Machine Learning-Based Classification to Disentangle EEG Responses to TMS and Auditory Input.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Transcranial magnetic stimulation (TMS) combined with electroencephalography (EEG) enables the study of cortical physiology via transcranial-evoked potentials (TEPs).
• Auditory evoked potentials (AEPs) from the TMS click can contaminate TEPs, making differentiation challenging with traditional statistical methods.
• Existing methods struggle to reliably separate TEPs from AEPs, particularly with imperfect auditory suppression.

Purpose of the Study:

• To investigate the efficacy of machine learning algorithms in distinguishing TEPs from AEPs.
• To assess the performance of machine learning in classifying TEPs under conditions of auditory contamination.
• To evaluate the potential of machine learning for isolating pure TEP signals.

Main Methods:

• Utilized machine learning algorithms to classify signals from healthy subjects under three conditions: masked TMS (TEPs), unmasked TMS (TEPs + AEPs), and AEPs alone.
• Trained and tested classifiers at both single-subject and group levels.
• Compared classification accuracy using average versus single-trial TEPs.

Main Results:

• The machine learning classifier achieved reliable results at the single-subject level, identifying differences not previously detected.
• Classification accuracy decreased at the group level and when comparing three conditions versus two.
• Higher classification accuracy was observed when using averaged TEPs compared to single-trial data.

Conclusions:

• Machine learning presents a promising tool for disentangling TEPs from contaminating AEPs.
• This proof-of-concept study demonstrates the potential of AI in refining neurophysiological signal analysis.
• The findings suggest machine learning can improve the isolation of pure TEPs in TMS-EEG research.