Affective state and emotion inducing imagery classification for brain-computer interfaces

  • Alain Desire Bigirimana

Student thesis: Doctoral Thesis

Abstract

The ongoing effort in brain-computer interface (BCI) research has established that factors such as fatigue level, spatial ability, and affective state (i.e., emotions, moods, and stress responses) among others influence the BCI performance. Furthermore, while there has been significant progress in BCI deployment, studies indicate that a non-negligible proportion of users are unable to intentionally modulate their brain signal to control a motor imagery based BCI. The research presented in this thesis aims to augment BCI with affective state. There is a growing body of literature for affective state classification using brain signals. Identifying the user’s affective state should enable calibration of the BCI system for optimal performance. The research in this thesis investigates a novel BCI control strategy, emotion-inducing imagery (EII) which involves imagining fictional or recalling mnemonic emotional events. The research in this thesis was carried out through offline dataset analysis and a series of online experiments which resulted in five main contributions paving the way for BCI augmentation with affective state.
In the first contribution, the research in thesis proposes a hybrid method of independent components analysis (ICA) and wavelet transform for the pre-processing of electroencephalogram (EEG) and magnetoencephalogram (MEG) data for affective state
classification and demonstrates that this method enhances affective state classification. For the second contribution, the research introduces EII as a potential alternative BCI control strategy and presents evidence that EIIs are associated with distinct neural correlates. For the third contribution, the work in this thesis resulted in evidence that MI outperforms EII for the majority of the participants, however, the research also presented evidence that some users can consistently achieve higher performance with EII than with MI demonstrating that EII might be a viable alternative imagery for some users. In the fourth contribution, the research showed that different EEG rhythms during the period prior to using BCI and the classification accuracies across participants during EIIBCI or MI-BCI use are significantly correlated. Finally, for the fifth contribution, the research established that a signal processing framework combining neural-time-seriesprediction-pre-processing (NTSPP), filter bank common spatial patterns (FBCSP), and hemispheric asymmetry (ASYM) leads to higher performance than the individual framework in terms of classification accuracy.
Date of AwardJul 2020
Original languageEnglish
SupervisorNazmul Siddique (Supervisor) & Damien Coyle (Supervisor)

Cite this

'