Custo por atraso de janela: uma nova visão sobre o desempenho de interfaces cérebro-máquina

Abstract

Brain-Computer Interfaces (BCIs) have several applications in the areas of entertainment, health, and quality of life. For this reason, their research has been increasingly encouraged. These incentives for the development of BCIs occur primarily through competitions. However, because this form of incentive is predominant, it has caused a bias in the evaluation of BCIs. Currently, BCIs are evaluated mainly on the basis of their classification capacity at a fixed point in time. Thus, all the temporal characteristics present in the BCI applications are neglected. The main objective of this work is to propose new metrics that incorporate the temporal properties of BCIs. In addition, it analyzes the different types of temporal and spatial filters for BCIs together with the use of convolutional neural networks. To evaluate the new proposed metrics, 48 different models for BCIs were created and tested on five data sets from the literature. The results showed that the new metrics brought about new interpretations of the performance of the BCIs. Including showing that models like EEGNet, which are considered slightly superior to the others, would present some limitations in real applications that have not yet been pointed out in the literature.