A Transfer Learning Framework for Creating Subject-Specific Musculoskeletal Models of the Hand

项目总结 恢复手功能对于许多临床疾病来说仍然是一个难以实现的目标，包括中风、骨关节炎、 四肢瘫痪、截肢和创伤性损伤。手的解剖学复杂性使得恢复手的功能 尤其具有挑战性，因为更改手中的任何一个参数都可能产生级联效应， 很难预测，但对控制是必不可少的。在本提案中，作为向个性化提供信息的关键一步 对于手的治疗，我们将研究不同学科的差异如何影响手的功能。完成 这一建议将依赖于三个数据集的收集，这些数据集旨在提供不同水平的生物力学 详细信息，需要付出不同程度的努力才能收集。简而言之，这些数据集包括(1)模拟数据集 包含500,000个模拟，完全描述了与手力产生有关的所有肌肉骨骼参数， (2)描述运动学、动力学和肌肉活动所需的密集的生物力学数据集 (3)描述人口统计学的稀疏、临床启发的数据集， 1000名成人手功能的人体测量学和临床测量。在目标1中，我们将利用前两个 数据集来设计数据驱动的分析框架，以确定最重要的生物力学 参数(S)，并映射这些参数如何影响手力产生。完成这一目标将 阐明调节手力产生的生物力学机制并评估其使用能力 用模拟数据而不是实验数据来识别这些机制。在目标2中，我们将充分利用这三项 数据集，以创建能够高效、准确地预测特定学科的迁移学习框架 肌肉力量产生参数来自于易于收集的临床数据。我们特别关注肌肉力量- 生成参数因为这些参数仍然难以快速准确地估计，所以 已知在不同人群中有所不同，并与力量等功能指标高度相关。完成这项工作 AIM将提供一种新的方法来快速估计特定于受试者的肌肉骨骼参数，从而 使得能够有效地创建特定于对象的模型，并且潜在地催化在临床中使用这些模型 布景。总体而言，这项研究的结果可以增强我们提供个性化诊断和 手部损伤患者的预后。

PROJECT SUMMARY Restoring hand function remains an elusive goal for many clinical conditions, including stroke, osteoarthritis, tetraplegia, amputation, and traumatic injury. The hand’s anatomical complexity makes restoring hand function particularly challenging because altering any one parameter in the hand can have cascading effects that are difficult to predict, but essential to control. In this proposal, as a critical step toward informing personalized treatments for the hand, we will study how subject-specific differences influence hand function. Completion of this proposal will rely on collection of three datasets that are designed to provide varying levels of biomechanical detail and require varying levels of effort to collect. Briefly, these datasets include (1) a simulation dataset containing 500,000 simulations fully describing all musculoskeletal parameters involved in hand force production, (2) a dense, biomechanical datasets that describes the kinematics, kinetics, and muscle activity required for hand force production in 30 adults, and (3) a sparse, clinically-inspired dataset that describes demographics, anthropometrics, and clinical metrics of hand function in 1000 adults. In Aim 1, we will leverage the first two datasets to design a data-driven analysis framework that identifies the most important biomechanical parameter(s) and maps how those parameters influence hand force production. Completion of this aim will elucidate the biomechanical mechanisms that modulate hand force production and evaluate the ability to use simulation data, instead of experimental data, to identify these mechanisms. In Aim 2, we will leverage all three datasets to create a transfer learning framework capable of efficiently and accurately predicting subject-specific muscle force-generating parameters from easy to collect clinical data. We specifically focus on muscle force- generating parameters because these parameters remain challenging to quickly and accurately estimate, are known to vary across the population, and are highly related to functional metrics like strength. Completion of this aim will provide a new approach for rapidly estimating subject-specific musculoskeletal parameters, thereby enabling efficient creation of subject-specific models and potentially catalyzing use of such models in a clinical setting. Overall, the results from this study could enhance our ability to provide personalized diagnoses and prognoses for individuals suffering from hand impairments.