Brain-Computer Interfacing

Many studies over the past two decades have shown that people and animals can use brain signals to convey their intent to a computer using brain-computer interfaces (BCIs). BCI systems measure specific features of brain activity and translate them into control signals that drive an output. The sensor modalities that have most commonly been used in BCI studies have been electroencephalographic (EEG) recordings from the scalp and single-neuron recordings from within the cortex.

Over the past decade, our laboratory has pioneered the use of electrocorticographic (ECoG) activity recorded directly from the surface of the brain. ECoG has attracted substantial and increasing interest, because it has been shown to reflect specific details of actual and imagined actions, and because its technical characteristics should readily support robust and chronic implementations of BCI systems in humans. Specifically, compared to signals acquired from the scalp (electroencephalography (EEG)) and intraparenchymal single neuronal recordings, ECoG recordings have characteristics that make them especially suited for basic neuroscience research and resulting translational opportunities. These characteristics include high spatial resolution and signal fidelity, resistance to noise, and substantial robustness over long recording periods. Thus, ECoG recordings appear to strike an ideal balance between fidelity and clinical practicality. Over the past decade, we have been exploring the promise of ECoG recordings for BCI applications in a series of collaborative studies.

Leuthardt et al. 1 was the first report of online ECoG-based BCI operation. In four subjects, it used different actual or imagined motor actions to chose the ECoG features to be used for online control of one-dimensional cursor movement to a target located at the bottom or top of a computer screen. Over brief training periods of only 3-24 min, and using features associated with different actual or imagined actions, the four subjects achieved online success rates of 74-100% (with 50 percent expected by chance). Furthermore, offline analyses of data gathered from the same subjects while they were using a joystick to control two-dimensional cursor movement indicated, for the first time in humans, that ECoG features at frequencies up to 180 Hz encode substantial information about both dimensions of movement.

The online one-dimensional BCI control reported in this initial report was confirmed and extended by several other studies using similar experimental protocols. Wilson et al. 2 reported comparable control using closer electrode spacing (i.e., 5 mm) spacing and ECoG features associated with sensory (rather than movement) imagery. Van Steensel et al. 3 showed that ECoG recorded over left dorsolateral prefrontal cortex, an area involved in working memory, can also support rapid acquisition of movement control. Miller et al. 4 showed that motor imagery-based BCI control using locations in motor cortex can produce ECoG changes that exceed those produced by actual movements. Furthermore, in a study of potential importance for the development of practical long-term ECoG-based BCIs, Leuthardt et al. 5 found that an epidurally-placed electrode could also support effective control (i.e., 100% accuracy using an electrode placed over premotor cortex).

Going beyond one-dimensional control, Schalk et al., 2008 showed in the first and to date only two-dimensional ECoG BCI study 6 that an ECoG-based BCI allowed five subjects to use imagined or actual motor actions to control a computer cursor in two dimensions. Over a brief training period of 12-36 min, each subject acquired substantial control of particular ECoG features recorded from several electrodes in a single array over one hemisphere. These features supported success rates of 53-73% in a two-dimensional four-target center-out task in which chance accuracy was 25%. Acquisition of comparable levels of two-dimensional control using EEG typically requires substantially more training.

In a study of item selection (rather than movement control), Brunner et al. 7 tested in one subject an ECoG-based matrix speller comparable to that developed for use with EEG. The subject achieved spelling rates (i.e., 17 characters/min (69 bits/min) sustained, 22 characters/min (113 bits/min) peak) several times higher than those typically reported for EEG using the same approach.

In the first ECoG BCI study that used the language network, Leuthardt et al. 8 showed that ECoG allows for accurate discrimination of different overt and imagined phoneme articulations (68%-91% accuracy in a binary task) with less than 15 minutes of training. Of importance to eventual clinical implementation of ECoG-based BCI systems, in one of the subjects, these results were achieved using recordings from a microarray consisting of 1 mm spaced microwires.

In summary, the ECoG-based BCI studies in our and in other laboratories to date strongly encourage further research using the ECoG platform for brain-computer interfacing.

You are here

Home » Research » Brain-Computer Interfacing