Evaluation of different training strategies for motor-imagery brain-machine interfaces

Last years have seen an increasing interest in the field of human-machine interaction (HMI), especially thanks to new methods to acquire signals and new improvements on processing algo- rithms. For people affected by deabilitant diseases that make movement difficult or impossibile (as for paraplegic and tetraplegic patients), the interaction between human and machines is typ- ically possible using the brain-machine interfaces (BMIs) by which, using specific patterns of neural activity, a person is able to send a command to an external device. In order to perform a correct utilization of BMI, users should undergo a training period during which their electroen- cephalogram data are acquired and a classifier is created, leading them to control an external device. Generally, most researchers neglect the importance of subject learning (e.g. both sub- ject and decoder learn from each other) and they prefer to focus on machine learning techniques to increase decoder performances: as it is known from literature, the brain constantly modifies itself (neuroplasticity) and if a continuous re-calibration of decoders is performed, the subject could not learn enough to use BMI in daily life because he could not be able to keep his neural patterns stable; moreover despite they are considerably studied in the literature, BMIs are not yet developed enough to be processed and used by all, especially given the high mental fatigue and the high training period to maximize their performances. Therefore, it is necessary to set defined standards to improve training strategies and fully exploit the potential of individuals, thus allowing an extension of the use of BMIs outside the research field: to achieve these condi- tions, this thesis aims to demonstrate that different typologies of training can lead to increased performances in BMIs usage, showing at the same time how the neurophysiological patterns evolve during this period and how subjects get used to maintaining such parameters constant. In point of fact, the main intuition is based on the psychological evaluation of people and how 2 they manage to change their characteristics depending on habits: if the main objective will be to use, as experienced in this elaborate, a wheelchair, then those who immediately train with it, will be better and consistent in its use, compared to others who train in different ways. To give to the reader a general view of this environment and to provide him the keys to a deeper understanding of the goal, this work has been divided in four principal areas: in the first one, it is given a brief recap of the brain and how it works with a detailed description of BMIs and the methods to create them; the second part is devoted to define the experimental protocol of the project, highlighting the different strategies chosen for the experiment and the importance of mutual learning during all the days of the trials; in the third part, there will be explained the methodologies of data processing and the algorithm used for the classification; finally, in the last chapter, all the acquired data are evaluated with multiple statistical and numeral approaches in order to demonstrate the differences and the advantages of a training typology respect to the other one, including also informations that are generally not covered in the literature. Indeed, the choice of the best training protocol can increase significantly both subject and decoder performances, decreasing the days required to master the usage of BMI and simplifying the approach: if a new standard and easier learning protocol is defined, then this technology could also be used by people with motor impairment, whose features are usually not stable, without the need of an hard work.