Efficient Methods for Multi-Domain Biomechanical Simulations

描述（由申请人提供）：在计算生物力学中，有两个发展良好但独立的建模领域：用于身体运动的多体动力学和用于组织变形的有限元建模。然而，许多临床问题跨越这两个领域。全身解剖结构、质量分布和步态模式通常不在有限元模型中表示，但这些是影响肌肉骨骼系统中组织应力的重要现实因素，这可能导致骨关节炎和糖尿病足溃疡等临床问题。另一方面，运动模拟缺乏组织变形的表示，组织变形是机械引起的疼痛和其他感觉反馈（或缺乏感觉反馈）的指标，并且将导致步态的可观察变化。这些神经肌肉骨骼整合机制的探索只能通过多域模拟来完成。用于多域建模的当前技术是不够的，因为前向动态运动模拟通常在时间上沿着许多小步的序列进行沿着。有限元模型速度太慢，无法在每个步骤中找到解决方案。人们可能会煞费苦心地产生一个单一的运动模拟，但不是预测运动优化所需的数千个模拟，这是肌肉骨骼动力学的最新技术。这已经成为我们自己和其他人研究的瓶颈。因此，我们的第一个目标是实现一个通用的，自我完善，代理建模方案，其目的是在给定的误差容限内重现一个底层的基于物理的有限元模型，但在一个低得多的计算成本。自细化特性是再现典型节理或节理复合体有限元模型的多维输入输出空间的关键。我们的第二个目的是通过将足部的有限元模型连接到完整的肌肉骨骼步态模拟来证明这些工具的实用性，该模拟将测试峰值足底压力（糖尿病足溃疡的指标）的假设，可以通过在步态期间选择特定的最佳肌肉协调模式来降低安全阈值。拟议的研究将通过提供基本的替代建模算法来推进斯坦福大学生物医学计算中心的计算环境，这些算法可能适用于其他基于多尺度物理的问题，并扩展中心在神经肌肉生物力学方面的努力。

DESCRIPTION (provided by applicant): In computational biomechanics, there are two well-developed but separate modeling domains: multi body dynamics for body movements, and finite element modeling for tissue deformations. Many clinical problems, however, span both domains. Whole body anatomy, mass distribution, and gait pattern are not typically represented in finite element models, yet these are important real-world factors that affect tissue stresses in the musculoskeletal system, which may contribute to clinical problems such as osteoarthritis and diabetic foot ulceration. Movement simulations, on the other hand, lack a representation of tissue deformations, which are indicators of mechanically induced pain and other sensory feedback (or the lack thereof) and will cause observable changes in gait. Exploration of these neuromusculoskeletal integrative mechanisms can only be accomplished by multi-domain simulations. Current techniques for multi-domain modeling are insufficient because forward dynamic movement simulations typically proceed along a sequence of many small steps in time. Finite element models are too slow to allow a solution at each of these steps. One may painstakingly produce a single movement simulation, but not the thousands of simulations that are required for predictive movement optimizations that are the state of the art in musculoskeletal dynamics. This has become a bottleneck for our own research, as well as for others. Our first aim, therefore, is to implement a generic, self-refining, surrogate modeling scheme, which aims to reproduce an underlying physics-based finite element model within a given error tolerance, but at a far lower computational cost. The self-refining feature is the key to reproduce the multi-dimensional input-output space of a typical finite element model of a joint or joint complex. Our second aim is to demonstrate the utility of these tools by connecting a finite element model of the foot to a complete musculoskeletal gait simulation, which will test the hypothesis that peak plantar pressures (an indicator of diabetic foot ulceration), can be lowered under safety thresholds by selecting a specific optimal muscle coordination pattern during gait. The proposed research will advance the computational environment at the Stanford Center for Biomedical Computation by providing basic surrogate modeling algorithms that are potentially applicable to other multiscale physics-based problems and also extend Center's efforts in neuromuscular biomechanics.