Data-Driven Methods for Analyzing TMS-Evoked EEG Responses

Johanna Metsomaa

Research output: ThesisDoctoral ThesisCollection of Articles

Abstract

Transcranial magnetic stimulation (TMS) is a technique with which one can non-invasively activate a selected area in the brain. Consequently, signals arising from the TMS-evoked neural activity can be measured on the scalp by electroencephalography (EEG). These measurements can be used to understand brain functions in various physiological and pathophysiological conditions. There are many possibilities for interpreting EEG. A single EEG channel measures a weighted sum of activity in several brain regions as well as artifacts and noise. Therefore, EEG data are commonly recorded using several channels, which should be optimally utilized to uncover the underlying interesting brain activity. Physical and statistical modelling of the data can be used to identify and interpret neural processes based on EEG. In addition to the EEG signals arising from brain activity, TMS often gives rise to severe stimulus-induced artifacts in EEG, which can completely mask the interesting neural signals. The TMS-related artifacts arise from several origins and their physical modelling is not feasible. In this Thesis, we used and developed blind source separation (BSS) techniques, which are based solely on statistical modelling of the data. The aim was to identify and separate artifacts and neural activity in TMS-evoked EEG using two BSS methods: independent component analysis (ICA) and momentary-uncorrelated component analysis (MUCA). There are many sources of uncertainty when applying these methods to the EEG data: (1) The identification of artifactual and neural processes/activity is highly subjective. (2) When the data have high-amplitude artifacts, BSS becomes unstable. (3) The TMS-evoked EEG is non-stationary (time-dependent), which is commonly not taken into account in the statistical models of the BSS methods. All of these issues have been discussed and solutions for the problems have been presented in this work. We also developed methods to clean noisy EEG, which are designed to utilize multi-sensor EEG measurement optimally based on the Wiener estimation framework and physical model for the EEG signals originating from the brain. Altogether, the presented methods are designed for identifying and correcting contaminated TMS-evoked EEG signals and uncovering underlying neural activity. The applicability of the techniques also extends to EEG and magnetoencephalography (MEG) responses related to any other stimuli or events. Finally, the introduced artifact and noise suppression techniques are suitable for eliminating various kinds of distortions in EEG/MEG.
Translated title of the contributionDatasta ohjautuvat menetelmät TMS:n aiheuttamien EEG-vasteiden analysoinnissa
Original languageEnglish
QualificationDoctor's degree
Awarding Institution
  • Aalto University
Supervisors/Advisors
  • Ilmoniemi, Risto, Supervising Professor
  • Ilmoniemi, Risto, Thesis Advisor
  • Sarvas, Jukka, Thesis Advisor
Publisher
Print ISBNs978-952-60-7396-5
Electronic ISBNs978-952-60-7395-8
Publication statusPublished - 2017
MoE publication typeG5 Doctoral dissertation (article)

Keywords

  • artifacts
  • blind source separation
  • electroencephalography
  • event-related potentials
  • independent component analysis
  • transcranial magnetic stimulation

Fingerprint

Dive into the research topics of 'Data-Driven Methods for Analyzing TMS-Evoked EEG Responses'. Together they form a unique fingerprint.

Cite this