Tools
Tools
Falcon Caro, A 
ORCID: https://orcid.org/0000-0002-1085-7716,
2025.
Electroencephalogram hyperscanning-based brain-computer interfacing.
PhD, Nottingham Trent University.
Preview |
Text
Alicia FalconCaro 2025.pdf - Published version Download (13MB) | Preview |
Abstract
Electroencephalography (EEG) hyperscanning refers to simultaneous EEG recordings from multiple subjects. This scenario has become popular for studying group performances by social science activists to investigate human competing or collaborative actions. However, the use of EEG hyperscanning had not become practical in brain-computer interfacing (BCI) prior to this research.
In this thesis, we investigate the use of EEG hyperscanning for two BCI scenarios; one where the BCI tasks are performed in an uncontrolled environment in which the brain is naturally engaged in multi-task activities, and the objective is to enhance the BCI training accuracy. The other one is to assess how well a subject’s brain can follow another brain. The latter scenario can be useful in brain rehabilitation, i.e., after stroke, or for investigating brain plasticity where new motor region substitutes the inactive zone.
For the first scenario, a new formulation for common spatial patterns for EEG hyperscanning (namely hyperCSP) has been proposed. It learns a rotating matrix to best derive the desired (common between multiple subjects) motor task and remove all the undesired (uncommon between the subjects) ones caused by the uncontrolled environment. It has been demonstrated that in an uncontrolled environment, learning from two subjects can significantly enhance the motor classification rates, achieving a best classification accuracy of 0.82 using an SVM classifier.
For the second scenario, the concept of adaptive cooperative networking has been exploited to investigate the collaboration or similarity in function between multiple brains. A brain functional connectivity-informed single task diffusion adaptation has been used for this purpose, achieving the lowest error compared to competing methods. In this scenario, the EEG sensors for each subject are considered as the agents of a connected network, that cooperate to achieve a desired task prescribed by a second subject. The outcome of this achievement can boost brain rehabilitation process where the collaboration or difference between a healthy and a subject under rehabilitation is investigated.
Finally, given that the brain can be considered as a connected network, a prolonged physical task can be classified by assessing the variation in such network through cooperative networking using diffusion adaptation. In this study, the 3-D body movements are segmented and 1-D orthogonal vectors, in this case Bessel functions, are allocated to the segments. These orthogonal vectors are used as the targets for classification of the brain tasks, obtaining a best inter-subject validation accuracy of 0.83 under a non-ideal situation. The outcome of this research paves the path for the use of BCI with no feedback (forward BCI) for developing brain-driven cybernetic devices.
The research in this thesis is concluded by suggesting combination of hyperCSP and other proposed methods for obtaining inclusive BCI systems and a pipeline for real-time applications.
Each chapter of this thesis is followed by extensive sets of experiments and involves in-house data recordings (some made publicly available).
| Item Type: | Thesis |
|---|---|
| Creators: | Falcon Caro, A. |
| Contributors: | Name Role NTU ID ORCID |
| Date: | February 2025 |
| Record created by: | Jeremy Silvester |
| Date Added: | 15 Aug 2025 10:01 |
| Last Modified: | 15 Aug 2025 10:01 |
| URI: | https://irep.ntu.ac.uk/id/eprint/54215 |
Actions (login required)
 |
Edit View |
Statistics
Views
Views per month over past year
Downloads
Downloads per month over past year
