By Luigi Bianchi, Tor Vergata University Rome
Brain-Computer Interfaces (BCI) allow people to interact with the environment for either communicating or controlling external devices without using the natural pathways of nerves and muscles . By inducing endogenously or exogenously recognizable brain states, a user intention can be deduced by a special machine that can then drive an external peripheral.
Several different protocols have been proposed over the years, and several brain signals have been analyzed such as EEG, MEG, ECoG, fNIRS and fMRI.
BCI constitutes a highly multidisciplinary research field that has gained great interest in the last two decades, in which several research areas are involved such as engineering, computer science, robotics, neurology, neurophysiology, psychology and rehabilitation. Moreover, the experts must interact not only among themselves but also with patients, health professionals and medical doctors to design or tune a system in the most efficient way. This richness of expertise, however, has some drawbacks because different vocabularies and points of view are used to deal with the same model or BCI system element, and this can easily lead to misunderstandings. Since the early days, it was clear that the large variety of BCI systems could generate confusion: for this reason, in 2003, Mason et al  proposed a general static (e.g. no timing issues among modules were dealt) functional model, which is illustrated in Fig. 1: the two relevant main components are the Transducer and the Control Interface. The transducer, in short, is responsible of detecting brain states and its output (a logical symbol, LS, which is the classifier output in general has no semantic meaning) constitutes the input for the control interface, which is responsible of encoding sequences of LSs into a Semantic Symbol SS such as a spelling device that converts classifier’s outputs into a character of the English alphabet.
Fig. 1 – Mason’s functional model of a Brain-Computer Interface
However, even if this functional model were widely adopted, how can we measure BCI performances? Typically, computer scientists are more interested at increasing brain states classification accuracy whereas Amyotrophic Lateral Sclerosis (ALS) patients are usually demanding to maximize their communication speed. Even if the two ways of expressing the performances of BCIs seem comparable, they are actually not: in the first case only the identification of brain patterns is involved, that occurs at the output of the transducer, while in the second case also the control interface play a relevant role (e.g. the choice of the used alphabet) that affects the performances of a system. This simple fact could make it difficult, if not impossible, to compare different systems and is caused by the lack of standardized procedures.
In addition, clear and widely accepted definitions of simple characteristics such as “trial”, “session”, “run”, “real-time”, are missing, which very often differ among research laboratories, manufacturers and the available frameworks, making the description of a system confusing.
In 2008, Quitadamo et al.  extended Mason’s model that evolved from static to dynamic, thus dealing with timing issues and synchronization among the various elements, by means of a detailed description made in Unified Modeling Language (UML). In this work she demonstrated that it could be successfully applied to five different commonly used BCI protocols: P300, SSVEP, Motor Imagery, Slow Cortical Potentials and fMRI mental tasks. The great advantage of such implementation was that all the systems shared the same terminology and metrics and that it could be possible to unify their description, making it easy to compare and describe different systems. However, even if several BCI system frameworks were made available over the years, none of them but  fully adopted it, making it virtually impossible to share resources among different implementations and very often to compare the performances of the various systems.
As a consequence of all the different visions of what a BCI is, it seems impossible today to imagine converging towards common definitions and methods which allows a painless sharing of resources. The scenery is complex, with different models, methods and frameworks and consequently different file formats that make the cooperation among different laboratories very difficult.
Today the existence of BCI standards is mandatory and their adoption cannot be delayed anymore. This process, however, should be implemented smoothly in order to minimize the effort of making standard compliant to the actually available systems and to maximize the perception adhering to them will provide great advantages to patients, users, manufacturers and the scientific community.
The clear starting point of the standardization process is the definition and adoption of a common BCI functional model that will then open the way to the definition of file formats and tools for designing, describing, optimizing, evaluating, comparing and tuning systems that could be shared among caregivers, health professionals, researchers and engineers. IEEE Standards Association and Brain-Computer Interface Society can clearly play a fundamental role to achieve this goal. Previous experiences demonstrated that it is possible to share a common BCI model and terminology across a wide range of BCIs providing the aforementioned advantages.
A roadmap has been also proposed in  showing that relevant benefits can be easily obtained with little effort, even if limited to offline analysis, systems configuration and in general non real-time BCI behavior. This last, which requires a relevant effort to adapt existing systems to a common dynamic implementation of a BCI, could be however addressed in a successive phase.
- J. Wolpaw, E.W. Wolpaw, “Brain-Computer Interfaces: Principles and Practice”, Oxford University Press, Oxford, 2012.
- S. G. Mason, G. E. Birch, “A general framework for brain-computer interface design”, IEEE Trans Neural Syst Rehabil Eng., vol. 11(1): 70-85, Mar 2003.
- L.R. Quitadamo, M.G. Marciani, G.C. Cardarilli, L. Bianchi. “Describing different brain computer interface systems through a unique model: a UML implementation.” Neuroinformatics. 2008 Summer;6(2):81-96.
- P. Brunner, L. Bianchi, C. Guger, F. Cincotti, G. Schalk., “Current trends in hardware and software for braincomputer interfaces (BCIs).” J. Neural Eng. 2011 Apr;8(2):025001 L. Bianchi, “Brain-Computer Interface Systems: Why a Standard Model is Essential”, in: 2018 IEEE Life Sciences Conference (LSC). p. 134-137, Piscataway (NJ):IEEE, Montreal, QC, Canada, 28-30 Oct. 2018.
Luigi Bianchi, PhD.
Department of Civil Engineering and Computer Science
Tor Vergata University Rome