Online P300 Brain-Computer Interface Based On SCF Paradigm

Brain-computer interfaces (BCI) can provide physically disabled patients with a direct communication and control channel between the brain and external devices. Event-related P300 potential is one type of electroencephalography (EEG) pattern that is widely used in non-invasive BCI research. P300 potential has the advantage of stability and requires no subject training. Most P300-based BCI research focuses on the classical P300 Speller which adopts a row/column flash (RCF) paradigm to elicit P300 potentials. However, recent research found that P300 Speller is too complicated for certain subjects and inapplicable under some circumstances. Moreover, stimulating parameter, P300 recognition, online adaption and many other problems are calling for deeper and more systematic studies in P300 BCI.To eliminate the constrains of P300 Speller and study other issues in P300 BCI research, current work investigated P300 BCI based on single character flash (SCF) paradigm and further explored issues in interactive interface, feature extraction, classification, online adaption and so on, aiming to improve performance of the online P300 BCI based on SCF paradigm. These include:1. Current work designed a configurable P300 BCI based on SCF paradigm, where single character flashed as the stimulus to elicit P300 potentials, differing from the classical RCF paradigm. On the basis of the platform, a P300 BCI based Chinese typewriter was implemented. Its 7-oddball SCF paradigm utilized the facts that Chinese characters consist of strokes and fives type of strokes constitute all Chinese characters. This way, a quick and efficient P300 Chinese input model was put forward for the first time.2. A personalized configuration strategy of the P300 stimulating parameter to was put forward improve the interactive interface. Using the developed P300 BCI based on SCF paradigm, current study investigated P300 stimulating parameter configuration. Experimental configuration modes were set and tested with online feedback. Results showed that all the modes could cause different amplitudes and latencies of P300 wave, and thus lead to diverse online accuracies. So a personalized configuration strategy of stimulating parameter in terms of the characteristic of each subject was proposed.3. In feature extraction of the SCF based P300 BCI, ocular artifacts removal by linear regression and EEG channel selection by rough set were proposed. Three strategies of ocular artifacts removal by linear regression were introduced and tested, which proved to be useful in the P300 BCI. EEG channel selection by rough set was proposed for the first time. It requires no classifier modeling and training to determine the importance of each EEG channel, and thus provides a quick and effective procedure to optimize EEG channel and reduce signal dimension. Preliminary results were consistent with the traditional experience based channel selection in previous P300 BCI studies.4. Recognition and averaging problems under the SCF paradigm of P300 BCI were also studies, aiming to higher classification accuracy of P300 detection. Then an online overlapping adaptive P300 model was proposed, which effectively overcame the problem that processing time delay exists in inter-trial intervals. Fisher's linear discriminant analysis (FLDA), support vector machine (SVM), as well as bagging algorithm were implemented and achieved better P300 detection performance under the SCF paradigm. Relations between averaging modes, number of averaging and performance were also studied. An accuracy of 93.07% averaging over 4 trials was obtained by FLDA trained from single trial samples. Based on these, an online overlapping adaptive P300 model was put forward, which shortened processing time delay of the adaptive averaging by overlapping stimuli and feedback on the user interface, resulting in a better online adaption.5. A hybrid EEG control strategy was proposed to expand function of P300 BCI. By time sharing strategy of detecting both P300 potential and motor imagery related sensorimotor rhythms in one BCI system, the hybrid control strategy provides discrete, high-dimensional control commands via P300 potential, and low dimensional, continuous commands via motor imagery patterns. This control strategy was applied to a virtual environment navigation and control system, and it effectively expanded control functions of conventional single motor imagery or P300 potential based BCI in the virtual environment.Compared to most other P300 BCI studies based on offline analysis, current work not only developed an online P300 BCI based on SCF paradigm, but also designed various testing experiments with online feedback to validate methods and models proposed in the work. Moreover, the experimental tasks were context-sensitive, which conformed to realistic scenario of P300 BCI application.