Combination of Canonical Correlation Analysis and Empirical Mode Decomposition Applied to Denoising the Labor Electrohysterogram
ABSTRACT The electrohysterogram (EHG) is often corrupted by electronic and electromagnetic noise as well as movement artifacts, skeletal electromyogram, and ECGs from both mother and fetus. The interfering signals are sporadic and/or have spectra overlapping the spectra of the signals of interest rendering classical filtering ineffective. In the absence of efficient methods for denoising the monopolar EHG signal, bipolar methods are usually used. In this paper, we propose a novel combination of blind source separation using canonical correlation analysis (BSS_CCA) and empirical mode decomposition (EMD) methods to denoise monopolar EHG. We first extract the uterine bursts by using BSS_CCA then the biggest part of any residual noise is removed from the bursts by EMD. Our algorithm, called CCA_EMD, was compared with wavelet filtering and independent component analysis. We also compared CCA_EMD with the corresponding bipolar signals to demonstrate that the new method gives signals that have not been degraded by the new method. The proposed method successfully removed artifacts from the signal without altering the underlying uterine activity as observed by bipolar methods. The CCA_EMD algorithm performed considerably better than the comparison methods.
[show abstract] [hide abstract]
ABSTRACT: The electrohysterogram (EHG) signal is mainly corrupted by the mother's electrocardiogram (ECG), which remains present despite analog filtering during acquisition. Wavelets are a powerful denoising tool and have already proved their efficiency on the EHG. In this paper, we propose a new method that employs the redundant wavelet packet transform. We first study wavelet packet coefficient histograms and propose an algorithm to automatically detect the histogram mode number. Using a new criterion, we compute a best basis adapted to the denoising. After EHG wavelet packet coefficient thresholding in the selected basis, the inverse transform is applied. The ECG seems to be very efficiently removed.IEEE Transactions on Biomedical Engineering 09/2000; 47(8):1010-7. · 2.28 Impact Factor
Chapter: Abdominal EHG on a 4 by 4 grid: mapping and presenting the propagation of uterine contractions[show abstract] [hide abstract]
ABSTRACT: Numerous studies have observed and analyzed the external electrical activity of the uterus associated with contractions and labor. Most of these studies have involved the use of only 3 to 5 electrodes and little effort has been made to investigate the electrical activity concurrently at different locations. In this paper we present the results from measurements of contractions in labor using a 16 electrodes grid. We tried out various methods of presenting and analyzing this data and found this to be a non-trivial task. Here we present both an animation of the evolution of the electric potential, as well as a temporal correlation presentation. The results from a limited sample are in many ways surprising and may provide a new insight in to possible mechanism underlying uterine contractions.12/2006: pages 139-143;
Article: Interactions between Uterine EMG at Different Sites Investigated Using Wavelet Analysis: Comparison of Pregnancy and Labor Contractions.EURASIP J. Adv. Sig. Proc. 01/2010; 2010.
Abstract? the electrohysterogram (EHG) is often corrupted by
electronic and electromagnetic noise as well as movement
artifacts, skeletal EMG and electrocardiograms from both
mother and fetus. The interfering signals are sporadic and/or
have spectra overlapping the spectra of the signals of interest
rendering classical filtering ineffective. In the absence of efficient
methods for denoising the monopolar EHG signal, bipolar
methods are usually used. In this paper, we propose a novel
combination of Blind Source Separation using Canonical
Correlation Analysis (BSS_CCA)
Decomposition (EMD) methods to denoise monopolar EHG. We
first extract the uterine bursts by using BSS_CCA then the
biggest part of any residual noise is removed from the bursts by
EMD. Our algorithm, called CCA_EMD, was compared with
wavelet filtering and independent component analysis (ICA). We
also compared CCA_EMD with the corresponding bipolar
signals to demonstrate that the new method gives signals that
have not been degraded by the new method. The proposed
method successfully removed artifacts from the signal without
altering the underlying uterine activity as observed by bipolar
methods. The CCA_EMD algorithm performed considerably
better than the comparison methods.
Index Terms? uterine EMG;; canonical correlation analysis;;
empirical mode decomposition;; preterm labor.
and Empirical Mode
lectrohysterogram (EHG) is the uterine electromyogram
signal recorded externally on pregnant women. This
signal has long been expected to have important clinical
applications, especially in relation to predicting and
preventing pre-term labor. The EHG is a problematic signal
that has very low frequency and is very low amplitude as
compared to various sources of contaminating noise. In this
paper we describe a large step in helping to isolate the signal
from noise, and thus to overcome important barriers that have
prevented exploiting the potential of the signal to provide
information on the genesis and evolution of human labor.
During recordings, the EHG is frequently contaminated by
different artifacts which may often complicate the
interpretation of the EHG. The typical noise origins in the
EHG are maternal and fetal electrocardiograms (ECG),
abdominal muscle electromyogram (EMG), maternal and fetal
movement artifacts and electronic noise from the surrounding
The EHG is a noisy signal even recorded by bipolar electrodes
that are fairly close together. Wavelet filtering has been used
successfully on bipolar EHG for removing maternal and fetal
ECG as well as stationary electric noises . Most techniques
of wavelet filtering assume that the noise is of low amplitude
and stationary when compared to the signal of interest.
However, the noise in monopolar EHG is not stationary and
usually of higher amplitude than the signal of interest. In
addition, many sources of noise have main frequency
components that are close to the frequency of the signal of
interest. The EHG has most of its energy in the frequency
band of 0.1 to 1.5 Hz, as compared to maternal cardiac
frequency (typically 1.2 Hz), maternal respiration (typically
0.2 Hz) and fetal respiration (typically 1 Hz). As the noise is
in the same frequency bands and/or is of high amplitude and
sporadic, it cannot be rejected by classical or wavelet filters.
Recently, researchers have focused their efforts on
multichannel EHG in the hope of applying propagation
analysis to predict preterm labor. An example of this type of
approach is the work on signals obtained from a 16 electrode
matrix that have been recorded in Iceland in the last few years
. All the related studies have been done by using bipolar
signals to reduce noise effect [3-5]. Spatial resolution is very
important for signal propagation analysis and the use of a
bipolar signal reduces this resolution by almost half. Another
important limitation in using bipolar electrodes in a matrix is
that any analysis of the direction of propagation becomes
meaningless or biased, as the direction of propagation along
the line between the two bipolar electrodes is privileged.
Bipolar measurements of electric phenomena are basically a
way of rejecting the part of the signal that is measured on both
electrodes and keeping only the part that is dissimilar to the
two electrodes. The assumption underlying this technique is
????? ???? ???????? ????????al activity is dissimilar to the two
electrodes and that the common part is from further afield and
thus not relevant. It is highly uncertain that this holds for the
Combination of Canonical Correlation Analysis and
Empirical Mode Decomposition applied to denoising the
Hassan M., Boudaoud S., Terrien J., Karlsson B. and Marque C.
case of the EHG. It is therefore likely that important
information is lost by using bipolar signals only.
Blind Source Separation (BSS) methods such as Independent
Component Analysis (ICA) and Principal Component
Analysis (PCA) are increasingly being used in biomedical
signal processing involving analysis of multivariate time series
data such as EEG [6, 7].
Recently, a new method for muscle artifact elimination in
scalp EEG has been developed that does not have some of the
disadvantages of ICA . The method is based on the
statistical Canonical correlation analysis (CCA) method
applied as a blind source separation (BSS) technique, called as
BSS_CCA. This method has shown considerably better
performance than ICA in some applications [8, 9].
In this paper the aim of CCA is the extraction of the uterine
bursts based on the hypothesis that the bursts have higher
autocorrelation coefficients than the noise. Thus the expected
result from applying CCA is the extraction of uterine burst.
The bursts observed after applying CCA only, contain fewer
artifacts than the original signals but still contain the artifacts
that have high autocorrelation. Empirical mode decomposition
(EMD) was chosen as a second step to remove this noise from
the extracted burst. EMD technique was introduced by Huang
et al.  to analyze nonstationary and nonlinear signals.
EMD has become a very important tool to analyze biomedical
signals. The use of EMD for analyzing esophageal
manometric data in gastro esophageal reflux disease indicated
good performance in removing different kind of artifacts
(respiratory, motion...) from electrogastrogram (EGG) signals
[11, 12]. The EMD approach also proved to be efficient in
removing artifacts from ECG signals . For EEG data,
EMD algorithm has also been employed . In this work it
was demonstrated how reconstruction using the Hilbert-Huang
transform can successfully be applied to contaminated EEG
data for the purposes of removing unwanted ocular artifacts.
More recently, Wu and Huang have introduced a noise
assisted version of the EMD method, called Ensemble
Empirical mode decomposition (EEMD) . This method
has shown better performance than EMD as it extracts the
IMFs in a manner so that the mode mixing disadvantage of the
EMD method is corrected. We did not implement it in this
work, as EEMD is much slower than EMD and has not been
used, to our knowledge, in signal processing for clinical
Wavelet transform (WT) denoising methods could also be
good candidates for the second step of the processing
presented here. However, in this first attempt we chose EMD
rather than WT methods for two reasons: i) EMD is a data-
driven algorithm: it decomposes the signal in a natural way
without prior knowledge about the signal of interest embedded
in the noise which is not the case the wavelet transform;; ii)
EMD demonstrated better performance than WT when
combined with ICA in denoising EEG signals .
The aim of this work is to combine the use of BSS_CCA and
EMD algorithms to design a new tool, called CCA_EMD, to
remove the main interferences embedded in monopolar
abdominal EHG recordings. The method is applied to real
signals recorded on women during pregnancy and labor. A
comparison to bipolar signals obtained in the same matrix is
made. A quantitative comparison between CCA_EMD,
wavelet filtering, bipolar signals and ICA is also presented.
II. MATERIELS AND METHODS
A. Recording protocol
The EHG signal recording was performed using a 16-channel
multi-purpose physiological signal recorder, most commonly
used for investigating sleep disorders (Embla A10). We used
reusable Ag/AgCl electrodes. The measurements were
performed at the Landspitali University hospital in Iceland,
following a protocol approved by the relevant ethical
committee (VSN 02-0006-V2).
The signal sampling rate was 200 Hz on 16 bits. The recording
device has an anti-aliasing filter with a low pass cut-off
frequency of 100 Hz. The concurrent tocodynamometer paper
trace (TOCO) was digitized in order to ease the identification
of contractions. Recordings from six women during labor are
used in this work.
B. Canonical Correlation Aanalysis (CCA)
In BSS approach, the observed multichannel signals are
assumed to reflect a linear combination of several sources
which are associated with underlying physiological processes,
artifacts and noise. The BSS approach aims to recover a set of
unobserved source signals by using only a set of observed
mixtures of sources. The observed time series X(t) =[x1(t);;
x2(t);; . . . ;; xK(t)]T , is the result of an unknown mixture of a set
of unknown source signals S(t) = [s1(t);; s2(t);; . . . ;; sK(t)]T with
t = 1;; . . .;;N, where N the number of samples, K the number of
sensors and T is the transpose operator. The mixing is
assumed to be linear, thus reducing the mixing to a matrix
( )X t
where A is the unknown mixing matrix. The aim is to estimate
the mixing matrix and recover the original source signals S(t).
This could be done by introducing the de-mixing matrix W
such that it approximates the unknown source signals in S(t),
by a scaling factor:
( )Z t WX t
Ideally, W is the inverse of the unknown mixing matrix A, up
to scaling and permutation. There are many ways to solve the
BSS problem depending on the definition of contrast
functions. The ICA method tries to make the estimated
( )AS t
sources as non-Gaussian as possible. However, in PCA and
most of the ICA algorithms, the temporal correlations are not
taken into consideration for solving contrast functions. CCA
solves this BSS problem by forcing the sources to be
maximally autocorrelated and mutually uncorrelated, while the
mixing matrix is assumed to be square .
Consider the observed data matrix X(t) and its temporally
delayed version Y(t) = X(t-1). The CCA method obtains two
sets of basis vectors, one for x and the other for y, such that the
correlations between the projections of the variables onto
these basis vectors are mutually maximized.
The total covariance matrix is given by
where CXX and CYY are the auto-covariance matrices of X and Y
respectively, CXY is the cross- covariance matrix (CXY= CT
and E is the expectation operation.
The canonical correlation between X and Y can be calculated
by solving these equations:
C C C Cww
C C C Cww
with the canonical correlation coefficient ? as the square
root of the eigen-value, and wX and wY as eigenvectors. Since
the solutions are related, only one of the eigen-value equations
needs to be solved to get the demixing matrix w. The CCA
gives the source signals that are uncorrelated with each other,
maximally autocorrelated and ordered by decreasing
When BSS-CCA is applied to the EHG, the sources
contributing to the EHG and noise are obtained. The artifacts
can be removed by setting equal to zero, the columns
representing the activations of the related sources, before the
with Z(t) the sources obtained by BSS-CCA, and
mixing matrix, with its columns related to artifact sources, set
C. Empirical Mode Decomposition (EMD)
The empirical mode decomposition (EMD) was proposed by
Huang et al. as a new signal decomposition method for
nonlinear and nonstationary signals . The EMD
decomposes a signal into a collection of oscillatory modes,
called intrinsic mode functions (IMF), which represent fast to
slow oscillations in the signal. Each IMF can be viewed as a
subband of a signal. Therefore, the EMD can be viewed as
subband signal decomposition.
Given a signal x(t), the effective algorithm of EMD can be
summarized as follows :
1. Identify all extrema of x(t)
2. Interpolate along the point of x(t) identified in the
first step, in order to form an upper emax(t) and lower
3. Compute the mean m(t)=( emin(t) + emax(t))/2
4. Extract the detail d(t)=x(t)-m(t)
5. Iterate on the residual m(t)
In practice, the above procedure has to be refined by a sifting
process  which amounts to first iterating steps 1 to 4 upon
the detail signal d(t), until this latter can be considered as zero-
mean according to some stopping criterion. Once this is
achieved, the detail is referred to as an Intrinsic Mode
Function (IMF), the corresponding residual is computed and
step 5 applies. By construction, the number of extrema is
decreased when going from one residual to the next, and the
whole decomposition is guaranteed to be completed with a
finite number of modes.
Denoising by EMD is in general carried out by partial
signal reconstruction, which is based on the fact that noise
components lie in the first several IMFs.
D. CCA/EMD combination
We assume that the BSS is the best way to extract the uterine
bursts and based on the hypothesis that the sources of uterine
bursts have higher autocorrelation than the sources
corresponding to the artifacts, we choose the CCA method as a
way to extract the uterine bursts and in the same time
eliminate all the low autocorrelated noise. The sources of
device noise (electronic artifacts) are high autocorrelated and
then it is not possible to remove it by using only the CCA
method. It has been demonstrated that EMD shows good
performance in removing this kind of noise . For this and
other reasons, we chose to use EMD as the complementary
tool to remove the residual electronic noise. We call this
combination the CCA_EMD algorithm.
E. Comparative study
For performance evaluation purposes, the proposed technique
is compared with two other techniques that we have already
used for denoising EHG signals in our group.
Wavelet filtering: The specific algorithm which is based
on the redundant wavelet packet transform was
developed by Leman et al. and has already been used to
remove artifacts from bipolar EHG signal. By applying
this algorithm, the corrupting ECG seems to be totally
removed while all other artifacts (fetal movements,
electronic noise) are still presented .
FastICA: This is a BSS approach that uses similar
techniques as CCA_EMD and is therefore useful for
comparison. The selection of the ICA components
accounting for the artifact removing was based on visual
inspection and the signal was reconstructed, excluding
the components related to the artifact.
For a quantitative comparison between the different methods,
signal to noise ratio (SNR) was used as a criterion. For each
contraction and each EHG channel, the SNR was estimated by
computing the energy of the burst over the energy of the base
lines present before and after the EHG burst. The SNRs
obtained by CCA_EMD were compared with those obtained
with bipolar signal (vertical differentiation), ICA and the
wavelet method. For statistical comparison, we used the two
tailed sign test with a minimal significant level of 0.05.
Figure 1A shows a typical example of raw monopolar EHG
signals recorded from a woman during labor. The canonical
components (CC) obtained by applying the BSS_CCA method
are shown in Figure 1B. The fetal movements pump spikes
(located in the 16th CC) and part of electronic noises (located
in the 14th and the 15th CC) are well distinguished from the
components related to the uterine activity.
A very important step to take into account here is the choice of
the CCs corresponding to the artifacts, in order to then remove
them before signal reconstruction. We should detect the
?????????? ?????????????? ??? ???? ??????????? ????? ?????????
The methodology we propose to choose the optimal threshold
value can be described as following:
Calculate the CCA components and the associated
We choose a threshold ranging between 0 and 1 (with
0.1 steps), then remove the CCs below this value and
reconstruct the signals.
We compute the original bipolar signals (BipOrg) from
two raw channels X and Y, and the bipolar signal
obtained from the same channels after the two preceding
processing steps (BipDen). We then get two versions of
the bipolar signals, one created from the raw signals
directly (BipORG) and the other by eliminating all the
CCs below the given threshold of autocorrellation.
We then compute the correlation between BipOrg and
We repeat these steps for the 20 contractions from six
women and then we calculate the average and standard
deviation at each autocorrelation value, for each given
Fig.1. A) Original raw signals from a woman in labor B) corresponding CCA
The curve in figure 2A indicates that 0.5 is the optimal
threshold value for eliminating noisy CCs as we got the
highest correlation between BipOrg and BipDen for this value.
Computation of the mean square error (MSE) instead of the
correlation confirms this, as the lowest MSE is observed at 0.5
(results not shown).
Figure 2B presents the autocorrelation coefficients curve
computed for the 20 contractions from six women during
labor, and the autocorrelation coefficients curve for the CC of
the contraction presented in figure 1. The threshold (presented
as a dashed horizontal line) is fixed at 0.5. All the CCs below
this value are excluded. For the presented contraction (solid
line) we reject the last three CCs. We can notice that this
threshold corresponds to the last four CCs on the curve
representing the autocorrelation coefficients averaged over the
6 women during labor (Figure 2 dashed curve).
20 4060 80100120140160
Fig.2. A) The correlation coefficient between Bip Org (raw bipolar signal) and
Bipden (processed bipolar signal) in function of the autocorrelation threshold
B) Solid curve: autocorrelation coefficients from the contraction shown in
Figures 1. Dashed curve: averaged autocorrelation curve obtained from the 6
women in labor. The horizontal dashed line represent the threshold value.
By using a threshold value equal to 0.5 and removing before
reconstruction of the EHG the components below this value,
we obtain the intermediate denoised EHGs shown in Fig.3.
Notice that all the fetal movements, maternal/fetal ECG and
part of electronic noises have been removed from the original
signal. After BSS_CCA denoising, some noise remains that
presents high autocorrelation coefficient (specially the
electronic noise coming from the devices). These artifacts are
not completely removed by BSS_CCA. To remove this noise
we apply EMD to the signals previously denoised by
Fig.3. Uterine bursts extracted by CCA only. The different curves represent
the 16 simultaneous recorded channels for one contraction.
Figure 4 shows the final signals after CCA_EMD. Here, based
on visual inspection, partial reconstruction is applied by
removing the first three IMFs that we consider to be high
frequency noise. In figure 4, we can clearly differentiate
between the baseline and the uterine bursts, which will greatly
ease the segmentation of uterine bursts.
Fig.4. Signals denoised by CCA _EMD. The two vertical lines indicate the
start and the end of the uterine bursts based on the TOCO trace.
Figure 5 presents an example of one of the channels extracted
from the signals shown in Figures 1, 3 and 4, in order to better
evidence the way each step of CCA_EMD improves the
signal, from a very noisy monopolar signal with various types
of artifacts, to a clearly visible uterine burst. This result is
based on a threshold autocorrelation value of 0.5 as described
00.10.20.3 0.40.50.60.7 0.8 0.91
Over 6 women
20406080 100120140 160
Fig.5. Top: Original signal;; Middle: signal after CCA processing;; Bottom:
denoised signal after CCA_EMD.
Some residual noise from pump spikes and high frequency
electronic noise is clearly visible on the middle panel of figure
5. This residual noise is then completely erased from the burst
by applying the EMD method (figure 5 bottom). As the
bipolar signal is our only available reference, we then compare
the raw bipolar signal (computed as the difference between the
raw channel-1 and raw channel-2, BipOrg) and the denoised
bipolar signal (computed as the difference between denoised
channel-1 and denoised channel-2, BipDen).
In figure 6 we present a typical example that indicates how
CCA_EMD enhances the uterine activity information at least
as well as the bipolar method. The two bipolar signals
(original and denoised) are quite similar and there is no
distortion (RMS error=10-4, correlation coefficient=0.74).
Time-frequency representations (Scalograms) show that the
CCA_EMD has not visibly affected the uterine burst
frequency content when compared to bipolar signals.
Figure 7 shows the quantitative difference between the SNR of
monopolar, bipolar, CCA_EMD, FastICA and the Wavelet
algorithm described in . The values represent the median of
SNR over the 16 channels for the six women during labor. The
monopolar raw signals present a poor SNR (median SNR = -
3.4 dB). The results indicate clearly that CCA_EMD has the
highest median SNR equal to 6.74 dB compared to 3.16 dB for
wavelet, 5.02 dB for FastICA and 5.2 dB for bipolar signals.
Fig.6. From top to bottom: monopolar channel-1;; monopolar channel-2;;
corresponding bipolar signals: raw data (Left) and denoised data (Right)..;;
scalograms of the bipolar signal computed from raw data (Left) and Denoised
Fig.7. The SNR for monopolar signal, wavelet, ICA, bipolar and CCA_EMD
204060 80100 120 140160
2040 60 80100120 140160
2040 60 80100 120 140160
20 4060 80100 120140160
2040 6080 100120140160
204060 80100 120140160
20 4060 80 100120 140160
2040 6080 100120140 160
20406080 100 120140 160
204060 80 100120 140 160
MonopolarWavelet ICA Bipolar CCA_EMD
IV. DISCUSSION AND CONCLUSION
In this work a novel combination of two important methods
has been used in order to completely remove artifacts from a
monopolar EHG signal. The method consists of two steps.
First, BSS_CCA is used to extract the uterine bursts in the
presence of high intensity noise with overlapping frequency
band. Then the signals are given a final cleaning by applying
In addition, the method described is fast and computationally
economical and can be used as a preprocessing step to
facilitate segmentation of uterine EMG bursts. Furthermore,
the noise activities are present in the least autocorrelated CCA
components, which indicates that it is possible with this
technique to separate noise sources (small autocorrelation)
from the uterine activity (higher autocorrelation).
Concerning the CCA method, we have in this work proposed a
method that analyses the autocorrelation coefficient curve to
estimate the ones that have to be eliminated, based on a
similarity to a reference signal (BipOrg). In addition, the
estimation of the threshold value by using criteria such as
Akaike Information Criteria could be an important parallel
way to optimize the threshold selection.
The removal of artifacts by EMD was based on visual
inspection of the IMFs. For this method to be useful in real
time and/or in a clinical setting, the selection of the proper
IMFs has to be automated based also on the use of some
This paper introduces the first combination between CCA and
another method applied to a particular signal. In future work,
comparison between CCA_EMD and other combination
method could be interesting, i.e. the use of wavelet transform
instead of EMD, testing EEMD instead of EMD and the
comparison with other existing combination such as
We are presently working on the comparison of bipolar and
denoised monopolar EHG signals to investigate the
propagation of the uterine electrical activity. Preliminary
results show a clear superiority of monopolar analysis as it has
better spatial resolution and it does not bias the measure of
propagation directions as bipolar methods usually do.
Finally, we conclude that this work has opened the door for
the use of monopolar EHG recordings to investigate uterine
contractions. This will, in our opinion, have important
applications in studying the genesis of human labor and to
develop ways of predicting preterm labor.
This project is financed by the Icelandic centre for research
RANNÍS and the French National Center for University and
School (CNOUS). The authors would like to especially thank
Thora Steingrimsdottir and Ásgeir Alexandersson for their
help in the acquisition of EHG signals.
 H. Leman and C. Marque, "Rejection of the maternal
electrocardiogram in the electrohysterogram signal," IEEE Trans
Biomed Eng, vol. 47, pp. 1010-7, Aug 2000.
B. Karlsson, J. Terrien, V. Guðmundsson, T. Steingrímsdóttir, and
C. Marque, "Abdominal EHG on a 4 by 4 grid: mapping and
presenting the propagation of uterine contractions," in 11th
Mediterranean Conference on Medical and Biological Engineering
and Computing, Ljubljana, Slovenia, 2007, pp. 139-143.
M. Hassan, J. Terrien, B. Karlsson, and C. Marque, "Spatial
analysis of uterine EMG signals: evidence of increased in
synchronization with term," in Conf Proc IEEE Eng Med Biol Soc,
2009, pp. 6296-9.
M. Hassan, J. Terrien, B. Karlsson, and C. Marque, "Interactions
between Uterine EMG at Different Sites Investigated Using
Wavelet Analysis: Comparison of Pregnancy and Labor
Contractions," EURASIP Journal on Advances in Signal
Processing, vol. 2010, p. 9, 2010.
J. Terrien, T. Steingrimsdottir, B. Karlsson, and C. Marque,
"Synchronization between EMG at Different Uterine Locations
Investigated Using Time-Frequency Ridge Reconstruction:
Comparison of Pregnancy and Labor Contractions," EURASIP
Journal on Advances in Signal Processing, vol. 2010, p. 10, 2010.
P. LeVan, E. Urrestarazu, and J. Gotman, "A system for automatic
artifact removal in ictal scalp EEG based on independent
component analysis and Bayesian
Neurophysiol, vol. 117, pp. 912-27, Apr 2006.
C. J. James and O. J. Gibson, "Temporally constrained ICA: an
application to artifact rejection in electromagnetic brain signal
analysis," IEEE Trans Biomed Eng, vol. 50, pp. 1108-16, Sep
W. De Clercq, A. Vergult, B. Vanrumste, W. Van Paesschen, and
S. Van Huffel, "Canonical correlation analysis applied to remove
muscle artifacts from the electroencephalogram," IEEE Trans
Biomed Eng, vol. 53, pp. 2583-7, Dec 2006.
M. De Vos, A. Vergult, L. De Lathauwer, W. De Clercq, S. Van
Huffel, P. Dupont, A. Palmini, and W. Van Paesschen, "Canonical
decomposition of ictal scalp EEG reliably detects the seizure onset
zone," Neuroimage, vol. 37, pp. 844-54, Sep 1 2007.
N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, H. H. Shin, Q.
Zheng, N. C. Yen, C. C. Tung, and H. H. Liu, "The Empirical
Mode Decomposition and the Hilbert spectrum for nonlinear and
non-stationary time series analysis," Proc. Royal Soc. , vol. 454,
pp. 903-995, 1998.
H. Liang, Z. Lin, and R. W. McCallum, "Artifact reduction in
electrogastrogram based on empirical mode decomposition
method," Med Biol Eng Comput, vol. 38, pp. 35-41, Jan 2000.
H. Liang, Q. H. Lin, and J. D. Chen, "Application of the empirical
mode decomposition to the analysis of esophageal manometric
data in gastroesophageal reflux disease," IEEE Trans Biomed Eng,
vol. 52, pp. 1692-701, Oct 2005.
M. Blanco-Velasco, B. Weng, and K. E. Barner, "ECG signal
denoising and baseline wander correction based on the empirical
mode decomposition," Computers in Biology and Medicine, vol.
38, pp. 1-13, 2008.
D. Looney, L. Li, T. M. Rutkowski, D. P. Mandic, and A.
Cichocki, "Artifacts Removal from EEG Using EMD," in
International Conference on Cognitive Neurodynamics. ICCN,
Z. Wu and N. Huang, "Ensemble Empirical Mode Decomposition:
a Noise-Assisted Data Analysis Method," Advances in Adaptive
Data Analysis, vol. 1, pp. 1-41, 2009.
B. Mijovic, M. De Vos, I. Gligorijevic, J. Taelman, and S. Van
Huffel, "Source separation from single-channel recordings by
combining empirical-mode decomposition and independent
component analysis," IEEE Trans Biomed Eng, vol. 57, pp. 2188-
O. Friman, M. Borga, P. Lundberg, and H. Knutsson, "Exploratory
fMRI analysis by autocorrelation maximization," Neuroimage, vol.
16, pp. 454-64, Jun 2002.