Biomedical Engineering Reference
In-Depth Information
I
II
III
IV
V
100
μ
30 min
7th min
127th min
157th min
37th min
187th min
67th min
97th min
217th min
100
μ
2s
Figure 3.28
A 4-hour EEG recording in a rat brain injury experiment. Five regions (I-V) correspond
to different phases of the experiment. I: baseline (20 minutes); II: asphyxia (5 minutes); III: silent
phase after asphyxia (15 minutes); IV: early recovery (90 minutes); and V: late recovery (110 min-
utes). The high-amplitude signal preceding period III is an artifact due to cardiopulmonary resuscita-
tion manipulations. The lower panel details waveforms at the indicated time, 10 seconds each, from
the EEG recording above. (
From:
[97]. © 2003 Biomedical Engineering Society. Reprinted with
permission.)
cal EEG waveforms following a hypoxic-ischemic brain injury. Taking the
amplitudes in the time domain, we demonstrate how to estimate the entropy from
an raw EEG data
s
(
n
), where
n
1, ...,
N
, which could be easily extended to fre-
quency- and time-frequency domains. The probability distribution, {
P
i
} in (3.61),
(3.64), and (3.66), can be estimated simply by a normalized histogram or more
accurate kernel functions.
=
3.3.2.1 Histogram-Based Probability Distribution
A histogram is the simplest way to obtain the approximate probability distribution.
The range of EEG signals is usually equally divided into
M
interconnected and
nonoverlapping intervals, and the probability {
P
i
} of the
i
th bin (
I
i
) is simply defined
as the ratio of the number of samples falling into
I
i
to the length of the signal
N
:
()
NI
N
i
P
=
,
for
i
=
1
,
,
M
(3.67)
i
This histogram-based method is simple and easy for computer processing. The
distribution {
P
i
} is strongly dependent on the number of bins and the partitioning
approaches.
Search WWH ::
Custom Search