Closed loop BCI system for Cybathlon 2020

Abstract

ABSTRACT We developed a Brain-Computer Interface (BCI) System for the BCI discipline of Cybathlon 2020 competition, where participants with tetraplegia (pilots) control a computer game with mental commands. To extract features from one-second-long electroencephalographic (EEG) signals, we calculated the absolute of the Fast-Fourier Transformation amplitude (FFTabs) and introduced two methods: Feature Average and Feature Range. The former calculates the average of the FFTabs for a specific frequency band, while the later generates multiple Feature Averages for non-overlapping 2 Hz wide frequency bins. The resulting features were fed to a Support Vector Machine classifier and tested on the PhysioNet database and our dataset containing 16 offline experiments recorded with the help of 2 pilots. 27 gameplay trials (out of 59) with our pilots reached the 240-second qualification time limit, which demonstrates the usability of our system in real-time circumstances. We critically compared the Feature Average of canonical frequency bands (alpha, beta, gamma, and theta) with our suggested range30 and range40 methods. On the PhysioNet dataset, the range40 method combined with an ensemble SVM classifier significantly reached the highest accuracy level (0.4607), with a 4-class classification; moreover, it outperformed the state-of-the-art EEGNet.