Decoding Motor Excitability in TMS using EEG-Features: An Exploratory Machine Learning Approach.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Transcranial magnetic stimulation (TMS) effectiveness can be enhanced by synchronizing with brain activity.
• Current methods use static parameters, ignoring individual brain differences and dynamic states, thus limiting therapeutic outcomes.

Purpose of the Study:

• To develop a machine learning framework for predicting individual motor excitability states using pre-stimulus electroencephalography (EEG) features.
• To identify personalized biomarkers for optimizing TMS interventions.

Main Methods:

• A supervised machine learning approach was used, combining established biomarkers with spectral and connectivity measures.
• Multi-scale feature selection within a nested cross-validation scheme was implemented.
• Validation was performed across multiple classifiers, feature sets, and protocols in 50 healthy participants.

Main Results:

• The framework achieved a mean prediction accuracy of 71 ± 7% for individual motor excitability states.
• Hierarchical clustering revealed two distinct subgroups based on predictive EEG features (alpha/low-beta vs. gamma bands).
• One subgroup showed features in sensorimotor regions (alpha/low-beta), while the other showed features in parietal regions (gamma bands).

Conclusions:

• The data-driven framework successfully identifies personalized motor excitability biomarkers.
• This approach holds potential for optimizing TMS interventions in clinical and research settings.
• The framework offers a versatile platform for biomarker discovery in neuromodulation.