Pan Zhengye, Liu Lushuai, Li Xingman, Ma Yunchao
College of Physical Education and Sports, Beijing Normal University, Beijing, China.
Heliyon. 2024 Apr 14;10(8):e29673. doi: 10.1016/j.heliyon.2024.e29673. eCollection 2024 Apr 30.
Although the spatio-temporal structure of muscle activation in cutting have been studied extensively, including time-varying motor primitives and time-invariant motor modules under various conditions, the factorisation methods suitable for cutting are unclear, and the extent to which each factorisation method loses information about movement during dimensionality reduction is uncertain.
To clarify the extent to which NMF, PCA and ICA retain information about movement when downscaling, and to explore the factorisation method suitable for cutting.
The kinematic data during cutting was captured with a Vicon motion capture system, from which the kinematic features of the pelvic centre of mass were calculated. NMF, PCA and ICA were used to obtain muscle synergies based on sEMG of the cutting at different angles, respectively. A back propagation neural network was constructed using temporal component of synergy as input and the kinematics data of pelvic as output. Calculation of the Hurst index using fractal analysis based on the temporal component of muscle synergy.
The quality of sEMG reconstruction is significantly higher with ICA ( < 0.01) than with NMF and PCA for the cutting. The NMF reconstruction has a high degree of preservation of movement, whereas the ICA loses movement information about the most of the swing phase, and the PCA loses information related to the change of direction. Hurst index less than 0.5 for all three angles of cutting.
NMF is suitable for extracting muscle synergies in all directions of cutting. Information related to movement may be lost when using PCA and ICA to extract the synergy of cutting. The high individual variability of muscle synergy in cutting may be responsible for the loss of movement information in muscle synergy.
尽管在切削过程中肌肉激活的时空结构已得到广泛研究，包括各种条件下的时变运动基元和时不变运动模块，但适用于切削的分解方法尚不清楚，且每种分解方法在降维过程中丢失运动信息的程度也不确定。
为了阐明非负矩阵分解（NMF）、主成分分析（PCA）和独立成分分析（ICA）在降维时保留运动信息的程度，并探索适用于切削的分解方法。
使用Vicon运动捕捉系统采集切削过程中的运动学数据，并计算骨盆质心的运动学特征。分别使用NMF、PCA和ICA基于不同角度切削的表面肌电图（sEMG）来获得肌肉协同作用。构建一个反向传播神经网络，将协同作用的时间成分作为输入，骨盆的运动学数据作为输出。基于肌肉协同作用的时间成分，使用分形分析计算赫斯特指数。
对于切削，ICA的sEMG重建质量（<0.01）显著高于NMF和PCA。NMF重建对运动具有高度的保留性，而ICA在大部分摆动阶段丢失运动信息，PCA则丢失与方向变化相关的信息。所有三个切削角度的赫斯特指数均小于0.5。
NMF适用于提取切削各个方向的肌肉协同作用。使用PCA和ICA提取切削协同作用时可能会丢失与运动相关的信息。切削过程中肌肉协同作用的高度个体变异性可能是肌肉协同作用中运动信息丢失的原因。
