Design And Optimization On Evoked Potential-Based Brain-Computer Interface

Brain-computer interface(BCI) is a novel communication technique that allows direct connection between the brain and a computer whilst bypassing the participation of peripheral nerves and muscles. Evoked potential-based BCI(EP-BCI) is a branch of BCI research that has attracted significant attention due to its comparatively superior performance in terms of simplicity of implementation, efficiency, and multi-targetted information transfer. Although in recent years great progress has been made in EP-BCI research and the information transfer rate(ITR) has been significantly enhanced; the required performance of EP-BCI for real-life applications is still a distant objective. To achieve practical application, our principal aim was to identify novel and effective BCI approaches to enhance the system performance. This study primarily focuses on the BCIs based on event-related potential(ERP) and steady-state visually evoked potential(SSVEP), which includes multimodal stimuli, multiple brain potentials, and the human factors in the BCI performance. The contents and contribution of the dissertation are summarized as follows.Exploring auditory-tactile-based bimodal stimuli for vision-independent BCI design. To verify the feasibility of multimodal stimuli for enhancing EP-BCI performance, this study proposed a bimodal ERP-BCI that simultaneously employs auditory and tactile stimuli. With this approach, we designed a direction-congruent bimodal paradigm by randomly and simultaneously presenting auditory and tactile stimuli from the same direction. As there is no visual interaction required of the user, we term this bimodal paradigm a vision-independent BCI. Furthermore, the EEG signal channels and number of trials were tailored to each user to improve online performance. The experimental results showed that the average online ITR of the bimodal approach improved by 45.43% and 51.05% over that attained with the individual auditory and tactile approaches, respectively. To the best of our knowledge, the proposed bimodal BCI system boasts the most competitive online ITR for a gaze-independent ERP-BCI to date(10.77 bits/min). These findings suggest that the proposed bimodal approach provides a new direction for vision-independent ERP-BCI research.Exploring real-time feedback and dynamic optimization for SSVEP-BCI. To enhance the system performance, this study proposed a dynamically optimized SSVEP-BCI with a real-time feedback mechanism. In this approach, we implemented a feedback mechanism based on real-time SSVEP signal processing, to increase attention on the visual stimuli. Secondly, we introduced a row/column(RC) paradigm into the SSVEP-BCI to establish a SSVEP-BCI speller with 36 items that uses only six frequencies. In addition, a dynamic optimization approach was designed for setting the stimulus time, and was compared with the fixed optimization approach via online experiments. The experimental results suggested that the proposed SSVEP-BCI speller significantly enhanced the performance compared with the traditional approach. Specifically, the real-time feedback increased the spelling accuracy by improving the participants? attention on the target stimuli. The dynamic optimization approach yielded a higher practical ITR(PITR) than the fixed optimization approach.Exploring and optimal design on hybrid BCI based on P300 and SSVEP. In EP-BCI research field, most previous studies have only relied on the single-mode brain potential, which approaches a ceiling in system performance. To further enhance the BCI performance, we proposed three hybrid approaches by combining P300 and SSVEP. Additional details of these approaches are described below.We initially demonstrated the feasibility of simultaneously evoking the P300 and SSVEP potentials. We then progressed to propose a novel hybrid approach based on the incorporation of SSVEP into the P300 paradigm. In this approach, the hybrid stimuli mechanism consisted of random flashes and periodic flickers of the symbols. The target symbols in the hybrid BCI speller were detected by the fusion of three-dimensional time-frequency features. The experimental results of single trials using the proposed BCI speller exhibited an online spelling accuracy of 93.85%. The pilot studies suggested that the proposed BCI speller could achieve a more accurate and stable system performance compared with the conventional P300 speller.Secondly, to increase the spelling speed, we proposed a novel faster hybrid BCI approach based on the parallel spelling of P300 and SSVEP. In this approach, the target item was identified by 2-D coordinates that were simultaneously realized by the P300 and SSVEP sub-spellers. The sub-area/location(SL) and RC faster spelling paradigms were designed based on this approach. Furthermore, the PITR was used to determine the optimal number of stimulus trials for each subject during online spelling. The experimental results demonstrated that our hybrid BCI approach could achieve higher spelling speed compared with the conventional P300 and SSVEP spellers. The average PITR achieved using the RC technique was 53.06 bits/min.Finally, to achieve more reasonable use of P300 and SSVEP information, we proposed a hybrid BCI with 64 selectable items based on the fusion of P300 and SSVEP brain signals. To achieve this, RC P300 and two-step SSVEP paradigms were integrated to create two hybrid paradigms, which we denoted as the double RC(DRC) and four-dimensional(4-D) spellers, respectively. We further proposed a maximum-probability estimation(MPE) fusion approach to combine the P300 and SSVEP on a score level. The experimental results indicated that the 4-D hybrid paradigm outperformed the DRC paradigm and that the MPE fusion achieved higher accuracy compared with four other widely used score fusion techniques. Importantly, twelve of the thirteen participants, using the 4-D paradigm with single trials achieved an online accuracy of over 90%(the average accuracy was 95.18%).