Editing as much as possible
An effective division-of-work strategy for MI-cBCI
Abstract
Collaborative brain computer interface (cBCI) has been demonstrated to be an effective tool to enhance the accuracy of brain control recognition through the increase of the user dimensions. However, current cBCI systems only employ the same-work strategy for collaboration. In other words, the electroencephalography (EEG) features are integrated from the same instruction executed by all users to improve the recognition performance. These systems can barely fully involve advantages of the group collaboration and their performances need further optimization. For this reason, a novel cBCI system suitable for division-of-work for multi-person was proposed, and its performance was compared with those of conventional same-work type ones.
In this study, a set of motor imagery (MI) -BCI division-of-work strategies (5 types in total) suitable for 5 persons and 6 instructions were designed. By planning control instructions and cooperation modes of each user, the features (under the centralized architecture) or decision-making (under distributed architecture) information of multiple people were integrated to improve the recognition accuracy rate of MI-cBCI. The EEG information from 19 subjects that independently performed 6-instruction MI was collected in this study and five of them were randomly selected as a cBCI user group. Off-line data processing was employed to simulate the on-line recognition process of the division-of-work type MI-cBCI. Under all division-of-work strategies, the minimum average accuracy rates of MI-cBCI were significantly higher than that of the single-person level. For both centralized and distributed architectures, the recognition accuracy rates of MI-cBCI were maximized under the third division-of-work strategy and higher than that of conventional same-work type system.
To the best of our knowledge, this study applied the division-of-work strategy to the cBCI system for the first time. The results demonstrated that the division-of-work type MI-cBCI was a collaborative strategy that can illustrate the group advantages. It can enhance the recognition performance of the system and reduce the work load of a single person, thus providing support and references for future applications of BCI technology in social work.
1. Introduction
Brain-computer interface (BCI) allows the direct interaction between brains and external devices [3]. Compared with other human output pathways, such as body motion and voice, it provides highly efficient human-computer interaction channels. BCI achieves the expressions of active brain intentions and the monitor of physiological status by detecting and analyzing neural activity information. According to different sensing modes, BCI can be classified into multiple types, such as MRI, NIR, and EEG et al., where each type has advantages and suitable applications. For instance, due to the high time resolution and portability of EEG-BCI, it is more suitable for active control for external devices in daily operations.
Based on the purpose of external control, EEG-BCI can be divided into two types: active BCI without the requirement of external inductive stimulus, represented by motor imagery brain-computer interface (MI-BCI), and reactive BCI that needs to be stimulated by related events, such as SSVEP-BCI based on visual stimulus [4]. However, active BCI has not been widely used in social works mainly due to the following two reasons: (1) Low information transmission rate. The signal-to-noise ratio of current EEG sensing technology is very low, so that the system is incapable to extract enough effective neural response features in a short time, resulting in poor decoding performance. For elaborate control operations, sufficient human-computer hybrid intelligence is usually required in order to achieve operations with high precision, large instructions, short time delay and long-term, which can hardly be fulfilled by current BCI systems. (2) Weak group connection. The social characteristics of human determines that the form of human-computer interaction should not be limited to single person- single computer independent operation. Instead, it should be multiple persons-multiple computers collaboration operation. Single-person BCI can hardly involve intergroup connection and complementarity, so that it is not adaptable to a large scale group operations. Therefore, a collaborative brain computer interface (cBCI) has been proposed, which is suitable for group interaction and improves the recognition performance through the increase of the user dimensions. Its advantages not only lie in the effective fusion of group EEG features to improve decoding accuracy and robustness, but also can take full advantages of group interaction, thus improving the decision confidence of human-computer hybrid intelligence in advanced tasks [5].