Object Movement Motor Imagery for EEG based BCI System using Convolutional Neural Networks

Brain-Computer Interface systems aim to send commands to the computer or a certain device through brain activity. Motor Imagery, part of BCI research field, aims to generate commands by imagining the movements of limbs without performing any real movement. So it is a state in which the subject forces himself/herself into the feeling of performing a particular action. The users provide training data through a particular recording technology, such as EEG, and after the recording, different techniques and algorithms are used to learn patterns from the data in order to map them with a label or output. In this paper, we are interested to investigate, whether it is possible to generate similar results as in the case of imagining the movement of limbs, by imagining the movement of an external object. To investigate this we show to our subject a video of a box that moves through a plane in three different directions, right, left, and forward. The mental task consists of imagining we are moving the box with our brain activity. Since embodiment provides greater accuracy in motor imagery, this kind of visual feedback, when there is no obvious congruity with any daily neural command, brings a greater difficulty than usual motor imagery. The recorded EEG data are split into training and testing subsets, and are fed to a simple convolutional neural network, that tries to learn the different patterns and to classify them. The network demonstrates the possibility to achieve meaningful accuracy for our object movement imagery.