Back to search

National Technical University of Athens (NTUA)

Processing and analysis of EEG data recordings with the application of machine learning methods

Abstract

dc:description

Cognition refers to all conscious mental activity, involved in thinking, remembering, and reasoning. As such, the different brain functions range from volitional movement to language, imagination and planning. Conventionally, brain research focuses on the perception of the external world by examining how information is processed by analyzing biomedical records based on statistical or quantitative characteristics. However, the complex attributes and interactions in the human brain pose several challenges in order to effectively elucidate the underlying neural substrates that govern higher-order cerebral functions. To address this, recent studies have suggested the inclusion of machine learning designs as a way for an effective and accurate analysis of the complicated properties of brain signals, unveiling (hidden) cognitive associations and characteristics. In this PhD dissertation, advanced machine learning approaches for processing and analysis of electroencephalographic (EEG) signals are proposed, exploiting the signal properties in the time and/or frequency domains. Towards this direction, three high-density EEG experiments were employed in order to study higher-order cognitive functions, forming the basis for real-world applications especially in the field of Brain-Computer Interfaces (BCI). The recorded EEG data were investigated using event-related potentials (ERPs), spectral decomposition and brain networks, while new methodologies were developed for the modeling, reconstruction and analysis of neurological structures providing valid results. Moreover, the features extracted from each experiment (representing different properties of the various brain structures) were combined with different classification and feature selection techniques, isolating a small subset of important interactions which achieved high accuracy in the discrimination between the different conditions. Of note is that the deciphering of the brain electrical activity in a global manner (irrelevant of task and subject variability) is a major requirement for the adaptation of a common framework to real world BCI applications. As such, this Doctoral Thesis focuses on addressing this through expanding the current state-of-the art in neurocognitive adaptations of Machine Learning.Specifically, experiment 1 (Chapter 3) investigates the impact of different time windows in time-locked ERP components after an error was committed, in an auditory task with varying complexity conditions. As such, signal characteristics (like amplitude, latency etc.) that are representative of error-elicited ERP components were utilized in a feature selection and classification framework examining the effect of EEG signal modulations through time. Experiment 2 (Chapter 4) regards the task-independent assessment of high vs low mental workload in two different working memory paradigms. Within this scope, a brain network design was combined with EEG spectral characteristics to extract the global traits that are representative of mental load, while different classification schemes were paired with a feature selection algorithm to evaluate the distinguishability properties of the fused feature subsets. Further examination of the selected features revealed common task-independent patterns regarding their spectral and localization properties. Lastly, experiment 3 (Chapter 5) investigates the neural mechanisms of mental workload and their modulation in conditions simulating real-world scenarios. On this premise, a well-controlled EEG flight simulation experiment was conducted in computer screen (2D) and virtual reality (3D) conditions with multiple levels of difficulty. The resulting data were subsequently processed using electrical source imaging (ESI) and parcellated into cortical regions constructing spectral source space networks. Machine learning algorithms were then applied to compare the connectivity alterations and identify the brain networks reorganization patterns in regard to each workload level and condition.The outcomes provide new insights in the underlying cognitive mechanisms, while provide support to experts in the relevant medical fields, subsequently allowing effective monitoring of brain activity. In addition, the combination of machine learning analysis and processing with neuroscience pave the way for Human-Machine Interface (HMI) and BCI applications in both scientific and medical fields. The experimental works and analysis in the aforementioned experiments form the theoretical basis for HMI/BCI, the implications and taxonomy of which is also discussed in this PhD (Chapter 6).

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2021

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kakkos, Ioannis
  • Κάκκος, Ιωάννης

Subjects

dc:subject × 13

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/49988

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Kakkos, Ioannis; Κάκκος, Ιωάννης. Processing and analysis of EEG data recordings with the application of machine learning methods. National Technical University of Athens (NTUA), 2021. http://hdl.handle.net/10442/hedi/49988