TMS-EEG decay artifact: an adaptive algorithm for signal detrending

TMS-EEG represent one of the most promising methods in the investigation of the brain dynamics. However, during EEG, the discharge of TMS may generate a decay artifact that can last for tens of milliseconds. Such artifact represents a problem for the analysis of the TMS-evoked potentials (TEPs). So far, two main strategies of correction have been proposed involving the use of a linear detrend or independent component analysis (ICA). However, none of these solutions may be considered optimal: firstly, because in most of the cases the decay artifact shows a non-linear trend; secondly, because the ICA correction (1) might be influenced by individual researcher’s choices and (2) might cause the removal of physiological responses. Our aim is to verify the feasibility of a new adaptive detrend by comparing it with the ICA correction. Thirty-six healthy volunteers were stimulated with 50 TMS pulses over the left M1. The peak-to-peak amplitude and the morphology of the TEPs were compared among three conditions: RAW (no correction of the decay artifact was applied); ICA (the decay components were extracted and removed by ICA); ALG (the decay artifact was corrected through the use of an adaptive algorithm). To assess whether the artifact correction significantly affected also the physiological responses to TMS, we examined the differences in the -100 +400 ms time window after the TMS pulse, across the three conditions, by means of a non-parametric, cluster-based, permutation statistical test. Then, we compared the peak-to-peak TEPs amplitude within the detected time windows. The grand-averaged EEG response revealed 5 main peaks: P30, N45, P60, N100 and P180. Significant differences (i.e. Monte Carlo p-values < 0.05) among the three conditions were detected in a cluster nearby the TMS coil, and specifically over FC1 (all the components); CP1 (P30/N45 and N45/P60) and FC2 (N45/P60 and P60/N100). Repeated-measures ANOVA revealed a higher peak-to-peak amplitude in 5 of the 8 TEPs after ICA correction, compared to the RAW and ALG conditions. Our results showed that the ICA correction significantly affected the amplitude and the morphology of most of the analyzed TEPs. On the other hand, when our algorithm was used, the amplitude and the morphology of the peaks did not differ from the original signal (i.e. RAW condition). The present results showed that our adaptive detrend is a reliable solution for the correction of the TMS-evoked decay artefact, especially considering that, contrary to ICA, (1) it is not dependent from the number of recording channels, (2) it does not affect the physiological responses and (3) it is completely independent from the experimenter’s choices.