Single-Channel EEG Analysis - Quantitative EEG Analysis Methods and Clinical Applications - page 101

which is determined by the statistical properties. Discussions in the literature cover
the estimation of q [114]; however, it is still not clear how to extract the value of q
from recorded raw data such as that found in an EEG. Capurro and colleagues
[115] found that q was able to enhance the spiky components in the EEG; that is, a
larger q will result in a better signal (spike) to noise (background slow waves) ratio.
For the same EEG signal in Figure 3.31(a), Figures 3.36(b-d) show TDE changes
under different entropic indexes ( q 1.5, 3.0, and 5.0). Regardless of the scale of the

TDE, we can still find the change of comparison between the “spike” and back-
ground “slow waves.” Therefore, by tuning the value of q , we are able to make the
TDE focus on “slow waves” (smaller q ) or “spiky components” (larger q ). Empiri-
cally, we recommend a medium value of q

=

3.0 in the study of EEG signals follow-
ing hypoxic-ischemic brain injury when both slow wave and spiky activities are
present; whereas for the spontaneous EEG signal, smaller entropic index (e.g., q

=

=

1.5) or Shannon entropy is suggested.

3.3.3.5 Quantitative Analysis of the Spike Detection Performance of Tsallis Entropy
To quantify the performance of Tsallis entropy in “spike detection,” we introduce a
measure called spike gain improvement (SGI):

MP
S

−

sig v

SGI = (3.74)

sig

10

0

EEG following brain injury

− 10

0 2000 4000 6000 8000 10000

(a)

1.5

1

Tsallis TDE (q=1.5)

0.5

2000

0 2000 4000 6000 8000 10000

(b)

0.4

Tsallis TDE (q=3.0)

0.2

0 2000 4000 6000 8000 10000

(c)

0.25

Tsallis TDE (q=5.0)

0.2

0 2000 4000 6000 8000 10000

(d)

Figure 3.36 The role of nonextensive parameter q in TDE: (a) 40-second EEG segment selected
from the recovery of brain asphyxia, which includes three typical bursts in recovery phase; and (b-d)
TDE plots for different nonextensive parameter ( q = 1.5, 3.0, and 5.0). The size of the sliding window
is fixed at w = 128. The sliding step is one sample ( Δ = 1). Partition number M = 10.