Deep Learning-Based Classification of Finger Movements using tEEG and EEG Signals

Mapping electroencephalography (EEG) signals obtained during hand and finger movements to left or right hands has applications in many fields including motor imagery brain-computer interface (MI-BCI) systems. MI-BCI systems are innovative technology to assist patients with severe motor impairments, however, novice users can face the "BCI-illiteracy" issue and fail to control the system. A step towards assisting such patients is to correctly classify their EEG signals collected while the patients imagine or perform hand/finger movements. This work considers using the tripolar electroencephalography (tEEG) instead of the traditional EEG due to higher signal-to-noise (SNR), better spatial resolution, better artifact noise rejection feature, and ability to capture high-frequency features of tEEG, compared to conventional EEG. In this paper, we compare the performance of tEEG and EEG signals in classifying motor imagery (MI) of right and left index fingers by using a deep learning algorithm: convolutional neural network (CNN). The results show that in 5 out of 7 subjects, CNN is able to perform classification with high accuracy, with the highest accuracy being 96.4% for tEEG data.