Estimating the Primary Activation in TMS-Evoked EEG recordings through Source-Based Filtering

62 

Abstract Submission 

Xavier Corominas-Teruel¹, Tuomas Mutanen², Carlo Leto¹, Cécile Gallea¹, Martina Bracco¹, Antoni Valero-Cabré¹

¹Sorbonne Université, Paris Brain Institute, Inserm, CNRS, APHP, Hôpital de la Pitié Salpêtrière, Paris, France, ²Department of Neuroscience and Biomedical Engineering, Aalto University School of Science, Espoo, Finland

Xavier Corominas-Teruel  
Sorbonne Université, Paris Brain Institute, Inserm, CNRS, APHP, Hôpital de la Pitié Salpêtrière
Paris, France

Tuomas Mutanen  
Department of Neuroscience and Biomedical Engineering, Aalto University School of Science
Espoo, Finland

Carlo Leto  
Sorbonne Université, Paris Brain Institute, Inserm, CNRS, APHP, Hôpital de la Pitié Salpêtrière
Paris, France

Cécile Gallea  
Sorbonne Université, Paris Brain Institute, Inserm, CNRS, APHP, Hôpital de la Pitié Salpêtrière
Paris, France

Martina Bracco  
Sorbonne Université, Paris Brain Institute, Inserm, CNRS, APHP, Hôpital de la Pitié Salpêtrière
Paris, France

Antoni Valero-Cabré  
Sorbonne Université, Paris Brain Institute, Inserm, CNRS, APHP, Hôpital de la Pitié Salpêtrière
Paris, France

Transcranial Magnetic Stimulation (TMS) is a well-established non-invasive technology used both for mapping human cognitive processes and therapeutic neuromodulation. Concurrent electroencephalography (EEG) enables to assess the neural impact of TMS [1]. However, due to the limited spatial resolution of EEG, TMS-evoked EEG activity remains spatially unspecific as it represents the sum of mixed spatial sources directly or indirectly activated by TMS [2]. The integration of magnetic resonance imaging (MRI)-based neuronavigation into TMS procedures offers precise information on the TMS coil position during the recordings, enabling individualized modelling of the distribution of TMS-delivered currents (E-fields, Fig. 1A) [3]. Here, we present and test a novel analysis pipeline for TMS-EEG datasets. This approach combines the distribution of TMS-generated E-fields with head and brain structural features extracted from individual MRIs. The aim is to utilize the E-field as prior information to extract the local cortical TMS-evoked activity at the stimulated site more accurately from the recorded EEG

The analysis pipeline (Fig.1B) is based upon previous frameworks for the design of spatial filters for EEG/MEG data based on cross-talk functions (DeFleCT)[4]. We first reconstructed individual finite element head models (FEM) in a cohort of n=18 participants with Simnibs4.0[3] and ISO2MESH[5] and computed the lead field matrix through the Helsinki BEM framework[6]. We then estimated the E-field distribution induced by a TMS pulse in this same forward model using Simnibs4.3 and outlined the cortical region most impacted by it (>70% of the maximal TMS-induced E-field strength). Finally, a spatial filter with 2 minimization constraints (one for noise and another for distributed sources decreasing cross-talk leakage within the ROI) was applied [4] to the EEG data to obtain activity in the cortical area directly impacted by TMS. To characterize the pipeline's performance, we processed EEG datasets from this same cohort, stimulated with 80 TMS pulses (Magstim Rapid2) delivered at 60% of the maximal stimulator output (MSO) to the left primary motor cortex (M1). Non-parametric cluster-based statistics (dependent t-test;[7]) were used to compare TMS-evoked potentials (TEP, from -100 to 600 ms – with respect to the TMS pulse) obtained with our spatial filter, and the same data but at sensor-level from a set of pre-defined electrodes (C1, C3, CP1, CP3).

Our analyses revealed important corrections for spatially filtered compared to non-spatially-filtered data operating in the temporal and spatial domains. More specifically spatially-filtered data were characterized by a consistent lower power amplitude (μV) particularly from ~100 ms with respect to the TMS pulse (t=2.5 p=0.04, t=4.1 p=0.003, Fig.1C). The sensor-level topographies of our TEP (from ~100 ms onwards) suggest signs of possible multisensory activation (Fig.1C.1)[8]. Our pipeline suppresses the contribution of those responses in the filtered signal (Fig.1C.2), originating from the targeted cortical ROI, revealing a damping wave pattern at the directly simulated region.

The spatial resolution of EEG remains a major challenge for combined TMS-EEG experiments. We here present a framework and a pipeline, which provides the community with a tool to estimate more accurately the focal EEG-evoked signals in the cortical region of interest, e.g. at the area experiencing the largest E-fields. Importantly, this can be adapted to varying research goals and integrate multimodal datasets (fMRI, DTI, fNIRS etc.) by projecting its readouts into compatible forward models or by using them in real-time close-loop implementations[9]. Our outcomes set the stage to disentangle complex patterns of TMS-evoked/induced brain dynamics previously inaccessible and foster their implementation in multiple experimental and clinically applied scenarios in which non-invasive brain stimulation shows promise.

Non-invasive Magnetic/TMS ¹

EEG/MEG Modeling and Analysis ²

Exploratory Modeling and Artifact Removal

Methods Development

Workflows

Electroencephaolography (EEG)

ELECTROPHYSIOLOGY

Modeling

Transcranial Magnetic Stimulation (TMS)

Workflows