Information Technology Reference
In-Depth Information
time periods during which features were resolved, namely, the epileptiform, PLED,
and triphasic waves. It can be seen that the ICA results remained unchanged un-
der various data lengths where the same CJD-related patterns repeatedly appeared.
Specifically, the PLED presents in the first 20 s within the first minute and in the
17th-35th seconds in the fifth minute of the epoch (see the yellow bars in the 1st
and 5th windows in (a), 1st and 4th windows in (b), 1st and 2nd windows in (c),
1st and 2nd windows in (d) and in (e)). The epileptiform were detected within the
3rd window in (a), 2nd and 3rd window in (b), 2nd and 3rd window in (c), 1st and
2nd window in (d) and in (e) (see orange bars). Finally, the triphasic waves can be
observed across from the 2nd to the 5th windows in (a), which also appeared in the
1st-4th windows in (b), 1st and 2nd windows in (c), 1st and 2nd windows in (d)
and in (e) (see green bars). It should be noted that not only the temporal features
preserved the same waveforms and durations, but also the three corresponding spa-
tial maps remained resemble (see Fig. 4.7). Similar results have been obtained from
other patients (not shown).
4.4.3 Feature Extraction by PCA
It has been reported that the use of ICA under the assumption of source indepen-
dence can separate more realistically neurophysiologic signals in comparison with
the principal component analysis (PCA) [10, 12]. Since the EEG signals induced by
eyeblinking or contaminated by electrical noise usually present far larger variances
than physiological signals, the covariance-based PCA decomposing procedure is in-
ferior to ICA for resolving meaningful brain activities. As shown in the Fig. 4.8b
where the same time window in Fig. 4.4a was selected, the temporal waveforms
of the first four principal components (eigenvectors corresponding to the first four
largest eigenvalues) merely exhibit the preservation of the most power of the origi-
nal signals. None of them extracted the evident eyeblinking artifacts or CJD-related
features from the raw EEG as compared to the ICA results in Fig. 4.4.
4.5 Discussions
This study aims to extract the CJD-related waveforms in conjunction with the spa-
tial dominances from the EEG recordings for the early diagnosis of CJD. Our re-
sults demonstrate that ICA is an effective tool for distinguishing FIRDA, PLEDs
and PSWCs from EEG recordings in the early stage of CJD (Figs. 4.2d, e, 4.3d, e,
4.4d, e, and 4.5) with dominance in each corresponding spatial map being revealed.
In comparison with the raw EEG data in the shaded areas in Figs. 4.2b, 4.3b, and
4.4b, where the CJD-related waveforms were severely smeared by the large poten-
tials of eye movements, three PLEDs, four triphasic waves, and two epileptiforms
can be evidently recovered in the shaded areas of IC3 in Fig. 4.2d, IC4 in Fig. 4.3d,
