Ballistic Finger Imagery for Asynchronous BCI Control

Abstract

In this work a novel paradigm and algorithm are developed in order to decode ballistic finger imagery of both the left and right index fingers in real time. The novel paradigm is specifically developed for rapid decision making and high information transfer rates. This is done through a hybrid classifier as well as the use of tripolar electro-encephalography (TEEG). TEEG cortical signals are also compared with concurrently recorded traditional EEG signals (Au), as well as high density EEG signals (Ag/AgCl). Online results are shown to be stable across 10 subjects over three sessions. Spatio-temporal analysis is performed in both the sensor and electro-cortical source domain offline and additional features are derived through a combination of data-mining and neuroscientific insight to develop a robust TEEG ERP derived classifier which achieves 88.8% accuracy in a subject independent monte-carlo cross validation simulation. TEEG signals are found to have similar waveforms and spatial maps as EEG signals, but are found to be more independent with low or often negative correlations with other channels, as well as diverse covariance matrices.