Brain-computer interfaces function by gathering data on biophysiological activity in the brain, which is then analyzed by software to extract and classify data features. Several different imaging modalities are available for signal acquisition; broadly, these can be grouped into two categories, which are invasive (requiring surgery) and non-invasive techniques.
Invasive techniques include electrocorticography and intracortical neuron recording. Both approaches require a craniotomy to place electrodes directly on the surface of the brain, or inside the gray matter of the brain. Although invasive modalities provide a high degree of spatial and temporal resolution, their use in humans has been very limited due to the practical concerns of health and safety.
A variety of non-invasive approaches are also capable of acquiring useful biophysiological signals from the brain. These include electroencephalography (EEG), magnetoencephalography (MEG), functional magnetic resonance imaging (fMRI) and near-infrared spectroscopy (NIRS).
EEG measures electric activity in the brain caused by the flow of electric currents through neurons. MEG uses magnetic induction to monitor the brain’s magnetic activity. fMRI detects changes in blood flow and oxygenation levels during heightened periods of neural activity. NIRS uses infrared light to detect fluctuations in cerebral metabolism.
Of the four non-invasive modalities, EEG is by far the most popular approach. This is due to EEG’s relatively low cost, high-portability and ease-of-use. EEG is capable of acquiring a variety of signals from the brain, including the frequency bands delta (δ), theta (θ), alpha (α), beta (β) and gamma (γ). Although these wave frequencies bear some correlation to general states of brain activity, they have not proven useful as mechanisms for direct control. For that reason, most BCIs employ a different group of brain signals that have demonstrated higher utility as control signals.