Information Technology Reference
In-Depth Information
Table 3.2: Statistical feature - channel selection results
Band
Channel (step 2)
Channel (step 4)
Target
Delta (
δ
)
2
-
-
Theta (
θ
)
, 5
2
2
Alpha (
α
) , 5
5
5
Beta (
β
)
-
-
-
Gamma1
(
gamma
1
)
2, 5
2, 5
2, 5
Gamma2
(
gamma
2
)
2, 5
2, 5
2, 5
methodology with its second statistical feature selection scheme can efficiently iso-
late the channels and the correct band activations. Traditional FT spectral analysis
methods pose intrinsic limitations on encapsulating the time variation of the signal.
Beyond traditional spectral analysis, the WT enables the consideration of time spe-
cific significant regions as in Step 3 of the proposed methodology. WT is proved to
be a useful measure to detect time-varying spectral power and performs better than
traditional time-frequency methods in identifying activity, especially on a shorter
temporal scale in high frequencies, which could indicate neuronal synchronous
activation in some cortical regions. This is an advantage to previous methodolo-
gies, since high-frequency bands are weak and difficult to evaluate using spectral
methods.
A qualitative reasoning arising from the application of this methodology to ac-
tual data is discussed in [14], where the certain methodology is applied to a com-
plex mathematical reasoning task. Finally, the presented method reveals additional
signal characteristics, since it captures not only its average power but also the
time-localized activation of the signal.
3.5 Conclusion
The proposed algorithmic approach emphasizes the idea of selecting EEG features
based on their statistical significance and further supports the use of time-scale WT
domain in order to select significant EEG segments capable of describing the most
prominent task-related changes.
Results suggest that the proposed methodology is capable of identifying regions
of increased activity during the specified target task. The entire process is automated
in the sense that different feature types can be adaptively (according to the data pro-
file) extracted and further refined in a way “transparent” to the user. Such processes
may be transferred to a clinical environment if the methods prove to be valuable
for the diagnosis of certain pathologies by comparing any routine EEG against a
database of pathological ones.
