Introduction - matlab expo 2016

Separation between fetal and maternal heart sound
Jean-Michelle ENTENA - University of Applied Sciences
Supervisors: FH-Prof. DI Dr. Lars MEHNEN, FH-Prof. DI Dr. Johannes MARTINEK
Introduction
The sound of the heart can provide useful information for heart
diagnosis. Cardiologists can for example detect heart murmurs
or the localization of the heart damage by listening to the heart
sound. Therefore, the inclusion of the analysis of heart sounds is
becoming more and more popular when monitoring the well-being
of the fetus.
However, frequency overlapping of the fetal heart sound with internal interference signals such as the maternal heart sound occurs.
The aim of this thesis is to develop a statistical “tool” and to test
whether it is feasible to achieve a clear separation of the two different heart signals using the Dictaphone (Model: Olympus VN8600PC) and two stereo stethoscopes with integrated microphones (Model: omni solder pad 6mm, Stock nr. 724-3131, brand: RS).
Furthermore, it is of particular interest, how accurate the obtained
results are, when using the FastICA algorithm by Hyvärinen.
Materials & Methods
In this section a short introduction to the
measurement set-up and the methods
used in this feasibility study is given.
In Fig.1 the components of the measurement chain can be seen: test subject,
two stethoscopes, two microphones and
a Dictaphone.
Real-time recordings of the maternal and
fetal heart sounds from 8 probands between the 25th and 38th gestational week
are used for the analysis in MATLAB.
To achieve a clear separation of the two
heart sounds Independent Component
Analysis (ICA) is performed with the FastICA algorithm by Hyvärinen et al.
As two stethoscopes are provided a multi
-channel Independent Component Analysis (ICA) is performed.
Independent Component Analysis (ICA)
The ICA is based on the idea that a separation of a mixed signal can be achieved,
when the original sources are independent
and have non-gaussian distribution.
Several statistical methods (envelope detection (see Fig.2), Fast Fourier Transform
(FFT) (see Fig.3), autocorrelation, correlation and Wavelet Transform) are used to
segment and analyze the outcome of ICA.
Fig.2: Extraction of envelope curves from proband4; title - number of channel, x-axis
- time in sec, y-axis - amplitude.
The purpose of the envelope detection was to eliminate the strong variances of the
audio signals and to capture just the shape of the signals recorded.
Fig.1: Measurement chain: test subject, two stethoscopes, integrated microphones, Dictaphone
Results
The main finding of this work is that there are indications for the detection of fetal
heart sounds with the low-cost phonocardiogram device; distinctive peaks were
observed in the relevant frequency bands
of the FFT-analysis (see Fig.3). However,
the results also show that the signals are
still contaminated with a lot of noise and
overlapping with maternal heart sounds
might still exist.
Fig.3: Extraction of FFT-results of envelope curves from proband4; title- number
of channel, x-axis - frequency in Hz, y-axis - amplitude in power. The relevant frequency band lies approx. between 0.5 and 3.5Hz. The frequency 0 is excluded in
the graphs as noise accumulates at this frequency.
References
A. Hyvärinen et al. Independent Component Analysis: Algorithms and Applications. Neural Networks, 13(4-5):411-430, 2000.
Discussion
The detection of the fetal heart sounds
could not be assumed in all probands.
Only in one proband out of 8 this assumption could be made. The analysis of the
features of the probands shows that the
fat tissue on the abdomen of the proband
is probably one main feature that influences the detection of the fetal heart sounds.
Additionally, the postioning of the stethoscopes might represent another important
factor with regard to the detection of the
fetal heart sounds and should be investigated in more detail.
Other aspects, which could be improved
in further studies with the aim of separating the fetal and maternal heart sounds
with a low-cost phonocardiogram device,
are as follows:
* inclusion of more probands to enable to
design an own mother wavelet
* recording of addtional bio-signals such
as fECG or ultrasound Dopler as a reference signal