Information Technology Reference
In-Depth Information
In this work we attempt to retrieve additional information (as compared to tra-
ditional spectral analysis methods) by making use of the time profile of the EEG
signal during the target task under study. The motivation for this work stems from
the fact that the WT method is able to extract not only the spectral activations but
also the time segments at which they occur. It constitutes the cornerstone of our
feature extraction scheme and is used for analyzing task-related or control EEG
signals by effectively capturing the power spectrum (PS) of each frequency band
and channel. In particular, it encodes the activation differences between the mental
states of interest. The subsequent feature selection steps apply test statistics on the
extracted “time-averaged” PS features. In addition, our approach introduces an extra
refinement step that makes further use of the time profile provided by the WT as to
derive and encode the temporally activated brain regions and bands. The proposed
EEG feature extraction and selection method may also be applied to other similar
nonstationary biological signal analysis problems.
3.2 Methods
3.2.1 Methodology Overview
Two different cognitive tasks are assumed for simplicity: the control and the tar-
get ones that involve a modulated rather than random activity. The latter task en-
capsulates the crucial information for extracting both the frequency bands and the
location of brain activity, in terms of channel references or groups of channels (re-
lated to specific brain areas) as an index of cerebral engagement in certain men-
tal tasks. The testing hypothesis suggests that the target task induces activity on
certain brain lobes, reflected on the associated electrodes in a way significantly
different compared to a control task. The WT constitutes the cornerstone of fea-
ture extraction and is used in analyzing task-related or control EEG signals by
effectively capturing the power spectrum (PS) of each band and channel, partic-
ularly encoding the activation differences between the tasks. From the technical
point of view, statistics is used to extract and select salient features, testing for
significance in both the time and scale domains of the signal. The feature selec-
tion steps apply test statistics on the extracted “time-averaged” PS features, but
in addition our approach introduces an extra refinement step that makes further
use of the time profile of the WT, as to derive and encode the temporally acti-
vated brain regions and bands. Test statistics form appropriate means for the de-
sign of feature selection criteria strictly based on statistical significance; they are
simple to implement and often perform better than other heuristic selection meth-
ods. To that respect, we base our selection on statistical tests that rely on statistical
properties of the feature data under consideration. Hopefully the identified chan-
nels and lobes may elucidate any neurophysiological pathways involved in brain
function.
