Shoulder musculo-skeletal modelling: from muscle path refinement to optimal control based on direct multiple shooting

肩部疾病的发病率仅次于背痛。由于肩部肌肉冗余和多关节肌肉的存在，如果没有肌肉力的估计，很难确定每块肌肉对关节活动性和稳定性的贡献。这种分析有助于改善神经和骨科疾病的诊断和治疗。由于肌肉力量的动态测量在临床环境中是不可实现的，因此肌电图（EMG）和运动学的非侵入性协议将与肌肉骨骼（MSK）模型相结合来估计肌肉力量。使用这种方法，我们最近设计了一种创新的肩部术后矫形器。然而，从我们和其他人的经验来看，目前的MSK模型已经被发现在更个性化的临床和工业应用中失败。在力量能力和肌肉协调策略方面，模型不是特定于受试者的。此外，到目前为止，建模工作还没有集中在肩袖肌肉上，而最常见和最昂贵的肩部疾病与这些从肩胛骨到肱骨的深层肌肉有关。我们的5年目标是确定个性化的肩部肌肉参数，特别强调肩袖肌肉（特定目标，SO1），利用肌电图和运动学数据来估计职业任务中准确的肌肉力量（SO2）。最优控制和辨识问题将采用直接多次射击方法来解决。这种方法在机器人学和生物力学中得到了很好的应用，但从未应用于MSK模型。一个补充目标（SO3）是在MSK模型中改进肩袖的几何形状。SO1：为了确定肌肉参数，受试者将在测力计上进行一系列等速和等速最大自主收缩。多样化的运动是必不可少的，因为没有一个单一的运动能完全激活所有的肩部肌肉。SO2：与现有的肌肉力估计方法（假设无噪声运动学）相比，我们方法的新颖之处在于，在重复标准化任务时，最优解仍在受试者内部可变性的约束下，联合优化运动学和肌电数据。关节运动学的预期误差也将增加到可变性中，因为在皮肤上放置标记的临床方案会导致肩骨运动学的实质性误差。在过去的3年里，我们专注于估计肩关节的运动学，但其精度仍然未知。基于皮肤标记的运动学将与使用金标准的骨骼运动学进行比较，即在锁骨，肩胛骨和肱骨上钉入销钉。SO3：最后，应该使用弹簧网来获得更多的生理肌肉路径，而不是以非生理方式在肱骨头上分散的独立动作线。网格应该复制的几何形状的肌腱相互指间形成一个袖在肱骨头。尸体的肩膀将被用来验证几种手臂配置的模型。我们的方法将为特定学科的技术提供最佳的肌肉力量估计。本课题是利用最先进的设备和最优的多体系统控制算法，培养4名研究生和10名本科生实验人体生物力学的绝佳机会。我们的方法有望成为最先进的肩部MSK建模，特别是研究肩袖损伤。完善仿真模型和改进肩部最优控制方法是促进肩部矫形器发展、改善肩部肌肉撕裂手术修复和预防工作场所肩部损伤项目的基石。

Shoulder disorder prevalence stands just after back pain. Due to shoulder muscle redundancy and the presence of polyarticular muscles, the contribution of each muscle to joint mobility versus stability is difficulty to determine without an estimate of muscle forces. Such analyses can contribute to improved diagnosis and treatment of neurological and orthopedic diseases. Since dynamic measure of muscle forces are unfeasible in clinical settings, non-invasive protocols with electromyography (EMG) and kinematics are to be combined with musculoskeletal (MSK) models to estimate muscle forces. Using this approach, we recently design an innovative shoulder post-operative orthosis. However, from our and other’s experience, current MSK models have be found to fail in more personalized clinical and industrial applications. Models are not subject-specific in terms of strength capacity and muscle coordination strategies. Furthermore, to date modeling efforts have not focused on the rotator cuff muscles, while the most frequent and costly shoulder complaints are related to these deep muscles, which run from the scapula to the humerus. Our 5-year objective is to identify personalized shoulder muscle parameters with a particular emphasis on rotator cuff muscles (specific objective, SO1) to estimate accurate muscle force in occupational tasks (SO2), using both EMG and kinematic data. Both optimal control and identification problems will be solved using a direct multiple shooting method. This method is well-tried in robotics and biomechanics but has never been applied to MSK models. A complementary objective (SO3) is to refine the rotator cuff geometry in the MSK model. SO1: To identify muscle parameters, subjects will perform series of isometric and isokinetic maximal voluntary contractions on a dynamometer. Diversity of exercises is essential since no single movement fully activates all shoulder muscles. SO2: In contrast to existing methods for estimating muscle forces, which assume noiseless kinematics, the novelty of our approach is to jointly optimize kinematic and EMG data, under the constraint that the optimal solution remains within intra-subject variability when repeating a standardized task. An expected error in joint kinematics will also be added to variability, since clinical protocols with markers put on the skin lead to substantial errors in shoulder bone kinematics. While in the last 3 years we focused on estimating of shoulder joint kinematics, the accuracy is still unknown. Skin marker-based kinematics will be compared to skeletal kinematics using a gold standard, namely pins screwed in clavicle, scapula and humerus. SO3: Finally, more physiological muscle paths should be obtained using a mesh of springs instead of independent lines of action that spread on the humeral head in a non-physiological manner. The mesh should reproduce the geometry of the tendons which interdigitate with each other to form a cuff around the humeral head. Cadaveric shoulders will be used to validate the model in several arm configurations. Our method will provide the best estimate of muscle forces for a subject-specific technique. This proposal is an outstanding opportunity to train 4 graduate students and 10 undergraduates in both experimental human biomechanics with the most advanced equipment and optimal control algorithms of multibody systems. Our method is expected to become the state-of-the-art in shoulder MSK modeling, particularly to study rotator cuff injuries. Refining simulation models and improving methods for shoulder optimal control are the cornerstone of projects leading to shoulder orthotic development, improvements in surgical restoration of shoulder muscle tears and in the prevention of shoulder injuries in the workplace.