Biomedical Engineering Reference
In-Depth Information
At the center of motor BMI technology is the ability to infer the user's intent directly from
the CNS with sufficient resolution and accuracy to seamlessly control an external robotic device.
A conceptual drawing of a motor BMI is shown in Figure
1.6
where neural activity is recorded,
conditioned, transmitted (wired or wirelessly), and translated directly into control commands of a
prosthetic arm or cursor of a computer. As described earlier, variety of signal processing techniques
have been used to analyze the control commands collected from EEG [
70
], EcOG [
35
], LFPs [
71
],
and ensembles of single neurons (SUA) [
72
].
There are two basic types of motor BMIs: command and trajectory control BMIs. Research
in command BMIs started in the 1970s with EEG [
73
] where subjects learned (and practice ex-
tensively) to control their regional brain activity in a predetermined fashion so that it could be ro-
bustly classified by a pattern recognition algorithm and converted into a set of discrete commands.
Often, EEG-based interfaces use evoked potentials that study the impulse response of the brain as
a dynamic system [
74-76
]. Event-related potentials are very important as clinical tools, but they
are based on repeatability across stimuli, often abstract imagery, and may not be the most natural
interface for real-time motor control. Therefore, the paradigm for communication is mostly based
on selection of cursor actions (up/down, left/right) on a computer display. The computer presents
to the users a set of possibilities, and they choose one of them through the cursor movement, until
FIgURE 1.6:
Motor BMIs derive intent from brain signals to command and control devices such as
computers and prosthetics.
