Machine Learning-Based Classiﬁcation to Disentangle EEG Responses to TMS and Auditory Input - Proof-of-concept study

Aug 4, 2026

Upload by Aurora

Transcranial magnetic stimulation (TMS) combined with electroencephalography (EEG) enables investigation of cortical physiology through transcranial-evoked potentials (TEPs), yet these signals can be contaminated by auditory evoked potentials (AEPs) generated by the TMS click. Common statistical approaches may struggle when AEP suppression is suboptimal. Using machine-learning algorithms, the study compares masked TEPs, AEPs, and non-masked TEPs in healthy participants, showing reliable single-subject classification and lower group-level performance under cross-subject training and multi-condition comparisons, with improved accuracy for averaged versus single-trial TEPs.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Introduction

TEPs and AEPs contamination in TMS-EEG

Methods (as described in abstract)

Machine learning classification across stimulation conditions

Results (as described in abstract)

Single-subject reliability vs group-level accuracy

Conclusion (as described in abstract)

Separating pure TEPs from sensory-contaminated responses

brain sciences Article Machine Learning-Based Classiﬁcation to Disentangle EEG Responses to TMS and Auditory Input Andrea Cristofari 1, Marianna De Santis 2, Stefano Lucidi 2, John Rothwell 3, Elias P. Casula 4 and Lorenzo Rocchi 3,5, * Citation: Cristofari, A.; De Santis, M.; Lucidi, S.; Rothwell, J.; Casula, E.P.; Rocchi, L. Machine Learning-Based Classiﬁcation to Disentangle EEG Responses to TMS and Auditory Input. Brain Sci. 2023, 13, 866 . [https://doi.org/10.3390/](https://doi.org/10.3390/)[ ](https://doi.org/10.3390/)brainsci13060866 Academic Editor: David Papo Received: 7 May 2023 Revised: 21 May 2023 Accepted: 25 May 2023 Published: 27 May 2023 Copyright: © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license ([https://](https://)[ ](https://)[creativecommons.org/licenses/by/](creativecommons.org/licenses/by/)[ ](creativecommons.org/licenses/by/)[4.0/](4.0/)) . 1 Department of Civil Engineering and Computer Science Engineering,“Tor Vergata” University of Rome, 00133 Rome, Italy; [andrea.cristofari@uniroma2.it](andrea.cristofari@uniroma2.it) 2 Department of Computer, Automatic and Management Engineering,“Sapienza” University of Rome, 00185 Rome, Italy; [marianna.desantis@uniroma1.it](marianna.desantis@uniroma1.it) (M.D.S.); [lucidi@diag.uniroma1.it](lucidi@diag.uniroma1.it) (S.L.) 3 Department of Clinical and Movement Neurosciences, UCL Queen Square Institute of Neurology, University College London, London WC1N 3BG, UK; [j.rothwell@ucl.ac.uk](j.rothwell@ucl.ac.uk) 4 Department of System Medicine,“Tor Vergata” University of Rome, 00133 Rome, Italy; [elias.casula@gmail.com](elias.casula@gmail.com) 5 Department of Medical Sciences and Public Health, University of Cagliari, Cittadella Universitaria di Monserrato, 09042 Cagliari, Italy * Correspondence: [l.rocchi@ucl.ac.uk](l.rocchi@ucl.ac.uk) Abstract: The combination of transcranial magnetic stimulation (TMS) and electroencephalography (EEG) offers an unparalleled opportunity to study cortical physiology by characterizing brain electrical responses to external perturbation, called transcranial-evoked potentials (TEPs) . Although these reﬂect cortical post-synaptic potentials, they can be contaminated by auditory evoked potentials (AEPs) due to the TMS click, which partly show a similar spatial and temporal scalp distribution. Therefore, TEPs and AEPs can be difﬁcult to disentangle by common statistical methods, especially in conditions of suboptimal AEP suppression. In this work, we explored the ability of machine learning algorithms to distinguish TEPs recorded with masking of the TMS click, AEPs and non-masked TEPs in a sample of healthy subjects. Overall, our classiﬁer provided reliable results at the singlesubject level, even for signals where differences were not shown in previous works. Classiﬁcation accuracy (CA) was lower at the group level, when different subjects were used for training and test phases, and when three stimulation conditions instead of two were compared. Lastly, CA was higher when average, rather than single-trial TEPs, were used. In conclusion, this proof-of-concept study proposes machine learning as a promising tool to separate pure TEPs from those contaminated by sensory input. Keywords: transcranial magnetic stimulation; electroencephalography; TMS-EEG; evoked potentials; machine learning; neural networks 1. Introduction The combination of transcranial magnetic stimulation (TMS) and electroencephalography (EEG) has become an increasingly used approach to assess cortical physiology in healthy humans [1–4] and patients affected by disorders of the central nervous system [5–7] . There is considerable evidence to support the notion that EEG signals following TMS, either measured as transcranial-evoked potentials (TEPs) or oscillations, mostly reﬂect the summation of excitat

FAQs

Why is it difficult to disentangle TEPs and AEPs in TMS-EEG data?

TMS produces a click that can elicit auditory evoked potentials (AEPs) with partly similar spatial and temporal scalp distributions. When auditory suppression is not optimal, common statistical methods may not separate TEPs and AEPs well.

What role does machine learning play in the study?

Machine learning algorithms are used to distinguish TEPs recorded with masking of the TMS click, AEPs, and non-masked TEPs. The work evaluates how well classification performs under different conditions.

How do classification results differ between single-subject and group analyses?

The classifier provides reliable results at the single-subject level. Group-level accuracy is lower when training and test involve different subjects and when comparing three stimulation conditions instead of two.

Aug 4, 2026

Upload by Aurora