Response to “Contribution of EEG signals to brain-machine interfaces”

In response to “Contribution of EEG signals to brain-machine interfaces” by Ordikhani-Seyedlar and Doser (2018), we agree that electroencephalogram (EEG) recordings serve an important role in brain-machine interface (BMI) research. As they observe, we intentionally omitted noninvasive BMIs from our review, including EEG, magnetoencephalography, functional magnetic resonance imaging, and functional near-infrared spectroscopy. We did this both to limit the scope of the review to a tractable size and because, in our judgment, noninvasive techniques have different sets of limitations and thus may need different methods of evaluation than invasively recorded signals. We chose to evaluate three primary criteria: amount of movement-related information, signal longevity, and signal stability. Of these criteria, only movement-related information can be applied to noninvasively recorded signals in the same way as invasively recorded signals. In contrast, longevity and stability take on slightly different meanings when considering noninvasive signals. For example, we define the longevity of a signal source as the amount of time it can be reliably recorded. This is a prime differentiator between single-unit spikes and local field potentials (LFPs) and is an important avenue of inquiry for signals like electrocorticography and epidural field potentials. In contrast, noninvasive acquisition of signals does not provoke any response by the body. Therefore signal longevity, by our definition, has little meaning when applied to noninvasive signals. One could argue that the longevity of EEG (for example) is dependent on the electrode-skin conductivity properties, which in turn depend on whether and how a conductive gel is applied, etc. In that case, longevity becomes less a property of the signals themselves than a property of the recording technique and the precision of the electrode application. We also note that the Bhagat et al. (2016) and Meng et al. (2016) studies cited by Ordikhani-Seyedlar and Doser really are more relevant to questions of signal stability, rather than signal longevity.

Signal stability, as we define it (Flint et al. 2016; or the review itself), is likewise a property of individual recording channels. In this case, the question is: to what degree does a single channel of a signal maintain its relationship to movement, when measured over time? Here again, our definition of the metric is better suited to invasively recorded signals, where the sensors are fixed in place with respect to the tissue, throughout the length of the study. Stability in EEG, in contrast, depends not only upon the signal stability itself, but also upon the consistency of application of electrodes from day-to-day; non-signal related variables could include impedance and electrode location.

EEG is a valuable research tool, and we reiterate that it plays an important role in BMI research. We would, however, contest the statement that “the user acceptability of noninvasive methods is obviously higher.” People with paralysis have stated their desire to have an implanted BMI if it will restore significant arm or hand function (>60% of respondents in two surveys: Blabe et al. 2015; Collinger et al. 2013). When conceiving our review’s subject matter, we made the decision to focus our efforts on a relatively narrow set of criteria, which were well suited to comparison across invasively recorded signal types. Including noninvasively acquired signals would have required greatly broadening our scope, and would have altered our intended focus. We chose not to do so, but we are gratified that our work has motivated a useful and important discussion.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

M.W.S. and R.D.F. prepared figures; M.W.S. and R.D.F. drafted manuscript; M.W.S. and R.D.F. edited and revised manuscript; M.W.S. and R.D.F. approved final version of manuscript.