Machine learning approaches to predict whether MEPs can be elicited via TMS.

Area of Science:

• Neuroscience
• Biomedical Engineering

Background:

• Transcranial magnetic stimulation (TMS) is a key tool for evaluating motor cortex and cortico-muscular pathway function.
• Motor-evoked potentials (MEPs) are measurable outputs of TMS, influenced by coil parameters and intensity.
• Understanding these parameters is crucial for consistent TMS application.

Purpose of the Study:

• To investigate the predictive power of TMS setup parameters on MEP presence using machine learning.
• To compare the efficacy of within-subject versus between-subject machine learning models for MEP prediction.

Main Methods:

• Machine learning models were trained to predict MEP presence based on TMS parameters.
• Both within-subject and between-subject training designs were employed.
• Bagging ensemble methods were evaluated as a primary approach.

Main Results:

• Prediction accuracies for MEP presence reached up to 90% within subjects and 72% between subjects.
• Average accuracies were 77% (within-subject) and 65% (between-subject).
• Bagging ensembles demonstrated the highest suitability for MEP prediction.

Conclusions:

• Machine learning prediction of MEPs using TMS parameters is feasible within individual subjects.
• Limited between-subject accuracy highlights challenges for generalizing this predictive model to diverse research or clinical settings.
• Further development is needed to overcome inter-subject variability for wider applicability.