Non-invasive Mental Task-based Brain-Computer Interfaces (MT-BCIs) enable their users to interact with the environment through their brain activity alone (measured using electroencephalography for example), by performing mental tasks such as mental calculation or motor imagery. Current developments in technology hint at a wide range of possible applications, both in the clinical and non-clinical domains. MT-BCIs can be used to control (neuro)prostheses or interact with video games, among many other applications. They can also be used to restore cognitive and motor abilities for stroke rehabilitation, or even improve athletic performance.
Nonetheless, the expected transfer of MT-BCIs from the lab to the marketplace will be greatly impeded if all resources are allocated to technological aspects alone. We cannot neglect the Human End-User that sits in the centre of the loop. Indeed, self-regulating one’s brain activity through mental tasks to interact is an acquired skill that requires appropriate training. Yet several studies have shown that current training procedures do not enable MT-BCI users to reach adequate levels of performance. Therefore, one significant challenge for the community is that of improving end-user training.
To do so, another fundamental challenge must be taken into account: we need to understand the processes that underlie MT-BCI performance and user learning. It is currently estimated that 10 to 30% of people cannot control an MT-BCI. These people are often referred to as “BCI inefficient”. But the concept of “BCI inefficiency” is debated. Does it really exist? Or, are low performances due to insufficient training, training procedures that are unsuited to these users or is the BCI data processing not sensitive enough? The currently available literature does not allow for a definitive answer to these questions as most published studies either include a limited number of participants (i.e., 10 to 20 participants) and/or training sessions (i.e., 1 or 2). We still have very little insight into what the MT-BCI learning curve looks like, and into which factors (including both user-related and machine-related factors) influence this learning curve. Finding answers will require a large number of experiments, involving a large number of participants taking part in multiple training sessions. It is not feasible for one research lab or even a small consortium to undertake such experiments alone. Therefore, an unprecedented coordinated effort from the research community is necessary.
We are convinced that combining forces will allow us to characterise in detail MT-BCI user learning, and thereby provide a mandatory step toward transferring BCIs “out of the lab”. This is why we gathered an international, interdisciplinary consortium of BCI researchers from more than 20 different labs across Europe and Japan, including pioneers in the field. This collaboration will enable us to collect considerable amounts of data (at least 100 participants for 20 training sessions each) and establish a large open database. Based on this precious resource, we could then lead sound analyses to answer the previously mentioned questions. Using this data, our consortium could offer solutions on how to improve MT-BCI training procedures using innovative approaches (e.g., personalisation using intelligent tutoring systems) and technologies (e.g., virtual reality). The CHIST-ERA programme represents a unique opportunity to conduct this ambitious project, which will foster innovation in our field and strengthen our community.
C. Jeunet, C. Benaroch, F. Cabestaing, R. Chavarriaga, E. Colamarino, M.-C. Corsi, D. Coyle, F. De Vico Fallani, S. Enriquez-Geppert, P. Figueirédo, M. Grosse-Wentrup, S. Kleih, S. Kober, A. Kübler, F. Lotte, E. Maby, D. Mattia, J. Mattout, G.R. Müller-Putz, S. Perdikis, L. Pillette, A. Riccio, S. Rimbert, A. Roc, R. N. Roy, R. Scherer, P. Seguin, H. Si-Mohammed, T. Tanaka, M. Tangermann, L. Tonin, A. Vourvopoulos, A. Vuckovic, G. Wood, S. Wriessnegger