ArticlePDF Available

Automated detection and elimination of periodic ECG artifacts in EEG using the energy interval histogram method


Abstract and Figures

An automated method for electrocardiogram (ECG)-artifact detection and elimination is proposed for application to a single-channel electroencephalogram (EEG) without a separate ECG channel for reference. The method is based on three characteristics of ECG artifacts: the spike-like property, the periodicity and the lack of correlation with the EEG. The method involves a two-step process: ECG artifact detection using the energy interval histogram (EIH) method and ECG artifact elimination using a modification of ensemble average subtraction. We applied a smoothed nonlinear energy operator to the contaminated EEG, which significantly emphasized the ECG artifacts compared with the background EEG. The EIH method was initially proposed to estimate the rate of false positives (FPs) and false negatives (FNs) that were necessary to determine the optimal threshold for the detection of the ECG artifact. As a postprocessing step, we used two types of threshold adjusting algorithms that were based on the periodicity of the ECG R-peaks. The technique was applied to four whole-night sleep EEG recordings from four subjects with severe obstructive sleep apnea syndrome, from which a total of 132878 heartbeats were monitored over 31.8 h. We found that ECG artifacts were successfully detected and eliminated with FP = 0.017 and FN = 0.074 for the epochs where the elimination process is necessarily required.
Content may be subject to copyright.
Automated Detection and Elimination of
Periodic ECG Artifacts in EEG Using
the Energy Interval Histogram Method
Hae-Jeong Park , Do-Un Jeong, and Kwang-Suk Park, Member, IEEE
Abstract—An automatedmethod for electrocardiogram (ECG)-
artifact detection and elimination is proposed for application to
a single-channel electroencephalogram (EEG) without a separate
ECG channel for reference. The method is based on three char-
acteristics of ECG artifacts: the spike-like property, the period-
icity and the lack of correlation with the EEG. The method in-
volves a two-step process: ECG artifact detection using the energy
interval histogram (EIH) method and ECG artifact elimination
using a modification of ensemble average subtraction. We applied
a smoothed nonlinear energy operator to the contaminated EEG,
which significantly emphasized the ECG artifacts compared with
the background EEG. The EIH method was initially proposed to
estimate the rate of false positives (FPs) and false negatives (FNs)
that were necessary to determine the optimal threshold for the de-
tection of the ECG artifact. As a postprocessing step, we used two
types of threshold adjusting algorithms that were based on the pe-
riodicity of the ECG R-peaks. The technique was applied to four
whole-night sleep EEG recordings from four subjects with severe
obstructive sleep apnea syndrome, from which a total of 132878
heartbeats were monitored over 31.8 h. We found that ECG arti-
facts were successfully detected and eliminated with FP
and FN
0.074 for the epochs where the elimination process is
necessarily required.
Index Terms—Electrocardiogram (ECG) artifacts, energy
interval histogram, ensemble average subtraction, nonlinear
energy operator.
HE need for ambulatory electroencephalographic moni-
toring has increased in both clinical practice and research,
in areas such as sleep/wake state or epilepsy monitoring [1], [2].
However, long-term recordings are vulnerable to various arti-
facts. In particular, cardiac activity may have pronounced ef-
fects on the electroencephalogram (EEG) because of its rela-
tively high electrical energy, especially upon the noncephalic
reference recordings of EEG.
Manuscript received March 20, 2001; revised July 2, 2002. This work was
the result of research activity of the Advanced Biometric Research Center sup-
ported by the Korea Science and Engineering Foundation. Asterisk indicates
corresponding author.
H.-J. Park was with the Advanced Biometric Research Center, Seoul Na-
tional University College of Medicine, Seoul, Korea. He is now with the De-
partment of Psychiatry, Harvard Medical School, Boston, MA 02115 USA.
D.-U. Jeong is with the Department of Psychiatry, Seoul National University
College of Medicine and the Clinical Research Institute, Seoul National Uni-
versity Hospital, Seoul 110-744, Korea.
K.-S. Park is with the Department of Biomedical Engineering, Seoul
National University College of Medicine, Seoul 110-744, Korea (e-mail:
Digital Object Identifier 10.1109/TBME.2002.805482
Algorithms have been proposed to eliminate electrocardio-
gram (ECG) artifacts from the EEG. Nakamura and Shibasaki
[3] proposed an ECG artifact elimination algorithm, which we
call the ensemble average subtraction (EAS) method, whereby
ECG-contaminated EEG series are synchronously segmented
with respect to the timing of consecutive ECG R-peaks. By
subtracting the ensemble average across EEG segments from
the contaminated EEG, the algorithm eliminates ECG artifacts.
EAS is based on the strict assumptions of homogeneity across
segments and Gaussian property of the EEG [3], [4].
Using a different concept, the independent component anal-
ysis (ICA) method was also applied to eliminate ECG artifacts
using multichannel signals [7]. Previously, we adopted adaptive
noise canceling theory [5] to eliminate such ECG artifacts using
a reference ECG channel [6].
It should be noted that these algorithms use consecutive
R-waves in a separate ECG channel as a reference, and there-
fore, cannot be applied when an ECG channel is not available.
Several ambulatory monitoring systems used for studying
sleep/wake states do not record ECG waveforms. Ambulatory
sleep/wake recordings use a reduced number of essential
channels, compared with the laboratory polysomnographic
units. EEG, electrooculogram (EOG), and chin electromyo-
gram (EMG) are necessary to assess the brain state, and nasal
airflow, respiratory effort, oxygen saturation, and heart rate
to monitor respiration and circulation. Recording heart rate is
frequently preferred to recording the ECG waveform in order
to reduce the data size when the ECG waveform is not a main
concern. Therefore, a new method of eliminating ECG artifacts
from the EEG is required when an ECG channel is unavailable.
In this paper, we propose an automated method for detecting
ECG artifacts in a single-channel EEG, and a method for elim-
inating them.
The proposed method for elimination of ECG artifacts in-
volves a two-step process: 1) ECG artifact detection using the
energy interval histogram (EIH) method, and 2) ECG artifact
elimination using a modification of the EAS method. In this
paper, we will focus primarily on the ECG artifact detection
method and then briefly deal with ECG artifact elimination.
A. Detection Procedure: Energy Interval Histogram Method
The smoothednonlinear energyoperator (SNEO) was used to
emphasize the ECG R-waves that corrupt the pure EEG signals,
0018-9294/02$17.00 © 2002 IEEE
and the EIH technique was developed to estimate the optimal
threshold using threshold-adjusting algorithms.
Step 1: Emphasizing the ECG Artifacts Using SNEO: The
SNEO [8], which uses the Teager–Kaiser Energy Operator
[9]–[11], is regarded as an efficient tool for detecting spike-like
signals because of its sensitivity to instantaneous changes in
frequency-dependent energy.
For a discrete-time series, the nonlinear energy operator
and SNEO can be defined as follows [8]:
is the convolution operator and is a smoothing
window function. SNEO is dependent on the square of both the
amplitude and the frequency of the signal, and shows high en-
ergy for a high-frequency spike. For a linear combination of
source signal
and spike artifact , i.e., ,
and are uncorrelated, the expected energy applying
is expressed by (3) (for detailed derivation, refer to
For spike-dominant positions,
while for nonspike positions, ,
. Using this property of SNEO, the problem
of detecting an ECG spike in the presence of the EEG back-
ground is reduced to finding an appropriate threshold
separates the spike regions from background signal regions by
the equation
Mukhopadhyay and Ray [8], defined the threshold of SNEO
as the mean energy multiplied by a scaling factor
as follows:
The scaling factor
is initially determined by experiment
and used as a constant thereafter. This threshold method cannot
be adapted to nonstationary situationswhere the spike energy of
the ECG artifacts in the EEG is variable and no precise knowl-
edge is available on the energy distributions of the spikes and
the background activities. Therefore, we developed a new auto-
mated threshold-selection and threshold-adjusting algorithm, as
described in the following steps.
Step 2: EIH for Estimating the Optimal Threshold: For de-
tection problems in general, the optimal threshold is determined
to minimize false negatives (FNs) while maintaining false pos-
itives (FPs) within a reasonably low limit [12]. In the present
application, FPs are more crucial than FNs, because the subse-
quent EAS procedure can be severely disrupted by false alarms.
Therefore, an optimal threshold should be chosen to minimize
FPs at a reasonable FN level. However, it is not easy to derive
the FP and FN rates and the corresponding optimal threshold
mathematically, because we have no exact knowledge on the a
priori probability density functions of the EEG and ECG arti-
fact energy. Therefore, we used a heuristic approach to estimate
FP and FN rates.
After detecting peaks from the smoothed signal energy (
we applied a series of thresholds (
) to these peaks (denoted as
Fig. 1. An illustration of energy intervals. Intervals between peaks were
calculated from the instantaneous energy distribution
(lower figure)
emphasized using the SNEO of contaminated EEG (upper figure). The circle
marks (
) in the upper figure and square marks ( ) in the lower figure indicate
the real R-peak positions on the ECG. Only peaks above the threshold
used for interval calculation. Neighboring peak intervals h3 and h4 fall within
normal heart beat interval range
while h1, h2 fall within half the normal
heart beat interval range
, and h5 falls within twice the normal heart
beat interval range
at a threshold .
), whereby was varied from the maximal value of
to its minimal value. Only the peaks higher than the resulting
threshold value of
were used to calculate a histogram of peak
intervals. We called this histogram the EIH and have denoted
it as
, i.e., as a function of the threshold . An example
of energy intervals is illustrated in Fig. 1, where an ECG-con-
taminated EEG and its SNEO energy are also displayed. The
energy peaks above the threshold
form a series of intervals
(i.e., h1–h5).
Histogram bins of
were divided into three ranges:
normal heart beat interval range
, twice the normal heart
beat interval range
, and half the normal heart beat
interval range
. These ranges are defined with respect
to the expected normal heart beat range
as follows:
In case of multiepoch signals,
was initially given with
an arbitrary normal range value for the first epoch and was esti-
mated for subsequent epochs by the mean heart beat interval of
the previous epochs.
In Fig. 1, the intervals h3 and h4 fall within
h1 and h2 fall within
, and h5 falls within , at the
The numbers of intervals that fall into the aforementioned
threerangesare denotedas
, ,and .
At a high value of
, most intervals of the peaks fall in .
decreases, the intervals fall more into and
increases while decreases. As is reduced further
toward the minimal value, most intervals fall within
is maximized. For practical purposes, we counted
the number of intervals directly within the three ranges of (5)
instead of calculating the intervals in smaller-sized bins. Fig. 2
illustrates the course of EIH according to the threshold, at (a)
Fig. 2. An illustration of the energy intervalhistogram according to the
thresholds: (a)
, (b) , (c) , and
. As the threshold decreases, increases while
decreases. At , the relative number of intervals within is the
same as that within
, i.e., .
is the threshold that maximizes , i.e., divided by the
total number of intervals.
, (b) , (c) , and
. In this figure,
The relative histograms
, , and
indicate the ratios of , ,
with respect to the number of total inter-
vals within all three intervals of (5), which is denoted by
is highly correlated with the number of missed peaks (an
estimate of FN, abbreviated as eFN) and
is highly
correlated with the number of false alarms (an estimate of FP,
abbreviated eFP).
implies true positives (eTP). Using the
estimated values for
, , and ,
we considered four criteria for selecting the optimal threshold
represents the threshold at which the false alarm
rate eFP and the miss rate eFN have the same value.
is a threshold chosen to minimize eFP while maintaining eFN
within a reasonable limit
(0.1 in this study).
is the threshold that maximizes the eTP. Fig. 3 shows the EIH
function derived using the procedures previously described
and shows the relative number of intervals according to the
Step 3: Postprocessing of Detection: STEP 3-1: Reducing
false alarms using the next-spike selection algorithm.
Since random high-frequency noise frequently disturbed the
detection of periodicECG spikes, we reducedthese false alarms
using the nonperiodic characteristics of random noise. When
multiple peaks were detected within 1.5 times the mean heart-
Fig. 3. A relative energy interval histogram
as a function of threshold
. are functions of the relative number of intervals that fall within
predetermined ranges at the threshold
. The dark solid line is
the dash-dotted line is
, and the dashed line is . These
functions were derived by normalizing
, , and
with respect to .
beat interval from the reference spike, the most
likely position of the following spike was considered to be the
position delayed by
from the reference spike, due to the
periodic characteristic of ECG spike trains. The peak nearest
the expected position was selected as a new reference. With the
exception of this selected peak, other peaks within
of the reference peak were regarded as false alarms. When no
spikes were found within
from the reference, a new
reference was selected using an interval mask ranging from two
tofour times
afterthe current peakwith amask gapof
as follows: , .By
shiftingthis maskto consecutive peaksand countingthe number
of peaks that fall within the mask, the peak with the highest co-
incident number was selected as a new reference on theassump-
tion of periodicity.
Fig.4 illustratesthis procedure.Thepeaks detectedby thresh-
olding are marked with plus signs (
). Within from
the reference
, the most likely ECG-peak position is
represented by the peak that is closest to the expected peak po-
sition, i.e.,
. In Fig. 4, shows the nearest
interval to
and is selected as a new spike
. can be calculated using the same
procedure with the reference being
STEP 3-2: Reducing misses using the threshold-adjusting
In order to redetect missed spikes, we applied a threshold-ad-
justing technique around the expected spike regions with a
threshold window. This window has a triangular shape with
a minimal peak value equal to a constant multiplied by the
Fig. 4. An illustration of the next-spike selection algorithm. In the energy plot
of ECG-contaminated EEG, all intervals from the reference
( , , , and ) are displayed with arrows. The plus ( )
and circle (
) signed peaks are spike candidates higher than the threshold; the
circles indicate true spikes. Within
from the reference ,
the most probable position is the peak nearest the expected peak position,
. Of the intervals, the peak delayed by
is nearest
the expected heart beat peak and is regarded as a new spike.
considered to be false alarms and rejected. The peak located at
will be recalculated with a new reference and will be rejected as a
false alarm.
Fig.5. The threshold-adjustingalgorithm.Whenaspike ismissed, a triangular
window with a minimum of
at the mid-way position between spikes is
applied in an effort to redetect the missed spike. The square marks (
) indicate
peaks detected by the thresholding method initially, while the circle mark (
indicates a missed spike, to be re-detected by lowering the threshold.
threshold in the expected region. This is shown in
Fig. 5.
STEP 3-3: Reducing false alarms by removing points neigh-
boring the expected beats.
During STEP 3-2, a lowered threshold may cause the re-de-
tection of artifacts and increase the number of FPs. Therefore,
re-application ofSTEP 3-1 is requiredto reduce these extra FPs.
Fig. 6is an illustrationof the resultof each detectionstep, and
was obtained by plotting heart beat intervals versus the detected
heartbeats. Beat intervals much shorter than the mean beat in-
can be regarded as false alarms while beat inter-
vals two or three times longer than
can be regarded as
indicating misses. After thresholding spike energies in STEP 2,
falsealarmsand missesoccurredwith
in this example [Fig. 6(a)]. The first postprocessing step STEP
3-1 removed the false alarms of STEP 2 using the next-spike
selection algorithm [Fig. 6(b)]. STEP 3-2 redetected the misses
using the threshold-adjusting algorithm [Fig. 6(c)], using a tri-
Fig. 6. Results at four detection steps. The estimated beats and the intervals
between neighboring beats are plotted on the
and axes, respectively. (a)
Shows peaks thresholded by an optimal threshold which is derived from EIH
in STEP 2. (b) Shows the resultant peaks after application of the next-spike
selection algorithm, which reduced false alarms in STEP 3-1. (c) Result of the
threshold-adjusting algorithm for reducing misses in STEP 3-2. The final result
of STEP 3-3, in which the (d) next-spike selection algorithm reduces the false
alarms caused by STEP 3-2.
angle-weighted threshold window when the heart beat intervals
were longerthantwo orthree times
.STEP 3-3 reducedthe
false alarms generated by STEP 3-2 to give a final FP of 0.006
and a final FN of 0.006 [Fig. 6(d)].
B. Elimination of ECG Artifacts: EAS
We adopted EAS [3] with a small modification. For ECG-
contaminated EEGs,
, where is the
original EEG and
is the ECG spike, the R-peaks of the
reference ECG can be used as triggering points for averaging.
All EEG signals were segmented onto the time range between
200 ms prior to the current triggering point and 200 ms prior
to the next triggering point. By averaging these segments, an
estimate of the ECG artifact waveform can be derived by the
following equation:
is the number of the segments and denotes the seg-
ment index. On the assumption that the EEG has a zero-mean
Gaussian distribution, the first term of the equation can be re-
duced tozero, leavingonly thesecond term.The remainder [i.e.,
] is an ensemble average and indi-
cates an estimation of the ECG artifact. By subtracting this en-
semble average from the contaminated EEG
, the original
EEG waveform
can be estimated using
.Forthe nonstationarycasewhen theECG waveformvaries
with time or there are multiepoch events, the previously cal-
culated ECG ensemble average can be added, with a weight
, in a new averaging process as shown by the following
Fig. 7. Detection and elimination of simulated ECG-artifacts.
Thisensemble averageacross EEGsegmentsiscomputed, as-
suming that only ECG artifacts and no true cerebral activity are
time-locked to the R-wave in the recorded EEG. In this paper,
the bias errors between the exact spike peaks and the detected
peaks cannot be disregarded, and therefore, the EAS algorithm
ofNakamuraand Shibasaki[3]abovewasmodified. Beforesub-
tracting the averaged ECG component from the EEG, we re-
aligned the averaged ECG artifact segment to the real EEG seg-
ment with a time delay. The time delay
was derived to make
the cross-correlation between both segments maximal and can
be described as follows:
for (9)
is the length of the averaged ECG waveform and is
the maximum time shift around the peak. The resultant EEG is
calculated using
Fig. 7 illustrates the complete process used for ECG artifact
detection and elimination in the simulated signals. Artifact-free
EEG signals [Fig. 7(a)] and the time-synchronized ECG arti-
facts [Fig. 7(b)] were added to generate the simulated EEG
ECG [Fig. 7(c)]. Fig. 7(d) shows the energy of the EEG ECG
derived from the SNEO. The estimated EEG (eEEG) using our
algorithm is shown in Fig. 7(e). Fig. 7(f) shows the estimated
ECG artifacts derived by subtracting the eEEG from the simu-
lated EEG
ECG (top) and the difference between the original
EEG and estimated EEG (bottom).
A. Evaluation Sets
In order to evaluate the performance of the artifact detection
and eliminationalgorithm, we acquired six8-h EEGs (C3-A2 or
O2-A1) during sleep from one normal subject and five subjects
with obstructive sleep apnea syndrome (OSAS). One OSAS
recording was used for determining the optimal detection pa-
rameters, one normal recording was used for evaluating ECG
elimination performance, and the other four recordings were
used for evaluating the overall performance of our method.
All recordings of the OSAS subjects contained ECG artifacts
in the EEG, but no ECG artifacts were present in the normal
recording. In all cases, the ECG was recorded simultaneously
with the EEG, as a reference. Both ECG and EEG signals were
sampled at a frequency of 250 Hz. R-peaks of the reference
ECG, detected using a general R-peak detection algorithm [13]
were used as a target for evaluating the detection performance.
B. Performance Indexes
Spike-to-Background Signal Energy Ratio (SBR): SBR was
defined to be a function of mean spike energy normalized with
respect to the background signal energy. In practice, we defined
SBR as the ratio of the mean energy of the spike region to that
of the background signal as follows:
Fig. 8. Detection performance versus SBR in OSAS recordings (total 3814 epochs and 132 878 heartbeats during 3814 30 s). (a) Illustrates mean FP and FN
in the epochs of the given SBR. (b) Mean FP and FN in the epochs having higher SBR than the given SBR. (c) Relative number of epochs having higher SBR than
the given SBR. SBRs higher than 20, whereby FP and FN reached 0.02 and 0.1, accounted for 60% of the total epochs.
where and indicate the instantaneous
signal energy of the spike region and of the nonspike region
in the
th segment, with segment sizes of and ,
respectively. Each segment is composed of samples located
around the R-peak and
is the number of total segments in
the epoch.
FN Ratio : FN was defined as the ratio of the number of
missed spikes to the number of actual spikes.
FP Ratio: FP was defined as the ratio of the number of false
alarm spikes to the number of actual spikes.
In the cases of both FP and FN, the R-peaks of the ECG ref-
erence were regarded as actual spikes.
C. Performance Evaluation
Bartlett, Hamming, and rectangular windows were tested as
a smoothing window for the SNEO, and a rectangular window
of length of seven samples was found to produce the best results
in emphasizing spike-to-signal ratio maximally.
The Detection Performance According to Detection
Steps: We optimized the detection parameters by applying the
current method toECG-contaminated EEG samples (1068 heart
beats with mean heart rate 71.2 beats/min within a duration
of 15 min) derived from the EEG recording of one OSAS
subject, including sleep stage 1, stage 2, REM, and wakefulness
state. In order to evaluate the validity of threshold selection
by EIH at STEP2, we compared the estimated thresholds
, , and ) with an exper-
imentally determined optimal threshold, which was chosen
to minimize FP and FN using real ECG R-peak references.
The FP and FN of the ECG-referenced results were 0.08 and
0.09, respectively, and these values were equal to the maximal
performance that can be achieved by the thresholding method.
The high correlation
between the ECG-referenced,
the experimental optimal threshold and one type of estimated
derived by EIH supported the use of
EIH for estimating the optimal threshold.
Though FP and FN at STEP2 were dependent on the
threshold type, no significant differences were evident after
postprocessing. We selected
as an optimally esti-
mated threshold because
requires less calculation
time than the other thresholds. However, in particular situ-
ations when
could not be obtained, we used the
average value of the other threshold types, i.e.,
STEP3-1 reduced FP at the cost of a slight increase in FN.
STEP3-2 then decreased FN while STEP 3-3 readjusted the FP.
Due to the postprocessing algorithms, FP and FN were reduced
to 0.008 from 0.0763 and to 0.008 from 0.0109, respectively,
The Elimination Performance: The performance of elimi-
nation using EAS was evaluated using simulated signals that
were generated by adding weighted ECG spike trains to the ar-
tifact-freenormal EEG.Themean power errorbetween theorig-
inal signals and the ECG-eliminated signals, normalized to the
original signal power, was 10%. The elimination procedure re-
duced the SBR of the contaminated EEG from 25.3 to 3.1.
Applicationto the OSASRecordings: FourOSAS recordings
were used to evaluate the overall performance of the algorithm.
The mean sleep time of each subject was 7.94 h (i.e., on the
average 953.5 epochs of 30s each) and the total number of heart
beats was 132878 during a total period of 31.8 h with a mean
heart rate of 69.7 beats/min. The mean respiratory disturbance
index(RDI) was49.9 (counts/h),indicating severeOSAS. Sleep
stages were scored according to Rechtschaffen and Kales’ sleep
staging criteria [14] by a sleep expert. The total epoch numbers
of stage 1, stage 2, REM, and wakefulness were 530, 2244, 555,
and 485, respectively. Mean FP was 0.0424 and FN was 0.1504
for total epochs.
Fig.8 illustratesthe relationshipbetween SBRand theperfor-
mance indices FP and FN. Fig. 8(a) shows the mean FP and FN
in the epochs with the given SBR, and Fig. 8(b) shows the mean
FP andFNin theepochs havingSBR higherthan thegiven SBR.
Fig. 8(c) indicatesthe relative number, i.e., frequency, ofepochs
havingSBR higherthan thegivenSBR. As is shown inFig. 8(a),
an SBR value higher than 20 was required for FP and FN to
reach 0.02 and 0.1, respectively. Epochs having an SBR higher
than 20 account for 60% of the total epochs in Fig. 8(c). Inspec-
tion of epochs with SBR around 20 indicated that the EEG sig-
nalscontained othertypes ofartifact-bursts suchas EMG,which
made itdifficultto identify someECG artifacts. For epochs with
an SBR below ten, ECG artifacts could not be discriminated
easily due to other types of artifacts present, or due to a rela-
tively small ECG effect on the EEG. Therefore, ECG artifact
elimination in these epochs
was not thought to be
necessary. These epochs accounted for as much as 20% of the
epochs and it would make sense to exclude these epochs in the
performance evaluation. The detection performance at epochs
was and in Fig. 8(b).
For epochs of
, which was regarded as the boundary
of obviousness for ECG artifacts by visual inspection, the algo-
rithm achieved
and [Fig. 8(b)].
In Table I, results of the performance are listed according to
sleep stages. Wakefulness states showed the lowest SBR and
the highest FP and FN. This result can be explained by the fact
that in the sleep of OSAS, a wakefulness state usually follows
apneic events with deep exhalation, which is associated with
severe movement and muscle artifacts. During REM stages, the
SBR was at its highest and the number of FPs and FNs at their
lowest. Upon applying the algorithm, the SBR decreased from
59.2 to 6.5 on the average, which confirms the efficiency of the
We propose an ECG artifact detection and elimination
method for EEG without the need for an additional ECG
channel. The method is based on the following three charac-
teristics of ECG artifacts: that the ECG R-peak occurs as a
spike, that the ECG R-peak has periodicity, and that the ECG
is uncorrelated with the EEG.
The SNEO showed excellent performance in emphasizing
spike components by a simple calculation. The EIH method uti-
lized ECG periodicity when estimating FP and FN to determine
an optimal threshold. Moreover, the EIH method allowed the
experimentally obtained optimal threshold to be approximated.
It should be noted that the detection performance with the
experimental optimal threshold derived by using an ECG ref-
erence was much lower than that obtained by using the pro-
cedures of EIH method. This indicates the intrinsic limitation
associated with the fixed thresholding method, due to the pres-
ence of the changing environment caused by the background
EEG. In some ECG spike regions, the original EEG activity
may be of the opposite phase to the ECG activity and, thus,
may degrade the spike energy. Conversely, the original EEG ac-
tivity may be in-phase, which would increase the ECG spike
energy. These situations explain the diverse instantaneous en-
ergy distribution of ECG spikes. In addition, SNEO is rela-
tively sensitive to high-frequency noise, such as EMG signals
that arefrequently present in EEGsignals. Therefore, additional
algorithmic adjustment is required to overcome the degrada-
tion of ECG detection caused by the various environments. In
order to decrease the number of misses and false alarms, two
threshold-adjusting procedures wereapplied. One procedure in-
volved rejecting spikes of shorter interspike duration than the
expected heart beat interval. The other procedure redetected
missedeventsbyadjusting thethresholdin theexpectedpeak re-
gion using ECG periodicity information. These adjusting steps
increased the detection performance significantly.
The application to OSAS data produced satisfactory results.
According to the results obtained, the performance of our al-
gorithm was mainly dependent on SBR. In the low-SBR cases,
ECG artifacts tend to be immersed in other artifacts or are not
easily distinguished from the EEG and in such cases, there is no
need to apply the ECG-artifact elimination algorithm.
We did not evaluate the performance of the method system-
atically according to heart rate variability and ECG waveform
changes. However, recordings of OSAS involving various sleep
stages can be considered to present a somehow extreme case of
heart-rate variability and ECG waveform changes which may
be exacerbated in the presence of several types of heart dis-
orders. In the case of OSAS signals where the standard devi-
ation of heart beat intervals was extended to almost 25 % of the
mean heart beatinterval, we foundno significant correlation be-
tween the standard deviation of heart beat intervals and FP or
FN within the same SBR range. Therefore, we believe that the
algorithm has enough redundancy, provided the cardiac activity
maintains periodicity within the normal ranges.
ECG waveform variability, due to, for example, amplitude
variation of R-peaks, may not severely disrupt the detection
performance, if such variability remains within normal limits,
considering that the ECG waveform change is equivalent to a
change in the background EEG in the sense of additive energy.
However, the artifact-elimination performance of the EAS algo-
rithm is believed to be dependent on the ECG waveform vari-
ability even if the algorithm uses the adaptive averaging method
of (8).
We believe that the elimination algorithm is worth examining
further, even though this is not the main concern of this paper.
Some intrinsic limitations were found in the EAS algorithm of
[3], which was based on the assumption that the beat-to-beat
QRS waveform is constant and every beat is averaged with a
triggering reference of time-locked R peaks. In order to satisfy
the time-locking assumption, the bias between the estimated
R-peaks and the accurate position of R-peaks should be mini-
mized. However, in the non-ECG reference case, the bias may
be unavoidable, and, therefore, we used a temporal realignment
technique, using template matching in order to reduce the bias
One merit of the proposed method is its excellent detection
performance in terms of real contaminated signals. In epochs
where the elimination process is thought to be necessary
, the average FP was 0.017 and the average FN
was 0.074 (Table I), which is an acceptable result for clinical
signals. The other merit of the method is that it uses only one
channel of contaminated EEG. The simplicity of the algorithm
is another merit for successful implementation in practical
real-time situations. If this algorithm is applied to multichannel
EEGs, and if the time synchronization information between
channels were used, we would expect increased detection and
elimination performance in the low-SBR situation.
This paper proposes an efficient method for detecting and
eliminating periodic artifacts, and we believe that it has the po-
tential for more general application in systems that involve pe-
riodic or semiperiodic spike artifacts.
[1] R. Broughton, J. Fleming, and J. Fleetham, “Home assessment of sleep
disorders by portable monitoring,” J. Clin. Neurophys., vol. 13, no. 4,
pp. 272–284, 1996.
[2] B. Van Sweden, M. Koenderink, G. Windau, M. Van de Bor, F. Van Bel,
J. G. Van Dijk, and A. Wauquier, “Long-term EEG monitoring in the
early premature: Developmental and chronobiological aspects,” Elec-
troencephalogr. Clin. Neurophys., vol. 79, pp. 94–100, 1991.
[3] M. Nakamura and H. Shibasaki, “Elimination of EKG artifacts from
EEG records: Anew method of noncephalic referential EEGrecording,”
Electroencephalogr. Clin. Neurophys., vol. 66, pp. 89–92, 1987.
[4] J. L. Kenemans, P. C. M. Molenaar, and M. N. Verbaten, “Models
for estimation and removal of artifacts in biological signals,” in
Digital Biosignal Processing, Techniques in the Behavioral and Neural
Sciences, R. Weitkunat, Ed. Amsterdam, The Netherlands: Elsevier,
1991, vol. 5, ch. 9.
[5] B. Widrow and S. D. Stearns, “Adaptive interference canceling,” in
Adaptive Signal Processing. Upper Saddle River, NJ: Prentice-Hall,
1985, ch. 12, pp. 302–367.
[6] H. Park,J. Han, D.Jeong,and K.Park, “A studyon the eliminationof the
ECG artifact in the polysomnographic EEG and EOG using AR model,”
in Proc. 20th Ann. Int. Conf. IEEE Engineering Medicine Biology Soc.,
vol. 3, 1998, pp. 1632–1635.
[7] R. Everson and S. J. Roberts, “Independent component analysis,” in Ar-
tificial Neural Networks in Biomedicine, P. J. G. Lisboa, E. C. Ifeachor,
and P. S. Szczepaniak, Eds. New York: Springer-Verlag, 2000, ch. 12.
[8] S. Mukhopadhyay and G. C. Ray, “A new interpretation of nonlinear
energyoperatorand itsefficacy inspikedetection,” IEEETrans. Biomed.
Eng., vol. 45, pp. 180–187, Feb. 1998.
[9] J. F. Kaiser, “On a simple algorithm to calculate the energy of a
signal,” in Proc. IEEE Int. Conf. Acoustic Speech Signal Processing,
Albuquerque, NM, 1990, pp. 381–384.
[10] R. B. Dunn, T. F. Quatieri, and J. F. Kaiser, “Detection of transient sig-
nals using energy operator,” in Proc. IEEE Int. Conf. Acoustic Speech
Signal Processing, vol. 3, 1993, pp. 145–148.
[11] P.Maragos,J. F.Kaiser,and T.F. Quatieri, “Onamplitudeandfrequency
demodulation using energy operators,” IEEE Trans. Signal Processing,
vol. 41, pp. 1532–1550, Apr. 1993.
[12] H. L. Van Trees, Detection, Estimation, and Modulation Theory: Part
I. New York: Wiley, 1968.
[13] O. Pahlm and L. Sornmo, “Software QRS detection in ambulatory mon-
itoring,” Med. Biol. Eng. Comput., pp. 289–297, 1984.
[14] A. Rechtschaffen and A. Kales, Eds., “A manual of standardized termi-
nology,” in Techniques and Scoring System for Sleep Stages of Human
Subjects. Los Angeles, CA: BIS/BRI, UCLA, 1968.
Hae-Jeong Park was born in KyungNam, Korea, in
1970. He received the B.S. degree in electrical engi-
neeringand theM.S.and Ph.D.degrees inbiomedical
engineering, from Seoul National University, Seoul,
Korea, in 1993, 1995, and 2000, respectively.
He is currently working as a Research Fellow
at the Clinical Neuroscience Division, Labora-
tory of Neuroscience, Boston VA Health Care
System—Brockton Division, Department of Psy-
chiatry and Surgical Planning Laboratory, MRI
Division, Department of Radiology, Brigham and
Women’s Hospital, Harvard Medical School, Boston, MA. His research
interests include biomedical signal processing, biomedical image analysis, and
Do-Un Jeong received the M.D. and Ph.D. degrees
from the Seoul National University, Seoul, Korea, in
1976 and 1988, respectively.
He is currently a Professor of Psychiatry at the
Seoul National University and a Researcher at the
Clinical Research Institute of the Seoul National
University Hospital and at the Neuroscience Re-
search Institute of the Seoul National University.
He was trained in sleep medicine and physiology
and is a Diplomat of the American Board of Sleep
Medicine. He is the Director of Division of Sleep
Studies at the Seoul National University Hospital and doing research in sleep
disorders, chronobiology, and signal processing.
Dr. Jeong is a Fellow of the American Academy of Sleep Medicine and is
currently President of the Korean Academy of Sleep Medicine.
Kwang-SukParkwas bornin Seoul, Korea, in 1957.
He receivedthe B.S., M.S., and Ph.D degrees inelec-
tronic engineering, especially for biomedical engi-
neering, from the Seoul National University, Seoul,
Korea, in 1980, 1983, and 1985 respectively.
He is currently the Professor and Chairman
of the Department of Biomedical Engineering,
Seoul National University College of Medicine.
His research interests include biomedical signal
processing, biomedical image analysis, and medical
... So it is impossible to use simple filtering techniques to remove the ECG from ECG contaminated EEG. There are possible algorithms proposed for effective removal of ECG from EEG since last few decades based upon energy of ECG signals (Nakamura and Shibasaki, 1987;Park et al., 2002;Waser and Garn, 2013), independent component analysis (ICA)-based methods (Zhou, 2002;Devuyst et al., 2008;Hamaneh et al., 2014), noise cancellation using adaptive filters (AFs) (Cho et al., 2007;Sheniha et al., 2013;Rajan, 2014a, 2014b;Jafarifarmand and Badamchizadeh, 2013;Sahul et al., 1995), and few other. This section discusses each techniques proposed by the authors along with the advantages and limitations in terms of extent of artefact removal, computation time and also in terms of input signal requirement. ...
... But its disadvantage lies in additional ECG channel requirement and EEG signals differing from the assumption model. Source: Nakamura and Shibasaki (1987) Another QRS wave energy-based method was proposed by Park et al. (2002) using energy interval histogram (EIH). Here the authors used a smoothed nonlinear energy operator (SNEO), i.e., Teager-Kaiser energy operator to detect the spike like signals (Kaiser, 1990). ...
... Source: Park et al. (2002) In Waser and Garn (2013), a modified version of Pan-Tompkin algorithm (Pan and Tompkins, 1985) used along with two stage linear regression (Draper and Smith, 2014) to eliminate the ECG interference from the EEG, using a reference synchronous ECG channel. ...
... RECOMMENDATIONS .......................................................................................... 56 5.3 CONCLUDING REMARKS .............................................................................................................. 57 BIBLIOGRAPHY ..............................................................................................................................58 APPENDIX A: EXPERIMENTAL PROTOCOL ............................................................................61Word Count:12,131 ...
Full-text available
This dissertation describes research whose the purpose is to remove artefacts from the EEG signals of preterm neonates, thereby facilitating the prediction of carbon dioxide levels dissolved in the blood; this was achieved using key statistics obtained from the EEG. The EEG signals of neonates were recorded over a 24 – 36 hours period using LabVIEW (National Instruments) software and a XLTEK EMU40EX breakout head box with a 7- electrodes configuration. Raw EEG signals of neonates were processed using discrete wavelet transform multiresolution analysis to remove artefacts. A nonlinear energy operator was used to calculate the energy of the artefact-free neonate’s EEG signals. Threshold detection was used to find the bursts and the interburst intervals. Spectral power analysis was performed on every burst-interburst interval cycle to calculate the relative powers in the delta (0.5-3.5 Hz), theta (4.0-7.5 Hz), alpha (8-12.5 Hz), and beta (13-30 Hz) frequency bands in each burst-interburst interval cycle of the neonate’s EEG signals. A two minute arithmetic average of the interburst intervals, and the burst-interburst interval relative powers were used to develop a linear regression equation for the prediction of the carbon dioxide level. The EEG signals of different gestational age neonates were processed offline. Results were compared with the readings of ABL800 FLEX blood gas analyser (a blood gas machine) with a measuring range of 0.67 – 33.3 kilopascals. For the clean neonate’s EEG signals, the absolute prediction error lay between 0 – 1.0 kilopascals.
... Non-invasive EEG signal acquisition occurs using electrodes placed on the scalp. The recorded signal represents the summation of post-synaptic electrical potentials from the underlying and surrounding brain structures (Teplan, 2002;Sanei and Chambers, 2013), together with electrical noise from the surrounding environment, recording equipment, and ongoing electrophysiological activity from the eyes (Dement and Kleitman, 1957;Overton and Shagass, 1969;Schlögl et al., 2007), heart (Stephenson and Gibbs, 1951;Park et al., 2002), muscles (Goncharova et al., 2003;Whitham et al., 2007;Muthukumaraswamy, 2013). Electroencephalographic signals measured from the scalp show microvolt-scale fluctuations (peak-peak range: 0.5 µV-100 µV; Teplan, 2002), while noise contamination can occur at the millivolt scale (1,000× greater amplitude). ...
Full-text available
Walking or running in real-world environments requires dynamic multisensory processing within the brain. Studying supraspinal neural pathways during human locomotion provides opportunities to better understand complex neural circuity that may become compromised due to aging, neurological disorder, or disease. Knowledge gained from studies examining human electrical brain dynamics during gait can also lay foundations for developing locomotor neurotechnologies for rehabilitation or human performance. Technical barriers have largely prohibited neuroimaging during gait, but the portability and precise temporal resolution of non-invasive electroencephalography (EEG) have expanded human neuromotor research into increasingly dynamic tasks. In this narrative mini-review, we provide a (1) brief introduction and overview of modern neuroimaging technologies and then identify considerations for (2) mobile EEG hardware, (3) and data processing, (4) including technical challenges and possible solutions. Finally, we summarize (5) knowledge gained from human locomotor control studies that have used mobile EEG, and (6) discuss future directions for real-world neuroimaging research.
... Since there have been many studies that have applied various methods or algorithms, including the utilization of principal component analysis (PCA), support vector machine classifiers or energy interval histograms to overcome or discriminate the artifact detection [61][62][63], these methods/algorithms may be applied in the development of future medical devices. Table 1 presents some clinical studies (EEG-based and non-EEG based system) applying these artifact elimination methods/algorithms which have been successful in reducing their false positive rates while still maintaining their seizure sensitivity rate [21,25,26,34,35,37]. ...
Background: Epilepsy is a devastating neurological disorder that affects nearly 70 million people worldwide. Epilepsy causes uncontrollable, unprovoked and unpredictable seizures that reduce the quality of life of those afflicted, with 1-9 epileptic patient deaths per 1000 patients occurring annually due to sudden unexpected death in epilepsy (SUDEP). Predicting the onset of seizures and managing them may help patients from harming themselves and may improve their well-being. For a long time, electroencephalography (EEG) devices have been the mainstay for seizure detection and monitoring. This systematic review aimed to elucidate and critically evaluate the latest advancements in medical devices, besides EEG, that have been proposed for the management and prediction of epileptic seizures. A literature search was performed on three databases, PubMed, Scopus and EMBASE. Methods: Following title/abstract screening by two independent reviewers, 27 articles were selected for critical analysis in this review. Results: These articles revealed ambulatory, non-invasive and wearable medical devices, such as the in-ear EEG devices; the accelerometer-based devices and the subcutaneous implanted EEG devices might be more acceptable than traditional EEG systems. In addition, extracerebral signalbased devices may be more efficient than EEG-based systems, especially when combined with an intervention trigger. Although further studies may still be required to improve and validate these proposed systems before commercialization, these findings may give hope to epileptic patients, particularly those with refractory epilepsy, to predict and manage their seizures. Conclusion: The use of medical devices for epilepsy may improve patients' independence and quality of life and possibly prevent sudden unexpected death in epilepsy (SUDEP).
... The anterior fontanelle, which is close to the standard positions of the Cz and Fz electrodes, is generally the main cause of cardiac-related interference in neonatal EEG signals, including interference from the pulsatile activity of the underlying arterial vessels. For these reasons, the algorithms developed for the removal of cardiac artefacts in the adult EEG [15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31] cannot straightforwardly be applied to neonatal EEG recordings. ...
Full-text available
Electrical cardiac and pulsatile interference is very difficult to remove from electroencephalographic (EEG) signals, especially if recorded in neonates, for which a small number of EEG channels is used. Several methods were proposed, including Blind Source Separation (BSS) methods that required the use of artificial cardiac-related signals to improve the separation of artefactual components. To optimize the separation of cardiac-related artefactual components, we propose a method based on Independent Component Analysis (ICA) that exploits specific features of the real electrocardiographic (ECG) signals that were simultaneously recorded with the neonatal EEG. A total of forty EEG segments from 19-channel neonatal EEG recordings with and without seizures were used to test and validate the performance of our method. We observed a significant reduction in the number of independent components (ICs) containing cardiac-related interferences, with a consequent improvement in the automated classification of the separated ICs. The comparison with the expert labeling of the ICs separately containing electrical cardiac and pulsatile interference led to an accuracy = 0.99, a false omission rate = 0.01 and a sensitivity = 0.93, outperforming existing methods. Furthermore, we verified that true brain activity was preserved in neonatal EEG signals reconstructed after the removal of artefactual ICs, demonstrating the effectiveness of our method and its safe applicability in a clinical context.
... Eyeblink and muscle artifact components were detected and extracted using an independent component analysis (ICA) algorithm (FastICA) [33], [34]. The ADJUST algorithm [35] was then used to identify and eliminate contaminated ICs based on the unsupervised method and then verified manually. Next, the reconstructed EEG data were re-referenced to the averaged reference. ...
Full-text available
Transcutaneous electrical nerve stimulation (TENS) has been reported to alleviate pain in chronic pain patients. Currently, there is limited knowledge how TENS affects can cause cortical neuromodulation and lead to modulation of non-painful and painful sensations. Our aim was therefore to investigate the effect of conventional, high-frequency TENS on cortical activation and perceived sensations in healthy subjects. We recorded somatosensory evoked potentials (SEPs) and perceived sensations following high-frequency TENS (100 Hz) in 40 healthy subjects (sham and intervention group). The effect of TENS was examined up to an hour after the intervention phase, and results revealed significant cortical inhibition. We found that the magnitude of N100, P200 waves, and theta and alpha band power was significantly suppressed following the TENS intervention. These changes were associated with a simultaneous reduction in the perceived intensity and the size of the area where the sensation was felt. Although phantom limb pain relief previously has been associated with an inhibition of cortical activity, the efficacy of the present TENS intervention to induce such cortical inhibition and cause pain relief should be verified in a future clinical trial.
... On the contrary, it decreases the slowly varying parts of this signal (i.e., P and T waves). Based on this capability of TEO to accentuate the R peaks, Park et al. [48] have explored TEO to enhance the QRS regions, and further to eliminate the ECG signal from the electroencephalogram (EEG) one. However, while these important TEO characteristics, very few number of publications can be found in the literature that suggest the use of TEO for ECG signal processing [55]. ...
Full-text available
The electrocardiogram is an important tool that is widely used for diagnosis of many cardiovascular diseases. In this context, QRS complex detection is a very crucial step in the ECG diagnosis system. The major aim of this work is to develop a novel method for QRS complex detection under various ECG signal morphologies as well as under different ECG recording conditions, including numerous noise sources and varying QRS waveforms. The proposed algorithm is based principally on the stationary wavelet transform (SWT) and Teager energy operator (TEO). In our scheme, SWT is first used for ECG signal preprocessing and QRS complex frequency content localization. Subsequently, a novel process for R peak detection based on TEO and a moving average (MA) filter is introduced. More precisely, SWT is coupled with TEO and the MA filter to construct a smoothed detection mask. Then, after the mask segmentation and adaptive thresholding steps, R peak times are identified using the maxima detected on the created mask and employing a reference ECG signal. At this stage, efficient decision rules are applied for reducing the number of false alarms. In the experiments, we validate the proposed method on the well-known annotated MIT-BIH arrhythmia database (MITDB). The experimental results show that the newly proposed algorithm provides satisfactory detection performances compared to the recent state-of-the-art methods, with an average sensitivity of 99.84%, average positive predictivity (P+) of \(99.87\%\), detection error rate of 0.30% and an overall detection accuracy of 99.70%. Also, the proposed method presents a low computational time complexity with an average processing time of 12 s on each ECG record from MITDB.
Full-text available
Background and motivation: Time-frequency representation (TFR) of a signal finds its application in numerous fields for non-stationary multicomponent signal analysis. Due to underlying difficulties and improvement scope in the current methodology, developing a new time-frequency method can improve spectral analysis of real-life signals and further can be extended to practical applications. Materials and methods: The proposed new method swarm-sparse decomposition method (SSDM) is an advanced version of swarm decomposition (SWD) for decomposing nonstationary multicomponent signals into a finite number of oscillatory components (OCs). Benefiting from sparse spectrum and SWD, the proposed SSDM method delivers optimal estimation of boundary frequencies in the sparse spectrum, resulting in improved filter banks. In addition to SSDM, we have also proposed the spectrum approximator function, i.e., fused least absolute shrinkage and selection operation to modify sparse spectrum and get significant OCs. The performance of the proposed SSDM has been evaluated by TFR analysis and compared to SWD and Hilbert-Huang transform methods. Also, it has been tested for automated sleep apnea classification using a convolutional neural network (CNN) and bi-directional long-short term memory (BiLSTM) on the publicly available EEG database. Results: The proposed SSDM-TFR-CNN and SSDM-feature-fusion-BiLSTM frameworks outperformed all the compared methods used for sleep apnea detection and achieved the highest classification accuracy of 96.24% and 95.86%, respectively, in the subject-independent crossvalidation scheme. Conclusion: Simulation result shows that the proposed SSDM method delivers substantial improvement in time-frequency analysis. Our developed sleep apnea detection model could be a vital aid in clinical solutions.
Brain activity recordings outside clinical or laboratory settings using mobile EEG systems have gained popular interest allowing for realistic long-term monitoring and eventually leading to identification of possible biomarkers for diseases. The less obtrusive, minimized systems (e.g., single-channel EEG, no ECG reference) have the drawback of artifact contamination with varying intensity that are particularly difficult to identify and remove. We developed brMEGA, the first open-source algorithm for automated detection and removal of cardiogenic artifacts using non-linear time-frequency analysis and machine learning to (1) detect whether and where cardiogenic artifacts exist, and (2) remove those artifacts. We compare our algorithm against visual artifact identification and a previously established approach and validate it in one real and semi-real datasets. We demonstrated that brMEGA successfully identifies and substantially removes cardiogenic artifacts in single-channel EEG recordings. Moreover, recovery of cardiogenic artifacts, if present, gives the opportunity for future extraction of heart rate features without ECG measurement.
Full-text available
Brain activity recordings outside clinical or laboratory settings using mobile EEG systems have recently gained popular interest allowing for realistic long-term monitoring and eventually leading to identification of possible biomarkers for diseases. The less obtrusive, minimized systems (e.g. single-channel EEG, no ECG reference) have the drawback of artifact contamination with varying intensity that are particularly difficult to identify and remove. We developed brMEGA, the first algorithm for automated detection and removal of cardiogenic artifacts using non-linear time-frequency analysis and machine learning to (1) detect whether and where cardiogenic artifacts exist, and (2) remove those artifacts. We compare our algorithm against visual artifact identification and a previously established approach and validate it in one real and semi-real datasets. We demonstrated that brMEGA successfully identifies and substantially removes cardiogenic artifacts in single-channel EEG recordings. Moreover, recovery of cardiogenic artifacts gives the opportunity for future extraction of heart rate features without ECG measurement.
Full-text available
Mutants of sperm whale myoglobin were constructed at position 29 (B10 in helix notation) to examine the effects of distal pocket size on the rates of ligand binding and autooxidation. Leu29 was replaced with Ala, Val, and Phe using the synthetic gene and Escherichia coli expression system of Springer and Sligar (Springer, B. A., and Sligar, S. G. (1987) Proc. Natl. Acad. Sci. U. S. A. 84, 8961-8965). Structures of the ferric forms of Val29 and Phe29, and the oxy form of Phe29 myoglobin were determined to 1.7 A by x-ray crystallography. The ferric mutant proteins are remarkably isomorphous with the wild type protein except in the immediate vicinity of residue 29. Thus, the protein structure in the distal pocket of myoglobin can accommodate either a large "hole" (i.e. Ala or Val) or a large side chain (i.e. Phe) at position 29 without perturbation of tertiary structure. Phe29 oxymyoglobin is also identical to the native oxy protein in terms of overall structure and interactions between the bound O2 and His64, Val68, Phe43, and Ile107. The distance between the nearest side chain atom of residue 29 and the second atom of the bound oxygen molecule is 3.2 A in the Phe29 protein and 4.9 A in native myoglobin. The equilibrium constants for O2 binding to Ala29, Val29, and Leu29 (native) myoglobin are the same, approximately 1.0 x 10(6) M-1 at 20 degrees C, whereas that for the Phe29 protein is markedly greater, 15 x 10(6) M-1. This increase in affinity is due primarily to a 10-fold decrease in the O2 dissociation rate constant for the Phe29 mutant and appears to be the result of stabilizing interactions between the negative portion of the bound O2 dipole and the partially positive edge of the phenyl ring. Increasing the size of residue 29 causes large decreases in the rate of autooxidation of myoglobin: k(ox) = 0.24, 0.23, 0.055, and 0.005 h-1 for Ala29, Val29, Leu29 (native), and Phe29 myoglobin, respectively, in air at 37 degrees C. Thus, the Leu29----Phe mutation produces a reduced protein that is remarkably stable and is expressed in E. coli as 100% MbO2. The selective pressure to conserve Leu29 at the B10 position probably represents a compromise between reducing the rate of autooxidation and maintaining a large enough O2 dissociation rate constant to allow rapid oxygen release during respiration.
Conference Paper
A simple algorithm is derived that permits on-the-fly calculation of the energy required to generate, in a certain sense, a signal. The results of applying this algorithm to a number of well-known signals are shown. Some of the invariance and noise properties of the algorithm are derived and verified by simulation. The implementation of the algorithm and its application to speech processing are briefly discussed
Long-term cassette EEG monitoring in the neonatal intensive care unit has established prognostic criteria regarding the developmental outcome by quantifying seizure activity. The clinical significance of the organization of continuous and discontinuous EEG patterns in the early premature is still an open question. This report presents quantified EEG data from repeated 24 h records during the first week of life in premature infants (conceptional age less than 32 weeks) with and without ultrasound evidence of intracerebral hemorrhage. The repartition and evolution of EEG background activity is not a reliable parameter regarding pathology. The continuity index is rather a maturational variable and its ultradian fluctuation is an early expression of the "basic rest activity cycle" (BRAC) rhythm.
We have obtained the oxygen-17 nuclear magnetic resonance (NMR) spectra of a variety of C17O-labeled heme proteins, including sperm whale (Physeter catodon) myoglobin, two synthetic sperm whale myoglobin mutants (His E7----Val E7; His E7----Phe E7), adult human hemoglobin, rabbit (Oryctolagus cuniculus) hemoglobin, horseradish (Cochlearia armoracia) peroxidase (E.C. isoenzymes A and C, and Caldariomyces fumago chloroperoxidase (E.C., in some cases as a function of pH, and have determined their isotropic 17O NMR chemical shifts, delta i, and spin-lattice relaxation times, T1. We have also obtained similar results on a picket fence prophyrin, [5,10,15,20-tetrakis(alpha, alpha, alpha, alpha, alpha-pivalamidophenyl)porphyrinato]iron(II) (1-MeIm)CO, both in solution and in the solid state. Our results show an excellent correlation between the infrared C-O vibrational frequencies, v(C-O), and delta i, between v(C-O) and the 17O nuclear quadrupole coupling constant (e2qQ/h, derived from T1), and as expected between e2qQ/h and delta i. Taken together with the work of others on the 13C NMR of 13CO-labeled proteins, where we find an excellent correlation between delta i(13C) and v(Fe-C), our results suggest that IR and NMR measurements reflect the same interaction, which is thought to be primarily the degree of pi-back-bonding from Fe d to CO pi* orbitals, as outlined previously [Li, X.-Y., & Spiro, T.G. (1988) J. Am. Chem. Soc. 110, 6024]. The modulation of this interaction by the local charge field of the distal heme residue (histidine, glutamine, arginine, and possibly lysine) in a variety of species and mutants, as reflected in the NMR and IR measurements, is discussed, as is the effect of cysteine as the proximal heme ligand.