Making brain-computer interfaces better : improving usability through post-processing

Abstract

A c r o n y m s BCI brain-computer interface. EEG electroencephalography.ERP event-related potential.FMRI functional magnetic resonance imaging. FNIRS functional near-infrared spectroscopy.HCI human-computer interaction.MEG magnetoencephalography. MI motor imagery.PET positron emission tomography.SCP slow cortical potential.SSVEP steady-state visually-evoked potential. G l o s s a r yApplication A computer program that helps a user with some particular task or goal. AsynchronousThe user can provide input to the system at any time. See also synchronous. BCI cycleThe cycle of interaction that takes place within and between the user and the system, including the processing. BCI pipelineThe sequence of data processing steps from the user to the BCIcontrolled application. This simplified BCI system view ignores the effect the BCI pipeline can have on the user, which in turn will affect the system. Biosemi ActiveTwo A high-grade EEG system, using a cap for positioning and gel for conduction. 256 electrodes can be mounted in one cap. The sampling rate can be a maximum of 16 kHz.Brain-computer interface A system that recognizes mental tasks or states based on the user's brain activity. This allows you to control (or otherwise affect) devices (or applications) directly with your brain. Cerebral cortexThe part of the brain closest to the scalp, of which we measure the neuronal activity with EEG. The cortex consists of four lobes which are roughly related to planning and motivation, integration of sensory information, sound and verbal memory, and sight. Control interfaceThe control interface translates the logical control signal to a semantic control signal: something meaningful in terms of the application. ElectroencephalographyThe recording of voltage changes along the scalp. These changes are the result from activity of groups of neurons in the cortex.Emotiv EPOC A commercially available, wireless head set for measuring EEG that is easy to use. It comes with 14 electrodes (plus CMS and DRL) at a 128Hz sampling rate. Event-related potential A brain response that occurs related to some specific event (stimulus).Feature extraction After pre-processing, the purpose of feature extraction is to magnify those characteristics that are most distinctive for mental state detection, and to suppress or remove the rest.Feature translation When the most distinctive features have been derived, they can be translated to some logical meaning, which expresses what mental activity has been detected. This feature translation is commonly achieved through regression or classification.Feedback Information from the system to the user. Feedback can inform the user on changes in the application state, but also on what user input the BCI detected.Ground truth A label indicating the actual class of a certain data sample. This allows us to either train a classifier to label such samples as that class, or to determine the performance of a classifier by comparing the classification results with this known ground truth. Human-computer interactionThe in...