A data-driven optimization framework for improving the adaptation of the neuromuscular system in brain pathology

该项目旨在建立一个新的计算机框架，可用于解释神经肌肉骨骼系统对中风，脑瘫或多发性硬化症等脑病理的适应机制。这些病理显著限制了受影响受试者的运动能力，然而，不幸的是，不存在令人满意的治疗。最终目标是支持开发新的和改进现有的治疗应用。 基于身体旨在以最佳方式处理给定条件的概念，我们打算使用约束优化的数学技术来实现这一目标。通过采用神经肌肉骨骼系统的系统多尺度模型，我们期望这样的方法可以对真实的生理系统做出有意义的预测。然而，考虑到这种框架在建模、计算和数学方面的高复杂性，这种方法从未被尝试过。为了实现这一愿景，我们的目标是将我们的多尺度神经肌肉模型和我们的3D连续机械肌肉骨骼模型结合起来，预计将做出以下贡献：我们将整合运动控制和大脑病变的新数学模型。此外，现有的神经肌肉建模工具箱需要通过异源反馈回路、重塑过程和肌肉代谢来丰富。 为了提供一个灵活的仿真和优化框架，可交换的组件，我们打算建立一个分区的仿真框架。这需要新的技术和数值耦合方法，以及处理大脑病理学短期和长期反应之间的多尺度特性的概念。基于这些新模型的优化再次需要模型-数学-HPC协同设计：（i）反映神经肌肉系统的高级目标和优化参数的目标函数，即，神经肌肉骨骼模型中的自由度，其表示对给定扰动的允许的短期或长期适应;（ii）需要开发优化框架的其他组件，特别是替代模型以降低计算成本，伴随如果我们遵循拉格朗日方法，以及外部优化框架本身的实现。对于潜在的未来临床应用（超出本项目的范围），需要考虑对我们的组合优化框架的进一步数据处理挑战。所有任务都需要PI组中收集的专业知识在模型，数值，HPC和数据之间的协同设计意义上的密切互动。

This projects aims to establish a novel in silico framework that can be used to explain adaptation mechanisms in the neuro-musculoskeletal system in response to brain pathology such as stroke, cerebral palsy or multiple sclerosis. These pathologies significantly limit motor abilities in affected subjects, however, satisfactory treatments do not exist unfortunately. The ultimate goal is to support the development of novel and the improvement of existing therapeutic applications. Based on the concept that the body aims to adapt in such a way as to optimally deal with the given conditions, we intend to use mathematical techniques from constrained optimization to tackle this goal. By employing a systemic multi-scale model of the neuro-musculoskeletal system, we expect that such an approach can make meaningful predictions for the real physiological system. However, given the high complexity that such a framework demands with respect to modeling, computation and mathematics, such an approach has never been attempted. To achieve this vision we aim to unite our multi-scale neuromuscular model and our 3D continuum-mechanical musculoskeletal model, for which the following contributions are foreseen:We will integrate new mathematical models of motor control and brain lesions. In addition, the existing neuromuscular modeling toolbox need to be enriched by heteronymous feedback circuits, remodeling processes and muscle metabolism. To provide a flexible simulation and optimization framework with exchangeable components, we intend to set up a partitioned simulation framework. This requires new technical and numerical coupling methods as well as concepts for handling multi-scale properties between short-term and long-term reactions to brain pathology.Optimization based on these new models again requires model-mathematics-HPC co-design: (i) objective functions that reflect the high-level goals of the neuromuscular system and optimization parameters, \ie, the degrees of freedom in the neuro-musculoskeletal model that represent the permissible short- or long-term adaptation to a given perturbation; (ii) further components of the optimization framework need to be developed, in particular surrogate models to reduce the computational cost, adjoints if we follow a Lagrangian approach, and the implementation of the outer optimization framework itself. For potential future clinical applications (beyond the scope of this project), further data handling challenges to our composable optimization framework need to be considered. All tasks require a close interaction between the expertise gathered in the groups of the PIs in the sense of co-design between models, numerics, HPC and data.