Tissue scaffolds have emerged as a promising solution for treatment of critical size bone defects, offering significant advantages over conventional strategies. One of the key functionalities of bone scaffolds is their ability to promote long-term bone ingrowth effectively. To enhance this functionality, we develop a novel dynamic optimisation framework to customise bone scaffolds for achieving maximum bone ingrowth outcomes over a certain period in this study. To improve the design efficiency, we extensively leverage machine learning (ML) techniques within our proposed dynamic optimisation framework. Specifically, two neural networks are integrated into a dynamic bone growth model, and another neural network is coupled with a genetic algorithm for dynamic optimisation process. To demonstrate the effectiveness and efficiency of the approach, we employ a sheep mandible reconstruction for treating a critical size bone defect as an illustraive example. To validate the finite element (FE) model established, we first conduct a mechanical test on the sheep mandible assembled with a tailored 3D printed scaffold made of Polyetherketone (PEK) material. Then, we compare three different optimisation schemes, namely uniform design, lateral gradient design, and vertical gradient design, with an empirical design under the same biomechanical conditions. A 18.5 % enhancement is found in the long-term bone ingrowth when the optimised scaffold is adopted in comparison with the empirical design, which is attributed to the fine-tuning of strut sizes within lattice scaffold structures for facilitating bone regeneration in the gradient regions. This study proposes a novel design framework by combining ML and time-dependent topology optimisation, which provides a new methodology for developing innovative tissue scaffolds with better clinical outcomes.

中文翻译：

---

使用机器学习技术动态优化分级组织支架

组织支架已成为治疗临界尺寸骨缺损的有前途的解决方案，与传统策略相比具有显着优势。骨支架的关键功能之一是能够有效促进长期骨向内生长。为了增强这一功能，我们开发了一种新颖的动态优化框架来定制骨支架，以在本研究的特定时期内实现最大的骨向内生长结果。为了提高设计效率，我们在提出的动态优化框架中广泛利用机器学习（ML）技术。具体来说，将两个神经网络集成到动态骨骼生长模型中，另一个神经网络与遗传算法结合进行动态优化过程。为了证明该方法的有效性和效率，我们采用羊下颌骨重建来治疗临界尺寸骨缺损作为说明性示例。为了验证建立的有限元 (FE) 模型，我们首先对使用聚醚酮 (PEK) 材料制成的定制 3D 打印支架组装的羊下颌骨进行机械测试。然后，我们将三种不同的优化方案，即均匀设计、横向梯度设计和垂直梯度设计与相同生物力学条件下的经验设计进行比较。与经验设计相比，采用优化支架时，长期骨向内生长提高了 18.5%，这归因于网格支架结构内支柱尺寸的微调，以促进梯度区域的骨再生。这项研究提出了一种新颖的设计框架，将机器学习和时间相关的拓扑优化相结合，为开发具有更好临床效果的创新组织支架提供了一种新方法。