Biomedical Engineering Reference
In-Depth Information
directional movements, that is, left-hand, right-hand, and foot imagery for moving
left, right, and forward, respectively.
Five right-handed volunteers (three males and two females, 22 to 27 years old)
participated in the study. They were chosen from the subjects who could success-
fully perform two-class online BCI control in our previous study [55]. The recording
was made using a BioSemi ActiveTwo EEG system. Thirty-two EEG channels were
measured at positions involving the primary motor area (M1) and the supplemen-
tary motor area (SMA) (see Figure 8.8). Signals were sampled at 256 Hz and
preprocessed by a 50-Hz notch filter to remove the power line interference, and a 4-
to 35-Hz bandpass filter to retain the EEG activity in the mu and beta bands.
Here we propose a three-phase approach to allow for better adaptation between
the brain and the computer algorithm. The detailed procedure is shown in Figure
8.9. For phase 1, a simple feature extraction and classification method was used for
Biosemi
EEG amplifier
Biosemi
EEG data server
Visual feedback
Figure 8.8
System configurations for an online BCI using the motor imagery paradigm. EEG signals
were recorded with electrodes over sensorimotor and surrounding areas. The amplified and digitized
EEGs were transmitted to a laptop computer, where the online BCI program translated it into screen
cursor movements for providing visual feedback for the subject.
1. Online
training
Bandpass
filtering
Data
interception
C3/C4
Power Feature
LDA
classifying
EEG
2. Offline
optimization
Parameter
optimization
CSP
training
LDA classifier
training
3. Online
control
Bandpass
filtering
Data
interception
Spatial
filtering
LDA
classifying
EEG
Figure 8.9
Flowchart of three-phase brain computer adaptation. The brain and BCI algorithm were
first coadapted in an initial training phase, then the BCI algorithm was optimized in the following
phase for better online control in the last phase.
Search WWH ::
Custom Search