Abstract

Recently, 3D-printed biodegradable scaffolds have shown great potential for bone repair in critical-size fractures. The differentiation of the cells on a scaffold is impacted among other factors by the surface deformation of the scaffold due to mechanical loading and the wall shear stresses imposed by the interstitial fluid flow. These factors are in turn significantly affected by the material properties, the geometry of the scaffold, as well as the loading and flow conditions. In this work, a numerical framework is proposed to study the influence of these factors on the expected osteochondral cell differentiation. The considered scaffold is rectangular with a 0/90 lay-down pattern and a four-layered strut made of polylactic acid with a 5% steel particle content. The distribution of the different types of cells on the scaffold surface is estimated through a scalar stimulus, calculated by using a mechanobioregulatory model. To reduce the simulation time for the computation of the stimulus, a probabilistic machine learning (ML)-based reduced-order model (ROM) is proposed. Then, a sensitivity analysis is performed using the Shapley additive explanations to examine the contribution of the various parameters to the framework stimulus predictions. In a final step, a multiobjective optimization procedure is implemented using genetic algorithms and the ROM, aiming to identify the material parameters and loading conditions that maximize the percentage of surface area populated by bone cells while minimizing the area corresponding to the other types of cells and the resorption condition. The results of the performed analysis highlight the potential of using ROMs for the scaffold design, by dramatically reducing the simulation time while enabling the efficient implementation of sensitivity analysis and optimization procedures.

中文翻译：

---

用于骨组织工程支架设计的可解释的基于机器学习的概率框架

摘要

最近，3D打印的可生物降解支架在临界尺寸骨折的骨修复方面显示出巨大的潜力。除其他因素外，支架上细胞的分化还受到由于机械载荷导致的支架表面变形和间质液流施加的壁剪切应力的影响。这些因素反过来又受到材料特性、支架几何形状以及负载和流动条件的显着影响。在这项工作中，提出了一个数值框架来研究这些因素对预期骨软骨细胞分化的影响。所考虑的脚手架是矩形的，具有 0/90 的铺设图案和由钢颗粒含量为 5% 的聚乳酸制成的四层支柱。通过使用机械生物调节模型计算的标量刺激来估计支架表面上不同类型细胞的分布。为了减少刺激计算的模拟时间，提出了一种基于概率机器学习（ML）的降阶模型（ROM）。然后，使用 Shapley 附加解释进行敏感性分析，以检查各种参数对框架刺激预测的贡献。最后一步，使用遗传算法和 ROM 实施多目标优化程序，旨在确定材料参数和负载条件，以最大化骨细胞填充的表面积百分比，同时最小化与其他类型细胞相对应的面积和吸收条件。执行分析的结果凸显了使用 ROM 进行支架设计的潜力，可以显着减少模拟时间，同时能够有效实施灵敏度分析和优化程序。