Sound-production Related Cognitive Tasks for Onset Detection in Self-Paced Brain-Computer Interfaces

Objective. The main goal of this research is proposing a novel method of onset detection for Self-Paced (SP) Brain-Computer Interfaces (BCIs) to increase usability and practicality of BCIs towards real-world uses from laboratory research settings. Approach. To achieve this goal, various Sound-Production Related Cognitive Tasks (SPRCTs) were tested against idle state in offline and simulated-online experiments. An online experiment was then conducted that turned a messenger dialogue on when a new message arrived by executing the Sound Imagery (SI) onset detection task in real-life scenarios (e.g. watching video, reading text). The SI task was chosen as an onset task because of its advantages over other tasks: 1) Intuitiveness. 2) Beneficial for people with motor disabilities. 3) No significant overlap with other common, spontaneous cognitive states becoming easier to use in daily-life situations. 4) No dependence on user’s mother language. Main results. The final online experimental results showed the new SI onset task had significantly better performance than the Motor Imagery (MI) approach. 84.04% (SI) vs 66.79% (MI) TFP score for sliding image scenario, 80.84% vs 61.07% for watching video task. Furthermore, the onset response speed showed the SI task being significantly faster than MI. In terms of usability, 75% of subjects answered SI was easier to use. Significance. The new SPRCT outperforms typical MI for SP onset detection BCIs (significantly better performance, faster onset response and easier usability), therefore it would be more easily used in daily-life situations. Another contribution of this thesis is a novel EMG artefact-contaminated EEG channel selection and handling method that showed significant class separation improvement against typical blind source separation techniques. A new performance evaluation metric for SP BCIs, called true-false positive score was also proposed as a standardised performance assessment method that considers idle period length, which was not considered in other typical metrics.

Name dc:type.qualificationname

phd

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of Essex

Year dc:date.issued

2017

Author dc:creator

• Song, Youngjae

• QA75 Electronic computers. Computer science

Language dc:language

en