Resumen de Towards smarter Brain Computer Interface (BCI): Study of electroencephalographic signal processing and classification techniques toward the use of intelligent and adaptive BCI
Flavio Vinicio Changoluisa Panchi
The brain is one of the most important organs in our body. It is the one that allows us to think and feel, the one that controls our heart rate or the digestion of the food we eat. However, to this day, it is still a little-known organ in its operation. The existing exploration techniques help us to know this organ, but we still do not know the fundamentals that govern it. A technology that seeks to contribute to the understanding of the brain and, in turn, influence it is the Brain-Computer Interface (BCI). Its structure is relatively simple: it acquires brain activity, processes it and translates it into actions. The functionality of BCI has been successfully tested in laboratories but without much success in real life. Several challenges still need to be overcome; among them is the non-stationary nature of brain activity, which generates inter- and intrasubject variability that makes it difficult to achieve accurate BCIs over time. This thesis addresses this challenge by investigating feature engineering techniques and artificial intelligence algorithms that allow adaptable BCIs to different real-life circumstances.
For this, we study the characteristics of the EEG signal of Event-Related Potentials (ERPs) and Working Memory (WM) in devices with wet and dry electrode technologies. The EEG signal was studied in both the time and frequency domain. The results obtained show that through characterization, it is possible to achieve an improvement in the performance and adaptability of BCIs. One of the applications of this study is the selection of electrodes that allows improving the performance metrics of a BCI, for example, accuracy, ITR, BCI-Utility. In addition, implemented in dry electrodes significantly improves performance compared to the selection of standard electrodes.
The original approach posed by this thesis is to take advantage of the characteristics of a signal to achieve adaptability of the BCIs. Therefore, from this thesis, we can conclude that proper characterization of the control signal and the use of low computational cost algorithms for its implementation in real-time achieve the adaptability of the BCI and improve the performance compared to standard methodologies used by default.
The findings derived from this thesis will facilitate the creation of low-cost BCIs and contribute to understanding the properties that facilitate their adaptability. The exploration and identification of characteristics is an alternative to understanding an adaptive BCI and all the processes involved. Although, currently, several technologies achieve an adaptation through algorithms that work like a black box (e.g., greedy or exhaustive search algorithms), the limitation of understanding the properties that adaptability involves remains.
