The concurrent use of transcranial magnetic stimulation with electroencephalography (TMS-EEG) is growing in popularity as a method for assessing various cortical properties such as excitability, oscillations and connectivity. However, this combination of methods is technically challenging, resulting in artifacts both during recording and following typical EEG analysis methods, which can distort the underlying neural signal. In this article, we review the causes of artifacts in EEG recordings resulting from TMS, as well as artifacts introduced during analysis (e.g. as the result of filtering over high-frequency, large amplitude artifacts). We then discuss methods for removing artifacts, and ways of designing pipelines to minimise analysis-related artifacts. Finally, we introduce the TMS-EEG signal analyser (TESA), an open-source extension for EEGLAB, which includes functions that are specific for TMS-EEG analysis, such as removing and interpolating the TMS pulse artifact, removing and minimising TMS-evoked muscle activity, and analysing TMS-evoked potentials. The aims of TESA are to provide users with easy access to current TMS-EEG analysis methods and to encourage direct comparisons of these methods and pipelines. It is hoped that providing open-source functions will aid in both improving and standardising analysis across the field of TMS-EEG research.

Transcranial magnetic stimulation (TMS) evokes neuronal activity in the targeted cortex and connected brain regions. The evoked brain response can be measured with electroencephalography (EEG). TMS combined with simultaneous EEG (TMS-EEG) is widely used for studying cortical reactivity and connectivity at high spatiotemporal resolution. Methodologically, the combination of TMS with EEG is challenging, and there are many open questions in the field. Different TMS-EEG equipment and approaches for data collection and analysis are used. The lack of standardization may affect reproducibility and limit the comparability of results produced in different research laboratories. In addition, there is controversy about the extent to which auditory and somatosensory inputs contribute to transcranially evoked EEG. This review provides a guide for researchers who wish to use TMS-EEG to study the reactivity of the human cortex. A worldwide panel of experts working on TMS-EEG covered all aspects that should be considered in TMS-EEG experiments, providing methodological recommendations (when possible) for effective TMS-EEG recordings and analysis. The panel identified and discussed the challenges of the technique, particularly regarding recording procedures, artifact correction, analysis, and interpretation of the transcranial evoked potentials (TEPs). Therefore, this work offers an extensive overview of TMS-EEG methodology and thus may promote standardization of experimental and computational procedures across groups.

Objectives. This paper aims to investigate the feasibility and the validity of applying deep convolutional neural networks (CNN) to identify motor unit (MU) spike trains and estimate the neural drive to muscles from high-density electromyography (HD-EMG) signals in real time. Two distinct deep CNNs are compared with the convolution kernel compensation (CKC) algorithm using simulated and experimentally recorded signals. The effects of window size and step size of the input HD-EMG signals are also investigated. Approach. The MU spike trains were first identified with the CKC algorithm. The HD-EMG signals and spike trains were used to train the deep CNN. Then, the deep CNN decomposed the HD-EMG signals into MU discharge times in real time. Two CNN approaches are compared with the CKC: (a) multiple single-output deep CNN (SO-DCNN) with one MU decomposed per network, and (b) one multiple-output deep CNN (MO-DCNN) to decompose all MUs (up to 23) with one network. Main results. The MO-DCNN outperformed the SO-DCNN in terms of training time (3.2–21.4 s epoch⁻¹ vs 6.5–47.8 s epoch⁻¹, respectively) and prediction time (0.04 vs 0.27 s sample⁻¹, respectively). The optimal window size and step size for MO-DCNN were 120 and 20 data points, respectively. It results in sensitivity of 98% and 85% with simulated and experimentally recorded HD-EMG signals, respectively. There is a high cross-correlation coefficient between the neural drive estimated with CKC and that estimated with MO-DCNN (range of r-value across conditions: 0.88–0.95). Significance. We demonstrate the feasibility and the validity of using deep CNN to accurately identify MU activity from HD-EMG with a latency lower than 80 ms, which falls within the lower bound of the human electromechanical delay. This method opens many opportunities for using the neural drive to interface humans with assistive devices.