively overcoming the shortcomings of low efficiency and insufficient optimization in existing design methods, and improving the reliability of design schemes.

[0035] To make the technical solution of the absorbable bio-valve design method based on artificial intelligence and biomechanics described in this invention clearer, the invention will be further described below in conjunction with specific embodiments. Attached Figure Description

[0036] Figure 1 This is a flowchart of the absorbable bio-valve design method based on artificial intelligence and biomechanics described in this invention. Detailed Implementation Example 1

[0037] This embodiment uses the aortic valve as an example to describe in detail the absorbable bioprosthetic valve design method based on artificial intelligence and biomechanics described in this invention. The design method flowchart in this embodiment can be found here. Figure 1 The specific steps of this method are as follows: S1: Construct a performance requirement system for the valve, decomposing functional requirements into functional indicators in the functional domain and mapping them into structural and material parameters in the physical domain, specifically including: (1) Define user requirements based on the clinical purpose of bioprosthetic valves and the anatomical characteristics of the aortic valve. For aortic valve replacement, bioprosthetic valves need to meet the treatment goals of cardiovascular surgery and the anatomical constraints of the aortic root. The user requirements for the design of absorbable aortic bioprosthetic valves can be summarized as follows: a. Provide a reliable one-way blood flow path immediately after surgery to ensure smooth left ventricular ejection without significant regurgitation; b. Effectively relieve symptoms caused by valvular disease and restore normal cardiac output and aortic pressure levels; c. Ensure that the artificial valve matches well with the anatomical structure of the aortic root, does not obstruct the coronary artery ostium, and is stably fixed in the annulus position; d. Provide sufficient mechanical support in the early stage of implantation and withstand repeated blood flow impacts until the cardiac tissue gradually adapts or new tissue grows in; e. Be able to be biodegraded and absorbed after completing the transitional function, avoiding long-term foreign body presence and reducing the need for secondary surgery; f. Have good biocompatibility and do not cause significant thrombosis or immune rejection.

[0038] (2) Based on the user requirements defined in step (1), determine the top-level performance requirements FR0 that the absorbable bioprosthetic valve must possess to meet the user requirements, and map them into the corresponding top-level design parameters DP0 in the physical domain. For aortic bioprosthetic valves, based on the above user requirements, the top-level functional requirement FR0 can be defined as: FR0: Possesses comprehensive properties that meet cardiovascular physiological requirements in vivo. The corresponding top-level design parameter DP0 is: DP0: Overall Design Scheme for Bioprosthetic Valves Furthermore, in this step, ethical constraints (Cs) must be defined within the functional domain for the design and validation of the bioprosthetic valve, based on the ethical requirements of the medical device regulatory standards system. These ethical constraints (Cs) are primarily based on the "Animal Protection Requirements" of the ISO 10993 standard system and are used to regulate animal experiments involved in the design validation. They include: a. Qualification review of experimental personnel; b. Compliance with regulations regarding preoperative, intraoperative, and postoperative animal care; c. Humane handling of animals; and d. Reduction of unnecessary repetitive experiments. By setting these ethical constraints, it can be ensured that the validation process of the design method of this invention complies with ethical requirements.

[0039] (3) After determining FR0 and DP0, the biomechanical performance requirements, material performance requirements and biological performance requirements that the bioprosthetic valve needs to meet are decomposed by mapping between the functional domain and the physical domain, and mapped to the corresponding structural design parameters and material design parameters, until the leaf-level performance index FRs and leaf-level design parameters DPs that can be directly reflected in mechanical, material and biological tests are obtained. The FR0 and DP0 obtained in step (2) are relatively general and need to be further decomposed and refined. For aortic bioprosthetic valves, the following requirements need to be met in the design and use process: (1) They have good hemodynamic and mechanical properties, provide a stable and unobstructed pathway for cardiac ejection, and ensure the structural stability of the valve during opening and closing; (2) The material maintains the necessary physicochemical properties during the service period of implantation, does not undergo premature degradation leading to a sudden drop in performance, and can be gradually degraded and absorbed after the established service period is completed. Therefore, FR0 is decomposed into: FR 11 It possesses excellent hemodynamic and mechanical properties; FR 12 It possesses controllable material degradation properties.

[0040] The above three items constitute the first layer of functional requirements for aortic valve bioprosthetics. In the physical domain, it is readily apparent that the valve's structural and material design both influence these requirements; therefore, the top-level design parameter DP0 can be decomposed accordingly as follows: DP 11 Structural design; DP 12 Materials design.

[0041] The above two items are the first layer of design parameters for bioprosthetic valves. Next, the design parameters (DP) will be further refined within the functional domains. 11 and DP 12 The corresponding functional requirements.

[0042] For FR 11 The specific performance requirements for aortic bioprosthetic valves include: (1) sufficient opening capacity: the valve should be able to open fully during ventricular contraction to minimize transvalvular pressure gradient and ensure sufficient ejection volume; (2) reliable closure tightness: the valve leaflets should be able to close quickly and tightly during ventricular diastole to minimize regurgitation; (3) structural strength to withstand cyclic loads: after implantation, the valve will be subjected to the periodic alternating pressure generated by each heartbeat, and should have sufficient strength and flexibility to avoid mechanical fatigue damage during the expected service life; (4) stable positioning and sealing: the valve should be able to be firmly fixed at the aortic annulus without displacement and ensure no leakage around the valve; (5) flow field friendliness: the valve structure should have a reasonable stress distribution at the aortic root, and the opening and closing process should not adversely affect the adjacent blood flow (such as coronary blood flow) to avoid eddies or other hydrodynamic abnormalities. Therefore, FR 11 This can be further refined as follows: FR 111 It has sufficient valve opening area (low transvalvular pressure gradient); FR 112 It possesses excellent leaflet sealing performance (low backflow); FR 113 It possesses sufficient structural strength and fatigue resistance; FR 114 It possesses reliable implantation positioning stability (no displacement or paravalvular leakage); FR 115 It has good flow field compliance (does not interfere with adjacent blood flow).

[0043] Based on the characteristics of absorbable materials and the long-term use requirements of implants, FR 12 This can be further refined to: (1) The mechanical properties of the material remain stable for a period of time after implantation and will not deteriorate rapidly due to environmental changes, thereby ensuring that the valve functions normally within its intended service life; (2) The material can gradually degrade after completing its supporting function, and its degradation rate is appropriate, so as not to lose structural integrity too quickly, nor to remain in the body for a long time too slowly. Therefore, FR 12 To be further refined as follows: FR 121 The mechanical properties of the material remain above the performance threshold during its service life; FR 122 The material's degradation behavior is controllable, and it can be safely absorbed after its service life.

[0044] The above items constitute the second layer of functional requirements for aortic valve bioprosthetics. Further decomposition and refinement of the design parameters affecting these functional requirements are then performed in the physical domain. Based on a structural survey of existing bioprosthetic valve devices, the main elements of aortic valve structural design can be summarized as follows: leaflet structure, support structure, and the connection method between the two. Therefore, regarding DP... 11 Further breakdown into: DP 111 Leaflet structure design; DP 112 Support ring (bracket) structural design; DP 113 : The connection design between the leaflets and the supporting structure.

[0045] Among them, the leaflet structure design DP 111 This involves determining the number, shape, and size parameters of the leaflets (e.g., leaflet arc length, curvature, thickness distribution, etc.); and designing the support ring structure using dynamic programming (DP). 112 This involves determining the geometric parameters (e.g., diameter, height, stent aperture layout, etc.) of the annular stent or suture ring used to fix the valve position and provide necessary support; and the design of the connection between the leaflet and the support structure (DP). 113 This involves the method and interface structure (such as the clamping method at the base of the leaflet, the form of connecting materials, etc.) to ensure that the leaflet is stably installed on the support structure and that the force is transmitted reasonably.

[0046] Similarly, material design DP 12 It is also necessary to refine the material composition and performance factors. Absorbable bioprosthetic valves are typically made of biodegradable polymers, and their material design parameters mainly include the material's composition, initial mechanical properties, microstructure, and degradation characteristics. Therefore, DP... 12 This can be further refined as follows: DP 121 Valve material composition parameter design; DP 122 Design of mechanical performance parameters for valve materials; DP 123 Design of valve material microstructure and degradation characteristics.

[0047] For DP 121 The material composition parameters, specifically the monomer ratios of the copolymers, are mainly considered in this embodiment for the biodegradable polymer material. For example, this embodiment uses polylactic acid-co-polyglycolic acid copolymer (PLGA) as the matrix material and improves flexibility by introducing soft-segment copolymers. The specific component ratios include the content of L-lactic acid monomer, glycolide monomer, and ε-caprolactone monomer. Therefore, DP... 121 This can be further refined as follows: DP 1211 L-lactic acid monomer content; DP 1212 : content of glycolide monomer; DP 1213 : Content of ε-caprolactone monomer.

[0048] For DP 122 The material's mechanical properties mainly include its elastic modulus, strength, and ductility. These parameters determine the valve's mechanical performance in its initial state, therefore DP 122 It can be further refined as follows: DP 1221 Initial Young's modulus of the material; DP 1222 : Material tensile strength; DP 1223 Material elongation at break; DP 1224 Material fatigue resistance indicators (such as fatigue life cycle count).

[0049] For DP 123 The microstructure and degradation characteristics of the material are mainly considered in light of factors affecting the degradation rate and mechanical changes, including polymer molecular weight and distribution, crystallinity, and hydrophilicity. Therefore, DP 123 It can be further refined as follows: DP 1231 : Polymer average molecular weight and molecular weight distribution; DP 1232 Polymer crystallinity; DP 1233 Material hydrophilicity.

[0050] Through the above decomposition and mapping of FR and DP, the correspondence and hierarchical structure between functional requirements and design parameters are established, laying the foundation for optimized design.

[0051] S2: Obtain the patient's clinical medical imaging data, and use deep learning methods to construct a three-dimensional reconstruction model of the original lesion structure of the human body and a parameterized structural model of the valve, respectively. The structural model responds to the structural parameter form defined in step S1.

[0052] The three-dimensional reconstruction model of the original pathological structure of the human body is used to accurately restore the spatial anatomical morphology of the aorta, aortic valve, and calcified plaques; the parametric structural model is used to design and fit according to the morphology of the target area and clinical functional requirements. The two models together constitute the spatial constraints and structural adaptation basis in the design method of this invention, supporting the functional requirement FR. 11 (Possessing good hemodynamic and mechanical properties, i.e., good mechanical and opening / closing properties) — and FR 114 (Reliable implantation stability) – FR 114 The achievement of reliable implantation positioning stability (no displacement or paravalvular leakage) corresponds to the structural design parameter DP. 11 DP leaflet design 111 DP bracket design 112 and connection method DP 113 .

[0053] The human lesion structure model is based on high-resolution clinical imaging data acquired before surgery. It employs deep learning methods to automatically segment and reconstruct various tissue structures, including the aorta, aortic valve, left ventricular outflow tract, and calcified plaques. The image data preferably uses CT angiography data; the original images are standardized, registered, and denoised before being input into the trained 3D image segmentation neural network model. The segmentation model outputs data including the aortic root Ω... aorta Aortic valve Ω valve Calcified plaque area Ω calc and left ventricular outflow tract Ω lvot Voxel labels with similar structures. The aortic valve leaflets consist of three leaflets, each represented by a parametric surface. The calcified region is represented by a polygonal block and intersects spatially with the leaflet model to assess the degree of valvular lesions.

[0054] The leaflet surface of the aortic valve model Generated through cubic parametric surface fitting, its form can be expressed as:

[0055] in, For the summation index, For leaflet index, For the first The surface curvature coefficient of the leaf petal. This is a normalized local coordinate system. The leaflet surface definition allows for parameter control of the leaflet's curvature, free edge length, thickness distribution, etc., facilitating subsequent integration with interventional valve design.

[0056] Calcified plaque regions are identified using a segmentation network trained on high-density areas. The model outputs the spatial location, volume, and contact area with the valve leaflet of the plaque, providing fundamental constraints for stent deployment simulation and avoidance area placement in interventional design. The resulting lesion structure model Ω native Expressed as:

[0057] Based on the above structural model, a parameterized structural model of the first absorbable bio-valve is further established. The model structure includes three leaflets, a ring-shaped support structure, and an optional skirt anchoring unit, with design parameters corresponding to DP. 111 ~DP 113 The sub-items include leaflet thickness, curvature, free edge length, support ring size, pitch and height, and local arrangement adapted to the calcified area.

[0058] The valve structure model is generated by a preset template generator, and a three-dimensional structural sample is formed driven by design parameters. Its spatial shape is aligned and fitted with Ωnative. The curved surface of each leaflet... Expressed as:

[0059] in, For the summation index, For leaflet index, The first in the leaflet design scheme The parameter coefficients of the leaflets can be gradually adjusted through an optimization process to meet the opening and closing performance (FR). 111 ) and stress distribution requirements (FR) 113 ).

[0060] The support structure model is created using a rotational surface generator. It automatically fits the inner diameter based on the lobe ring boundary, and parameters such as pitch, strut angle, and number of connecting ribs can be defined on the periphery to form spatial anchor points and developmental mechanical properties, ensuring compliance with FR (Front-Rate Layout) requirements. 114 Stability constraints are imposed. The contact relationship between the calcified region and the interventional structure is determined using the Boolean intersection method, generating an interference risk map to identify unfavorable layout areas and optimize the skirt design. This ultimately forms a complete interventional valve model Ω. implant :

[0061] in For the union operation, Represents the petal frame model. This indicates the anchoring structure between the support structure and the petal ring.

[0062] The two models are fitted together through rigid body registration and local elastic adaptation, serving as the initial spatial basis for subsequent mechanical simulation, model training, and reinforcement learning optimization. Through this parametric modeling method, this invention can generate individualized design models for different patients' anatomical morphologies, support fully automated optimization, and ensure the structural design parameters (DP) are optimized. 11 Response Functional Requirements FR 11 and FR 114 .

[0063] S3: Establish a first deep learning model simulating the interventional mechanical behavior of the valve; the training data of the first deep learning model comes from the finite element simulation results, with the structural model generated in step S2 and user-defined material parameters as input, and the core performance indicators extracted from the finite element simulation as output, to predict the mechanical response of the valve during the intervention process: (1) The lesion structure model Ω constructed in step S2 native and interventional valve structural model Ω implant Based on this, a complete valve-aortic system simulation model was established. The mechanical performance of the interventional device during implantation and circulatory initiation and closure was simulated and analyzed. Training data was then generated to establish a first deep learning model for efficient performance prediction. The simulation analysis results were used to validate the interventional design parameter DP. 11 Material Design Parameters (DP) 12 Does the valve structure meet the stress safety (FR) requirements under various operating conditions? 113 ), anchoring stability (FR) 114 ) and on / off functionality (FR 111 FR 112 Core indicators such as )

[0064] The mechanical performance simulation comprises two phases: the deployment and positioning process during valve stent implantation, and the cyclic stress response of the valve during its opening and closing process in the cardiac cycle. The former verifies the geometric fit and interference risks between the interventional valve structure and the existing anatomical structure, while the latter assesses the structural stability and functional performance of the valve under blood flow loads. In the simulation model for the implantation phase, Ω... implant The stent region is defined as an elastically expandable structure. By simulating the deployment process of an external delivery system, its mechanical response, including positioning deformation and pressure distribution at the anchoring contact surface, is observed under the Ωcalc environment of the aortic valve annulus and calcified plaque. If significant slippage, excessive stress concentration, or abnormal stent deformation occurs after deployment, the stent is considered to have a DP (displacement failure). 112 DP113 and DP 121 DP 122 Not meeting FR 114 Requirements: In the simulation of the opening and closing phases, physiological pressure boundary conditions are set to simulate two extreme conditions: ventricular systole (valve closure) and diastole (valve opening). Typical transvalvular pressure differentials are applied (e.g., 120 mmHg during systole and 10 mmHg during diastole). The opening area, regurgitation clearance, leaflet deformation, and maximum equivalent stress distribution under pressure load are evaluated. The various output performance parameters are used to determine the DP (Dynamic Performance). 111 and DP 121 DP 122 Does it meet FR? 111 FR 112 and FR 113 The mechanical requirements.

[0065] To improve design efficiency and achieve automated optimization, this step further constructs a model. The model uses structural design parameters DP. 11 With material parameters DP 12 Using the core performance indicators extracted from simulation as input, a mapping relationship between input and output is established. The model training employs supervised learning. Inputs include parameters such as leaflet curvature, stent pitch, and valve thickness. The network model's output Y(t) includes: Y(t) = [A open , R f ,σ max E leaf E anchor ]

[0066] Where A open R represents the valve opening area. f For the backflow rate, σ max For the maximum equivalent stress, E leaf E represents the equivalent stiffness of the leaflet. anchor This represents the average value of the anchoring contact force.

[0067] The model employs a multi-layer feedforward neural network structure, and the loss function LOSS is a multi-index weighted squared error.

[0068] in Y is the summation index for all samples, where n represents the total number of samples. j This represents the simulation analysis result of the j-th sample. This represents the network output result of the j-th sample.

[0069] The training data is generated from simulation analysis and, after standardization, is input into the model training process. The trained model can quickly predict valve performance under arbitrary parameter combinations, significantly improving the efficiency of design space search and providing performance feedback for the next optimization step.

[0070] S4: Construct a second deep learning model to predict the relationship between stress and material degradation behavior during long-term valve use. The training data for this second deep learning model comes from finite element simulations and in vitro experiments to characterize the degradation rate of the material under stress and the resulting evolution of the valve's mechanical properties, in order to predict whether the structural and material parameters defined in S1 meet the requirements. The second deep learning model is to construct a performance change model of the absorbable bioprosthetic valve during its service life, using deep learning methods to simulate the changes in valve structure and performance caused by material degradation, and further determine the design parameter DP. 111 With DP 123 Does the functional requirement meet the target service life? 12 (Evolution of material properties) and FR 11 (Functional continuity).

[0071] This step first involves setting the material design parameters (DP). 12 Factors such as the material's initial elastic modulus, degradation rate, and microstructure (e.g., crystallinity, hydrophilicity) are used as input to the deep learning model. These are combined with valve structural parameters DP. 11 To investigate the geometric factors affecting the degradation response (such as leaflet thickness gradient and coordinates of stress concentration feature points), a multidimensional time-series prediction model is established to simulate the changes in the mechanical properties of the valve in different regions over time t.

[0072] The deep learning model employs a sequence prediction architecture, with network input being [DP]. 11 ,DP 12 The output is the performance index P(t) at time t, specifically including: P(t) = [E(t), A(t), R] f (t),σ max (t)]

[0073] Where E(t) represents the change of the material's equivalent elastic modulus with time, A(t) is the opening area of ​​the valve at time t, and R... f (t) is the backflow rate, σ max (t) represents the maximum local stress value of the structure.

[0074] The model was trained using historical simulation and degradation experiment data. The samples included material properties at different time points derived from degradation experiments, as well as valve structural responses obtained from historical simulation sampling. The training objective was to minimize the mean square error between the predicted and actual values. The network optimization employed the Adam algorithm or a similar stable convergent optimizer for iterative updates.

[0075] Once the model is trained, it can be used to predict whether the performance evolution of a certain combination of design parameters will meet the requirements during the target service life, such as within 6 months. a) FR 121 Material properties are not lower than the safety threshold E min ,Right now ; b) FR 122 c) Performance degradation rate is controllable, i.e., dE / dt is stable without sudden drops; 11 Persistence: Opening area A(t) and backflow rate R f (t) consistently fluctuated within the clinically permissible range.

[0076] Therefore, this deep learning model can achieve performance prediction and usage risk assessment of the valve throughout its entire life cycle, which is a key aspect of realizing DP (Dynamic Performance Monitoring). 12 With FR 12 A key tool for closed-loop verification between different systems.

[0077] S5: Embed the first deep learning model and the second deep learning model into a reinforcement learning framework, and use reinforcement learning algorithms to automatically iteratively optimize the structural and material parameters of the valve, so as to simultaneously optimize the interventional performance, mechanical performance and biodegradation performance of the valve in a simulation environment, and realize end-to-end efficient optimization design from anatomical model input to design parameter output.

[0078] This step is a reinforcement learning-based design optimization step, aiming to utilize artificial intelligence methods to optimize the valve structure design parameters (DP). 11 Material Design Parameters (DP) 12 Perform collaborative optimization to achieve functional requirements FR under multiple constraints. 11 FR 12 and FR 114 The comprehensive requirements are met. The reinforcement learning method automatically explores the optimal design solution in a high-dimensional parameter space by constructing an interaction system between the agent and the environment.

[0079] The reinforcement learning framework consists of three parts: state space S, action space A, reward function R, and agent. Wherein: The state space S is used to describe the parameter combination corresponding to the current design scheme, and includes, but is not limited to, the following variables: a) Structural design parameters DP11 Relevant variable: leaflet thickness Leaflet curvature radius Support ring diameter support pitch skirt hem width ; b) Material Design Parameters (DP) 12 Relevant variable: initial elastic modulus Degradation rate constant Polymer crystallinity molecular weight Material hydrophilicity index

[0080] Therefore, the state S at any given time t It can be represented as a multidimensional vector:

[0081] This state serves as input to the model and the long-term performance model, used to predict the various performance metrics corresponding to the current parameter combination.

[0082] The action space A represents the adjustment strategy that the agent makes to the current design parameters in each iteration, specifically in the form of multi-dimensional parameter increase / decrease operations or continuous value adjustments. An action can be one or more of the following sets of operations: a) Regarding leaflet thickness b) Increase or decrease by a certain percentage; To affect the opening and closing stiffness; c) change the support pitch d) Adjusting the initial modulus of the material, which affects the development stress; (e) Adjusting the initial support properties; This alters the rate of evolution of material properties.

[0083] The actions are presented as continuous vectors, i.e. .

[0084] The reward function R(S) t ,a t The reward function is used to measure the overall performance of the current design in meeting various functional requirements and is the core objective function for reinforcement learning optimization. The reward function considers the following dimensions: a) Performance Implementation: If the open area A open Exceeding threshold A min If the backflow rate R is positive, a positive reward will be given; f If the stress is less than the set upper limit, a positive reward is given; if the maximum stress σ max If the stress is less than a certain percentage of the material's ultimate stress, it indicates that the structure is safe and a reward will be given. b) Material evolution: If the performance remains at E(t) ≥ E within 6 months of degradation. safe In this case, long-term rewards will be provided for the design solutions; c) Design stability: If the performance fluctuation is small after multiple iterations, it is considered a local optimal stable region and is given a reward for improvement; d) Structural adaptability: If it is compatible with Ω native If the spatial overlap is less than the threshold (i.e., there is no significant interference with calcified plaques), then positive feedback is given to the chimerism. e) Penalty mechanism: If the valve design causes significant regurgitation, structural damage, or a sharp drop in performance after degradation, a negative reward will be given; if the stent cannot be positioned or there is a serious risk of slippage, it will be considered an unsuitable solution and a larger penalty will be imposed.

[0085] reward function r t It can be represented as

[0086] Where Norm(·) represents the normalized performance index, and Fit(·) represents the geometric fit function, used to evaluate the degree of integration between the interventional valve model and the original human structure. The weights of each indicator can be dynamically adjusted according to the optimization objective. Based on the above definition, the reinforcement learning system can evaluate the design quality in each round of decision-making and maximize the cumulative reward value R. total :

[0087] in The summation is the time index, where T represents the predicted time series length, and r... t For the reward function, The discount factor controls the weight of future rewards, ultimately outputting a design parameter combination that performs optimally in the long run.

[0088] The intelligent agent, as the core computing unit of the reinforcement learning system, functions to: in each round of interaction, based on the current design state S... t Make parameter adjustment action a t Instant rewards for receiving environmental feedback t And update its strategy to improve the optimality of future decisions.

[0089] The agent's policy function Let θ be the probability distribution function, where θ is the set of learnable parameters of the policy function. The agent updates θ through iterative interaction with the environment, forming a decision policy that gradually converges. The policy function employs the Deep Deterministic Policy Gradient (DDPG) algorithm to achieve efficient search for the optimal solution in the continuous high-dimensional design parameter space, and to optimize the structural design parameters DP. 11 Material Design Parameters (DP) 12 Perform collaborative optimization to achieve functional requirements FR 11 FR 12 and FR 114 .

[0090] The DDPG algorithm is suitable for high-dimensional continuous parameter optimization scenarios. Its core consists of two deep neural networks: one is a policy network (Actor), which outputs the optimal design action in the current state; the other is a value network (Critic), which evaluates the performance of the current state-action combination.

[0091] The policy network is in its current design state S. t As input, output a continuous action vector a t This is used to adjust continuous design parameters such as valve thickness, stent size, and material modulus. In this embodiment, its structure adopts a three-layer feedforward neural network: a) Input layer: Design state vector b) Hidden layer 1: Fully connected layer, ReLU activation function, 128 nodes; c) Hidden layer 2: Fully connected layer, ReLU activation function, 64 nodes; d) Output layer: Action vector a t Use Tanh activation to constrain the range of motion.

[0092] The value network is in state S t and action a t The concatenated vector is taken as input, and the expected performance score Q(S) of the combination is output. t ,a t Its structure can be: a) Input layer: Concatenating input [S] t ,a t b) Hidden layer 1: Fully connected, ReLU activation function, 128 nodes; c) Hidden layer 2: Fully connected, ReLU activation function, 64 nodes; d) Output layer: Single-node linear output Q.

[0093] Based on this structure, the agent interacts with the model and the long-term performance model, calculates the reward function and updates the policy, and obtains a convergent and stable optimization policy function through repeated training. The final output design parameter combination shows excellent performance in terms of hemodynamic performance, structural strength, implantability and material evolution behavior, and meets the constraints of the multidimensional performance indicators described in this invention.

[0094] S6: Based on the optimization results of step S5, conduct simulation and experimental verification, construct a valve-heart chamber-aortic coupling simulation model and perform stent deployment, opening and closing cycle and material degradation simulation; prepare valve prototypes and conduct structural matching tests, opening and closing function tests and in vitro degradation experiments, and comprehensively evaluate the adaptability and reliability of the design parameters under the multidimensional functional requirements defined in S1.

[0095] Simultaneously optimize the interventional performance, mechanical performance and biodegradability of the valve in the simulation environment. The simulation verification steps are based on the design parameters of the aortic valve absorbable bio-valve obtained in this embodiment. A valve-aortic coupling model is constructed, and a commercial finite element platform is used to carry out multi-stage simulation verification to evaluate the functional performance of the designed scheme under clinical conditions. First, based on the geometric model of the human lesion structure and the parametric model of the interventional valve established in step S2, structural simulation is performed on the deployment and positioning process of the valve stent. By simulating the radial deployment behavior of the stent during release, the contact response between the stent and the valvular annulus and calcified plaques is analyzed to obtain key mechanical characteristics such as stent deformation, contact force distribution, and implantation stability, which are used to evaluate the structural design parameter DP. 112 DP 113 Material Design Parameters (DP) 121 DP 122 In meeting functional requirements FR 114 Performance in terms of (stable anchoring).

[0096] Subsequently, a fluid-structure interaction simulation model was constructed to simulate the opening and closing behavior of the valve during the two typical cardiac cycles of left ventricular systole and diastole. In this simulation, the valve leaflet structure model was embedded into the cardiovascular blood flow domain, and representative transvalvular pressure boundary conditions were applied to simulate the dynamic opening and closing process of the valve. The simulation outputs multiple indicators, including valve opening area, regurgitation flow, leaflet deformation distribution, and local maximum stress, as structural design parameters (DP). 111 Material Design Parameters (DP) 121 DP 122 Functional requirements FR 111 (Open / close function), FR 112 (Seamless fit) and FR 113 Evaluation criteria for (structural strength) response capability.

[0097] Furthermore, considering material degradation behavior, a time-evolution simulation process was established to analyze the structural mechanical properties of the valve at different usage time points (including the early, middle, and late stages of material degradation). By introducing a functional expression for the change of material properties over time, the material design parameter DP was... 123 The degradation rate, modulus evolution, and other characteristics are embedded into the simulation material model to achieve dynamic prediction of the changes in the mechanical properties of the valve leaflet. The simulation output includes changes in valve deformation caused by material modulus evolution, stress redistribution, and the stability of the sealing performance, which are then used to evaluate whether the design meets the requirements of FR (Front-to-Back) parameters. 121 (Performance retention) and FR 122 (Controllable evolution) and other long-term use functional requirements.

[0098] The experimental verification steps, based on the aortic valve design parameter optimization and simulation verification completed in this embodiment, further carried out relevant experimental verification procedures, and conducted a physical-level evaluation of the engineering feasibility and functional achievement capability of the design scheme.

[0099] First, based on the optimized output structural design parameters DP 11 With material parameters DP 12 A prototype valve structure was fabricated using absorbable polymer materials. The leaflets were processed using high-precision micro-injection molding technology, while the stent structure was completed using laser cutting and heat treatment molding processes. The prototype valve was implanted into an anatomical simulation model for structural matching tests. Visual inspection and imaging were used to observe its unfolding state within the valve annulus, its positional stability, and its contact relationship with calcified plaque areas, thus preliminarily verifying its rationality in terms of spatial adaptation and anchoring.

[0100] Subsequently, the prototype underwent opening and closing performance testing in a simulated cardiac circulation device. The simulation platform was configured with standard left ventricular systolic and diastolic pressure curves and equipped with a high-frequency camera and pressure sensors to synchronously record valve operation. During the test, it was observed whether the valve leaflet opening and closing rhythm was synchronized with pressure changes, and the maximum opening state, degree of closure tightness, and regurgitation performance of the valve were recorded to determine the impact of the leaflet structure and material design on FR. 111 (Open / close function) and FR 112 (Close-fitting) support capabilities.

[0101] Finally, a long-term in vitro degradation experiment was conducted on material samples from the same batch as the prototype material. The samples were immersed in simulated body fluid, and the evolution of key performance parameters such as mass, elastic modulus, and fracture strength was measured periodically. The degradation rate and performance retention trend during the service life were analyzed, and the results were compared with the output of the long-term performance prediction model in step (6) of this invention to further confirm the material parameter DP. 12 Supporting FR 121 (Performance retention) and FR122 The rationale regarding (controllable evolution).

[0102] Example 2 This embodiment, based on the method of the present invention, provides a detailed explanation of the absorbable bioprosthetic valve design method based on artificial intelligence and biomechanics, specifically for clinical conditions such as mitral regurgitation or calcified stenosis. The specific process of this method is as follows: S1: Construct a performance requirement system for the valve, decomposing the requirements into functional indicators in the functional domain and mapping them into structural and material parameters in the physical domain. The specific process is as follows: (1) Define user requirements based on the clinical purpose of bioprosthetic valves and the anatomical characteristics of the mitral valve. For patients with mitral regurgitation, bioprosthetic valves need to meet the treatment goals of valve repair or replacement surgery and the constraints of the mitral valve's anatomical structure. The user requirements for the design of absorbable bioprosthetic mitral valves can be summarized as follows: a. Provide a reliable one-way blood flow pathway immediately after surgery, ensuring smooth diastolic blood flow from the left atrium to the left ventricle, and complete valve closure without significant regurgitation during systole; b. Effectively relieve clinical symptoms caused by valvular regurgitation, significantly reduce or eliminate blood regurgitation, restore normal cardiac output, and improve the patient's hemodynamic status; c. Ensure good spatial matching between the artificial valve and the complex anatomical structure of the mitral valve (including leaflets, annulus, chordae tendineae, and papillary muscles), and that the supporting structure or artificial chordae tendineae structure can be firmly and stably anchored within the annulus and ventricular structures to avoid valve displacement or dislodgement; d. In the early stages of implantation, it provides sufficient structural and mechanical support to adapt to changes in the ventricular-atrial pressure gradient in the mitral valve region and the stress load generated by repeated opening and closing movements, maintaining structural integrity until new tissue gradually grows into the valve implantation area or autologous tissue repair is achieved; e. After completing its transitional function, it can be gradually biodegraded and absorbed, avoiding the impact of long-term foreign body presence on intracardiac structures and reducing the risk of subsequent reoperation; f. It possesses excellent biocompatibility, and does not induce significant thrombosis, hemolysis, or immune rejection after implantation, ensuring long-term safety and stable clinical efficacy.

[0103] (2) Based on the user requirements defined in step (1), determine the top-level performance requirements FR0 that the absorbable bioprosthetic valve must possess to meet the user requirements, and map them into the corresponding top-level design parameters DP0 in the physical domain. For the mitral valve absorbable bioprosthetic valve, based on the above user requirements, the top-level functional requirement FR0 can be defined as: having comprehensive performance that meets the physiological function requirements of the mitral valve in vivo.

[0104] The overall performance includes, but is not limited to, providing a reliable one-way blood flow channel during the cardiac cycle, stable opening and closing in high pressure differential circulation, close fit to the anatomical structure between the left atrium and left ventricle, and maintaining its mechanical and opening / closing functions during material degradation.

[0105] The corresponding top-level design parameter DP0 is: the overall design scheme of the absorbable biological mitral valve.

[0106] The overall design scheme includes the parametric geometric design of the artificial leaflet, the configuration scheme of the tendineae or alternative structures, the anchoring method and layout of the support structure, and the degradation characteristics and tissue compatibility of the selected materials.

[0107] Furthermore, this step also requires defining ethical constraints (Cs) within the functional domain for the design and validation of biological valves, based on the ethical requirements of the medical device regulatory standards system. These ethical constraints (Cs) are primarily based on the "Animal Protection Requirements" of the ISO 10993 standard system and are used to regulate animal experiments involved in the design and validation process. They include: a. Qualification review of experimental personnel; b. Preoperative, intraoperative, and postoperative animal care must comply with prescribed medical animal ethics standards; c. Humane methods must be used to handle experimental animals, avoiding unnecessary pain and trauma; d. Minimizing the number of experimental animals and avoiding duplicate experiments, employing necessary alternative methods.

[0108] (3) After clarifying FR0 and DP0 in step (2), it is necessary to further map between the functional domain and the physical domain, decompose the key performance requirements that the biological valve should possess layer by layer, and map them to the structural parameters and material parameters that can be specifically designed and implemented, so as to form the leaf-level performance requirements FRs and design parameters DPs that can be directly measured or verified in simulation calculations and in vitro experiments. Since the FR0 and DP0 obtained in step (2) are still top-level requirements and overall design schemes, and the content is relatively general, it is necessary to refine them in layers based on the tissue structure characteristics, functional characteristics and physiological environment of the mitral valve.

[0109] For absorbable bioprosthetic mitral valves, the following two core performance requirements must be considered during their design and application: First, they must possess good hemodynamic function and mechanical response during the cardiac cycle, meaning they must be able to fully open during diastole to provide a low-resistance blood flow channel, close rapidly during systole to prevent blood backflow, and maintain sufficient structural strength and fatigue resistance during long-term use; Second, the biodegradable materials constituting the valve must maintain stable mechanical performance support in the early stages of implantation, and then gradually degrade according to a set rhythm, being absorbed or replaced by tissue after completing structural support and functional transition, thus avoiding long-term foreign body risks.

[0110] Therefore, the top-level functional requirement FR0 is refined into the following first-level sub-requirements: FR 11 It possesses excellent hemodynamic and structural mechanical properties; FR 12 It possesses controllable material degradation properties.

[0111] Correspondingly, the top-level design parameter DP0 is refined as follows: DP 11 Structural design; DP 12 Materials design.

[0112] Among them, DP 11 Used to define the specific configuration of the valve structure, including the leaflets, supporting structures, and their connection and anchoring methods; DP 12 Used to define the composition, mechanical properties, and evolution characteristics of valve materials over time.

[0113] Furthermore, in order to achieve FR 11 The defined valve functional requirements are further refined into the following sub-items within the functional domain: FR 111 : Possesses sufficient valve opening capacity (low transvalvular pressure gradient); FR 112 It possesses good shut-off tightness (low backflow rate); FR 113 It possesses sufficient structural strength and fatigue resistance; FR 114 : Possesses reliable implantation stability (matches valve annulus and ventricular structure, avoiding displacement); FR 115 It possesses chordae tendineae biomechanical adaptability and synergistic ligament (maintaining leaflet closure).

[0114] Unlike the aortic valve, the mitral valve is a bicuspid structure and is connected to the papillary muscle via chordae tendineae; therefore, its opening and closing behavior depends more on the coordinated movement between the leaflets and the tension control of the chordae tendineae. In FR 115 The Chinese text explicitly states that the mechanical involvement of the chordae tendineae system is a key distinction in the functional breakdown of the mitral valve.

[0115] For the long-term performance requirements of materials FR 12 Similarly, it can be further refined into: FR 121 The mechanical properties of the material remain above the performance threshold during its service life; FR 122 The material's degradation behavior is controllable, and it can be gradually absorbed after its service life without leaving any residue.

[0116] In the physical domain, considering the unique configuration and load characteristics of the mitral valve structure, the structural design parameter DP is... 11 To be further refined as follows: DP 111 Leaflet structure design, including the relationship between the number of anterior and posterior leaflets, the length of the free edge, the leaflet thickness distribution, and the leaflet radius of curvature; DP112 Support ring structure design, including elliptical or saddle-shaped support structures for anchoring to the valve ring, with parameters including inner diameter, pitch, compliance, skirt width, etc.; DP 113 The design of the leaflet-support connection structure and chordae tendineae anchoring includes the connection interface between the leaflet and the support ring, the arrangement of the artificial chordae tendineae, the design of the attachment points, and the tension control method.

[0117] For material design parameters DP 12 Based on the compositional logic and engineering performance requirements of biodegradable polymer materials, the details are as follows: DP 121 Valve material composition parameter design; DP 122 Design of initial mechanical properties of valve materials; DP 123 Design of material microstructure and degradation behavior parameters.

[0118] Further refinement DP 1211 L-lactic acid monomer content; DP 1212 : content of glycolide monomer; DP 1213 ε-caprolactone monomer content; DP 1221 Initial Young's modulus; DP 1222 : Tensile strength; DP 1223 Elongation at break; DP 1224 Fatigue performance indicators (such as fatigue life, number of cycles); DP 1231 : Polymer average molecular weight and its distribution; DP 1232 Crystallinity; DP 1233 Hydrophilicity (e.g., water absorption rate, contact angle, etc.).

[0119] The above FR and DP form a hierarchical mapping relationship, that is: FR 111 With DP 111 (Related to valve orifice area and leaflet shape); FR 112 With DP 111 DP 113 (Related to the structure of the chordae tendineae and the design of the tendineae); FR 113 With DP 111 DP 112 (Structural strength) related; FR 114 With DP 112 DP 113 (Anchoring method and stability) are related; FR 115 With DP 113 (Related to chordae tendineae arrangement and tension adaptation); FR 121 FR 122 With DP12 The series (material composition, performance and degradation behavior) are directly related.

[0120] Through the above multi-level decomposition and mapping, a clear correspondence between functional requirements and design parameters is established, providing a structured foundation for subsequent modeling, simulation optimization, model training and experimental verification.

[0121] S2: Obtain the patient's clinical medical imaging data, and use deep learning methods to construct a three-dimensional reconstruction model of the original lesion structure of the human body and a parameterized structural model of the valve, respectively. The parameterized structural model responds to the anatomical structure and design structural parameter forms defined in step S1.

[0122] A three-dimensional reconstruction model of the original pathological structure of the human body is used to accurately restore the spatial anatomical morphology of the left atrium, left ventricle, anterior and posterior leaflets of the mitral valve, valve annulus, and chordae tendineae attachment structures; a parametric structural model is used for fitting design based on the morphological characteristics and clinical functional requirements of the target area. These two models together constitute the spatial constraints and structural adaptation basis of the design method of this invention, supporting the functional requirement FR. 11 (Good mechanical and opening / closing properties), FR 114 (Reliable implantation stability) and FR 115 The realization of (coordination synergy and tendineae stress adaptation) corresponds to the structural design parameter DP. 11 DP leaflet design 111 Support structure design DP 112 and connection method and anchoring design DP 113 .

[0123] The human pathological structure model is based on high-resolution clinical imaging data obtained from patients before surgery, and uses deep learning methods to analyze structures including the left atrium, left ventricle, and anterior mitral valve leaflet Ω. Aleaf , rear leaf Ω Pleaf Ω valve ring ring and chordae tendineae system Ω chord The system automatically segments and reconstructs various cardiac tissue structures, including the anterior leaflet Ω. The image data preferably uses cardiac CT or MRI images, which are standardized, spatially registered, and noise-suppressed before being input into a trained 3D image segmentation neural network model. The model output includes the anterior leaflet Ω. Aleaf posterior leaflet Ω Pleaf Ω valve ring ring Ω chord and left ventricular cavity Ω LV Voxel labels for anatomical regions.

[0124] The anterior and posterior leaflets are represented by parametric surfaces, with the anterior leaflet being... Later leaves The description, defined as follows:

[0125] in, For the summation index, For leaflet index, For the first The surface curvature coefficient of the leaf petal. For normalized local coordinates. Chordalamic system Ω chord The identification is accomplished by a trained local high-density path tracking network, which outputs the coordinates of the chordae tendineae attachment points, their path, and the connection region with the leaflets, assisting in constructing the biomechanical connection between the leaflets and papillary muscles. (Valve annulus Ω) ring Composed of closed B-splines, their spatial ellipticity, asymmetry, and curvature gradient are recorded as the reference for the fitting of the supporting structure. The complete lesion structure Ω is constructed from this. native Expressed as:

[0126] Based on this, a parameterized structural model of an absorbable biological mitral valve was further established. The model structure includes two leaflets, an elliptical support structure, and an optional artificial chordae tendineae anchoring unit, with design parameters corresponding to DP. 111 ~DP 113 The sub-items include leaflet thickness, length of the free edge of the front and rear leaflets and opening and closing angle, pitch of the support ring, radial stiffness and ellipticity of the leaflet ring, number and arrangement of tendon anchor points.

[0127] The valve structure model is generated by a structure template generator, forming a structural sample through design parameter vector input, and its spatial position is related to Ω. native Align and fit together. The curved surface of each petal. Expressed as:

[0128] in For the summation index, For leaflet index, To normalize local coordinates, The first in the leaflet design scheme The control coefficients of the leaflets can be iteratively adjusted through a reinforcement learning optimization process to meet the opening and closing performance (FR) requirements. 111 ), and compatibility (FR) 112 ) and structural stress distribution requirements (FR) 113 ).

[0129] The supporting structure model It is constructed by combining an elliptical surface of revolution with a radial elastic module, through a lobe ring Ω. ringThe surface points are automatically fitted with their inner diameter shape, and parameters such as pitch, support angle, and number of connecting ribs can be set on the periphery to construct spatial anchoring points and unfolding behavior, ensuring that FR is met during the simulation phase. 114 The stability requirements of the tendon cable anchoring structure. Composed of optional elastic columnar units, through Ω chord The path docking generates anchor point data, which, combined with the position of the leaflet edge interface, forms an effective mating support force to support the FR. 115 The required leaflet closure function must be matched with the mechanical properties of the chordae tendineae. Boolean operations are used to determine the contact and intersection between the valve structure and the lesion model, identify potential interference areas, chordae tendineae crossing paths, and high-stress areas at the leaflet base, and assist in the design of avoidance areas and reinforcement support strips.

[0130] Finally, a complete interventional valve model Ω is formed. implant Expressed as:

[0131] in To support the structural model, For tendon anchoring structure, This is the index for the union operation.

[0132] The two models are structurally connected through rigid body registration and local elastic interlocking methods, serving as the foundational geometric input for subsequent finite element mechanical simulations, long-term performance evolution modeling, and reinforcement learning strategy search. Using this parametric modeling method, this invention can rapidly construct individualized mitral valve structural models based on imaging data from different patients, and generate controllable and optimizable artificial valve structures in the design space, providing a basis for structural design parameters DP. 11 Response Functional Requirements FR 11 FR 114 and FR 115 Provide basic support.

[0133] S3: Establish a first deep learning model to simulate the interventional mechanical behavior of the valve; the training data of the first deep learning model comes from the finite element simulation results, with the structural model generated in step S2 and the user-defined material parameters as inputs, and the core performance indicators extracted from the finite element simulation as outputs, to predict the mechanical response of the valve during the intervention process.

[0134] Construction and training of the first deep learning model: using the lesion structure model Ω constructed in step S2. native With interventional valve structural model Ω implantBased on this, a complete simulation model of the valve-heart chamber system is established. The mechanical performance of the interventional device during implantation and the opening and closing cycle is simulated and analyzed. Training data is then generated to establish a model for efficient performance prediction. The simulation analysis results are used to verify the structural design parameter DP. 11 Material Design Parameters (DP) 12 Does the valve structure meet stress safety requirements under different operating conditions (FR)? 113 ), anchoring stability (FR) 114 ), on / off functionality (FR) 111 FR 112 ) and synergy (FR) 115 Core functional requirements such as ( ).

[0135] The mechanical simulation comprises two phases: the first is the deployment and positioning process during valve implantation; the second is the valve's opening and closing process and cyclic stress response during the cardiac cycle. The former is used to verify the geometric fit and anchoring stability between the interventional valve structure and the original mitral valve anatomy (annulus, chordae tendineae, papillary muscles, etc.), while the latter is used to evaluate the valve's structural response, opening and closing behavior, and long-term performance under physiological pressure changes.

[0136] In the simulation of the implantation stage, Ω implant The supporting structure is designed as an elastic expandable body, and its release process after being introduced via the apical path is simulated. The supporting structure and the valve annulus Ω... ring The contact behavior between them is modeled using a nonlinear contact algorithm. The simulation output includes the positioning deformation after unfolding and the tendineae attachment structure Ω. chord Key indicators include displacement changes, support anchoring force distribution, and contact pressure gradient. If, after support release, positional slippage, insufficient anchoring force, or localized excessive stress concentration occurs, the support design parameter DP should be assessed. 112 Anchoring structure parameters DP 113 or material design parameters DP 121 DP 122 Stability requirements not met (FR) 114 ) and the need for cooperation and coordination (FR) 115 ).

[0137] In the simulation of valve opening and closing, standard physiological cycle pressure boundary conditions are applied to simulate the two extreme conditions of ventricular systole and diastole (typical transvalvular pressure gradients such as 120 mmHg during systole and 10 mmHg during diastole). The simulation focuses on extracting the valve leaflet structure Ω. Aleaf With Ω Pleaf The maximum opening area under pressure, the occlusion integrity in the closed state, the regurgitation gap size, the maximum equivalent stress (Von-Mises) of the leaflet, the chordae tendineae tension distribution, and the annular support reaction force are all measured. These outputs are used to verify the leaflet structural design parameters DP.111 Material Design Parameters (DP) 121 DP 122 Functional requirements FR 111 (Fully open and close), FR 112 (Seamless anti-backflow), FR 113 (Structural safety) and FR 115 (Collaborative support capabilities).

[0138] To improve design optimization efficiency and support automated iterative updates, this step further establishes a deep learning model. This model replaces the complex finite element simulation process, enabling rapid performance prediction of various design combinations within the parameter space. The model input is the structural design parameter DP. 11 (e.g., leaflet thickness, leaflet curvature, chordae tendineae length, support segment distance) and material parameters DP 12 The eigenvector composed of (such as Young's modulus, degradation rate) outputs Y. t The set of performance metrics corresponding to each functional requirement: Y t =[A open ,R f ,σ max ,T chord E anchor ]

[0139] Among them, A open R is the maximum opening area of ​​the valve. f For the backflow rate, σ max For the maximum stress on the leaflet, T chord E represents the maximum tension of the chordae tendineae. anchor This is to provide the average anchoring contact stiffness between the support structure and the valve ring.

[0140] The model is trained using a multi-layer feedforward neural network structure, and a non-linear mapping relationship between input and output is constructed through supervised learning. The model's loss function, LOSS, is defined as a multi-objective weighted mean square error.

[0141] in, Here are the sample indices for the summation, and n is the total number of samples. The weighting coefficients for each performance indicator are set according to their clinical importance and design priority; and These represent the simulation result value and the predicted value, respectively.

[0142] Training samples are generated in batches using the simulation process described in the previous steps. After normalization and feature engineering, they are input into the neural network model for fitting training. Once trained, the model can predict performance under specified parameter combinations within milliseconds, significantly improving the efficiency of design space search. It can also be embedded in reinforcement learning optimization as a performance feedback channel.

[0143] S4: Construct a second deep learning model to predict the relationship between stress and material degradation behavior during long-term use of the valve. The training data for the second deep learning model comes from finite element simulation and in vitro experiments to characterize the change in the degradation rate of the material under stress and the resulting evolution of the valve's mechanical properties, and to predict whether the structural and material parameters defined in step S1 meet the requirements.

[0144] This step constructs a performance change model of the absorbable bioprosthetic valve during its service life, using deep learning methods to simulate changes in valve structure and performance caused by material degradation, and further determines the design parameter DP. 12 With DP 11 Does the functional requirement meet the target service life? 12 (Evolution of material properties) and FR 11 (Functional continuity).

[0145] This step first involves setting the material design parameters (DP). 12 Microstructural parameters, including the material's initial elastic modulus, degradation rate, polymer crystallinity, hydrophilicity, and molecular weight distribution, are used as inputs to the prediction model; simultaneously, structural design parameters DP are used. 11 Geometric factors influencing degradation behavior are input together, including leaflet thickness gradient, anterior and posterior leaflet free edge length, and stress concentration points in the connection area between the leaflet and the support structure. A time-series model is constructed to predict the evolution of the valve's mechanical response and opening / closing performance over its service life.

[0146] The prediction model employs a deep learning sequence modeling method based on a Long Short-Term Memory (LSTM) network architecture, with the model input X consisting of triples: X=[DP 11 ,DP 12 ,t]

[0147] The model output is the performance response vector P(t) at the predicted time t: P(t) = [E(t), A(t), R] f (t),σ max (t)]

[0148] Where E(t) is the equivalent elastic modulus of the material, reflecting the change in structural support performance over time; A(t) is the valve opening area, reflecting whether the valve opening and closing function continues; R f(t) represents the regurgitation rate, indicating the change in leaflet closure tightness over time; σ max (t) represents the maximum local stress value of the structure, used to assess potential fatigue and damage risks.

[0149] The model training employs a historical multi-source data fusion method, with sample data including the following two sources: a) Degradation experimental data: Mechanical tests of degradable polymer materials under standard conditions, recording the trends of modulus, strength and weight loss under different degradation cycles.

[0150] b) Multi-stage finite element simulation data: Mechanical simulations were performed under different degradation states (equivalent modulus descent gradients) to extract the valve structural response A(t), R. f (t), σ max Output metrics such as (t).

[0151] During training, standardization is used to normalize the input parameters, and the loss function LOSS uses the mean squared error (MSE) expression:

[0152] in, These are the model's predicted values. Here, i represents the experimental or simulated true value, n is the sample index for summation, t is the time index for summation, and T is the predicted time series length. The optimization process uses the Adam optimizer, automatically adjusting the learning rate to accelerate model convergence. After training, the model can perform calculations on any set of design parameters [DP]. 11 ,DP 12 At any time It enables rapid prediction of internal performance status.

[0153] The model output will be used to determine the following objectives: a) FR 121 Material performance stability: Throughout its entire life cycle, the material modulus remains within the safe threshold E. min That's all.

[0154] b) FR 122 Performance degradation controllability: The performance E(t) degradation process is continuously controllable, and the modulus change rate is stable, i.e. Bounded and continuous c)FR 11 Valve function maintenance: Valve opening area A(t) and regurgitation rate R f (t) Fluctuations remained within the clinically acceptable range during the usage period. min and R max Inside:

[0155] If the prediction results show that the performance index exceeds the safety threshold at any time, the DP can be adjusted by backtracking the input parameter vector. 11 or DP 12 Optimization and correction are carried out to achieve a closed-loop design feedback mechanism.

[0156] S5: Embed the first deep learning model and the second deep learning model into a reinforcement learning framework, and use reinforcement learning algorithms to automatically iteratively optimize the structural and material parameters of the valve, so as to simultaneously optimize the interventional performance, mechanical performance and biodegradation performance of the valve in a simulation environment, and realize end-to-end efficient optimization design from anatomical model input to design parameter output.

[0157] The design optimization steps based on reinforcement learning aim to utilize artificial intelligence methods to optimize valve structure design parameters (DP). 11 Material Design Parameters (DP) 12 Perform collaborative optimization to achieve functional requirements FR under multiple constraints. 11 (Structural opening and closing performance), FR 12 (Evolution of material properties) and FR 114 The reinforcement learning method comprehensively satisfies the requirements for (implantation stability). By constructing an interaction system between the agent and the environment, it automatically explores the optimal design solution in a high-dimensional continuous parameter space, forming an iteratively optimized intelligent decision-making process.

[0158] The reinforcement learning system comprises four core components: a state space S, an action space A, a reward function R, and an agent. The state space S is used to describe the parameter combination corresponding to the current design scheme, covering the structural design parameters DP. 11 Material Design Parameters (DP) 12 Relevant variables. The state vector includes the following components: a) Structural design parameters DP 11 Relevant variable: leaflet thickness Leaflet curvature radius Support ring diameter support pitch skirt hem width ; b) Material Design Parameters (DP) 12 Relevant variable: initial elastic modulus Degradation rate constant Polymer crystallinity molecular weight Material hydrophilicity index

[0159] Therefore, the state S at any given time t It can be represented as a multidimensional vector:

[0160] The aforementioned state vector serves as input to the constructed model and the performance evolution model, used to dynamically predict the performance indicators corresponding to this parameter combination, covering valve opening and closing function (FR). 11 ), Material Use Evolution (FR) 12 ) and structural stability (FR 114 Key evaluation points such as )

[0161] The action space A represents the adjustment method of the agent to the current design state in each round of optimization. The actions are multi-dimensional parameter increment / decrease operations or continuous value modifications, including the following forms: a) Regarding leaflet thickness b) Increase / decrease ratio; To affect the opening and closing flexibility; c) change the support pitch d) Adjusting the initial modulus of the material, which affects the development stress; (e) Adjusting the initial support properties; This alters the rate of evolution of material properties.

[0162] The actions are presented as continuous vectors, i.e. .

[0163] The reward function R(S) t ,a t The reward function is used to measure the overall performance of the current design in meeting various functional requirements and is the core objective function for reinforcement learning optimization. The reward function considers the following dimensions: a) Performance target dimension: If the valve opening area A open >A min If the backflow rate R is positive, a positive reward will be given; f <R max If the maximum stress σ is positive, a positive reward will be given; max <σ limit Deemed structurally safe, additional reward awarded; b) Material evolution: If the performance remains at E(t) ≥ E during the usage period. safe This would provide long-term rewards; c) Structural adaptability: If the intervention model Ω implant With the original structure Ω native If the spatial overlap is less than the threshold, then positive feedback is given to the chimerism. d) Penalty mechanism: If there is obvious structural slippage, severe regurgitation, valve rupture, or sudden performance degradation, a negative reward is implemented to exclude the solution.

[0164] The above objectives are expressed using a unified reward function r. t The formal representation is as follows

[0165] Where Norm(·) represents the normalized performance index, and Fit(·) represents the geometric fit function, used to evaluate the degree of integration between the interventional valve model and the original human structure. The weights for each indicator can be dynamically adjusted based on the optimization objective. The cumulative reward value R... total for

[0166] Where t is the time index of the summation, T is the prediction time series length, and r t For the reward function, The discount factor controls the weight of future rewards, ultimately outputting a design parameter combination that performs optimally in the long run.

[0167] The agent is the core computational module of the reinforcement learning optimization system, and its core task is to base its work on the current state S. t Predict the optimal action a t Receive reward r after performing the action t It updates its policy function parameter θ based on the interaction history to maximize the cumulative reward.

[0168] The agent policy function The proximal policy optimization (PPO) algorithm is used to implement iterative policy updates. The PPO algorithm is a reinforcement learning method based on policy gradients. By introducing a probability ratio clipping technique, it improves the stability of policy updates and is suitable for the continuous design variable optimization task involved in this invention.

[0169] The PPO algorithm mainly consists of two neural networks: a) Policy Network (Actor): Input current state S t Output the action change (such as the increase or decrease) corresponding to each design variable, i.e.:

[0170] θ actor The policy network parameters are represented by the following structure: Input layer: State vector S t Hidden Layer 1: Fully connected layer, 128 units, ReLU activation; Hidden Layer 2: Fully connected layer, 64 units, ReLU activation; Output Layer: Action Vector Tanh activation (constrains action range).

[0171] b) Value Network (Critic): Estimates the state-value function V(S) under the current policy. t ), used to evaluate the long-term expected return V(S) under the current strategy. t ):

[0172] θ critic The value network parameters are represented by the following structure: Input layer: State vector S t Hidden Layer 1: Fully connected layer, 128 units, ReLU activation; Hidden Layer 2: Fully connected layer, 64 units, ReLU activation; Output Layer: Single-node linear output V(S) t ).

[0173] The strategy updates the objective function L CLIP (θ) is:

[0174] Where θ represents the network parameters, r t (θ) is the probability ratio. For the estimation of the advantage function, The clipping threshold is set to 0.1, and clip(·) represents the value clipping function. This objective function encourages the policy to achieve higher rewards without deviating too far from the existing policy.

[0175] The training process for the agent policy is as follows: a) Initialize the policy network and value network; b) In conjunction with the mechanical model and long-term performance evolution model, sample multiple parameter combinations and evaluate rewards; c) Use the PPO algorithm to update the policy and maximize the cumulative reward R. total :

[0176] in For the time index of the summation, Indicates the predicted time series length. This is a discount factor used to adjust the weight of future rewards.

[0177] Ultimately, the agent can output a set of optimal design parameters. It meets the multidimensional performance constraints set by this invention in terms of valve opening and closing performance, material stability and implantability, forming an integrated solution space for structure-material synergistic optimization.

[0178] S6: Based on the optimization results of step S5, conduct simulation and experimental verification, construct a valve-heart chamber-aortic coupling simulation model and perform stent deployment, opening and closing cycle and material degradation simulation; prepare valve prototypes and conduct structural matching tests, opening and closing function tests and in vitro degradation experiments, and comprehensively evaluate the adaptability and reliability of the design parameters under the multidimensional functional requirements defined in S1.

[0179] The simulation verification step, based on the design parameters of the absorbable mitral valve obtained in this embodiment, constructs a valve-heart chamber coupling model and conducts multi-stage simulations using a commercial finite element analysis platform to comprehensively evaluate the functional performance of the designed scheme under typical clinical conditions. This simulation verification process is used to systematically determine the structural design parameter DP. 11 Material Design Parameters (DP) 12 Functional requirements under multiple boundary conditions 11 FR 12 and FR 114 Response capability.

[0180] First, based on the individualized lesion structure geometric model and interventional valve parametric model established in step S2, a valve stent implantation simulation model is constructed to conduct structural mechanics simulation analysis of the stent deployment and positioning process. The analysis simulates the radial deployment behavior of the stent along the apical release path, focusing on extracting the contact response, interlocking relationship, and spatial interference characteristics between the stent and the valve annulus and chordae tendineae. The simulation outputs the overall stent deformation, contact force distribution map, and anchoring position stability index, which are used to evaluate the supporting structure parameter DP. 112 Anchoring structure parameters DP 113 and material design parameters DP 121 DP 122 In meeting functional requirements FR 114 Performance in terms of (stable anchoring).

[0181] Secondly, a fluid-structure interaction simulation model was established to simulate the opening and closing behavior of the valve during the two key cardiac cycle phases of left ventricular systole and diastole. The simulation embeds the valve leaflet structure model into the blood flow simulation region between the left atrium and left ventricle, applying transvalvular pressure differential boundary conditions (typically 10 mmHg during diastole and 120 mmHg during systole) to simulate the dynamic opening and closing behavior of the valve leaflets under periodic loading. The simulation output includes: a) Maximum opening area A max b) Reverse flow Q regurg c) Distribution and degree of leaflet deformation and curling; d) Local maximum equivalent stress σ max e) Changes in chordae tendineae tension .

[0182] The above performance indicators serve as the design parameters for the leaflet structure (DP). 111 Material Design Parameters (DP) 121 DP 122 Functional requirements FR 111 (Open / close function), FR 112 (Closed fit) and FR 113 Evaluation criteria for (structural strength) response capability.

[0183] Furthermore, to evaluate the impact of material evolution over time on valve performance, an evolutionary simulation model incorporating the time dimension was constructed to analyze the mechanical response behavior of absorbable material valves at different time points of use (early implantation T0, mid-term T1, and late degradation T2). The model is based on the material parameter DP. 12 The evolution parameters, including elastic modulus E(t), degradation rate k, and initial elastic modulus E0, are established, and the evolution of material properties is represented by an explicit function E(t):

[0184] The time evolution simulation outputs the following timing metrics: a) Deformation of the leaflet and effective area of ​​the leaflet at each time point; b) Changes in load transfer path caused by changes in material modulus; c) Transition trajectory of local stress concentration area over time; d) Changes in stress redistribution of tendineae and fit integrity of the tight area.

[0185] The above output indicators comprehensively reflect the structural and functional changes caused by material degradation, and are used to verify the design parameter DP. 123 Does it meet FR? 121 (Machining properties retained during service life) and FR 122 (Controllability of material evolution process) and other long-term use requirements.

[0186] (9) The experimental verification steps, based on the optimization and simulation verification of the mitral valve absorbable bioprosthetic valve design parameters completed in this embodiment, further involve relevant experimental verification procedures to conduct a physical-level comprehensive evaluation of the engineering feasibility and functional achievement capability of the design scheme. The experimental verification process covers three aspects: structural prototype manufacturing, in vitro opening and closing testing, and material degradation experiments, aiming to systematically verify the structural design parameter DP. 11 Material Design Parameters (DP) 12 Functional requirements FR 11 FR 12 and FR 114 Its actual support capabilities.

[0187] First, based on the optimal combination of design parameters output by the reinforcement learning optimization module... A prototype valve structure was fabricated using medical-grade absorbable polymer materials. The leaflet structure was processed using a high-precision micro-injection molding process, and its thickness gradient, radius of curvature, and free edge shape were based on DP (Device Design). 111 The support structure is fabricated using laser cutting and heat treatment. Design parameters such as the support pitch and annular widening are based on DP. 112 Controlled molding; the chordal interface area uses an injection-molded flexible material for interface composite processing, and the anchoring features and skirt width are designed according to DP. 113 Implementation.

[0188] The manufactured valve prototype was implanted into a left atrium-left ventricle anatomical simulation model for structural matching tests to evaluate its spatial compatibility with the valve annulus, chordae tendineae, and ventricular cavity. The valve deployment state was simultaneously recorded using high-resolution video observation and X-ray projection imaging to verify the accuracy of valve positioning, the stability of anchoring, and to check for any abnormalities such as valve displacement, rotation, or interference with adjacent structures, thus evaluating dynamic plateau (DP). 112 Compared with DP113 to FR 114 (Stable anchoring) responsiveness.

[0189] Subsequently, the prototype structure underwent opening and closing performance testing in a standard simulated cardiac circulation device. The simulation platform was loaded with representative left atrial and left ventricular pressure waveforms (systolic: 120 mmHg, diastolic: 10 mmHg), and simultaneously equipped with a high-speed camera system, differential pressure sensor, and flow recorder. During the test, the synchronization between the valve leaflet opening and closing rhythm and pressure changes was dynamically observed; the effective opening area corresponding to the maximum valve opening state, the regurgitation flow rate in the closed state, and the leaflet adhesion integrity were recorded. The imaging results were used to determine whether the structural and material design parameters supported FR (Fluorescence Reduction). 111 (Open / close function) and FR 112 Performance in terms of (sealing)

[0190] Finally, long-term in vitro degradation experiments were conducted on the biodegradable polymer materials from the same batch as the prototype. Standardized sample strips were immersed in simulated body fluid (PBS solution) in a constant temperature environment (37°C). Samples were periodically taken to measure their mass, elastic modulus, tensile strength, and elongation at break, and performance change curves were constructed. The obtained experimental data were compared with the output of the material evolution model established in step (6) to verify the degradation rate, performance retention period, and degradation performance. 12 Whether the design values ​​are consistent, thereby assessing whether FR is met. 121 (Performance retention) and FR 122 (Evolutionary controllable) functional requirements.

[0191] The above embodiments are merely illustrative of several implementations of the present invention and should not be construed as limiting the scope of the present invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the claims.

Claims

1. A method for designing absorbable bioprosthetic valves based on artificial intelligence and biomechanics, characterized in that, Includes the following steps: S1: Construct a performance requirement system for valves, decompose functional requirements into functional indicators in the functional domain, and map them into structural parameters and material parameters in the physical domain; S2: Acquire the patient's clinical medical imaging data, and use deep learning methods to construct a three-dimensional reconstruction model of the original lesion structure of the human body and a parameterized structural model of the valve, respectively. The parameterized structural model responds to the structural parameters defined in S1. S3: Establish a first deep learning model to simulate the interventional mechanical behavior of the valve; the training data of the first deep learning model comes from the finite element simulation results, with the structural model generated in S2 and the material parameters defined in S1 as inputs, and the core performance indicators extracted from the finite element simulation as outputs, to predict the mechanical response of the valve during the intervention process; S4: Construct a second deep learning model to predict the relationship between stress and material degradation behavior during long-term use of the valve. The training data for the second deep learning model comes from finite element simulation and in vitro experiments to predict whether the structural and material parameters defined in S1 meet the requirements. S5: Embed the first deep learning model and the second deep learning model into a reinforcement learning framework, and use reinforcement learning algorithms to automatically iteratively optimize the structural parameters and material parameters of the valve, thereby optimizing the interventional performance, mechanical performance and biodegradability of the valve. S6: Based on the design results of S5, conduct simulation and experimental verification to evaluate the adaptability and reliability of the design parameters under the functional requirements defined in S1.

2. The method for designing absorbable bio-valve based on artificial intelligence and biomechanics according to claim 1, characterized in that, The structural parameters of the valve include the outer diameter of the stent after deployment, the stent length, the thickness of the stent support structure, the number and distribution of stent segments, the stent mesh size, the geometric parameters of the stent support ribs and connecting ribs, and the number, thickness, and shape and size parameters of the valve leaflets. The material parameters of the valve include the composition and content of each component of the stent material, the molecular weight distribution of the material, the Young's modulus, Poisson's ratio, tensile strength, and the degradation rate of the material. The performance indicators of the valve include the radial recoil rate, axial shortening rate after interventional implantation, the radial stiffness of the valve under cyclic loading, the stress safety factor, and the fatigue life, as well as the degradation rate, mechanical strength retention rate, and time required for complete degradation of the material during its service life.

3. The method for designing absorbable bio-valve based on artificial intelligence and biomechanics according to claim 3, characterized in that, The clinical medical imaging data used in step S2 includes computed tomography (CT) images and / or magnetic resonance imaging (MRI) images; the three-dimensional reconstruction model of the original human lesion structure includes a three-dimensional morphological model of the aorta, aortic valve leaflets, valve annulus, left ventricular outflow tract, and calcified plaque area; the parametric structural model of the valve includes at least three leaflet structures and a support ring structure.

4. The method for designing absorbable bio-valve based on artificial intelligence and biomechanics according to claim 3, characterized in that, The parametric structural model also includes a skirt anchoring structure. The parametric structural model is generated based on structural design parameters and spatially fits with the three-dimensional reconstruction model of the original human lesion structure. The valve leaflet structure is defined by a polynomial parametric surface. The support ring structure is generated using rotational geometry and spatially matched with the aortic valve annulus region through Boolean operations or a fitting method.

5. The method for designing absorbable bio-valve based on artificial intelligence and biomechanics according to claim 4, characterized in that, In step S3, the predicted mechanical response of the valve during the intervention process includes at least one of the following: valve structural stress safety, anchoring stability, opening and closing functionality, and occlusion coordination.

6. The method for designing absorbable bio-valve based on artificial intelligence and biomechanics according to claim 5, characterized in that, The first deep learning model is a regression-type deep neural network model, which is used to approximate the mechanical behavior of valve stents during interventional implantation. The deep learning model is trained on finite element simulation data of valve stent interventional processes under different structural and material parameters to predict the mechanical response index of the stent.

7. The method for designing absorbable bioprosthetic valves based on artificial intelligence and biomechanics according to claim 6, characterized in that, In S4, it is predicted whether the structural and material parameters defined in step S1 meet the requirements of material performance evolution and functional continuity.

8. The method for designing absorbable bio-valve based on artificial intelligence and biomechanics according to claim 7, characterized in that, The second deep learning model is trained by combining historical simulation and degradation experimental data to simulate changes in valve structure and performance caused by material degradation, in order to evaluate the valve's performance and usage risks throughout its entire life cycle.

9. The method for designing absorbable bioprosthetic valves based on artificial intelligence and biomechanics according to claim 8, characterized in that, The reinforcement learning framework in step S5 adopts an agent-environment interaction mode, in which the agent generates a valve structure design scheme based on the current design parameters.

10. The method for designing absorbable bio-valve based on artificial intelligence and biomechanics according to claim 9, characterized in that, In step S6, a valve-heart chamber-vascular coupling simulation model is constructed and simulations of stent deployment, opening and closing cycles, and material degradation are performed; a valve prototype is prepared and structural matching tests, opening and closing function tests, and in vitro degradation experiments are carried out to comprehensively evaluate the adaptability and reliability of the design parameters under the multidimensional functional requirements defined in S1.

Images