Investigation of EEG Neurophysiological Relationship to TMS Response in Mild Traumatic Brain Injury Patients

Transcranial magnetic stimulation (TMS) is utilized as a treatment method for a variety of neurological and psychiatric disorders. The standard dose parameter for the administration of TMS is known as the resting motor threshold (RMT). Between individuals, the RMT has traditionally been thought to reflect differences in neuroanatomy; however, the functional state of the brain has also been shown to have an influence on this value. In this study, 19 mild traumatic brain injury (mTBI) participants resting state electroencephalographs (EEG) were obtained before undergoing TMS treatment. Various time–frequency and power spectra values were derived from this EEG data to examine the relationship between RMT and neurophysiology. These EEG connectivity metrics were then used to establish relationships with RMT alone and account for differences in individual neuroanatomy. We found that the incorporation of EEG functional connectivity improved multiple regression model predictions of inter-participant RMT variability than neuroanatomy alone. Future investigations into the evaluation of EEG in predicting RMT variability should include a larger participant population with a healthy control.


I. INTRODUCTION
T RANSCRANIAL magnetic stimulation (TMS) is a non- invasive, neuromodulation technique that utilizes an electromagnetic coil to deliver short magnetic pulses to select regions of the brain.These pulses serve to stimulate neurons in the brain which, in turn, results in changes in cortical excitability [1].Repeated stimulation of this nature has been correlated with improvements in cognitive functioning and depression symptoms, and thus has been implemented as a form of treatment for varieties of neurological and psychiatric disorders [2], [3], [4].
In order to ascertain optimal stimulation strengths, the resting motor threshold (RMT) has to be determined for each patient.The RMT is used as a dosage parameter in TMS, as it reflects cortical excitability level for a specific subject.It is typically defined as the minimum amount of stimulation strength as a percentage of maximum stimulator output (MSO) required for a motor-evoked potential (MEP) response of at least 50 µV in electromyographic activity (EMG), or a visible motor twitch, in the first dorsal interossei muscle [5], [6].
In our previous study, TMS was applied to investigate changes in cognition and neuropsychological performance for individuals with mild-to-moderate traumatic brain injuries (mTBI) [7].Patients received 10 Hz of active or sham prefrontal cortex stimulation over a 20 min duration for five consecutive days.Following a 1 week wash-out period, the treatment was repeated with the alternate (active or sham) condition.In conjunction with the TMS administration, electroencephalograph (EEG) recordings were collected from each patient over the course of their resting-state periods prior to stimulation.
The TMS treatment outcomes of these patients have been published by Franke et al. [7].However, in-depth analysis of EEG parameters related to oscillatory activity and functional connectivity in the brain may help determine dose-response relationship.In this article, we are seeking to determine the best EEG parameter in establishing this relationship, utilizing both neurophysiological activity and functional connectivity of the brain alongside neuroanatomical parameters.We hypothesize that incorporating neurophysiological activity in a multiple linear regression model may establish a more accurate correlation between RMT and neuroanatomical parameters than a linear regression model between RMT and neuroanatomical parameters that do not account for functional state or connectivity in the brain.

II. METHODOLOGY
This current study used the 19 mTBI participants who completed the aforementioned study and had their T1-weighted magnetic resonance imaging taken.The present study branches further into the analysis of the 62-channel EEG data.MAT-LAB (Mathworks, R2022a) software was utilized to carry out analysis on the vast amount of resting state data collected from the study.0018-9464 © 2023 IEEE.Personal use is permitted, but republication/redistribution requires IEEE permission.
Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
For this study, the raw data was preprocessed through Fieldtrip and EEGLab, MATLAB toolboxes that allow for visualization and manipulation of electrophysiological information [8].To remove artifacts embedded in the data without disrupting the relevant information, an automated independent component analysis was conducted.This effectively cleaned the data, removing instances of oculomotor artifacts and other forms of excessive noise in the trials [9].For an in-depth investigation of brain connectivity and functional connectivity measures, this study primarily relied on the FieldTrip and Brain Connectivity MATLAB software [10], [11].Additional analysis of aperiodic and periodic neural activity was parameterized using specialized Python-based packages that were loaded into MATLAB to maintain a clean, consistent pipeline for all the data.
We began our analysis with an exploration into classical measures of power and time-frequency analysis in relation to functional connectivity.Functional connectivity is described as the statistical dependencies between signals from multiple regions in the brain [12], [13].EEG can be used to determine the degree of connectivity between different channels through spectral power analysis.Since the RMT is delivered the motor cortex, we focused our analysis on the motor and motor-related brain regions (parietal and fronto-polar EEG channels).We sought to better establish a clear correlation in these regions of interest.
In our study, the average functional connectivity strength was determined across all 62 nodes of each patient's EEG recording using FieldTrip Toolbox routines.Time series were normalized by their standard deviation.Then, by implementing a spectral density estimation technique known as multi-tapered fast Fourier transform, the data was decomposed into the frequency domain as a discrete signal, with independent tapers calculated over 2 s epochs to estimate the power across frequency [10].Time-frequency analysis was conducted on the data in the interest range of 0.5-55 Hz, creating power bands that were divided into Delta (0.5-3 Hz), Theta (3-8 Hz), Alpha (8-13 Hz), Beta (20-22 Hz), Gamma (30-55 Hz), and Broadband (0.1-55 Hz) frequencies.To assess the functional connectivity between EEG channel time series, we computed the weighted phase lag index (WPLI) [14], [15].
After running a FieldTrip connectivity analysis script, the Brain Connectivity Toolbox was implemented to create visualizations of the compiled results [16].Fig. 1 displays these varying connection orientations and lengths across a representative brain at each of the six frequency bands.
For a more thorough assessment of potential EEG measures correlated to RMT, we incorporated advanced measures of periodic and aperiodic neural activity analysis to supplement our data.Aperiodic activity is often overlooked in EEG data but holds important physiological information regarding neuron signaling and synaptic currents [17].Similar to other fractal activity, neurophysiological brain activity tends to have a 1/f-like distribution, with exponentially decreasing power across increasing frequencies [17], [18].Converging animal and human evidence have found that when a 1/f line is approximated to the log-log distribution of the brain's oscillatory activity, the slope of the line (i.e., the 1/f exponent)  is associated with the inhibition-excitation balance of neural activity, with "flatter" lines indicating more excitation relative to inhibition, and "steeper" lines indicating the opposite [19].
We used the fitting oscillations and one over f (FOOOF) Python 3.9.0package to parameterize the neural power spectra data across all frequencies and EEG channels [20].FOOOF was available as an environment in Anaconda software that automated the calculations necessary for this analysis.The process involves applying Welch's method of spectral density estimation, followed by Gaussian and Lorentzian functions to determine the power spectral density as a combination of the periodic and aperiodic components, respectively.Fig. 2 depicts the topography of these results.Given its relevance to cortical excitability, we focused our analysis only on the 1/f exponent that was derived from the FOOOF pipeline.
After amassing the functional connectivity and 1/f exponent measures from all 19 patients, the 62 EEG channels were reduced to a key region of focus primarily on the right side of the brain that aligned approximately with the motor cortex, with supplementary nodes located in the frontal and parietal lobes for comparison.The nodes initially chosen for further analysis were C2, C4, C6, FC2, FC4, FC6 FP1, FP2, P3, and P4.As our research is centered around EEG relationships to RMT, this region held the most relevance to our analysis [21].Simple linear regression models were developed for each of the parameters at the given nodes in comparison to RMT.Following the initial assessment of these nodes, the functional connectivity data were expanded to incorporate couplings of lateral and interhemispheric nodes.This included the node pairings C3-C4, P3-C4, P4-C3, FP1-FP2, and FC2-C4.With these additional nodes, rather than looking solely into the inter-connectivity between the regions of interest and the overall brain, we instead assessed the intra-connectivity between the pairs.
To gain a more nuanced understanding of the data in relation to MRI model-derived parameters, multiple regression models were developed using the software R (The R Foundation, v4.2.1).Site-specific-induced electric field strength (EFS) values were determined by Lewis et al. [22] and included in statistical modeling to correlate between RMT and EEG parameters when accounting for neuroanatomy.
Linear regression modeling reaffirmed the positive linear relationship found in Lewis et al. [22] between brain scalp distance (BSD) and RMT (R 2 = 0.28, p = 0.02).Applying the functional connectivity data collected from this study, a negative correlation was found between RMT and the connectivity average at FC4, FC6, P3, and P4 in the Delta powerband (Table I) (Fig. 3).The other powerbands did not show any direct influence of EEG on RMT.A negative correlation was also found with 1/f exponent in relation to RMT at nodes P3 (R 2 = 0.225, p = 0.040) and P4 (R 2 = 0.328, p = 0.010) (Fig. 4).
To test our hypothesis, multiple regression modeling was carried out on EEG, neuroanatomy, and TMS response parameters.The modeling resulted in 43.9% (R 2 , p = 0.01) of RMT inter-participant variance being explained by neuroanatomical factors.The modeling also revealed significant relationships between RMT and the interaction between EFS and connectivity in the Delta powerband for the C4, C6, FC4, FC6, and FP1 nodes (Table II).These relationships were negative but only at a low simulated EFS (Fig. 5).The other nodes within the Delta power band, along with the remaining five power

bands, did not show significant relationships when accounting for EFS or BSD.
There was a negative relationship between RMT and C3-C4 connectivity but only at a high simulated EFS (Fig. 6).The overall multiple regression model explained 40.5% (R 2 , p = 0.02) of interparticipant RMT variability.Multiple regression modeling did not show relationships between the node pairings P3-C4, P4-C3, FP1-FP2, and FC2-C4 when accounting for neuroanatomy.
Multiple regression modeling of the interactions between EFS and 1/f exponent indicated notable correlations to RMT.The strongest correlations were found at nodes FP1, FP2, FC2, and FC4 (Table III) (Fig. 7).

IV. DISCUSSION
In our research, we sought to better establish a correlation between RMT and MRI model-derived parameters with the Fig. 5.
Effect of Delta connectivity at the FP1 node on RMT.Delta FP1 connectivity showed the expected negative correlation with RMT in participants with a low simulated peak EFS magnitude.Fig. 6.Effect of connectivity between the C3 and C4 nodes on RMT.C3C4 connectivity showed the expected negative correlation with RMT in participants with a high simulated peak EFS magnitude.addition of neurophysiological and functional connectivity parameters.Our central hypothesis was that incorporating EEG activity and functional connectivity in multiple linear regression modeling with individual neuroanatomy was expected to account for more of the inter-participant RMT variability than individual neuroanatomy alone.This hypothesis was largely supported as evidenced by the presence of much larger coefficients of determination in multiple regression models with EEG functional connectivity over the R 2 = 0.439 found from neuroanatomy alone.
The negative relationship between EEG connectivity and RMT at low simulated EFS was expected with multiple regression modeling.Participants with increased connectivity in EEG sensors were expected to have lower RMTs corresponding to less of the MSO needed to initiate gray matter depolarization.This relationship depended on a low value for the simulated induced EFS.Thus, when the simulated EFS was small, EEG functional connectivity at these nodes predicted RMT with a negative linear trend.However, when the simulated EFS was larger, the relationships between RMT and EEG functional connectivity were reduced and required further investigation.Previous investigations into functional connectivity in the motor cortex have also  demonstrated a negative influence of functional connectivity on RMT [6].
The findings speak to a functional relevance of Delta oscillations in motor cortex excitability, whereby greater Delta synchrony may relate to greater cortical excitability, and thereby lower RMT values.It agrees with a previous TMS study which observed that greater coupling of Delta band oscillations across the bilateral centro-parietal-occipital cortices prior to the TMS pulse led to significantly larger MEPs compared to when Delta coupling was low [23].Other studies have suggested that slow frequency oscillations arise from functional disconnections of related cortical areas, including regions which contribute to the inhibitory control of the motor cortex [24].It could be speculated that increased connectivity in the Delta frequency band could represent a state of a functional unbinding of the motor cortex from frontal and parietal inhibitory control regions.
The negative relationship between RMT and the 1/f exponent within the parietal and frontal area lobe suggests that higher neuronal inhibitory activity relative to excitatory activity in that region resulted in a lower RMT.This would imply an inhibitory relationship between the parietal and motor cortices, in which disinhibition of the motor cortex via reduced parietal activity could be a basis for lower RMT.However, this is speculative and further research would be needed to unravel this relationship more closely.In applying a multiple regression model, similar to the connectivity results, a lower EFS value resulted in a negative correlation between the exponent at each of the selected nodes and RMT.
The original study was powered for the cognitive outcomes, and so all TMS and EEG outcomes should be considered secondary and interpreted accordingly, with caution.As such, this study is limited by a small sample size (n = 19) and the lack of a healthy cohort.A limited number of electrodes could be studied due to sample size and so any localization of effects is unclear.The exploratory nature of analyses could mean a high rate of Type I error.Future investigations into the influence of EEG on RMT in mTBI should incorporate a larger participant population with healthy control participants.This study was limited by the lack of diffusion tensor imaging (DTI).DTI allows for the creation of white matter-derived fiber tracts which have previously been shown to influence inter-participant RMT [6].
V. CONCLUSION This preliminary study investigated how resting state EEG influences RMT when accounting for other MRI model-derived parameters in 19 mTBI participants.Our results suggest EEG has a role in accounting for inter-participant RMT variability; however, the extent of this relationship needs to be determined.Future investigations into resting-state EEG as a predictor of RMT should include a larger participant population as well as a healthy control.

Fig. 1 .
Fig. 1.Topographical representation of functional connectivity throughout the brain at different frequency bands for mTBI Subject 3.

Fig. 3 .
Fig.3.Delta connectivity at the FC4 node correlated with RMT as a percentage of maximum stimulator output (%MSO) with a regression coefficient R 2 = 0.49 and p-value < 0.001.

Fig. 4 .
Fig. 4. 1/f Exponent at the P4 node correlated with RMT as a percentage of maximum stimulator output (%MSO) with a regression coefficient R 2 = 0.33 and p-value = 0.01.

Fig. 7 .
Fig. 7. Effect of 1/f exponent at the FP1 node on RMT.1/f exponent FP1 showed the expected negative correlation with RMT in participants with a low simulated peak EFS magnitude.

TABLE I SUMMARY
OF LINEAR REGRESSION MODELING (CONNECTIVITY)

TABLE II SUMMARY
OF MULTIPLE REGRESSION MODELING (CONNECTIVITY)

TABLE III SUMMARY
OF MULTIPLE REGRESSION MODELING (1/f EXPONENT)