ARTICLE IN PRESS

Computers in Biology and Medicine 40 (2010) 223–230

Contents lists available at ScienceDirect

Computers in Biology and Medicine

journal homepage: www.elsevier.com/locate/cbm

Uterine electromyogram database and processing function interface:

An open standard analysis platform for electrohysterogram signals
Jérémy Terrien a,!, Catherine Marque b, Jean Gondry c, Thora Steingrimsdottir d, Brynjar Karlsson a,e
a
Department of Biomedical Engineering, Reykjavik University, Kringlan 1, 103 Reykjavik, Iceland
b
UMR CNRS 6600, Biomécanique et Bioingénierie, Université de Technologie de Compie! gne, Compie! gne, France
c
Centre of Gynecology and Obstetric of Amiens, Amiens, France
d
Landspitali University Hospital/Department ob-gyn, Reykjavik, Iceland
e
University of Iceland, Institute of Physiology, Reykjavik, Iceland

a r t i c l e in fo abstract

Article history: The uterine electromyogram or electrohysterogram (EHG) is one of the most promising biophysical
Received 7 November 2008 markers of preterm labor. At this time no recording parameter standard exists for EHG recordings which
Accepted 29 November 2009 can be a problem for the establishment of international multicentric trials. In this paper, we present a
management and processing system dedicated to storing and processing EHG signals. This system can
Keywords: process EHG signals recorded in different experimental conditions i.e. different sampling frequencies.
Data management system The signal management is performed using an easy to use graphical user interface. Other available
Uterine electromyogram functions include visualization, preprocessing and analysis of EHG signals. The proposed processing
Preterm labor functions provide temporal, spectral and time-scale parameters obtained from the EHG bibliography.
Spectral analysis
The obtained results from real signals recorded in two different hospitals in two different countries are
Time-scale analysis
in accordance with the literature and demonstrate the potential of the proposed system. The
incorporation of new functions is easy, due to a standardization of the EHG data formats.
& 2009 Elsevier Ltd. All rights reserved.

1. Introduction
Premature labor is one of the most important public health
problems in Europe and other developed country as it represents
nearly 7% of all births. It is the main cause of morbidity and
mortality of newborns. Early detection of a preterm labor is
important for its prevention for example in insuring tocolytic drug
efficacy. Continuous efforts are made to find new biochemical or
biophysical markers of preterm labor threat [1]. One of the most
promising is the analysis of the electrical activity of the uterus.
Uterine electromyogram (EMG) recorded externally in women,
the so called electrohysterogram (EHG), has been proved to be
representative of uterine contractility. The analysis of such a
signal may allow the prediction of a preterm labor as soon as 28
weeks of gestation (WG) [2].
There is no EHG recording standard. The different international
research teams which work on the analysis of the EHG use very
different recording parameters. The most important differences
concern analog filter cutoff frequencies, electrode configuration
and sampling frequency. The comparison of the different results
of the literature is therefore difficult or sometimes impossible.
! Corresponding author. Tel.: + 354 599 6595; fax: + 354 599 6301.
E-mail addresses: jeremy@ru.is (J. Terrien), catherine.marque@utc.fr
(C. Marque), gondry.jean@chu-amiens.fr (J. Gondry), thoraste@lsh.is
(T. Steingrimsdottir), brynjar@ru.is (B. Karlsson).
Moreover, the analysis tools used on these signals are also very
different. During the first international workshop on pregnant
uterine smooth muscle EMG activity held in Ljubljana, Slovenia in
2006, the members of the EHG consortium highlighted that it
seems important to develop a recording standard and to create an
international EHG database, in order to increase our knowledge of
uterine contractility and to enable the use of this signal routinely
for the detection of preterm labor.
In this paper, we present software, developed under the
Matlab environment (The Mathworks, Inc.) and called EHG
Computing Engine, dedicated to the analysis of the EHG signal.
This software, associated to a MySQL database, can retrieve EHG
signals based on physiologic criteria but also by taking into
account recording constraints. The pre-processing of EHG signals
is performed automatically in order to be able to use the same
algorithms to process EHG signals recorded with different
recording parameters. The data can be managed using a graphical
user interface, so no previous knowledge of SQL is required. The
addition of other specific processing functions is also very easy by
adding Matlab scripts with standard input data format. As
example of processing functions, we propose mainly two func-
tions which have shown in the literature to be useful in the
prediction of preterm labor and in the description of the normal
evolution of the EHG characteristics during pregnancy [4–9]. No
add-on Matlab toolboxes are required to run the proposed
processing functions.

0010-4825/$ - see front matter & 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compbiomed.2009.11.019
 ARTICLE IN PRESS
224 J. Terrien et al. / Computers in Biology and Medicine 40 (2010) 223–230
Subject PhysiologicalSituation
- Subject ID 1 0 ... n
- SituationType
- Nationality
- SituationInformation
- ...
- ...
Event
- EventType
- EventNumber
- EventData
- ...
Fig. 1. Global structure of the EHG database.
2. Materials and methods
2.1. Database description
We proposed a database structure adapted to commonly used
patient and experimental information. The main tables are dedicated
to patient information, experiment description with associated
recording parameters, recorded signals and identified event signals
(Fig. 1). The EHG signal is in fact composed of contractile events like
Braxton-Hicks or Alvarez contractions but also non contractile ones
like fetal movements. One patient can be associated with a number of
physiological situations (e.g. several pregnancies). One physiological
situation can be associated with numerous experiments (repeated
measurements for the same situation). One experiment can be
associated to numerous recordings, in the case of multi channel
recording. Finally one recording can be associated with several
physiological events. The total number of tables is 12. This relatively
high number of tables is suitable for an easy modification of the tables
related to specific needs such as electrodes configuration or analog
settings (i.e. band-pass filter characteristics). The current table
structure stores the most commonly used information necessary to
identify classify and process EHG data. As a complement of Fig. 1, the
8 remaining tables and associated entries are given below:
! Placenta: Placental insertion description (anterior, posteriory)
! Research team: Name and information of the research team
(Institution name, countryy)
! Event type: Names of known EHG events (Braxton-Hicks,
Alvarez, Fetal movementy)
! Elemental electrode configuration: Names of electrode config-
urations (monopolar, bipolary)
! Electrode configuration: Description of the electrode config-
uration (Number of channels, inter electrode distancey)
! Analog settings: Parameter of the analog setting of the
acquisition device (Signal unit and filter characteristics)
! Pregnancy: Information concerning a particular pregnancy
(Date of beginning, term at deliveryy)
2.2. Database management and data retrieving
The interface between the MySQL DBMS and Matlab software
is developed using the free software MyM written in C.1 This
interface enables an SQL query to be sent from Matlab to the
1
MyMSoftware can be downloaded from http://www.mathworks.com/ma
tlabcentral/fileexchange/.
Experiment
1 0 ... n
- Analogic
Configuration
- ExperimentDate
- SamplingFrequency
- ...
0 ... n
Recording
0 ... n 1 - ChannelNumber
- ChannelName
- ChannelRecordingType
- ...
MySQL server. The database management is performed graphi-
cally under a custom made Matlab Graphical User Interface (GUI),
transformed into the appropriate SQL queries and send to the
server. The user can insert, delete or view table lines (plot for
signals) through the graphical user interface. Concerning the
addition of recordings or events into the database, another user
interface allows binary file loading from hard disk. The binary
format of the data should be indicated in a corresponding
recording table. The default format is floating point (32 bits).
A specific interface is dedicated to EHG extraction and
processing (Fig. 2). The user can choose a physiological criterion,
like the term class (pregnancy, pre-labor and labor), or for
example to make an automatic sorting of signals by weeks of
gestation (WG) or weeks before delivery (WBD). Effects of the
placenta on the EHG characteristics have been recently confirmed
[9]. The placental position during the recordings was therefore
added as a selectable criterion. The current selectable criteria are:
! The class of term or type of sorting: {Pregnancy; Pre-Labor;
Labor; Post-Partum; As group; WG; WBD}
! Placental position: {Anterior; Posterior; As group; Grouped}
‘‘As group’’ indicates that the different recordings are divided up
with respect to the associated criterion (placenta and/or class of
term). They are then processed and displayed separately.
‘‘Grouped’’ indicates that all recordings, whatever the placental
situation or class of term, will be processed and displayed in any
order.
This interface also allows the user to enter the recording
characteristics of the recordings to be included in the processing
run. Currently, these characteristics concern sampling frequency,
analog filter characteristics and recording configuration (bipolar,
monopolary). All the signals that do not fulfill these minimal
characteristics are excluded from the data set that is retrieved
from the database. The remaining signals are then automatically
filtered and down sampled, if needed before further processing, in
order to obtain uniform signals characteristics. The acquisition
parameters directly influence important signal parameters like
the temporal and spectral resolution as well as the signal to noise
ratio. In order to obtain comparable results, in a particular study,
it is important to process signals which have the same
characteristics. When running a large series or a computationally
expensive algorithm, it is also vital to be able to control the size of
the data through decimation of the signal to the minimal
sampling rate that still contains the frequency information in
the signals.
 ARTICLE IN PRESS
J. Terrien et al. / Computers in Biology and Medicine 40 (2010) 223–230
Fig. 2. The user interface for selecting the physiological criteria, recording
parameters constraints, and the choice of the processing function.
From a more global point of view, the study of the influence of
the acquisition parameters on the different studies might help to
define the optimal parameters and thus a standard in the
acquisition of the EHG which is currently lacking.
Despite of the narrow-band EHG filtering by analog and/or
digital filters, the maternal ECG and some electronic noises are
still observable in recordings due to their overlapping spectra.
Nonlinear filtering methods have been developed specifically for
removing maternal ECG and electronic noise in EHG signals
[10,11]. The method which has been proven to be more efficient
on real signals is based on an undecimated wavelet packet
transform of EHG events [11]. This algorithm is applied on all the
signals retrieved from the database before further analysis.
These preprocessing steps result in a set of signals that have
the same filter and sampling characteristics and fulfill a number
of criteria, both for physiological conditions and measurement
parameters. The chosen processing function is then applied to this
set of signals. The user can request the total number of
contraction events that fulfill the different conditions before the
processing stage.
2.3. Signal processing functions
In principle any signal processing function that can be
implemented in Matlab can be applied to the signals that are
stored in and retrieved from the proposed database structure.
Several processing methods have been applied to EHG signals in
the past in order either to differentiate normal from pathological
evolution of the uterine contractility during gestation, or to make
a prediction of preterm labor. The most common extracted
parameters found in the literature are derived from the temporal
and frequency domains, but more recently from the time-
frequency or time-scale domain [5,7,12].
As an example, and to illustrate the processing part of our
database, we implemented classic functions for the calculation of
common EHG temporal parameters like energy, signal duration
etc. The possibilities of the proposed system is however not
limited to these parameters or processing. In addition, we
implemented EHG spectrum estimation functions based on
Fourier transform and continuous wavelet transform (frequency
marginal). We used Welch’s periodogram and complex Morlet
wavelet (default central frequency o0 = 2p) for calculation of the
EHG scalogram (squared modulus of the continuous wavelet
transform). These methods have been used to predict preterm
labor [4,5,7,12]. The different results of the literature show mainly
225
Table 1
Recording parameters of the different EHG recordings.
Device (Hz) French Icelandic
H–P frequency 0.2 0
L–P frequency 8 100
Sampling frequency 16 200
H–P frequency: high pass cutoff frequency, L–P frequency: low pass cutoff
frequency.
a shift of the energy of the EHG toward high frequency along
gestation [3,4,8,9] but also in response to chemical agents that
increase uterine contractility [13]. The extraction/parameteriza-
tion of the local maxima or ridges of the EHG scalogram have been
successfully used in both pregnancy characterization and preterm
labor prediction [5,12]. The extraction of the scalogram ridges by
the method developed by Leman and Marque [14] is therefore
implemented. The extracted ridges are plotted and saved on disk
for further analysis.
The different parameters used by these methods can be
changed by the user through different GUI. The results are
displayed automatically according to the various physiological
criteria defined by the user, as tables, graphics and charts. The
results are also stored on disk in both Matlab and text format for
post-processing.
2.4. Real signals
In order to demonstrate and to test the functionality of the
proposed system, several recordings previously recorded with
two different devices, were stored in the database and then
process by the proposed system. The measurements were
performed at the Landspitali University hospital in Iceland and
at the Centre of Gynecology and Obstetric of Amiens in France,
using protocols approved by the relevant ethical committee of
each country.
After obtaining informed consent of the patient, the skin was
carefully prepared using an abrasive paste and alcoholic solution.
After that, two electrodes were placed on the abdominal wall
of the women (interelectrode distance: 2.1 cm), always on the
uterine median axis, midway between the fundus and the
symphysis. Reference electrodes were placed on the hip of
the women. The recording characteristics of the two different
devices used by each research team are summarized Table 1. We
used 126 EHG signals, recorded on 25 women (30.8 73.8 WG).
The recording constraints used for pre-processing, prior to
processing contraction events, correspond to the characteristics
of the French device that are more restrictive than the ones of the
Icelandic device.
3. Processing examples
The pre-processing stage correctly retrieve, sort and trans-
formed the EHG signals to fit the common signal parameters
defined by the user. The standardization of the signal character-
istics allowed identical processing of all the signals and ensured
the comparability of the results in spite of the different recording
devices used.
The evolution of the mean duration of contractile events
according to weeks of gestation at recording does not demon-
strate a particular pattern (Fig. 3). The only trend that could be
noticed is a decrease in the contractile EHG length as a function of
weeks before delivery as expected from the literature (Fig. 4) [15].
Indeed, it is well known that pregnancy contractions are mainly
 ARTICLE IN PRESS
226 J. Terrien et al. / Computers in Biology and Medicine 40 (2010) 223–230

160 pregnancy (Fig. 7). The same analyses, but on contractions sorted
by weeks before delivery, WBD (Fig. 8), shows the same patterns
140 for both FWH and FWL as in Fig. 6. The evolution of FWH as a
function of WBD is however more pronounced. The relationship
between WBD and the frequency position of FWH, obtained after
120
post-processing of the results, is also more clearly identifiable
(Fig. 9). The frequency position of the maximal energy of FWH
Signal duration (s)

100 seems to be a good indicator of imminence of delivery in women.

The high correlation coefficient obtained in the analysis of
80 contractions sorted by WBD seems to indicate that all women
present a similar increase in the frequency position of FWH until
delivery. The analysis of a global evolution of this parameter as a
60
function of weeks of gestation is however corrupted by the
different terms at delivery (i.e. preterm/term labor). Relationship
40 between the frequency content of the EHG and the gestational age
at the recording time has already been described in the literature
20 [3,4]. The nature of this relationship depends on the study. The
linear relation is presented here mainly to give an example of the
0 possible post-processing of our system results by ordinary
24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 analysis software.
WG

Fig. 3. Evolution of the mean duration of contractile events in seconds (plus 4. System usability scenarios
standard deviation, cross) sorted by weeks of gestation (WG).
The results presented here have been obtained on signals
recorded in two hospitals in two countries. Even if it is not a true
180 multicenter trial, the results are encouraging and give an idea of
the potential of the proposed system. To go beyond simply
160 demonstrating that the system works, we reviewed some
considerations that we think helpful for researchers in the field
140 that want to demonstrate the potential usefulness of routine of
EHG recording for pregnancy monitoring.
120
Signal duration (s)

100 4.1. Acquisition parameters considerations

80 In order to share and process EHG data coming from different

60
40
20
0 frequency components, i.e. FWL and FWH. All the teams use
9 8 7 6 5 4
WBD
Fig. 4. Evolution of the mean duration of contractile events in seconds (plus
standard deviation, cross) sorted in weeks before delivery (WBD).
polyphasic, i.e. composed of several electrical bursts, and labor
contractions mainly monophasic, i.e. composed on only one short
electrical burst, therefore leading to longer contraction burst
duration. This change may be due to an increase in the
propagation of the electrical activity of the uterus with the
advance of gestation.
The EHG is mainly composed of two frequency components
called FWH (fast wave high) and FWL (fast wave low) [15]. FWL is
situated in the low frequencies and FWH in the higher frequencies
(Fig. 5). The analysis of the EHG signal scalogram, shows an
increase in the frequency position and energy of FWH with the
advance of gestation, while FWL stays in very low frequencies
(Fig. 6). We then analyzed the linear relationship, using Excel,
between the frequency position of FWH and the gestational age at
recording. The frequency position of the peak of maximal energy
of FWH increases along gestation and reaches 0.5 Hz in late
sources they have to be transformed to the same acquisition
parameters which may be impossible for particular initial
parameters. We reviewed the acquisition parameters of the main
international research teams working in the field Table 2. We can
divide these parameters in two groups according to their spectral
content. As explain previously the EHG is composed of two
3 2 1 0 acquisition parameters that include the frequency range of the
FWH component. Only the French, Dutch and Icelandic teams use
parameters that also allow the observation of FWL. FWH may be
related to the excitability of the uterus. It has been successfully
used form the prediction of preterm labor on woman [4,7,8,17].
Our system is able to transform all the signals of these teams to
the more restrictive acquisition parameters and can thus help to
confirm the previously published results with a bigger signal
population by performing what can vaguely be termed meta-
analysis using data from many centers.
FWL may be related to the propagation of the electrical activity
of the uterus. This frequency component is situated in very low
frequencies. It has been shown that this frequency component
evolves in a circadian manner and carries thus specific informa-
tion of the uterine contractility that varies in a similar manner
[20]. A simultaneous recording of internal and external EHG also
shown that FWL is recordable externally even if it is modified by
the low pass filtering effect of the conductor volume [21]. We
believe in the usefulness of FWL in the analysis of the uterine
contractility and thus suggest using a high pass frequency low
enough to be able record this frequency component. Concerning
the sampling frequency, it influenced the temporal resolution of
the signal. A high temporal resolution is particularly useful for
 ARTICLE IN PRESS
J. Terrien et al. / Computers in Biology and Medicine 40 (2010) 223–230 227

0.05

A.U.
-0.05
0 20 40 60 80 100
time (s)

0.04
FWL FWH

0.02

0
0 0.5 1 1.5
Hz

Fig. 5. Example of a contractile event (upper panel) and associated relative energy spectrum (lower panel).

4.2. Clinical considerations

In the context of pregnancy monitoring, our system might help

the research community to define the best algorithm suitable for
preterm labor prediction but also identifying the important
parameters that influence the diagnosis.
Several algorithms have been proposed to predict preterm
0.3 labor with potentially good results. It is however difficult to
compare the performances of these methods since the results
0.2 have been obtained on different populations and more impor-
20
tantly with different acquisition parameters. The possibility to
compare the results of the different algorithms on the same
0.1 25
database of signals with homogenous acquisition parameters will
30 ensure an objective comparison of methods. The performance of a
WG

0 new algorithm will be easily compared against published ones.

0 0.2 35
0.4 0.6 0.8 1 as the prediction of onset of epileptic attacks. The establishment
1.2
Hz
Fig. 6. Evolution of the mean relative energy spectrum, calculated on scalogram of
contractile events (n= 110), along gestation (WG: week of gestation). The evolution
of the position of FWH is emphasized by the arrow.
time delay estimations between two or more EHG channel in a
propagation speed analysis. Even if several studies have tried to
estimate this propagation speed, the usefulness of this parameter
for preterm labor prediction has never been clarified. The
propagation of the electrical activity of the uterus is most likely
an important parameter for the efficiency of contractions and
might thus be an interesting parameter for labor prediction.
From a general point of view, we think that when we consider
the huge underlying public health problem, no part of the signal
can be discarded until it has been shown that it is not useful,
based on a larger database that can normally be obtained in a
single center. We therefore suggest to use a very low high-pass
frequency filter for signal acquisition and a sufficiently high
sampling frequency in order to obtain a good temporal resolution.
The ability of our system to transform the data to ‘‘virtual’’
acquisition parameters will help to define the optimal parameters
for this kind of measurements.
This approach is already successfully used for others signals, such
of scientific competitions, like the BCI competition for EEG
analysis, might also increase the chances of the development
and publishing of relevant methods by researchers who are not
directly concerned by EHG processing but have expertise in pure
signal analysis or decision making problems.
The other aspect that we think important is the influences of
clinical parameters on the analysis of EHG for pregnancy
monitoring. Research questions of this type include ‘‘what is the
minimal term at which good EHG recording can be realized?’’, ‘‘is
EHG monitoring suitable for high body mass index patient?’’,
‘‘does placenta position affect prediction performances?’’y These
questions have to be answered before EHG monitoring can be
used as a routine technique in a clinical setting and they can only
be solved by a statistical evaluation on a large database.
5. Discussion
The software presented in this work is able to manage EHG
signals recorded using different setups, devices and configurations.
The signal processing functions can automatically be applied
to all the signals in the associated database that fulfill predefined
physiological criteria, i.e. placental position, and recording
 ARTICLE IN PRESS
228 J. Terrien et al. / Computers in Biology and Medicine 40 (2010) 223–230

0.65

0.6

0.55

0.5

0.45
Hz

0.4

0.35

0.3

0.25

0.2
22 24 26 28 30 32 34 36 38 40
WG

Fig. 7. Evolution of the mean FWH peak frequency (PFFWH) calculated on scalogram of contractile events along gestation (WG: week of gestation). Equation of the fitting
line: PFFWH = 0.0106 WG+ 0.0937 Hz, R2 = 0.71.

group. The possibility of visualizing the individual results for

one recording or a sub-group of recordings, amongst the
population, could be a great benefit in the detection of outliers
or for the analysis of the effects of a particular treatment (i.e.
tocolytic drugs). A definition of any other grouping criteria
could also be helpful.
! Multi scale analysis: When the results are displayed as a
0.2 function of time, they are averaged within classes of one week
(i.e. week of gestation or week before delivery). It is known
0.1 that uterine contractility presents circadian and perhaps
0 ultradian patterns [20,22]. To evaluate a new drug for example,
0 it is common to follow its effects for several hours before and
after administration. Multi time scale functionality could be
0 5
useful in those cases. The present visualization methods can
0.5 however give relevant information about a new processing
BD

10
H

algorithm or parameter and its relevance in the analysis of

W
z

1 uterine contractility during gestation in a quick and easy way.

Fig. 8. Evolution of the mean relative energy spectrum calculated on scalogram of
contractile events (n =55) sorted by weeks before delivery (WBD). The evolution of
the position of FWH is emphasized by the arrow.
constraints i.e. minimal sampling frequency. A graphical user
interface permits the user to consult and modify the database
without knowledge of SQL. To illustrate the signal processing
capability of the software, temporal and spectral parameters
estimation functions were added that do not require specific add-
on toolboxes for Matlab. These functions, commonly used for EHG
processing, allowed us to easily obtain results similar to those
described in the literature, from a series of EHG recordings
obtained from two different places with very different acquisition
parameters. The use of Matlab scripts for the processing offers an
excellent customizability of the software on top of the advantages
of a standardized storage format and basic processing.
The current version of the proposed system has several
limitations but do not reduce it’s usefulness in many usual
applications. These limitations concern:
! Supplementary data, individual data processing: At this time,
the different results for each individual are averaged in each
! Statistical comparisons: The results of the different processing
function are currently only visualized on screen and stored on
disk for further analysis. Addition of a statistical module can
show directly the presence of a significant behavior, or
differences in different physiological situations. An exhaustive
implementation of different statistical tests seems however
unrealistic at this time and are too far from the scope of the
present software.
Other functionalities of this software will be added by the
specific needs or suggestions of the different future users of the
uterine EMG community. The proposed database structure could
also be a base for the future International EHG database.
6. Conclusion
There is currently no recording standard for EHG signals. Each
research team uses its own recording parameters and devices.
Before proposing a recording standard, it could be useful to
benefit from EHG signals already recorded, in order to increase
our knowledge on the uterine contractility mechanism and to find
pertinent parameters for the prediction of preterm labor in
humans. In many other fields of research in biomedicine,
 ARTICLE IN PRESS
J. Terrien et al. / Computers in Biology and Medicine 40 (2010) 223–230
9 8 7 6
Fig. 9. Evolution of the mean FWH peak frequency (PFFWH) calculated on scalogram of contractile events sorted in weeks before delivery (WBD). Equation of the fitting line:
PFFWH = " 0.0224 WBD + 0.5086, R2 =0.95.
Table 2
Recording parameters of the main research teams in the EHG field.
Team H–P L–P
frequency frequency
(Hz) (Hz)
Icelandic [16] 0 100
French [7] 0.02 8 16
American [17] 0.3 50
Dutch [18] 0.01 75
Slovenian [19] 0.1 4 20
H–P frequency: high pass cutoff frequency, L–P frequency: low pass cutoff
frequency.
dedicated processing toolboxes have become standards and now
include complete processing environments and databases con-
tributed by scientists and research groups worldwide. We think
that our system can become a valuable tool in evaluating new
EHG processing tools for clinical purposes. Researchers that
develop EHG analysis algorithms can then make these algorithms
public by using our system, for verification and evaluation by
other teams on other clinical data.
We have shown that the proposed software is suitable for
automatic processing of EHG signals based on an automatic
filtering of the signals, in a MySQL database based on physiolo-
gical criteria and recording parameter constraints. It enables the
processing of EHG signals obtained with different recording
protocols or devices. This software and associated database
structure can allow research teams to store and process, in an
easy way, EHG related data until an International recording
standard and database are adopted by the international EHG
researcher community. Concurrently, research groups which
already work with Matlab could add their own processing tools
and easily test them on a much larger number of contractions
representing a wider range of physiological situations. In this
work, we have proposed well known processing functions which
have been extensively applied. We have shown that our toolbox
easily reproduces well known results concerning the evolution of
uterine contractility during gestation until labor.
The proposed system is freely available under the GNU public
license for noncommercial use and open source development
(http://ehg.ru.is/EHGCompEngine.htm).
229
0.65
0.6
0.55
0.5
0.45
Hz
0.4
0.35
0.3
0.25
0.2
5 4 3 2 1 0
WBD
Installation instructions, documentations, functions to gener-
ate synthetic data, and several real EHG data are also available on
the web site.
Sampling
frequency
(Hz)
7. Summary
200
200
The uterine electromyogram or electrohysterogram (EHG) is
1000 one of the most promising biophysical markers of preterm labor.
At this time no recording parameter standard exists for EHG
recordings which can be a problem for the establishment of
international multi-centric trials. In this paper, we present a
management and processing system dedicated to storing and
processing EHG signals.
Research into the use of EHG for pregnancy monitoring has
been increasing in the last few decades. EHG researchers have
acknowledged in an international workshop the need for a
recording standard as it exists for other biomedical signals like
electroencephalogram or electrocardiogram for example. The
proposed system will allow each EHG research team to share or
simply use their signal database with automatic processing of the
data according to specific requirements i.e. old recording or future
standard acquisition parameters. Another aspect, for which our
system was designed, is to facilitate research into the clinical use
of EHG as a marker of preterm labor. In spite of numerous
potentially clinically relevant EHG parameters, the EHG signal is
currently not used in clinical practice. Validation in a large scale
multi-centric study of the parameters proposed in the literature is
needed to ascertain their relevance for obstetrical monitoring. For
that purpose, well known EHG processing methods were added to
the system in order to automatically process the signals of the
associated database. The methods already implemented are only a
sample of the methods found in the bibliography of the field. The
system allows the addition of any processing method, new or
already published, for its evaluation on the database.
In conclusion, the proposed system is designed to help the
research community interested in uterine contractility, as well as
helping research into clinical usefulness, by providing easy to use
management interfaces and automatic processing methods. We
demonstrated the potential of the proposed system by reprodu-
cing classical results obtained from signal databases of two
different research teams.
 ARTICLE IN PRESS
230 J. Terrien et al. / Computers in Biology and Medicine 40 (2010) 223–230
Conflict of interest statement
None declared.
Acknowledgement
This work was supported by the Icelandic Center for Research
RANNIS.
References
[1] SAFE Network: /http://www.safenoe.org/cocoon/safeorg/info/public/index.
xhtmlS.
[2] C. Marque, J. Terrien, S. Rihana, G. Germain, Preterm labour detection by use
of a biophysical marker: the uterine electrical activity, BMC Pregnancy
Childbirth 7 (Supplementary 1) (2007) 1.
[3] R.E. Garfield, W.L. Maner, L.B. MacKay, et al., Comparing uterine electro-
myography activity of antepartum patients versus term labor patients, Am. J.
Obstet. Gynecol. 193 (2005) 23–29.
[4] W.L. Maner, R.E. Garfield, H. Maul, et al., Predicting term and preterm delivery
with transabdominal uterine electromyography, Obstet. Gynecol. 101 (2003)
1254–1260.
[5] M.L. Voisine, C. Marque, H. Leman, J. Gondry, Analysis of uterine electro-
myogram for the monitoring of preterm birth. Instrumentation development
and data analysis, in: Fourth BSI International Workshop, Como, Italie, 2002.
[6] H. Eswaran, J.D. Wilson, P. Murphy, et al., Application of wavelet transform to
uterine electromyographic signals recorded using abdominal surface electro-
des, J. Matern. Fetal. Neonatal. Med. 11 (2002) 158–166.
[7] H. Leman, C. Marque, J. Gondry, Use of the electrohysterogram signal for
characterization of contractions during pregnancy, IEEE Trans. Biomed. Eng.
46 (1999) 1222–1229.
[8] R.E. Garfield, G. Saade, C. Buhimschi, et al., Control and assessment of the
uterus and cervix during pregnancy and labour, Hum. Reprod. Update 4
(1998) 673–695.
[9] J. Terrien, C. Marque, G. Germain, Study of the local influence of the placenta
on the characteristics of monkey uterine electrical activity, in: 25th Annual
International Conference of the IEEE MBS, Cancun Mexico, 2003.
[10] P. Carre, H. Leman, C. Fernandez, C. Marque, Denoising of the uterine EHG by
an undecimated wavelet transform, IEEE Trans. Biomed. Eng. 45 (1998)
1104–1113.
[11] H. Leman, C. Marque, Rejection of the maternal electrocardiogram in the
electrohysterogram signal, IEEE Trans. Biomed. Eng. 47 (2000) 1010–1017.
[12] J. Terrien, C. Marque, B. Karlsson, Spectral characterization of human EHG
frequency components based on the extraction and reconstruction of the
ridges in the scalogram, in: 29th Annual International Conference of the IEEE
EMBS, Lyon, France, 2007.
[13] M. Doret, R. Bukowski, M. Longo, et al., Uterine electromyography
characteristics for early diagnosis of mifepristone-induced preterm labor,
Obstet. Gynecol. 105 (2005) 822–830.
[14] H. Leman, C. Marque, Ridge Extraction from the Scalogram of the Uterine
Electromyogram, TFTS, Pittsburg, USA, 1998.
[15] D. Devedeux, C. Marque, S. Mansour, et al., Uterine electromyography: a
critical review, Am. J. Obstet. Gynecol. 169 (1993) 1636–1653.
[16] B. Karlsson, J. Terrien, V. Gujmundsson, et al. 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.
[17] R.E. Garfield, H. Maul, W. Maner, et al., Uterine electromyography and light-
induced fluorescence in the management of term and preterm labor, J. Soc.
Gynecol. Investig. 9 (2002) 265–275.
[18] C. Rabotti, M. Mischi, J. van Laar, et al. Myometrium electromechanical
modeling for internal uterine pressure estimation by electrohysterography,
in: 31st Annual Conference of the IEEE EMBS, Minneapolis, Minnesota, 2009.
[19] I. Verdenik, M. Pajntar, B. Leskosek, Uterine electrical activity as predictor of
preterm birth in women with preterm contractions, Eur. J. Obstet. Gynecol.
Reprod. Biol. 95 (2001) 149–153.
[20] J. Terrien, C. Marque, G. Germain, Circadian Rhythms of the Spectral
Characteristics of Uterine EMG: A Case Study, ICCAC, Antalya, Turkey, 2005.
[21] S. Mansour, D. Devedeux, G. Germain, et al., Uterine EMG spectral analysis
and relationship to mechanical activity in pregnant monkeys, Med. Biol. Eng.
Comput. 34 (1996) 115–121.
[22] A.M. Germain, G.J. Valenzuela, M. Ivankovic, et al., Relationship of circadian
rhythms of uterine activity with term and preterm delivery, Am. J. Obstet.
Gynecol. 168 (1993) 1271–1277.