Related Experiment Video
Updated: May 3, 2026
06:14
The Combination of Transcranial Alternating Current Stimulation and Electroencephalogram
Published on: October 10, 2025
801
Exploratory decoding of TMS-EEG: Predicting TEP response to intermittent and continuous theta burst stimulation
Arne Callaert1, Kristl Vonck2, Diego Nieves Avendaño1
1IDLab, Department of Electronics and Information Systems, Ghent University-imec, Ghent, Belgium.
Neuroimage
|May 1, 2026
Summary
Researchers used machine learning to predict how individuals respond to theta burst stimulation (TBS) based on resting-state EEG. This could lead to personalized neuromodulation for neurological and psychiatric disorders.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Neuromodulation
Background:
- Repetitive transcranial magnetic stimulation (rTMS), particularly theta burst stimulation (TBS), shows therapeutic potential for neurological and psychiatric disorders.
- Clinical application of TBS is hindered by significant inter-subject and inter-session variability in its effects.
- Combining TMS with electroencephalography (EEG) to analyze TMS-evoked potentials (TEPs) offers a method to assess cortical responses and personalize stimulation.
Purpose of the Study:
- To investigate whether pre-stimulation resting-state EEG features can predict individual changes in TEP component amplitudes after intermittent (iTBS) and continuous (cTBS) protocols.
- To evaluate the predictive performance of linear (Lasso) and nonlinear (CatBoost) machine learning models in forecasting TEP responses to TBS.
- To identify key neurophysiological predictors from resting-state EEG that influence TBS outcomes.
Main Methods:
- A randomized, single-blind crossover study involving fifteen healthy male participants.
- Application of iTBS and cTBS to the left primary motor cortex.
- Development and comparison of Lasso and CatBoost regression models to predict changes in six TEP components (N15, P30, N45, P60, N100, P180) using pre-stimulation resting-state EEG data.
Main Results:
- Both Lasso and CatBoost models demonstrated predictive accuracy exceeding random chance, indicating the ability to capture meaningful patterns in cortical responses.
- The optimal predictive model varied depending on the specific TEP component and TBS protocol.
- Feature importance analysis highlighted the significant predictive roles of spectral power and connectivity measures, particularly involving frontal, motor, and parietal regions.
Conclusions:
- Pre-stimulation resting-state EEG features can predict individual TEP responses to TBS, offering a potential pathway for personalized neuromodulation.
- Machine learning models provide a framework for understanding and predicting neurophysiological changes induced by TBS.
- Further validation in larger cohorts is necessary to confirm these exploratory findings and advance clinical translation.
