Faculty of Biological sciences, School of Biomedical sciences, University of Leeds, Leeds, United Kingdom.
J Neural Eng. 2022 Feb 18;19(1). doi: 10.1088/1741-2552/ac5150.
. Current approaches to muscle synergy extraction rely on linear dimensionality reduction algorithms that make specific assumptions on the underlying signals. However, to capture nonlinear time varying, large-scale but also muscle-specific interactions, a more generalised approach is required.. Here we developed a novel framework for muscle synergy extraction that relaxes model assumptions by using a combination of information and network theory and dimensionality reduction. We first quantify informational dynamics between muscles, time-samples or muscle-time pairings using a novel mutual information formulation. We then model these pairwise interactions as multiplex networks and identify modules representing the network architecture. We employ this modularity criterion as the input parameter for dimensionality reduction, which verifiably extracts the identified modules, and also to characterise salient structures within each module.. This novel framework captures spatial, temporal and spatiotemporal interactions across two benchmark datasets of reaching movements, producing distinct spatial groupings and both tonic and phasic temporal patterns. Readily interpretable muscle synergies spanning multiple spatial and temporal scales were identified, demonstrating significant task dependence, ability to capture trial-to-trial fluctuations and concordance across participants. Furthermore, our framework identifies submodular structures that represent the distributed networks of co-occurring signal interactions across scales.. The capabilities of this framework are illustrated through the concomitant continuity with previous research and novelty of the insights gained. Several previous limitations are circumvented including the extraction of functionally meaningful and multiplexed pairwise muscle couplings under relaxed model assumptions. The extracted synergies provide a holistic view of the movement while important details of task performance are readily interpretable. The identified muscle groupings transcend biomechanical constraints and the temporal patterns reveal characteristics of fundamental motor control mechanisms. We conclude that this framework opens new opportunities for muscle synergy research and can constitute a bridge between existing models and recent network-theoretic endeavours.
. 当前的肌肉协同作用提取方法依赖于线性降维算法，这些算法对底层信号做出了特定的假设。然而，为了捕捉非线性时变、大规模且具有肌肉特异性的相互作用，需要一种更为通用的方法。. 在这里，我们开发了一种新的肌肉协同作用提取框架，通过结合信息论和网络理论以及降维来放松模型假设。我们首先使用一种新的互信息公式来量化肌肉之间、时间样本或肌肉-时间配对之间的信息动态。然后，我们将这些成对的相互作用建模为多重网络，并识别表示网络结构的模块。我们将这种模块化标准用作降维的输入参数，该参数可验证地提取出所识别的模块，并用于描述每个模块内的显著结构。. 该新框架捕捉了两个基准抓握运动数据集之间的空间、时间和时空相互作用，产生了独特的空间分组以及紧张和相位的时间模式。识别出了可扩展到多个空间和时间尺度的易解释肌肉协同作用，表现出显著的任务依赖性、捕捉试验间波动的能力以及参与者之间的一致性。此外，我们的框架还识别了代表跨尺度共同信号相互作用的分布式网络的子模块化结构。. 通过与先前研究的连续性和所获得见解的新颖性，展示了该框架的能力。规避了几个先前的限制，包括在放松模型假设的情况下提取功能有意义和多重的成对肌肉耦合。提取的协同作用提供了运动的整体视图，同时可轻松解释任务表现的重要细节。所识别的肌肉分组超越了生物力学约束，时间模式揭示了基本运动控制机制的特征。我们得出的结论是，该框架为肌肉协同作用研究开辟了新的机会，并可以在现有模型和最近的网络理论努力之间架起桥梁。
