Please use this identifier to cite or link to this item:
http://hdl.handle.net/11375/26845
Full metadata record
DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Becker, Suzanna | - |
dc.contributor.advisor | Connolly, John | - |
dc.contributor.author | Mousapour, Leila | - |
dc.date.accessioned | 2021-08-30T21:16:32Z | - |
dc.date.available | 2021-08-30T21:16:32Z | - |
dc.date.issued | 2021 | - |
dc.identifier.uri | http://hdl.handle.net/11375/26845 | - |
dc.description.abstract | Brain-computer interfaces (BCIs) have incredible potential to allow people with limited communication or no ability for speech to be able to communicate. A BCI can acquire brain signals, analyze them, and interpret them into commands or words. A binary set of mental tasks can be mapped into two words, such as Yes and No, to enable a user to answer a binary tree of questions and sufficiently create a communication system. Although motor imagery tasks are the most established control signals used in the context of asynchronous BCI, they are not suitable for a non-negligible percentage of the users. This issue, known as BCI illiteracy, has been shown to improve when individuals are given greater latitude in the choice of mental tasks employed in operating a BCI. However, differentiating the activation pattern of non-conventional mental imageries (MI) is more challenging than motor imagery. Therefore, finding a good feature space in which machine learning and classification methods can be applied to the data is crucial. While the standard in EEG-based BCIs is to directly analyze the electrode space signals, the measurements are greatly contaminated by the volume conduction effect. To address this issue, a novel approach is to map the EEG signals from electrode space into spatial coordinates of the brain to achieve more distinctive features. Hence, this research is intended to investigate (1) the performance difference of a sensor vs. a source space-based motor-imagery BCI and (2) the effectiveness of source localization using beamforming in non-conventional mental imagery decoding for communication BCI. Firstly, the efficacy of feature extraction in sensor and source space was experimentally compared via using the Linearly Constrained Minimum Variance (LCMV) beamformer and common spatial patterns (CSP) in a two-class motor-imagery paradigm. The analysis suggests that the LCMV beamformer is informing the classifier with meaningful features and state-of-the-art classification accuracies are achieved by the proposed method. Secondly, to make individual optimization of BCI control strategies possible, on a participant by participant basis, the most separable control signals among six MI tasks were identified using spectral and connectivity measures extracted from beamformed sources without prior knowledge on the relevant brain regions/networks. It is demonstrated that beamforming can reveal at least one pair of highly classifiable mental commands specific to the participant which can further be employed in the online setup. Therefore, it is concluded that source space EEG analysis using beamforming, applied to a wide range of mental imagery tasks to select the most separable pair, constitutes a promising framework for further development of BCI systems for communication. | en_US |
dc.language.iso | en | en_US |
dc.title | SOURCE IMAGING IN BRAIN-COMPUTER INTERFACES FOR COMMUNICATION | en_US |
dc.type | Thesis | en_US |
dc.contributor.department | Computational Engineering and Science | en_US |
dc.description.degreetype | Thesis | en_US |
dc.description.degree | Master of Science (MSc) | en_US |
Appears in Collections: | Open Access Dissertations and Theses |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
Mousapour_Leila_202109_MSc.pdf | 10.2 MB | Adobe PDF | View/Open |
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.