A structured ICA-based process for removing auditory evoked potentials

Abstract

Transcranial magnetic stimulation (TMS)-evoked potentials (TEPs), recorded using electroencephalography (EEG), reflect a combination of TMS-induced cortical activity and multi-sensory responses to TMS. The auditory evoked potential (AEP) is a high-amplitude sensory potential—evoked by the “click” sound produced by every TMS pulse—that can dominate the TEP and obscure observation of other neural components. The AEP is peripherally evoked and therefore should not be stimulation site specific. We address the problem of disentangling the peripherally evoked AEP of the TEP from components evoked by cortical stimulation and ask whether removal of AEP enables more accurate isolation of TEP. We hypothesized that isolation of the AEP using Independent Components Analysis (ICA) would reveal features that are stimulation site specific and unique individual features. In order to improve the effectiveness of ICA for removal of AEP from the TEP, and thus more clearly separate the transcranial-evoked and non-specific TMS-modulated potentials, we merged sham and active TMS datasets representing multiple stimulation conditions, removed the resulting AEP component, and evaluated performance across different sham protocols and clinical populations using reduction in Global and Local Mean Field Power (GMFP/LMFP) and cosine similarity analysis. We show that removing AEPs significantly reduced GMFP and LMFP in the post-stimulation TEP (14 to 400 ms), driven by time windows consistent with the N100 and P200 temporal characteristics of AEPs. Cosine similarity analysis supports that removing AEPs reduces TEP similarity between subjects and reduces TEP similarity between stimulation conditions. Similarity is reduced most in a mid-latency window consistent with the N100 time-course, but nevertheless remains high in this time window. Residual TEP in this window has a time-course and topography unique from AEPs, which follow-up exploratory analyses suggest could be a modulation in the alpha band that is not stimulation site specific but is unique to individual subject. We show, using two datasets and two implementations of sham, evidence in cortical topography, TEP time-course, GMFP/LMFP and cosine similarity analyses that this procedure is effective and conservative in removing the AEP from TEP, and may thus better isolate TMS-evoked activity. We show TEP remaining in early, mid and late latencies. The early response is site and subject specific. Later response may be consistent with TMS-modulated alpha activity that is not site specific but is unique to the individual. TEP remaining after removal of AEP is unique and can provide insight into TMS-evoked potentials and other modulated oscillatory dynamics.

Experimental Suppression of TMS-EEG Sensory Potentials

“…Compared to prior studies that suggest that noise masking alone can minimize the sensory TEP [2,19], our study differs by intensity and brain target. Regarding intensity, compared to previous work that focused on subthreshold intensities (90% rMT: [2,19]), our suprathreshold stimulation protocol (120% rMT) better mimicked clinical stimulation parameters [64], but is more difficult to mask [18,40,41,65]. In regards to brain target, while other studies have explored sensory suppression after TMS to the primary motor [19] or premotor [2] targets, here we focus on the dlPFC, which may have different sensory contributions to the vertex N100-P200 compared with motor targets [43,66].…”

“…All EEG preprocessing and analyses were performed in MATLAB R2021a (Mathworks, Natick, MA, USA) using the EEGLAB v2021.1 toolbox [56] and custom scripts. Removal of artifactual EEG data was performed using a custom preprocessing pipeline, as is most common [57], but followed most closely with Ross et al [40] (steps prior to sensory removal), TMSEEG [58], and TESA [59]. Due to a marked impact of preprocessing pipelines on the TEP, as reported in [57], we took a conservative approach in all steps that required human judgement (with minimal data deletion) and describe all preprocessing steps used in detail with justification for each choice and supporting literature.…”

“…Alternatively, one can isolate the sensory contributions to the TEP using sensory-matched sham protocols. Although it is difficult to match the sensory experience of active TMS with sham TMS, the topography and time course of sensory potentials may be similar [18] enough to employ an ICA-based technique for removal [40,73]. Indeed, a combination of sensory suppression, such as ATTENUATE, coupled with sensory-matched sham TMS may be most effective for reducing the impact of sensory confounds while ensuring that residual sensory contributions to TEP can be more easily identified.…”

“…Recent evidence for effective standard masking is promising for subthreshold TMS to primary motor cortex (90% of the resting motor threshold (rMT) [2,19]). However, it has also been shown that these methods often do not fully suppress the sensory vertex N100-P200 [17,18,40–44], particularly for higher intensity protocols [17,40]. Rocchi et al .…”

“…Recent evidence for effective standard masking is promising for subthreshold TMS to primary motor cortex (90% of the resting motor threshold (rMT) [2,19]). However, it has also been shown that these methods often do not fully suppress the sensory vertex N100-P200 [17,18,[40][41][42][43][44], particularly for higher intensity protocols [17,40]. Rocchi et al [19] used over-the-ear protection in addition to noise masking to further minimize vertex N100-P200, with positive results for subthreshold M1 stimulation.…”