Platform Session – NIBS: Quantifying the stereotypy of TMS-evoked EEG potentials using a cosine similarity metric

Abstract

Introduction TMS-evoked EEG potentials (TEPs) can be used to probe connectivity and consciousness in healthy and diseased brains. A downside of this technique is that TMS causes neural artifacts, such as auditory and somatosensory-evoked responses, which may mask the true response to cortical stimulation, particularly at longer latencies. To avoid misinterpreting TEPs, methods for detecting neural artifact are required. One potentially helpful assumption is that neural artifacts have stereotyped topographies, while the true response to focal cortical stimulation will be spatially differentiated between stimulation conditions. Based on this assumption we used cosine similarity as a measure of response similarity and compared waveforms from real and sham stimulation at three scalp locations. Methods In 20 healthy volunteers, we delivered 150 active and 50 sham pulses to the right dorsolateral prefrontal cortex, left primary motor cortex, and left parietal cortex at 100% resting motor threshold. Sham stimulation was performed by placing a ∼2 cm spacer between the coil and scalp. TEPs were recorded with a 32-channel TMS-compatible EEG system and preprocessed using Fieldtrip® software. The derivatives of the preprocessed TEPs were calculated and then binarized for every time point at each channel. Cosine similarity at each time-point was calculated between conditions using all channels. Results The cosine similarity metric showed that for each comparison (e.g. motor vs. parietal, real vs. sham, etc.), the early ( ∼ 100 ms across all comparisons, coinciding with the N100 component. These quantitative results closely match qualitative assessment of the whole-brain topographies, lending credibility to our use of the cosine similarity technique as a measure of stereotypy. Conclusion Our results indicate that while early TEP components reflect site-specific responses to direct cortical stimulation, later responses become progressively stereotyped across stimulation sites, suggesting that these late components are neural artifact. Due to the exponential-like decay in differentiation seen in each comparison, we propose that only TEP components occurring earlier than 50–70 ms are safely interpretable as responses to cortical stimulation. Auditory and somatosensory stimulation controls and/or further refinements in signal processing may be necessary to extract true responses to cortical stimulation from later components.