Brain–Computer Interfaces - Index of

174 G.R. Müller-Putz et al.
channels each. They will be implanted to control simple basic movements such as
eating and grooming. Control signals will be obtained from implanted face and neck
electromyographic (EMG) sensors and from an additional position sensor placed on
the head of the user. Additionally, two external sensors are fixed at the forearm and
the upper arm to provide the actual position of the arm. A part of this concept was
already realized and implanted in nine arms in seven C5/C6 SCI individuals. The
study described in [9] showed that it is possible to control the neuroprostheses for
grasp-release function with EMG signals from strong, voluntarily activated muscles
with electrical stimulation in nearby muscles.
Rupp [28] presented a completely non invasive system to control grasp neuroprostheses
based on either surface or implanted electrodes. With this system,
it is possible to measure the voluntary EMG activity of very weak, partly paralysed
muscles that are directly involved in, but do not efficiently contribute to,
grasp function. Due to a special filter design, the system can detect nearby applied
stimulation pulses, remove the stimulation artefacts, and use the residual voluntary
EMG-activity to control the stimulation of the same muscle in the sense of “muscle
force amplification”.
2 Brain-Computer Interface for Control of Grasping
Neuroprostheses
To overcome the problems of limited degrees of freedom for control or controllers
that are not appropriate for daily activities outside the laboratory, Brain-Computer
Interfaces might provide an alternative control option in the future. The ideal solution
for voluntary control of a neuroprosthesis would be to directly record motor
commands from the scalp and transfer the converted control signals to the neuroprosthesis
itself, realizing a technical bypass around the interrupted nerve fiber tracts
in the spinal cord. A BCI in general is based on the measurement of the electrical
activity of the brain, in case of EEG in the range of μV [32]. In contrast, a neuroprosthesis
relies on the stimulation of nerves by electrical current pulses in the range
of up to 40 mA, which assumes an electrode-tissue resistance of 1 k�. One of the
challenges for combining these two methods is proving that it is possible to realize
an artefact free control system for neuroprosthesis with a BCI. In the last years, two
single case studies were performed by the Graz-Heidelberg group, achieving a one
degree of freedom control. The participating tetraplegic patients learned to operate
a self-paced 1-class (one mental state) BCI and thereby control a neuroprosthesis,
and hence control their grasp function [14, 15, 23].
The basic idea of a self-paced brain switch is shown in Fig. 2. In the beginning,
patients were trained to control the cue-based BCI with two types of motor imagery
(MI, here described in two features). Usually, a linear classifier (e.g., Fisher’s linear
discriminant analysis, LDA) fits a separation hyper plane in a way, e.g., to maximise
the distance of the means between the two classes (Fig. 2a). This requires analyzing
the classifier output time series, often as presented in Fig. 2b. One class (class 1)