Assessment of validity of a high-yield surface electromyogram decomposition

Abstract

BackgroundUsing recordings from a five-pin surface sensor array, a template-based surface electromyogram (sEMG) decomposition system has been developed to identify single motor unit discharge properties. However, the reliability of such template based decomposition results has not been thoroughly examined except by the developers. The focus of this study was to assess the validity of the motor unit decomposition technique, using EMG recordings from the first dorsal interosseous muscle of able-bodied human subjects.MethodsTwo tests were utilized. In the first test, a spike triggered averaging (STA) analysis was used to derive motor unit action potential (MUAP) parameters. We examined these STA derived MUAP shapes after firing times were perturbed by added timing noise. In the second test, a cross-correlation analysis was performed between the sEMG signal and MUAP trains constructed using STA estimates and their firing times.ResultsIn the first test, we found that MUAP shape features deteriorated significantly when rather small (0.6-2 ms) timing errors were added, affirming that the decomposed firing times are presumptively valid. The results of the second test reveal that the cross-correlation index between the EMG and MUAP trains increased monotonically up to 0.71 when the identified MUs were progressively added to reconstructed MUAP trains; however, this increment disappeared when the firing times or the MUAP templates were shifted randomly.ConclusionsBased on our STA selection criteria, our results suggest that the firing times and estimated MUAP shapes for each MU generated by the decomposition algorithms are presumptively valid.

The extraction of neural strategies from the surface EMG: an update

“…9C), as in Hu et al (57), the algorithm appeared to be accurate (high correlation in the spike-triggered potentials and in the sum of extracted spike trains). Moreover, there are an infinite number of algorithms that would satisfy the criteria established by Hu et al (57) without being accurate. This example is not meant to imply that any algorithm tested with this approach is necessarily inaccurate (as for our de novo "decomposition algorithm") but rather that the proposed test does not provide an adequate validation and does not resolve the issue of testing accuracy.…”

“…9B) or by the spike-train correlation method (Fig. 9C), as in Hu et al (57), the algorithm appeared to be accurate (high correlation in the spike-triggered potentials and in the sum of extracted spike trains). Moreover, there are an infinite number of algorithms that would satisfy the criteria established by Hu et al (57) without being accurate.…”

“…In both cases, there is the need for valid performance indexes, although their validity may be counterintuitive. For example, Hu et al (57) proposed intuitive criteria to judge the accuracy of an algorithm: 1) the averaged action potentials triggered by the decomposed trains of discharge times are highly correlated with the action potentials identified by the decomposition algorithm (but not if random variability is introduced in the triggers); and 2) the correlation between the original signal and the signal obtained by progressively adding the decomposed action potential trains is relatively high (but not after introducing variability in the identified discharge times). To assess the validity of these criteria, we implemented the approach in a de novo decomposition algorithm.…”

“…9. Validation of a new "decomposition" algorithm based on the criteria described by Hu et al (57). A: surface and intramuscular EMG signals were recorded concurrently during a contraction at ϳ2% MVC force.…”

Synchronization of motor unit firings: an epiphenomenon of firing rate characteristics not common inputs

“…The four channels of sEMG signals were decomposed into constituent MUAPTs using the dEMG algorithm described by De Luca et al ( ), improved in Nawab et al (2010, and independently verified by Hu et al (2013aHu et al ( , b, c, 2014. Further validation of our algorithm, using the two-source test, has been performed at different force levels ranging from 10 to 50% MVC and in different muscles, including the FDI and the tibialis anterior muscles (De Luca et al 2006Nawab et al 2010).…”