Brain-computer interface (BCI) is a direct communication pathway that decodes brain signals directly into actions. It is intended for restoring the abilities to move and communicate for individuals who have become paralyzed. In electroencephalography (EEG)-based BCI, signals on the surface of the brain are analyzed and decoded in real-time into computer commands. However, the signals are weak and noisy. Current BCIs are thus limited in speed, accuracy, and number of useful applications.
By using novel data-driven, statistical machine learning algorithms, we developed a high-performance BCI framework that is intuitive, robust, and require short training time. Specifically, we developed a BCI-controlled spelling device and a BCI based on intuitive motor imageries.
The data-driven techniques eliminate traditional ad-hoc methods of selecting channels and frequencies. Instead, the dimension reduction (classwise principal component analysis) and discriminant feature extraction (approximate information discriminant analysis) algorithms work on all data to identify the most salient features. This means that BCI users are free to use intuitive mental imageries, such as the imagery of walking to control ambulation, thereby cutting down the training time from 3-4 weeks to 10-15 minutes.
6 able-bodied (AB) individuals participated in using the BCI-Speller and all achieved spelling speed at least 0.5 bits/s faster than 5 other studies found in the literature. The fastest subject was able to achieve 3 bits/s, almost 3 times faster than the next highest performing system.
For the Motor BCI, 8 AB and 6 subjects with paraplegia due to spinal cord injury (SCI) used BCI to control a virtual reality walking simulator. In 10-15 minutes training, almost all subjects were able to attain purposeful control and were able to complete a goal-oriented test with high scores. In addition, 1 SCI subject also used BCI to control a robotic gait orthosis to walk on a treadmill.
These achievements show that the data-driven approach is an important component of BCI. With further research, BCI may eventually become widely accepted in neurorehabilitation and in restoring independence to and lower the medical and societal costs of paralyzed individuals.