Biomedical Engineering Reference
In-Depth Information
online feedback training, allowing for the initial adaptation of both the human brain
and the BCI algorithm. For phase 2, the recorded data from phase 1 were employed
to optimize the feature extraction and to refine the classifier parameters for each
individual, aiming at a better BCI algorithm through refined machine learning. For
the real testing phase, phase 3, three-class online control was achieved by coupling
the trained brain and optimized BCI algorithm.
8.3.3.2 Phase 1: Simple Classifier for Brain and Computer Online Adaptation
Figure 8.10 shows the paradigm of online BCI training with visual feedback. The
“left hand,” “right hand,” and “foot” movement imaginings were designated to
control three directional movements: left, right, and upward, respectively. The sub-
ject sat comfortably in an armchair, opposite a computer screen that displayed the
visual feedback. The duration of each trial was 8 seconds. During the first 2 seconds,
while the screen was blank, the subject was in relaxing state. At second 2, a visual
cue (arrow) was presented on the screen, indicating the imagery task to be
performed.
The arrow pointing left, right, and upward indicated the task of imagination of
left-hand, right-hand, and foot movement, respectively. At second 3, three progress
bars with different colors started to increase simultaneously from three different
directions. The value of each bar was determined by the accumulated classification
results from a linear discriminant analysis (LDA), and it was updated every 125 ms.
For example, if the current classification result is “foot,” then the “up” bar will
increase one step and the values of the other two bars will be retained. At second 8, a
true or false mark appeared to indicate the final result of the trial through calculat-
ing the maximum value of the three progress bars, and the subject was asked to relax
and wait for the next task. The experiment consisted of two or four sessions and
each session consisted of 90 trials (30 trials per class). The dataset comprising 360 or
180 trials (120 or 60 trials per class) was used for further offline analysis.
Feedback
Cue (arrow)
Movement imagination
Relax
0
1
2
3
4
5
6
7
8s
Foot
Left
Right
Figure 8.10
Paradigm of three-class online BCI training with visual feedback.
Search WWH ::
Custom Search