T. Uktveris “Multi-class EEG signal classification and acquisition system for brain-computer interface” doctoral dissertation defence

Author, Institution: Tomas Uktveris, Kaunas University of Technology

Science area, field of science: Natural Sciences, Informatics, N 009

Scientific Supervisor: Prof. Dr. Vacius Jusas (Kaunas University of Technology, Natural Sciences,
Informatics, N 009).

Dissertation Defence Board of Natural Sciences Field:
Prof. Dr. Habil. Rimantas Barauskas (Kaunas University of Technology, Natural Sciences,
Informatics, N 009) – chairman,
Prof. Dr. Habil. Gintautas Dzemyda (Vilnius University, Natural Sciences, Informatics, N 009),
Prof. Dr. Alfonsas Misevičius (Kaunas University of Technology, Natural Sciences, Informatics, N 009),
Prof. Dr. Gintaras Palubeckis (Kaunas University of Technology, Natural Sciences, Informatics, N 009),
Prof. Dr. Raimund Ubar (Tallinn University of Technology, Estonia,  Natural Sciences, Informatics – N 009).

The doctoral dissertation is available at the libraries of Kaunas University of Technology (K. Donelaičio g. 20, 44239 Kaunas), Vytautas Magnus University (K. Donelaičio g. 52, Kaunas) and Vilnius Gediminas Technical University (Saulėtekio al. 14, 10223 Vilnius).

Annotation:

The dissertation analyzes brain-computer interface (BCI) four-class motor imagery (MI) classification problem and the development of tools for the brain electroencephalogram (EEG) acquisition. Multiple feature extraction and classification methods have been investigated and tested using computational software and experimental analysis methods. A new feature extraction channel difference method has been proposed for the EEG data processing based on Bandpower and Laplace filtering approaches. The proposed algorithm gives a similar filtering performance to a well-known CSP (common spatial patterns) algorithm. Also, a new method for a single dimension (1D) feature vector adaptation to two-dimensional (2D) feature maps has been proposed. The algorithm has been successfully validated during the experiments. The convolutional neural networks (CNN) based classification method has been adapted to solve four-class MI problem, and the experimentally acquired results were close to the other state-of-the-art methods. Moreover, a stackable and modular EEG acquisition hardware system for MI has been developed to help record second four-class validation EEG dataset and spread BCI among the wider audience.