A novel P300 BCI speller based on the Triple RSVP paradigm

Zhimin Lin, Chi Zhang, Ying Zeng, Li Tong, Bin Yan
2018 Scientific Reports  
A brain-computer interface (BCI) is an advanced human-machine interaction technology. The BCI speller is a typical application that detects the stimulated source-induced EEG signal to identify the expected characters of the subjects. The current mainstream matrix-based BCI speller involves two problems that remain unsolved, namely, gaze-dependent and space-dependent problems. Some scholars have designed gaze-independent and space-independent spelling systems. However, this system still cannot
more » ... hieve a satisfactory information transfer rate (ITR). In this paper, we propose a novel triple RSVP speller with gaze-independent and space-independent characteristics and higher ITR. The triple RSVP speller uses rapid serial visual presentation (RSVP) paradigm, each time presents three different characters, and each character is presented three times to increase the ITR. The results of the experiments show the triple RSVP speller online average accuracy of 0.790 and average online ITR of 20.259 bit/min, where the system spelled at a speed of 10 s per character, and the stimulus presentation interface is a 90 × 195 pixel rectangle. Thus, the triple RSVP speller can be integrated into mobile smart devices (such as smartphones, smart watches, and others). A brain-computer interface (BCI) system based on electroencephalography (EEG) is a popular research direction in the field of human-computer interaction. The system can provide a direct communication channel to connect the human brain and computer. An EEG speller system is a typical brain-computer interface system. The speller system uses a clever paradigm to induce specific event-related potential (ERP) components (for example, P300 component). Then, according to the ERP components, a symbol of the expected subjects can be determined. The EEG speller system can re-establish the disabled communication, and normal people can use the technology to obtain convenient interactive methods 1-6 . At present, most EEG speller systems are based on modified P300 speller. The P300 component is a common ERP component, which shows a peak when small probability events are observed after approximately 300-500 ms 7-10 . The P300 component also exhibits significant waveform characteristics in the time domain 11 . A P300 detection algorithm is essential because it determines the accuracy and reliability of BCI systems. Thus, some scholars have attempted and proposed many P300 detection algorithms such as independent component analysis (ICA) 12 , common spatial pattern (CSP) 13 , xDawn 14,15 , hierarchical discriminant component analysis (HDCA) [16] [17] [18] [19] , sliding HDCA (sHDCA) 20,21 , and convolutional neural network (CNN) 22 . On this basis, many BCI system performances were improved significantly. Farwell and Donchin first proposed P300 alphabet speller system (FD speller) 23 . The system is a 6 × 6 matrix, and each element is a specified character. The system has a total of 36 characters (26 letters and other control characters) and uses a stepwise linear discriminant analysis (SWLDA) algorithm to detect P300 components. Krusienski et al. 24 compared the performances of various P300 detection algorithms and concluded that SWLDA and Fisher's linear discriminant (FDA) are suitable for the P300 Speller system. In the FD speller, the matrix rows or columns blink randomly, and when the locked symbol is hit by rows or columns (probability is 1/6), it induces a P300 component. The FD speller determines the expected symbol position using the detected P300 component. Cuntai Guan et al. 25 believed that if the probability of a character being hit is small, then the P300 component induced is evident and detection is easier. Thus, Guan et al. proposed a single-character display random flashing speller system (SC speller) based on the FD speller (the probability that the character was hit was 1/36). Furthermore, Townsend et al. 26 proposed a checkerboard paradigm speller system, that is, a random flash hits multiple characters, and then a special coding is employed to confirm the anticipant character of the subject. This method can achieve higher accuracy and mean bit rate. Brian Roark et al. 27, 28 proposed Huffman scanning, which uses Huffman coding to select the symbols to highlight, based on the FD speller. Qi LI et al. 29 proposed
doi:10.1038/s41598-018-21717-y pmid:29463870 pmcid:PMC5820322 fatcat:bmu37bgn6ffb5kqwolh5366b4q