IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. BME-33, NO. 12, DECEMBER 1986
Reduction of Heart Sounds from Lung Sounds by
Adaptive Filterng
VIJAY K. IYER, STUDENT MEMBER, IEEE, P. A. RAMAMOORTHY, MEMBER, IEEE,
HONG FAN, MEMBER, IEEE, AND YONGYUDH PLOYSONGSANG
Abstract-Auscultation of the chest is an attractive diagnostic method
used by physicians, owing to its simplicity and noninvasiveness. Hence,
there is interest in lung sound analysis using time and frequency do-
main techniques to increase its usefulness in diagnosis. The sounds rec-
orded or heard are, however, contaminated by incessant heart sounds
which interfere in the diagnosis based on, and analysis of, lung sounds.
A common method to minimize the effect of heart sounds is to filter the
sound with linear high-pass filters which, however, also eliminates the
overlapping spectrum of breath sounds. In this work we show how
adaptive filtering can be used to reduce heart sounds without signifi-
cantly affecting breath sounds. The technique is found to reduce the
heart sounds by 50-80 percent.
INTRODUCTION
EART sounds are an inherent interference in lung
Ilsound analysis, for the simple fact that the heart can-
not be made to cease beating. This is a problem in both
clinical auscultative interpretation and lung sounds re-
search, introducing pseudoperiodicity, masking the rele-
vant signal, and altering the energy distribution in the
spectrum. The time variance, nonlinearity, and transmis-
sion changes of the systems involved exclude direct sub-
traction of separately recorded heart sounds (during breath
holding) from the contaminated lung sounds as a solution.
The problem with linear fixed filtering is the spectral
overlap (- 50-150 Hz) of the signal and interference. The
choice of stopband edge for fixed high-pass filtering be-
comes a matter of compromise between retaining low fre-
quency energies of breath sounds and eliminating heart
sounds. Traditionally, researchers filter the sound below
some compromised and arbitrary value in the range of 50-
150 Hz with no strong scientific basis for the particular
choice [1], [2].
Adaptive filters have been used in many applications to
cancel noise and interference, and in parametric spectral
estimation. The underlying idea here is to design an adap-
tive filter based on some assumptions about the contami-
nating, contaminated, and desired signal. An estimate of
the desired signal is obtained as the'system adapts; there
are usually two phases of operation viz. a) training the
Manuscript received August 28, 1986.
V. K. Iyer, P. A. Ramamoorthy, and H. Fan are with the Department
of Electrical and Computer Engineering, College of Engineering, Univer-
sity of Cincinnati, Cincinnati, OH 45267.
Y. Ploysongsang is with the Department of Internal Medicine, College
of Medicine, University of Cincinnati, Cincinnati, OH 45267.
IEEE Log Number 8611207.
system to arrive at the filter coefficients' and b) operation
with the deduced coefficients to approximate the desired
response. However, in situations where the mechanism
producing interference is time-varying, dynamically up-
dating the coefficients to adapt to the dynamics of the time-
varying process becomes mandatory and the training and
operation become simultaneous. Such a time-varying
adaptive scheme is found to advantageously eliminate
heart sounds, in the mean square sense, from a mixture
of heart and breath sounds.
THE ADAPTIVE FILTER
Adaptive filters [3] are used in different configurations
depending on the application. For cancellation of additive
interference [3], [5], the digital implementation configu-
ration used is as shown in Fig. 1. The contaminated signal
is modeled as an additive mixture (d [n]) of the interfer-
ence (x' [n]) and the useful signal (s [n]). The interference
itself is modeled as a filtered version [filtered by H(z)] of
a reference signal (x[n]) that is available. The reference
signal is passed through the adaptive filter
Laiz-
A(z) =
J
The output of the adaptive filter (y[n]) is subtracted
from the contaminated signal to obtain the error signal
e[n]. The filters employed for adaptive filtering are almost
invariably finite impulse response (FIR) (with
bj
= 1;
j = 0,
bj
= 0; else.) because of their inherent stability
and mathematical tractability in algorithms for computa-
tion of the coefficients.
Widrow's least mean square (LMS) algorithm is used
to adjust the coefficients using e[n] as a correcting factor
and this minimizes the expected value E {e[n]2}. The ap-
plicability of the algorithm can be seen mathematically as
follows.
Widrow's LMS algorithm states
W[n + 1]
=
W[n] + KX[n] e[n]
where W[n], W[n + 1] are the column matrixes (N x 1)
of filter coefficients, before and after updating for the n
+ lth iteration, where N is the order of the adaptive FIR
filter, X[n] is the reference signal column matrix consist-
0018-9294/86/1200-1141$01.00
© 1986 IEEE
1141