Brain–Computer Interfaces - Index of

74 C. Neuper and G. Pfurtscheller
EEG EEG features/classification
feedback
signal
closed
loop system
visual feedback/cont. signal
visual feedback/BCI operation
transfomation algorithm
application interface
Fig. 4 Basic components of the BCI paradigm during training (upper loop) and during controlling
a device (application, lower loop). The closed loop system indicated by black lines corresponds to
an operant conditioning neurofeedback paradigm. BCIs also have other components, including
feature selection and classification, an output mechanism such as controlling continuous cursor
movement on the screen, and an additional transform algorithm to convert the classifier output to
a suitable control signal for device control
in the form of the visual feedback. In the latter case, moving distractions such as a
moving hand can interfere with the imagery task.
Unlike the direct visualization of the relevant EEG parameters in classical neurofeedback
paradigms, the use of a classifier entails controlling e.g. continuous cursor
movement on the screen according to the outcome of a classification procedure
(such as a time-varying distance function in the case of a linear discriminant analysis
(LDA)). This procedure provides the user with information about the separability of
the respective brain patterns, rather than representing a direct analogue of the brain
response.
Adaptive classification methods [63, 64] can add another challenge to improving
BCI control. Although such adaptive algorithms are intended to automatically
optimize control, they create a kind of moving target for self-regulation of brain patterns.
The neural activity pattern that worked at one time may become less effective
over time, and hence users may need to learn new patterns. Similarly, the translation
algorithm (application interface), which converts the classifier output into control
parameters to operate the specific device, introduces an additional processing stage.
This may further complicate the relationship between neural activity and the final
output control. The complex transformation of neural activity to output parameters
may make it hard to learn to control neural activity. This explains why some
BCI studies report unsuccessful learning [62]. For further discussion of these issues,
see [65].