Shoulder musculoskeletal modeling: from data-tracking to predictive simulations

我的长期目标是模拟仿生和最佳上肢运动，以改善肩部功能或减少肩部疾病的危险因素。在生物力学中，内部负荷（肌肉肌腱和关节力）对于理解人类如何运动和预测功能结果是必不可少的。这种负荷可以用神经-肌肉-骨骼[NMSK]模型来估计。然而，在NMSK建模中，肌肉冗余仍然是一个未解决的问题。我们的方法的独创性在于同时跟踪肌电图和皮肤标记来估计内部负荷。它很有前景，不仅用于数据跟踪模拟，而且用于预测模拟（即在没有实验数据的情况下生成最佳和生物形态的运动）。然而，技术转化为临床/人体工程学应用的主要障碍仍然存在；主要的与我的具体目标相对应：SO1)使用在线NMSK数据跟踪模拟提供内部负载的实时生物反馈；SO2)识别参与者特有的肌肉-肌腱特性；SO3)将数据跟踪算法转换为包含运动控制理论（肌肉协同作用和运动学理论）的预测模拟。***SO1)为了加速优化过程并向患者和临床提供反馈，将最优控制问题表示为一个非线性运动水平估计（50-100 ms时间跨度），由于变量较少，收敛性增强。由于并非所有的肌电信号都可以系统地测量，因此将使用长短期记忆神经网络从执行各种任务的人群（n=30）的数据中推断出缺失的肌电信号。***SO2)为了个性化肌肉肌腱的特性，学生将首先集中于识别最大的等速肌肉力，最佳长度，以及肌电图和神经兴奋之间的非线性形状因素，使用一系列（次）最大的努力在等速测力仪上进行。来自系统生物学的识别算法将适用于NMSK模型。***SO3)肌肉协同效应将首先从我们的大型肌电图数据库中提取。协同作用的使用将减少控制空间，并可以加强肌肉兴奋的生物形态（例如复制肩关节稳定的共同收缩）。此外，手的速度将根据运动学理论约束，以指导优化的现实解决方案。将使用先前收集的运动来验证最佳解决方案，以确定最相关的目标函数，约束和运动控制理论，以生成现实的运动。***我的发现计划提案将支持4名博士和10名本科生，他们将在多学科环境和最先进的基础设施中接受高级肌肉骨骼生物力学建模的培训。我们突破性的算法将成为临床，体育，艺术和人体工程学应用的基础。**

My long-term objective is to simulate biofidelic and optimal upper-limb movements to improve shoulder function or reduce risk factors for shoulder disorders. In biomechanics, internal loads (muscle-tendon and joint forces) are essential for understanding how humans move and for predicting functional outcome. Such loads can be estimated using neuro-musculoskeletal [NMSK] models. However, muscle redundancy remains an unsolved problem in NMSK modeling. The originality of our approach has been to track simultaneously EMG and skin markers to estimate the internal loads. It is promising, not only for data-tracking simulations, but also for predictive simulations (i.e. generating optimal and biofidelic movements, without experimental data). However, major obstacles remain for technology translation to clinical/ergonomic applications; the major ones correspond to my specific objectives: SO1) Provide real-time biofeedback of internal loads using an online NMSK data-tracking simulation; SO2) Identify participant-specific muscle-tendon properties; SO3) Transfer data-tracking algorithm to predictive simulations with inclusion of motor control theories (muscle synergies and the kinematic theory).***SO1) To speed-up the optimization process and provide feedback to patients and clinical, the optimal control problem will be expressed as a nonlinear moving horizon estimator (with 50-100 ms time span) with enhanced convergence due to fewer variables. Since not all EMGs can be systematically measured, missing EMG will be inferred using a long short-term memory neural network from data taken on a population (n=30) performing various tasks.***SO2) To personalize muscle-tendon properties, students will first focus on the identification of maximal isometric muscle forces, optimal lengths, and nonlinear shape factors between EMG and neural excitation using series of (sub)maximal efforts performed on an isokinetic dynamometer. Identification algorithms from systems biology will be adapted to NMSK models.***SO3) Muscle synergies will be first extracted from our large EMG database. Use of synergies will reduce the control space and could enforce biofidelic patterns of muscle excitations (e.g. to replicate the co-contraction for glenohumeral joint stability). Moreover, the velocity of the hand will be constrained according to the kinematic theory to guide the optimization toward realistic solutions. The optimal solutions will be validated using previously-collected movements to determine the most relevant objective functions, constraints and motor control theories for generating realistic movements. ***My Discovery Program proposal will support 4 PhD and 10 undergraduate students who will be trained on advanced musculoskeletal biomechanics modelling in a multidisciplinary environment and state-of-the art infrastructure. Our ground-breaking algorithms will be the foundation of clinical, sports, artistic and ergonomic applications. **