Wibawa A D, Verdonschot N, Halbertsma J P K, Burgerhof J G M, Diercks R L, Verkerke G J
Department of Multimedia and Network Engineering, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia.
Orthopaedics Research Laboratory, Radboud University Medical Centre, Nijmegen, the Netherlands; Department of Biomechanical Engineering, University of Twente, Enschede, the Netherlands.
J Biomech. 2016 Nov 7;49(15):3660-3666. doi: 10.1016/j.jbiomech.2016.09.041. Epub 2016 Oct 6.
This study focused on comparing muscle activities predicted by the Musculoskeletal Modeling System with EMG from ten healthy subjects who performed normal walking, one-legged forward hopping and side jumping. Eight EMG electrodes measured the activity of eight right leg muscles. Specific thresholds per muscle were applied on the EMG prior comparison. These thresholds were determined by equalizing the duration of EMG to AMS muscle activity. Three graph variables, number of onsets, offsets and hills were used to quantify the level of agreement by using Cohen׳s kappa analysis. The Pearson correlation coefficient was also calculated as a result comparison. Overall, visual inspection showed comparable activity patterns. However, when quantifying them some differences became apparent. The mean level of agreement of all tests was <0.20, meaning poor agreement. Pearson correlation showed better agreement compared to kappa analysis. In general, a more prescribed motion like FH and SJ showed a better agreement than NW. This explorative study shows that there are distinct differences between the model and EMG pattern. Those differences can be attributed to inevitable modeling limitation within the AMS framework like miscalculating the knee net moment, absence of co-contraction, simplified knee joint. Moreover, the delay between EMG and AMS has a clear effect on the comparison and this delay is obviously missing in the model. Despite those differences, this study can serve as a baseline measurement allowing progress in scientific work in order to reduce uncertainties with the aim to generate more reliable and robust musculoskeletal models in a valid manner.
本研究聚焦于比较肌肉骨骼建模系统预测的肌肉活动与来自10名健康受试者的肌电图（EMG），这些受试者进行了正常行走、单腿向前跳跃和侧向跳跃。八个EMG电极测量了右腿八块肌肉的活动。在进行比较之前，对每块肌肉的EMG应用了特定阈值。这些阈值是通过使EMG的持续时间与肌肉骨骼建模系统（AMS）的肌肉活动持续时间相等来确定的。使用三个图形变量，即起始次数、终止次数和波峰数，通过科恩kappa分析来量化一致性水平。还计算了皮尔逊相关系数作为结果比较。总体而言，目视检查显示出可比的活动模式。然而，在对它们进行量化时，一些差异变得明显。所有测试的平均一致性水平<0.20，这意味着一致性较差。与kappa分析相比，皮尔逊相关性显示出更好的一致性。一般来说，像向前跳跃（FH）和侧向跳跃（SJ）这样规定性更强的动作比正常行走（NW）显示出更好的一致性。这项探索性研究表明，模型和EMG模式之间存在明显差异。这些差异可归因于AMS框架内不可避免的建模限制，如膝关节净力矩计算错误、缺乏共同收缩、膝关节简化。此外，EMG和AMS之间的延迟对比较有明显影响，而模型中显然没有这种延迟。尽管存在这些差异，但本研究可作为基线测量，有助于科学工作取得进展，以减少不确定性，目标是以有效的方式生成更可靠、更稳健的肌肉骨骼模型。
