![](bg8.png)
4 CONCLUSIONS
The existing methods for the analysis of EMG-signals
are not precise enough for a reliable prediction of
the point of delivery. The new approach presented
in this contribution is based on the distinction be-
tween pulses (muscular contraction) and flat line sec-
tions during relaxation. A time-frequncy analysis re-
veals that only the pulses contain relevant information
whereas the flat line sections can be neglected. For
this reason, a semi-automatic pulse detection is de-
veloped. The physician controls the functionality of
the pulse detection by adapting two comprehensible
parameters, while the time-consuming work of pulse
extraction is done automatically. The first parame-
ter, the level value, influences the number of extracted
pulses, whereas the second parameter, the area value,
determines the length of the pulses. Therefore, the
physician integrates his current observations as well
as his medical experiences into the pulse detection.
The use of surface electrodes leads to deforma-
tions in the individual pulse shapes. In this ap-
proach, the pulses extracted by the pulse detection
are treated as realisations of a stationary stochastic
process. In order to derive a generalized pattern, a
stochastic analysis, the Karhunen-Lo`eve-Transform
(KLT), is carried out. The KLT is based on the eigen-
value/eigenvector problem of the covariance matrix.
While an eigenvector represents a generalised pattern,
the corresponding eigenvalue specifies the degree of
similarity with regard to the pulse ensemble. Eigen-
value and eigenvector yield to the eigenform. The
more dominant the first eigenform is, the better it rep-
resents the pulses of the ensemble.
Until now, the mean frequency has been used for
the prediction of the point of delivery. Although the
pulse detection reduces the frequency deviation sig-
nificantly, the mean frequency remains sensitive to
variations of the pulse detection’s parameters because
the individual pulses are randomly distorted by con-
ductivity effects. The first eigenform of the KLT is
less susceptible to parameter variations. Particular
in case of a dominant first eigenform, the mean fre-
quency becomes a reliable criterion.
Furthermore, a new characteristic value is devel-
oped: the symmetry value. It is derived from the
cross-correlation function of the first eigenforms of
the left and right EMG-channel. If the quality of
the electromyogram is high, the pulse ensembles of
the left and right channel will yield to quite identical
eigenforms and a symmetry value close to 100%. To-
gether with the eigenvalues of the KLT, the symmetry
value serves as a criterion for the measurement’s reli-
ability.
In the future, the pulse detection combined with
the stochastic analysis will be applied on a sufficiently
large amount of electromyograms taken from various
women during the last period of pregnancy. With
these results, the reliability of this new approach as
well as the improvement with regards to the present
methods will be be quantified.
REFERENCES
Jolliffe, I. T. (2002). Principal Component Analysis.
Springer, New York, 2nd edition.
Maul, H., Maner, W. L., Olson, G., Saade, G. R., and
Garfield, R. E. (2004). Non-invasive transabdominal
uterine electromyography correlates with the strength
of intrauterine pressure and is predictive of labor
and delivery. In The Journal of Maternal-Fetal and
Neonatal Medicine. Parthenon Publishing.
Mertins, A. (1999). Signal analysis: wavelets, filter banks,
time-frequency transforms and applications. Wiley,
Chichester, 2nd edition.
Oppenheim, A. and Schafer, R. W. (1999). Discrete-time
signal processing. Prentice Hall, Upper Saddle River,
NJ, 2nd edition.
Reicke, L., Kaiser, I., and Kroeger, M. (2006). Identifi-
cation of the running-state of railway wheelsets. In
ISMA2006, International Conference on Noise & Vi-
bration Engineering. Katholieke Universiteit Leuven,
Department of Mechanical Engineering.
BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing
384