Task Recognition in BCI via Short- and Long-Term Dynamic Entropy with Robotic Aid in Sight.

Application of time-series complexity measures (also known as entropies) has been quite useful in many Biomedical Engineering fields as in Brain–Computer Interfaces (BCI). Nevertheless, recently-proposed-entropy parameters referred to as dynamic complexity measures have not concretely been tested in BCI environment. The novelty of our research is evaluating the performance of Dynamic Bivariate Multiscale Entropy parameters (DBMSE) as alternative to typical classification algorithms in BCI, as common spatial patterns (CSP). After developing BCI training, the statistical results showed that CSP performed better than our initial DBMSE (DBMSE-I). This entropy uses a constant threshold r. However, after tunably including that quantity r, DBMSE-I improved to be almost as good as CSP. This updated entropy parameter was referred to as DBMSE-II. Since the limitations of these results were attributed to the shortness of the EEG records, a short-term-DBMSE (named here DBMSE-III) was developed. The findings revealed that DBMS-III had a slightly better performance than CSP outcomes, thus offering a new classifier alternative in BCI. Besides, we present entropy paths that reveal high/low complexity brain activity. This will be useful to identify EEG electrodes to move a robot within a preliminary assistance protocol in progress. Summing up: Besides knowing that DBMSE parameters are a good choice to show brain dynamics, DBMSE-III proved to be an alternative as classifier for CSP in our study, allowing interaction with a robot. In fact, we offer in this document a complete development from theory to execution of software and hardware. This experience will help support people with disabilities and their carers. [ABSTRACT FROM AUTHOR]