Research On Distributed Computing Model In Brain-Computer Interfaces

Brain-Computer Interfaces(BCIs)can provide a new pathway that can control external devices directly through brain without any use of peripheral nerves and muscles.It is widely used in scientific research,medicine,military and entertainment.The P300 Speller is one of the most popular BCI systems based on P300.It can help users select characters from a matrix of characters by detecting P300 event-related potentials.Our researches address the problem on low Information Transfer Rate(ITR)of P300 Speller.We propose a real-time distributed computing mechanism for P300 Speller BCI.The previous research used serial computation whether proposing new paradigms or improving algorithms.The method in this paper is different from the previous work which used serial computation.It proposes a real-time distributed computing mechanism based on Storm to improve ITR.Although there are many different paradigms of P300-based BCIs,the process of signal processing is basically consistent.In general,the process includes 5 steps:preprocess(),segment(),extract(),classify()and synthesize().This paper simplifies the technical details of signal processing so that we can focus our research on how to apply the steps of signal processing to the real-time distributed computing framework.In Storm,the mechanism assigns more computing resources to the steps whose computation complexity is high to perform parallel computation for higher computational efficiency.The steps whose computation complexity is low doesn’t need more computing resources,because it will lead to waste of resources.This paper also combines a threshold algorithm to adjust the times of repeated stimulus dynamically according to the subjects’performance so that it can reduce the time of spelling a character under a high accuracy.The experiment used P300 Speller paradigm and used ITR as a metric.Eight subjects completed the traditional P300 and parallel P300 which is mentioned in this paper.The results showed that the algorithm for P300 Speller could be computed faster on this mechanism proposed than it is done without this mechanism.In the case of using SWLDA,the parallel P300 improved the ITR to 11.51 bits/min on average,which was 63%higher than using the traditional method.In the case of using SWLDA and wavelet transform,the parallel P300 improved the ITR to 11.70 bits/min on average,which was 64%higher than using the traditional method.