Letter to the Editor

in Journal of Applied Biomechanics

Click name to view affiliation

Matthew S. TenanUnited States Army Research Laboratory

Search for other papers by Matthew S. Tenan in
Current site
Google Scholar
PubMed
Close
*
,
Andrew J. TweedellUnited States Army Research Laboratory

Search for other papers by Andrew J. Tweedell in
Current site
Google Scholar
PubMed
Close
*
, and
Courtney A. HaynesUnited States Army Research Laboratory

Search for other papers by Courtney A. Haynes in
Current site
Google Scholar
PubMed
Close
*

Our publication in the Journal of Applied Biomechanics examining various algorithms to determine intramuscular electromyography (EMG) onset1 has received considerable interest. We have similarly published work looking at various algorithms for surface EMG onset2 and posted all of our raw EMG data in a public repository (https://github.com/TenanATC/EMG). This “Open Science” approach has led to a number of thoughtful discussions with fellow researchers on the detection of EMG onset. The algorithm type that performed best in both our surface and intramuscular EMG studies was Bayesian changepoint (bcp) analysis.

Based on constructive conversations with researchers using our data and the algorithms assessed in our manuscripts, we would like to address a concern that has arisen with reproducing our bcp analysis. Specifically, our manuscripts utilized the version 4.0.0 bcp package in the R programming language. At the time of this letter, the bcp package has been updated to 4.0.3 by the maintainer, Dr. Xiaofei Wang. In version 4.0.0, our analyses continually demonstrated that, for detecting a single EMG onset, the “parameter of the prior on changepoint probabilities” or p0 argument should be “0” for optimal detection.1,2 Recent updates to the bcp package by Dr. Wang, described as “streamlining the C code” (personal communication), have resulted in the bcp algorithm returning a vector of zeros for the posterior probability of a changepoint when the argument p0 is set to “0.” In conversation with Dr. Wang, we agree that the current version of bcp (4.0.3) is likely more conceptually correct than the version used in our manuscripts (4.0.0). Indeed, we found it a bit peculiar that the bcp “p0 = 0” argument continually rendered the best EMG onset detection, but the purpose of our manuscripts was to examine the various algorithms in their current form for onset detection. We did not aim to critique or investigate the various algorithms themselves. When using version 4.0.0 of the bcp package for R, our finding that the “p0 = 0” argument produces the best onset detection for a single EMG onset is correct; an extremely small p0 does make conceptual sense for detecting a single EMG onset, in which one might expect an abrupt change in the time series (ie, rapid muscle contraction as opposed to a slow ramping contraction).

The question remains, “How should we consider using bcp analysis for EMG onset?” First, we would like to reiterate what we stated in our original manuscript: “While all top Bayesian algorithms in the present study used p0 = 0, it should not be expected that this setting is appropriate in all cases.” Second, pilot work by our group suggests that using the current bcp package (version 4.0.3) with the p0 argument set to an extremely small value (ie, <.0001) renders onsets similar to our manuscripts. Third, the R programming language is capable of loading previous versions of R packages using either the devtools package (ie, “install_version”) or directly installing an older package from the source (eg, directly from a website or from a local version). Therefore, there are a number of approaches to take when assessing the use of bcp algorithms to detect EMG onset, but we strongly encourage researchers to thoughtfully consider what algorithm settings are appropriate for their given data set and not blindly apply the results from our studies. Ultimately, we believe this present scenario to be a demonstration of the benefits of Open Science practices. Without posting our raw data and using open source software packages, these potential inconsistencies would not have been realized. We hope that this letter assists researchers in their pursuit of the best analytical approach for their data.

  • 1.

    Tenan MS, Tweedell AJ, Haynes CA. Iterative assessment of statistically-oriented and standard algorithms for determining muscle onset with intramuscular electromyography. J Appl Biomech. 2017;33:464468. PubMed ID: 28657852 doi:10.1123/jab.2016-0313

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 2.

    Tenan MS, Tweedell AJ, Haynes CA. Analysis of statistical and standard algorithms for detecting muscle onset with surface electromyography. PLoS One. 2017;12(5):0177312. PubMed ID: 28489897 doi:10.1371/journal.pone.0177312

    • Crossref
    • Search Google Scholar
    • Export Citation

Tenan, Tweedell, and Haynes are with Human Research and Engineering Directorate, Integrated Capability Enhancement Branch, United States Army Research Laboratory, Adelphi, MD.

Tenan (matthew.s.tenan.civ@mail.mil) is corresponding author.
  • Collapse
  • Expand
  • 1.

    Tenan MS, Tweedell AJ, Haynes CA. Iterative assessment of statistically-oriented and standard algorithms for determining muscle onset with intramuscular electromyography. J Appl Biomech. 2017;33:464468. PubMed ID: 28657852 doi:10.1123/jab.2016-0313

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 2.

    Tenan MS, Tweedell AJ, Haynes CA. Analysis of statistical and standard algorithms for detecting muscle onset with surface electromyography. PLoS One. 2017;12(5):0177312. PubMed ID: 28489897 doi:10.1371/journal.pone.0177312

    • Crossref
    • Search Google Scholar
    • Export Citation
All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 1214 623 170
PDF Downloads 140 48 6