003 1-3998/90/2802-0106$02.00/0
PEDIATRIC RESEARCH
Copyright 0 1990 International Pediatric Research Foundation. Inc.
Vol. 28. No. 2. 1990
Printc~drn U.S.A.
Automated Identification and Quantitation of
Four Patterns of Electrocortical Activity in the
Near-Term Fetal Lamb
MARY ELLEN McNERNEY AND HAZEL H. SZETO
Drparrmc~ntof Pl~armacolr;y?*,Cornell University Medical College. Nc\.cr York. New York 10021
ABSTRACT. Most previous efforts to characterize electrocortical states in the unanesthetized fetal lamb have
relied on descriptive measures of amplitude and frequency;
those that used quantitative parameters did not draw upon
data from a cohort of subjects for contiguous time segments. We have used Fourier analysis to quantitate amplitude and frequency characteristics of records of 2-3 h for
a group of fetuses. Four electrocortical states, including
two types of transitional activity, have been identified and
quantitated. Additionally, we have developed a method that
automates the assignment of data to these four electrocortical states. This has enabled us to monitor transitions
among states and indicates that epochs that appear homogeneous by descriptive measures may actually encompass several patterns of electrocortical activity. (Pediatr
Res 28: 106-1 10,1990)
Abbreviations
ECoG, electrocorticogram
HVSA, high-voltage, slow activity
LVFA, low-voltage, fast activity
FFT. Fast Fourier Transform
The fetal lamb begins to exhibit cyclic alternations among
discrete patterns of electrocortical activity at approximately 115
d (term = 145 d). Two predominant patterns, or states, have
been identified. These are variously termed HVSA and LVFA
(I), or simply, high-voltage electrocortical activity and low-voltage electrocortical activity (2). The properties of each state may
be defined by quantitation of both amplitude and frequency
components of the electrocortical waveform. Initial attempts to
characterize the developing ECoG relied upon descriptions of its
amplitude as obtained from visual analysis of polygraphic records. There were two problems with this approach. First, visual
analysis is hindered by subjectivity. Second, records were generally transcribed at low polygraphic paper speeds; this does not
permit the assessment of component frequencies. More recently,
attempts have been made to quantitate frequency characteristics
of the ECoG in the fetal lamb; these have focused on the two
preponderant states. The
has been used in our own (3) and
other (4, 5) laboratories for this purpose; this algorithm permits
Received January 25. 1990: accepted April 10, 1990.
Correspondence and reprint requests: Hazel H. Szeto. M.D., Ph.D.. Department
of Pharmacology. Cornell University Medical College. 1300 York Avenue, New
York. NY 1002 I .
Supponed in pan by the National Institute on Drug Abuse (DA02475-09).
M.E.M. is a recipient of a predoctoral fellowship from the Pharmaceutical Manufacturers' Association Foundation. H.H.S. is the recipient of a Research Scientist
Development Award from the National Institute on Drug Abuse (K02-DA0010006). ADAMHA.
the quantitation of the contribution to amplitude of each constituent frequency in a bioelectric signal.
Only the report from our laboratory has quantitated the spectral characteristics for a cohort of fetuses, however (3). Technical
restrictions at that time on data sampling abilities limited analysis
to a small number of temporally isolated segments of the ECoG
signal, which were selected by preliminary visual analysis. We
have recently developed the capability to perform continuous
off-line FFT for records of any length; the method is limited only
by data storage capacity. Additionally, we have devised a method
for automation of electrocortical state assignment that is based
on the amplitude and frequency information derived from the
FFT. We have examined continuous data segments of 2 h or
longer. This has enabled us to produce detailed analyses of the
frequency characteristics of HVSA and LVFA, as well as those
of transitional epochs. Moreover, it has enabled us to examine
the cycling from one state to another.
Electrocortical states, both preponderant and transitional, are
identified and characterized as a continuum whose extremes are
HVSA and LVFA. We propose that signals whose amplitude and
frequency properties are intermediate between those of HVSA
and LVFA are broadly divided into two additional states. We
have found that the latter are observed within epochs identified
visually as HVSA and LVFA, as well as during brief periods of
transition between HVSA and LVFA.
MATERIALS AND METHODS
Animal preparation. Seven fetal lambs were surgically instrumented for chronic intrauterine recording of electrocortical activity between 121 and 125 d ofgestation, in accord with guidelines approved by the institution for the care and use of the
animals. Details of the surgical procedure have been described
previously (6). Briefly, four stainless steel screws (size 0-80) were
implanted in the left and right parietal bones (approximately 8
and 25 m m distal to the coronal suture, and 15 mm lateral to
the sagittal suture) for recording ECoG activity. Leads were
tunnelled s.c. to the maternal flank and stored in a pouch.
Recording procedures. Ewes were allowed at least 72 h after
surgery before COmn~encementof fetal monitoring. The range of
gestational ages studied is 124-1 37 d; this span represents recordings obtained 3-1 3 d ~ o s t o ~ e r a t i v ePolygraphic
l~.
recordings
were carried out with the ewe standing or lying quietly in a cart.
,411 recordings, which were 2-3 h in length, were obtained at the
same time of day (0900 through 1200 h). The ewe had access to
food and water throughout the recording period.
acquisition and processing' The ECoG was recorded On
a model 2800s (Gould Cow., Cleveland, OH) analogue recorder
after amplification to about 250 mV with a bandpass filter of I100 HZ. The amplified, filtered analogue signal was recorded
cOncurrent~yonto FM tape ( x R - ~10; TEAC carp., ~ ~ ~ ~
CA) for Storage and off-line analysis.
Power spectral analysis. Analogue-to-digital conversion was,
106
!
107
ANALYSIS OF FOUR FETAL ECoG STATES
accomplished with a board (DT-2801A, Data Translation, Waltham, MA) resident in an AT microcomputer (IBM Instruments,
Inc., Danbury, CT). Data were continuously digitized off-line at
a rate of 256 Hz, and stored in the microcomputer in binary
format.
Raw data were first corrected for the presence of a linear trend
by subtracting from each data point the corresponding value of
the best-fit line estimated using all data points. The presence of
such a trend, if uncorrected, introduces artifactual power in the
low-frequency components (7, 8). After removal of the linear
trend, a Hamming window was applied. In the absence of a
suitable window, the FFT of a digitized series-such as the series
that results from our analog-to-digital conversion-would contain large artifacts in many frequencies of the spectrum (7, 8).
FFT' were performed on consecutive series of 1024 detrended,
windowed data points (4 s). The resolution of the FFT was '/4
Hz. The maximum detectable frequency was designated to be 64
Hz, although data for frequencies >32 Hz were discarded.
Five power spectra thus obtained were averaged to produce a
single mean power spectrum, or run. (Averaging of spectra is
necessary, as the error associated with each point in a single
power spectrum is 100%; the mean of n runs of data produces
an error term that is reduced to l/(n - 1)'" (7).) Mean power
spectra were written to a file.
Simultaneously, 50 and 90% spectral edges were calculated;
these are the frequencies below which 50 and 90% of the power
in a run reside. Spectral edges were also written to a file.
Subsequently,a number of spectral parameters were calculated
for each run. These included total power and percent power in
several EEG bandwidths, including those that are well-established markers in the adult: 1-4 Hz (delta), 5-7 Hz (theta), 8-10
Hz, 8- 13 Hz (alpha), 1 1-14 Hz and 15-32 Hz (beta).
Spectral parameter data from individual fetuses were pooled
within a given electrocortical state and mean spectral parameters
for each state were derived for the group. These were initially
evaluated for statistically significant differences between groups
with the Kolmogorov-Smirnov test for distributions (9). When
significant differences were observed, the nonparametric MannWhitney test (9, 10) was used to identify the source of the
deviation. Both Kolmogorov-Smirnov and Mann-Whitney tests
were implemented using the software package Systat V.4.0 (Systat, Inc., Evanston, IL).
The incidence of each state was calculated as:
Incidence = 100 x
Number of runs in state X
Total number of runs in record
Incidence data are reported for the group.
RESULTS
A total of 12 recordings were obtained from seven fetal lambs.
Fetal blood gases and pH were within normal limits on the day
of the study.
Figures 1, 2, and 3, as well as Table 2, are derived from a
single fetus; they are included to illustrate, in detail, some of the
preliminary results.
We have previously found that the 90% spectral edge is an
excellent index of electrocortical cycling in the fetal lamb (1 1).
Time series plots of this index were generated for each fetal
record. A representative polygraphic record, as well as the corresponding spectral edge time series, are shown in Figure 1 (top
and center panels).
When scrutinized in conjunction, the time series plot of the
90% spectral edge and the other spectral parameters listed above
provide a guide to the range of spectral parameters to be found
in stable episodes of HVSA and LVFA. We chose three parameters on which to base the automation of state assignments: total
power, percent power in the delta (1-4 Hz) bandwidth, and
percent power in the beta (15-32 Hz) bandwidth. The distinction
of HVSA and LVFA may be accomplished with any one of these
0
4
0
60
120
180
240
RUN NUMBER (TIME)
300
1
360
RUN NUMBER (TIME)
Fig. 1. Top panel: 120-min polygraphic recording of the ECoG of a
fetal lamb (gestational age 128 d). Full scale is approximately 250 ccV.
Center panel: time series plot of the 90% spectral edge for the interval
depicted by the polygraphic trace. Bottom panel: time series plot of the
state-to-state transitions for this interval.
criteria. Additionally, scrutiny of these parameters revealed that,
taken together, it was possible to distinguish different types of
transitional waveform; however, these could not be discriminated
solely on the basis of any single spectral parameter, whether
derived from amplitude (i.e. power) or frequency. Ultimately, it
was found that automated state assignment for all electrocortical
states could be accomplished by specification of numerical criteria for two of three parameters: 1 )total power within a given
run, 2) percent power in the 1-4 Hz bandwidth, and 3 ) percent
power in the 15-32 Hz bandwidth. Spectral parameters from
transitional signals, in particular, could be categorized into two
clusters when parameters 2 and 3 were considered in conjunction. We found that, in addition to the two preponderant electrocortical states, it is useful to consider the existence of two
transitional states.
Figures 2A and B depict two-dimensional histograms that
illustrate the clustering of spectral parameters according to electrocortical state for a single fetus. Figure 2A depicts total power
versus percent delta power; the criterion for total power that
distinguishes HVSA from a transitional state with similar percent
delta power characteristics is identified as 18 750. Similarly, in
Figure 2B (which illustrates percent delta power versus percent
beta power), it is possible to discern two major types of activity,
which correspond to the two preponderant states, as well as
intermediate patterns of activity, which we have broadly cate-
108
McNERNEY A N D SZETO
KIA
JpJ
I
I
r7-1-7
L A -1I
I
I
I
I
ri-1-1
E
W
8a
t i-k+-I-+-I-t
Q
LA-I-
4.
L 1 -1-
1I.
I
I
I
3- + + - I -
+
LA-I-!.
L_!-I-_!-I-I_I-L_iI
I 1 I
I
I
I
I
I
I
I
I
I
+ I - r 7 - 1 - t -1-7 1-T i - 1 - t
L 1-1-1
L A - C I - I - I - I - L J I
I
I
I
I
I
I
I
I
I
I
I
I
r i - r - r - I - T - I - r ir-7 - 1 - 1
L A - L A - I - I - I - L A LA-I- I
I
I
I
I
I
I
I
I
I
- Ir 7 I- 1 I- 7I'
I
k 4 -IC4-C4-I-+-I-C-I-
5
LL.
n
t-
z
W
0
w
W
a
+
L_!-I-'
I
I
I
I
-i-I-+-I-+-I-ti-kt-I-+-I-+I-t-*-I-+-
-l-l-i-r-L-l-L_r-i-I~I-I-I-LJ-LI-I-II
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
- 1 - I - T - I - T - I - P ~ - ~ ~ - I - T - I - ~ ~ - ~ ~ - I - T
25
50
75
100
TOTAL P O W E R / 5 0 0
B
0.--
--
IV
3000 -2000.
1000 .-
OF--
4
1
-
+
i
2
3
TIME IN SECONDS
PERCENT DELTA POWER
Fig. 2. Two-dimensional histograms (relative frequency of occurrcnce is reflected in the intensity of shading). ZA. total power es percent
power in the delta (1-4 Hz) waveband. .r c~ris:total power/500. v a-xis:
percent power in delta. 28. Percent power in the delta waveband vs
percent power in the beta (15-32 Hz) waveband. .v asis: percent power
in delta. I , (i.vis:percent power in beta. Solid 1ine.s in each figure indicate
criteria for state assignment.
gorized into two intermediate states. Careful visual scrutiny of
the raw data for every fetal record enabled an estimate of the
range of values for each parameter in each of the four states.
Automated assignment of runs to a given state was accomplished
by establishment of criteria (i.e. numerical cutoffs for each of the
three parameters listed above). The criteria used for this fetus
are: state I (HVSA); percent power in 1-4 Hz is >67.5% and
total power is > I 8 750; state 11 (transitional state "a"), percent
power in 1-4 Hz is >67.5% and total power is <18 750; state 111
(transitional state "b"). percent power in 1-4 Hz is <67.5% and
percent power in 15-32 Hz is <20%; and state IV (LVFA),
percent power in 1-4 Hz is <67.5% and percent power in 1532 Hz is >20%.
After the assignment of all runs in a fetal record to electrocortical states by these criteria, the states were compared to
determine whether they differed significantly. The KolmogorovSmirnov Test was used for the data from individual records to
compare states I and 11, states I1 and 111, and states 111 and IV
for each of three spectral parameters: total power, percent power
in 1-4 Hz, and percent power in 15-32 Hz. Cutoff criteria for
an individual fetal record were judged to be adequate if a) all
comparisons were statistically significant ( p < 0.05) and h) run-
FREQUENCY ( in Hz)
-
4
109
ANALYSIS O F FOUR FETAL ECoG STATES
Table 1. Cutoffcriteria for automated assignment offour
electrocortical states in fetal lamb
Criterion
73.18
+ SD
+ 2.41%
70-75
16.68
+ 3.16%
12-20
Mean
% Total power in 1-4 Hz
Range
for states I and 11
% Total power in 15-37
Hz for state IV
Powerinl-4Hzforstate
I (absolute)
18318+12311
5500-50000
Table 2. Spectral parameters oflour electrocortical states in
single fetal lamb (mean + SD)
% Total power
% Total power
State
Total vower
in 1-4 Hz
in 15-32 Hz
I (HVSA)
II
111
IV (LVFA)
27 336 + 5400
15 062 + 3495
8153 3092
4749 + 1200
85.04 + 3.34
76.9 4.0
56.1 + 10.6
35.57 + 6.26
3.43 + 1.15
6.05 2.05
14.69 + 3.65
29.08 + 5.44
+
+
+
Table 3. Spectral parameters offour electrocortical states in
population (mean f SEM)
% Power in band
Bandwidth
State I
State 11
State 111
State IV
Hz BANDWIDTH
Fig. 4. Power distribution among designated frequency bandwidths
for each of the four electrocortical states identitied.
by-run analyses of state assignments were in good agreement
with the visual record (see Fig. 1). The comparisons for the fetal
record depicted in Figure 2 were all significant at levels <0.00 1.
Runs representative of each of the four states during control
records are included here for illustrative purposes. Figure 3
depicts a 4-s interval of electrocortical signal for each state in
panel A and its associated power spectrum in panel B.
Criteria were individualized for each record, having been determined in the manner outlined above. Summary statistics of
the cutoff criteria for all fetal records are reproduced in Table 1.
Fig. 3. Representative signals ( A ) and Fourier transforms (B) for each
of the four electrocortical states identified. Top panels: (I). Secondpanels:
transitional state a (Ta or 11). Third panc%s: transitional state b (Tb or
111). R o ~ ~ opcrncl.\:
m
LVFA (IV).
0
4
8
12
16
20
FREQUENCY (in Hz)
24
32
28
Fig. 5. Average spectra derived for each of the four electrocortical
states identified. A. state I; A. state 11; 0.state 111; a,state IV.
Table 4. 90% s~ectraledae in states I-IV in ktal lamb
State
Spectral edge
(Mean SEM)
I (HVSA)
I1
111
IV (LVFA)
5.39 + 0.05
8.53 ? 0.10
13.28 0.1 1
16.68 0.09
+
+
+
(Note: The high degree of interindividual variability of the absolute power criterion precluded further use of absolute power
in population statistics.)
Once appropriate criteria had been selected for each fetal
record, files for individual records containing the relative power
data for 1-32 Hz, as well as spectral edge data, were analyzed by
the computer and new files containing the assigned runs for
states I, 11,111, and I V were created. The accuracy of automated
electrocortical staging was further verified visually, comparing
the assignment of runs to a state with polygraphic records for
that epoch of time.
The mean spectral parameters of each of the four states for
the prototypic fetus are reproduced in Table 2; these were calculated from all data recorded for a 2-h period.
Data from all fetal records were subsequently pooled to enable
the calculation of population statistics. Kolmogorov-Smirnov
comparisons similar to those canied out for individual records
were performed on pooled data; these confirmed the statistical
significance of differences between states I and 11, states I1 and
111, and states 111 and I V ( p < 0.001 in all cases). Mean population
data for relative power distribution for the four states are reported
in Table 3, and graphically depicted in Figure 4. Mann-Whitney
analyses reveal that the differences for delta (1-4 Hz) and beta
(15-32 Hz) parameters between states I and 11, states I1 and 111,
and states 111 and I V are highly statistically significant ( p <
0.0005).
Representative spectra for each of the four states are depicted
in Figure 5; these were derived by averaging data from all
animals. Standard errors are 4 % in all cases.
Average spectral edge data for each of the four electrocortical
states are outlined in Table 4.
The incidences of these states (mean f SEM) are as follows:
state I (HVSA), 31 f 7%; state 11, 8 f 0.7%; state 111, 15 2%;
and state I V (LVFA), 46 f 1.8%.
The bottom panel of Figure I outlines the time series of stateto-state transitions for the interval associated with the polygraphic trace and spectral edge plot.
+
DISCUSSION
A previous effort in our own laboratory to quantitate frequency
characteristics of the developing ECoG in the fetal lamb was
110
McNERNEY AND SZETO
hampered by the inability to process more than 4 s of contiguous
data (3). We have developed the capability for continuous data
digitization in an inexpensive, microcomputer-based system.
This permitted us to subject 12 2- to 3-h segments of ECoG
signal from seven late-term fetal lambs to Fourier analysis. This
type of extensive, quantitative analysis of developing electrocortical activity has not been previously reported. Furthermore, we
have automated the assignment of data segments to electrocortical states; this obviates the subjectivity that is associated with
visual assignment.
This approach has enabled us to identify and characterize four
electrocortical states in the range of gestational ages studied ( I 24137 d). Previous reports by other investigators who applied
Fourier analysis to the ECoG of the fetal lamb (4, 5) have been
limited to descriptions of the two preponderant states (HVSA
and LVFA) and have not detailed results for more than one
fetus. The four states that we have described represent a continuum in terms of amplitude and frequency. HVSA was designated
state I; approximately 85% of total power resides in the delta
band. Total power in this state is the greatest of the four states.
The major feature that distinguishes HVSA from state I1 is the
dramatic decline in total power in the latter; typically, it is only
33-40% of that in state I. Additionally, the power in delta is less
than that observed in HVSA and the power in remaining bands
is greater than that seen in HVSA.
The trends illustrated in state I1 are extended in state 111: there
is further reduction in total power (-20-25% that seen in state
I), further erosion of power in delta frequencies (-63%), and
enhancement of power in faster frequencies.
LVFA, designated state IV, has the least total power of the
four states; typically, this is -10% of that observed in state I.
Only 42% of total power resides in the delta band and there is a
substantial fraction (25%)observed in beta frequencies.
Data derived for 90% spectral edge for the four electrocortical
states exhibit a trend that is consistent with the frequency continuum observed. The value for state I was lowest. that for state
IV the greatest, and those for states I1 and I11 intermediate.
The incidences of states I and IV obtained by automated
analysis (3 1 and 46%, respectively) are somewhat lower than we
have previously reported for HVSA and LVFA (38 and 60%,
respectively) by visual analysis (6). However, when taken together, the total incidence of states I and I1 (39%); and states 111
and IV (61%) are virtually identical to our earlier findings.
Clearly, the automated method is capable of discriminating
spectral parameters that cannot be distinguished visually, although inspection of prototypic data (Fig. 3 and Table 2) reveals
that the differences among states I-IV are not particularly subtle.
States I1 and 111 were not exclusively observed during transitional epochs between HVSA and LVFA; they also occur within
intervals of states I and IV. It is not surprising that analysis of
state-to-state transitions reveals that epochs identified visually as
states I and IV are not as homogeneous as they appear. Rather,
there are routine excursions between states I and I1 and between
states 111 and IV. Furthermore, transitional epochs are not
sharply defined entities that occur between episodes of states I
and IV; in fact, they are most frequently interspersed with activity
that is quantitatively identified as state I or IV at the onset or
termination of one of these preponderant states. The overall
impression illustrated in Figure 1, bottom panel, is one of tem-
porary vacillation between states before stabilization in a new
preponderant state.
The quantitation of frequency and amplitude characteristics
for states I1 and 111 is important for three reasons. First, there
have been no attempts to quantitate the frequency and amplitude
characteristics of transitional electrocortical activity in the fetus.
Second, we report that waveforms of intermediate amplitude
and frequency are consistently found within longer epochs of
HVSA and LVFA; when records were visually analyzed, these
were overlooked. Finally, it is well known that a number of
pharmacologic and pathophysiologic perturbations in utero induce electrocortical states of intermediate amplitude and (previously) indeterminate frequency. The characterization of states I1
and 111 enables comparisons between physiologic conditions and
those established by such perturbations.
The significance of this work to the question of fetal sleep
staging cannot be determined definitively without concomitant
polygraphic records of postural muscle tone and eye movements
( 1 2); these recordings were not available for these studies. Nonetheless, our previous findings allow us to make a number of
inferences regarding states I and IV. State I (HVSA) seems to
represent the electrocortical component of the deeper planes of
quiet sleep. State IV probably corresponds to the presence of
arousal and/or rapid-eye-movement sleep; these two behavioral
states are both associated with LVFA, and can only be differentially distinguished by the presence of postural muscle tone and
rapid eye movements. Further investigation will attempt to corroborate these inferences, as well as identify the sleep stages that
correspond to states I1 and 111 in the normal developing fetus.
Acknowledgments. The authors thank Dr. Eugene Dwyer for
his assistance in software development. Additionally, we are
indebted to Dr. Antonio Sastre and especially to Dr. Paul Blank
for our education in the rigors of Fourier analysis.
REFERENCES
I
I . Clapp JF, Szeto HH, Abrams R, Larrow R, Mann LI 1980 Physiologic
variability and fetal electrocortical activity. Am J Obstet Gynecol 136:10451051
2. Dawes GS, Fox HE, Leduc BM, Liggins GC, Richards RT 1972 Respiratory
movementsand rapid eye movement sleep in the fetal lamb. J Physiol (Lond)
220:119-143
3. Szeto HH, Vo TD, Dwyer G, Dogramajian ME. Cox M, Senger G 1985 The
ontogeny of fetal lamb electrocortical activity: a power spectral analysis. Am
J Obstet Gynecol 153:462-468
4. Clewlow F, Dawes GS, Johnston BM, Walker DW 1983 Changes in breathing,
electrocortical and muscle activity in unanaesthetized fetal lambs with age.
J Physiol (Lond) 34 1 :463-476
5. Jost RG, Quilligan EJ, Yeh SY, Anderson GC 1972 Intrauterine electroencephalogram of the sheep fetus. Am J Obstet Gynecol 114:535-539
6. Szeto HH 1983 Effects of narcotic drugs on Rtal behavior activity: acute
methadone exposure. Am J Obstet Gynecol 146:21 1-2 17
7. Blackman RB, Tukey JW 1958 The Measurement of Power Spectra. Dover
Publications Inc, New York
8. Bloomfield P 1976 Fourier Analysis of Time Series: An Introduction. John
Wiley and Sons, New York
9. Daniel WB 1978 Biostatistics: A Foundation for Analysis in the Healtt
Sciences, 2nd ed. John Wiley and Sons, New York
10. Snedecor GW, Cochran WG 1967 Statistical Methods, 6th ed. The Iowa State
University Press, Ames, pp 130-1 32
I I. Szeto HH 1990 Spectral edge frequency as a simple quantitative measure o
the maturation of electrocortical activity. Pediatr Res 27:289-292
12. Szeto HH, Hinman DJ 1985 Prenatal development of sleepwake patterns ir
sheep. Sleep 8(4):347-355
I