Section: New Results

Understanding Mental Imagery-based Brain-Computer Interface user-training

Participants : Camille Jeunet, Fabien Lotte

Mental Imagery-based Brain-Computer Interface (MI-BCI) enable their users to send commands to computer by imagining mental tasks (i.e., by performing MI) that are recognized in their brain signals. This type of BCI requires user training, and this training is currently poorly understood, and we basically do not know, who can learn MI-BCI control, what is to learn and how to learn it efficiently. Moreover, we have shown that current MI-BCI training protocols were both theoretically and practically inappropriate, and that there is a lack of fundamental knowledge on BCI user training, which prevents us from designing better user training approach [12].

In order to address these points, we first proposed a review and classification of cognitive and psychological predictors of MI-BCI performance. Three categories were defined: the user-technology relationship, attention and spatial abilities. The user-technology relationship refers to personality traits and states that influence users' perception of the technology and consequently impact the way they will interact with the technology (i.e., their feeling of being in control, their self-efficacy, etc.). The attention category gathers, among others, users' attentional abilities, motivation and engagement towards the task. These elements are essential to learn, whatever the skill targeted. They are also closely related to the user-technology relationship (for instance, feeling in control will increase users' engagement towards the task, thus they will allocate more attentional resources to the task). Finally, spatial abilities are the ability to produce, manipulate and transform mental images, which is closely related to the ability to control an MI-BCI. The description of these categories and of their neurophysiological correlates enabled us to submit ideas to improve MI-BCI user-training. For instance, we explained how to reduce computer-anxiety and increase the sense of agency, notably through the use of a positively biased feedback for novice users. Also, we proposed solutions to raise and improve attention, e.g., using neurofeedback or meditation. Finally, we argued that spatial abilities could be trained to improve users' capacity to perform mental imagery and consequently, potentially improve their MI-BCI performance [17].

We also did a review of the literature of current training protocols (published as a book chapter in [41]) which suggests that these protocols are, at least theoretically, inappropriate to acquire a skill and thus that they could be one of the factors responsible for inefficient MI-BCI user-training. In particular, participants are most of the time provided with uni-modal and evaluative feedback while literature recommends multi-modal, informative and supporting feedback. Although instructive, these insights only provide theoretical considerations about the flaws associated with the feedback approaches used in MI-BCI. It was therefore necessary to concretely assess whether standard MI-BCI feedback is appropriate to train a skill, and to what extent the feedback impacts BCI performance and skill acquisition. In order to experimentally evaluate the extent to which such a feedback has an impact on their ability to acquire a skill, we used it to teach users to perform simple motor tasks. Results (N=53 participants) revealed that with this feedback, 17% did not manage to learn the skill. This suggests that current BCI feedback is most probably suboptimal to teach a skill. A sub-group of our participants (N=20) then took part in a motor-imagery based BCI experiment. Results showed that those who struggled during the first experiment improved in performance during the second, while the others did not. We hypothesised that these results are linked to the considerable cognitive resources required to process this feedback [16].

It should be noted that there are many connections between BCI user training, and neurofeedback training for clinical applications, both field aiming at training their users to perform self regulation of their brain activity. We have therefore shown how these two field share fundamental research questions on BCI user training, and how they can both benefit from each other [10].