Glaser V, Holobar A
Annu Int Conf IEEE Eng Med Biol Soc. 2017 Jul;2017:430-433. doi: 10.1109/EMBC.2017.8036854.
We discuss the adaptation of preexisting Convolution Kernel Compensation (CKC) surface electromyogram (EMG) decomposition technique to dynamic muscle contractions. In particular, three different algorithms for segmentation of motor unit (MU) spike trains into MU firings are discussed and mutually compared on synthetic dynamic surface EMG. The first segmentation algorithm employs a priori knowledge of the regularity of MU firings. The second one builds on K-means classification of MU spikes, whereas the third one combines both the regularity of MU firings and the previously introduced Pulse-to-Noise Ratio (PNR). On average, 5.5 ± 0.6 MUs were identified with sensitivity of 88.4 % ± 17.0 %, 83.8 % ± 16.7 % and 90.7 % ± 15.1 % for the first, the second and the third segmentation algorithm, respectively, demonstrating the feasibility of MU identification in moderate dynamic muscle contractions. In our tests, the third segmentation approach demonstrated superior accuracy in MU identification.
我们讨论了将现有的卷积核补偿（CKC）表面肌电图（EMG）分解技术应用于动态肌肉收缩的情况。具体而言，我们讨论了三种不同的将运动单元（MU）尖峰序列分割为MU放电的算法，并在合成动态表面肌电图上对它们进行了相互比较。第一种分割算法利用了MU放电规律性的先验知识。第二种算法基于MU尖峰的K均值分类，而第三种算法则结合了MU放电的规律性和先前引入的脉冲噪声比（PNR）。平均而言，对于第一种、第二种和第三种分割算法，分别识别出5.5±0.6个运动单元，灵敏度分别为88.4%±17.0%、83.8%±16.7%和90.7%±15.1%，这证明了在中等动态肌肉收缩中识别运动单元的可行性。在我们的测试中，第三种分割方法在运动单元识别方面表现出更高的准确性。
