Análise de janelamento em imaginação motora para interface cérebro-máquina

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

Universidade Federal de Juiz de Fora (UFJF)

DOI

Abstract

Various methods have been developed to record neural activity, with electroencephalography (EEG) being one of the most widely used due to its non-invasive and accessible nature. With the advancement of Brain-Computer Interfaces (BCIs), it has become possible to apply these brain signals in various domains, enabling direct interaction between the brain and external devices. One of the most studied paradigms in BCI is motor imagery (MI), in which the subject imagines movements without physically executing them. These signals are captured and processed so they can be interpreted by computational systems. In the literature, it is common to use fixed time windows starting around 2.5 seconds after the beginning of a trial to perform analyses and classifications. The aim of this work is to evaluate the main methods present in the literature for different temporal windowing strategies, investigating not only efficacy but also the efficiency and stability of methods over time. To this end, the following algorithms were tested: Common Spatial Patterns with LogPower and Naive Bayesian Parzen Window classifier, Filter Bank Common Spatial Patterns, and the EEGNet convolutional neural network, applied under different window configurations. The tests were conducted on widely used datasets from international BCI competitions. The results indicated a clear neurophysiological justification for the recurring use of windows around 2.5 seconds. This interval coincides with greater synchronization of beta frequencies, which are associated with motor imagery, thus enhancing signal discriminability. Furthermore, the study showed that using larger windows tends to improve results by providing more relevant temporal information for the classifiers. Another important finding was that combining multiple windows covering different time intervals not only improves the point performance of classifiers but also contributes to greater stability and quality of results over time. This suggests that the traditional approach of using a single fixed window may be limited and that exploring more dynamic and informative strategies holds promise for advancing BCI research.

Description

Citation

Endorsement

Review

Supplemented By

Referenced By

Creative Commons license

Except where otherwise noted, this item's license is described as Acesso Aberto