Multi-objective optimization of coronary artery stent design in ensemble surrogate model

Abstract

BACKGROUND: Percutaneous coronary intervention stent implantation is primarily used to treat coronary artery stenosis. However, current multi-objective stent optimization methods are limited by sample size constraints, resulting in insufficient prediction accuracy when balancing key performance indicators such as support and compliance, hindering the effectiveness of stent optimization design. OBJECTIVE: To establish an innovative optimization framework for coronary stents based on a ensemble surrogate model. METHODS: A three-dimensional parametric model of the vascular stent was constructed, and a mechanical response database was established through finite element simulation. A dynamic weight fusion strategy was adopted to integrate the global optimization characteristics of the Kriging model and the local nonlinear representation advantages of the radial basis function model. A ensemble surrogate model was constructed based on 20 groups of initial samples, and the non-dominated sorting genetic algorithm-II was used to optimize the parameter space. RESULTS AND CONCLUSION: Experimental results demonstrated that the ensemble surrogate model exhibited significant advantages in the finite sample setting. The coefficient of determination for the inverse prediction of the radial stiffness of the stent reached 0.974 2, a 4.4% improvement compared to the single model, validating the efficient modeling capability of the ensemble surrogate model in the finite sample setting. The prediction accuracy of the stent's bending stiffness also improved by 4.4% compared to the single radial basis function surrogate model. After optimization, the stent performance achieved dual-objective synergistic optimization. The inverse radial stiffness of the stent in the ensemble surrogate model group was reduced by 13.92% and 9.57% compared to the Kriging model group and the single radial basis function surrogate model group, respectively. The bending stiffness of the stent was optimized by 0.38% and 2.56% compared to the Kriging model group and the single radial basis function surrogate model group, respectively. The proposed ensemble surrogate model breaks through the performance limitations of traditional single models, providing a low-cost, high-precision solution for the 'rigid-flexible' synergistic optimization of coronary stents.

1. Introduction

Percutaneous coronary intervention is a core treatment for coronary atherosclerotic heart disease, widely used for coronary artery stenosis through intravascular intervention to achieve revascularization [1-4]. However, the mechanical properties of currently used stents present a prominent contradiction: the support and flexibility of stents are difficult to optimize synergistically. Traditional metal stents (e.g., 316L stainless steel) have strong support but poor flexibility, which may lead to malapposition and distal vascular injury. New materials such as high-nitrogen nickel-free stainless steel improve flexibility but may reduce support, potentially causing acute collapse. This results in effective improvement of hemodynamics after stent implantation, but complications such as in-stent restenosis still limit long-term efficacy [5]. Therefore, further improving stent performance, especially reducing complications during long-term use, has become a research focus.

Finite element analysis has become a preferred method for studying stent performance due to its speed and low cost [6-9]. The geometric configuration of stents significantly affects their performance. For example, PANT et al. [10] found that the width and axial length of the stent ring cross-section significantly influence stent performance. SONG et al. [11] showed that insufficient support or excessive recoil can lead to severe complications such as acute collapse, late disruption, and malposition, further increasing cardiovascular risk. LI Ning et al. [12] emphasized that good flexibility not only ensures smooth passage through complex vessels but also reduces malapposition and vessel wall injury, thereby lowering restenosis rates. However, PANT et al. [10] found a trade-off between support and flexibility: improving support often affects flexibility and vice versa. To enhance overall performance, multi-objective optimization of stent design is needed.

By constructing surrogate models, the dynamic relationship between design variables and mechanical properties can be accurately mapped, effectively reducing optimization costs, computational noise, and iteration cycles [13]. Common surrogate models include Kriging [14], radial basis function (RBF) [15], and support vector machine models [16]. Existing stent multi-objective optimization methods often rely on a single surrogate model and require large sample data, which has obvious limitations in practice: when sample size is insufficient or multiple objectives are coupled, a single surrogate model struggles to simultaneously handle multiple objectives and complex engineering constraints [17]. To address this, some scholars have developed ensemble surrogate models to achieve better optimization performance. By integrating the strengths of different models, ensemble surrogate models can significantly improve optimization accuracy with limited samples, effectively balance multiple performance objectives, and yield better solutions than single models [18-19]. For example, ZHENG Linqing et al. [20] applied ensemble surrogate models to construct part surface topography, reducing measurement points and improving model accuracy; CHEN Yang et al. [21] used ensemble surrogate models in the lightweight design of bridge crane box girders, significantly improving structural optimization efficiency while achieving mass reduction.