Faith and oscillations recovered: On analyzing EEG/MEG signals during tACS

Abstract

Despite recent success in analyzing brain oscillations recorded during transcranial alternating current stimulation (tACS), the field still requires further research to establish standards in artifact removal methods. This includes taking a step back from the removal of the tACS artifact and thoroughly characterizing the to-be-removed artifact. A recent study by Noury et al. (2016) contributed importantly to this endeavour by showing the existence of nonlinear artefacts in the tACS signal as seen by MEG and EEG. Unfortunately however this paper conveys the message that current artifact removal attempts have failed altogether and that—based on these available tools—brain oscillations recorded during tACS cannot be analyzed using MEG and EEG. Here we want to balance this overly pessimistic conclusion: In-depth reanalyses of our own data and phantom-head measurements indicate that nonlinearities can occur, but only when technical limits of the stimulator are reached. As such they are part of the “real” stimulation and not a specific MEG analysis problem. Future tACS studies should consider these technical limits to avoid any nonlinear modulations of the tACS artifact. We conclude that even with current approaches, brain oscillations recorded during tACS can be meaningfully studied in many practical cases.

Introduction

Transcranial alternating current stimulation (tACS) is a non-invasive brain stimulation technique used to demonstrate a causal relationship between brain oscillations and behavior (Herrmann et al., 2013). Until recently, brain activity during tACS could not be analyzed due to the massive tACS-elicited artifact. New studies presented different attempts at separating the artifact from brain activity in order to analyze the immediate effects of tACS. These artifact removal attempts, for example, PCA (Helfrich et al., 2014), temporal filtering (Voss et al., 2014), or beamforming (Neuling et al., 2015), share the assumption that the artifact is linearly imposed on the brain signal. However, a recent article by Noury et al. (2016) raised doubts on the linearity assumption and, in turn, on the validity of the aforementioned artifact removal attempts. While studying the exact characteristics of the artifact is important for any attempt to remove the artifact itself, the overall message conveyed by Noury et al. (2016) may leave the impression that investigating online effects is not possible at all. Even though, perhaps, it was not the intention of the authors, such a reading of the study could stall current research efforts into online tACS effects.

The main point of Noury et al. (2016) is a nonlinear tACS-artifact, amplitude-modulated (AM) in the frequency of the heartbeat and respiration rate, and visible as side-peaks in the power spectrum around the stimulation frequency. While the pure sinusoidal tACS artifact at the stimulation frequency is an expectable artifact, nonlinear tACS artifacts are undesirable. It is critical to understand if these nonlinearities are created at the level of the recording device or if the data shows the truth, in case the artifact is already created at the level of the stimulation. Noury et al. (2016) demonstrated that all current artifact rejection attempts fail to remove these side-peaks and, in turn, claim that spectral changes at the stimulation frequency due to tACS could merely be an artifact. However, the artifact removal results obtained by Noury et al. (2016), especially with beamformers, can partly be rejected as fallacies. This conclusion is derived from a critical reanalysis of our own data. We found how methodological subtleties as applied by Noury et al. (2016) can drastically deteriorate the artifact removal performance. Additionally, we applied tACS to a phantom to investigate if the side-peaks have a technical origin. To conclude, we argue in favor of recent artifact removal attempts and of how the future of tACS research can thrive on these accomplishments.