Chapter 7

ECG Signal Processing

Electrocardiographic analysis was one of the very first areas in medicine

where computer processing was introduced [1-3]. Early work mostly dealt
with the development of decision tree logic for ECG interpretation, mim-
icking the rules a cardiologist would apply. It soon became quite evident,
however, that the outcome of computer interpretation was critically depen-
dent on the accuracy of the measurements. As a result, the role of signal
processing has become increasingly important in producing accurate mea-
surements, especially when analyzing ECGs recorded under ambulatory or
strenuous conditions. In addition, theoretical advances in signal processing
have contributed significantly to a new understanding of the ECG signal
and, in particular, its dynamic properties.
So far, no system offers a "universal" type of ECG signal analysis, but
systems are designed to process signals recorded under particular conditions.
It is, therefore, customary to speak of systems for resting ECG interpreta-
tion, stress testing, ambulatory ECG monitoring, intensive care monitoring,
and so on. Common to all these systems is a set of algorithms which con-
dition the signal with respect to different types of noise, extract basic ECG
measurements of wave amplitudes and durations, and compress the data for
efficient storage or transmission. The block diagram in Figure 7.1 presents
this set of signal processing algorithms, i.e., filtering for noise reduction,
QRS detection, wave delineation, and data compression; their respective de-
scriptions define the scope of the present chapter. While these algorithms
are frequently implemented to operate in sequential order, information on
the occurrence time of a QRS complex, as produced by the QRS detector, is
sometimes incorporated into the other algorithms to improve performance.
The complexity of each algorithm varies from application to application so
that, for example, noise filtering performed in ambulatory monitoring is
much more sophisticated than that required in resting ECG analysis.

453
454 Chapter 7. ECG Signal Processing

Figure 7.1: Algorithms for basic ECG signal processing. The timing information
produced by the QRS detector may be fed to the blocks for noise filtering and data
compression (indicated by gray arrows) to improve their respective performance.
The output of the upper branch is the conditioned ECG signal and related temporal
information, including the occurrence time of each heartbeat and the onset and end
of each wave.

Before describing the details of the algorithms for basic ECG signal pro-
cessing, we will provide a brief introduction on their use and the analysis
which often follows.

E C G filtering. Considerable attention has been paid to the design of

filters, which may have linear or nonlinear structures, for the removal of
baseline wander and powerline interference, see Sections 7.1 and 7.2, respec-
tively. Both these types of disturbance imply the design of a narrowband
filter. Removal of noise due to muscle activity represents another important
filtering problem being much more difficult to handle because of the substan-
tim spectral overlap between the ECG and muscle noise (Section 7.3). This
circumstance is identical to the situation where the EEG signal is disturbed
by muscle noise, see page 74. In contrast to the EEG, muscle noise present
in the ECG can be reduced whenever it is appropriate to employ techniques
that benefit from the fact that the ECG is a recurrent signal. For example,
ensemble averaging techniques used for noise reduction of evoked potentials,
described in Chapter 4, can be successfully applied to time-aligned heart-
beats for reduction of muscle noise.
The filtering techniques described in Sections 7.1-7.3 are primarily used
for preprocessing of the signal and have as such been implemented in a wide
variety of systems for ECG analysis. It should be remembered, however, that
filtering of the ECG, as with any other type of biomedical signal, is contex-
tual and should be performed only when the desired information remains
undistorted. This important insight may be exemplified by filtering for the
removal of powerline interference. Such filtering is suitable in a system for
 455

the analysis of heart rate variability, whereas it is inappropriate in a system

for the analysis of late potentials, as late potentials spectrally overlap the
interference.
A major concern when filtering out noise is the degree to which the QRS
complexes influence the output of the filter. The QRS complex acts, in fact,
as an unwanted, large-amplitude impulse input to the filter. Since linear,
time-invariant filters are generally more sensitive to the presence of such
impulses, filters with a nonlinear structure may be preferable. In order to
assure that a filter does not introduce unacceptable distortion, its perfor-
mance should be assessed by means of simulated signals so that distortion
can be exactly quantified.

Q R S d e t e c t i o n . The presence of a QRS complex and its occurrence time

is basic information required in all types of ECG signal processing. The
design of a QRS detector, described in Section 7.4, is of crucial importance
since poor detection performance may propagate to subsequent processing
steps and, consequently, may limit the overall performance of the system.
Beats which remain undetected constitute a more severe error than do false
detections; the former type of error can be difficult to correct at a later
stage in the chain of processing algorithms, while, hopefully, false detections
can be eliminated by, for example, an algorithm for classification of QRS
morphologies.
Once the QRS complex has been detected, the T wave can be analyzed
since ventricular repolarization always follows depolarization. Conversely,
the P wave does not lend itself as easily to analysis since atrial and ven-
tricular rhythms may be independent of each other. In the vast majority of
cases, however, atrial and ventricular rhythms are associated so that P wave
detection may be based on a backward search in time beginning at the QRS
complex and ending at the end of the preceding T wave. The success rate of
the P wave detector is strongly dependent on the noise level of the ECG. As
a result, comprehensive rhythm interpretation, which assumes the availabil-
ity of P wave information, is precluded from those applications where the
ECG signal is relatively noisy. Further aspects of the problem of detecting
P waves can be found in, e.g., [4-8].

W a v e d e l i n e a t i o n . Since essential diagnostic information is contained in

the wave amplitudes and durations of a heartbeat, cf. Figure 6.10, wave delin-
eation represents an important step in ECG signal processing (Section 7.5).
The design of such delineation algorithms continues to receive attention with
the all-embracing goal of elegantly handling the fact that the signal ampli-
tude is low at the wave boundaries and often obscured by noise waves. The
456 Chapter 7. ECG Signal Processing

lack of universally acknowledged rules for finding the onset and end of ECG
waves is another factor that makes the design process challenging. Once
the onset and end of a wave have been determined, its duration and peak
amplitude can be readily computed.

D a t a c o m p r e s s i o n . The ECG signal exhibits a certain amount of redun-

dancy, as manifested by correlation between adjacent samples, the recurrence
of heartbeats with similar morphology, and the relative resemblance between
different leads. Considerable savings can be achieved in terms of storage ca-
pacity and transmission time by exploiting the different types of redundancy
so that each sample can be represented by fewer bits than in the original
signal. The use of data compression is, however, only acceptable as long
as the desired diagnostic information is preserved in the reconstructed sig-
nal. The major approaches to designing methods for ECG data compression
are presented in Section 7.6, together with various considerations on how to
evaluate performance.

F u r t h e r a n a l y s i s . The above aspects of basic ECG signal processing

are usually accompanied by further analysis dealing with morphology and
rhythm. Feature extraction is performed for the purpose of characterizing
the morphology of a QRS complex. Although the durations and amplitudes
that result from wave delineation contain important diagnostic information,
additional features are required to reliably group beats with similar mor-
phology into the same cluster. One approach to feature extraction is to
derive a set of "heuristic" features which, e.g., describe the area, polarity,
and slopes of the waves. Another, more robust, approach is to make use of
the coefficients that result from the correlation of each beat with either a
set of orthonormal basis functions or a set of QRS templates, being either
predefined or created dynamically during the analysis.
Based on the set of extracted features, clustering of QRS morphologies
can be performed. In its simplest form, clustering may be used to single
out beats that deviate from the predominant morphology, which is usually
that belonging to the normal sinus beat. Once this is done, beats belonging
to the "sinus cluster" can be subjected, for example, to ensemble averaging
or heart rate variability analysis. In other situations, there is reason to
study the entire range of beat clusters. Since clustering does not assign a
label with a physiological meaning to a beat, it may be necessary to classify
the beats according to their cardiac origin. The steps of feature extraction
and clustering were briefly described in Section 5.6.1 within the context of
electromyographic analysis and clustering of motor unit action potentials.
Section 7.1. Baseline Wander 457

Rhythm analysis is, for natural reasons, based on the pattern of RR inter-
vals, but must also embrace morphologic information since most arrhythmias
are manifested by a joint deviation in rhythm and morphology. The scope
of rhythm analysis is strongly application-dependent. For example, rhythm
analysis in a system for resting ECG interpretation is limited by the very
short duration of the recording and thus only deals with less complicated,
persistent arrhythmias, whereas a system for continuous arrhythmia moni-
toting is designed to detect life-threatening, transient arrhythmias such as
ventricular fibrillation.
In addition to these types of analysis, a wide variety of specialized signal
processing algorithms have been developed over the years [9]. A small se-
lection of such algorithms include those for noise reduction in stress testing,
detection of S T - T segment changes in ischemia monitoring, characterization
of heart rate variability (Chapter 8), detection and characterization of "un-
organized" arrhythmias such as atrial and ventricular fibrillation, serial com-
parison of E C G / V C G recordings, interpretation of pacemaker performance
("pacemaker ECG"), detection of late potentials, and dynamic analysis of
the repolarization phase including detection of T wave alternans.

7.1 Baseline Wander

Removal of baseline wander is required in order to minimize changes in beat
morphology which do not have cardiac origin. This is especially important
when subtle changes in the "low-frequency" S T - T segment are analyzed for
the diagnosis of ischemia, which may be observed, for example, during the
course of a stress test [10, 11]. The frequency content of baseline wander
is usually in the range below 0.5 Hz. However, increased movement of the
body during the latter stages of a stress test further increases the frequency
content of baseline wander, see Figure 7.2. Patients unable to perform a
traditional treadmill or ergometer stress test may still be able to perform
a stress test by either sitting, running an ergometer by hand, or using a
special rowing device. In such cases, baseline wander related to motion of
the arms severely distorts the ECG signal. The bandwidth of such baseline
wander is considerably larger than that caused by respiratory activity and
perspiration.
We will below describe the two major techniques employed for the re-
moval of baseline wander from the ECG, namely, linear filtering and poly-
nomial fitting. Linear filtering can be further divided into filtering based on
time-invariant or time-variant structures.
458 C h a p t e r 7. ECG Signal Processing

1ii.!i.! !..!.iii. i.. ................ : : : ! :ii! i!i ii!i 84

!!ii !i:i.i .... ".... :::
~ i i i i : : i i . l i i . i ~ : i i i i i l . } . i i i i i . i ! i . i l } . i i . ! ii: . i i i i i i i i : i i i i} 84 i . ! i i i i i . i i i i : i . :

~ i ....... i ....... i ........ i . . . . . . i ........ i ........ i ....... i. . . . . . : ....... i . . . . . . . . . . . . . . . . . . . . . . . i . . . . . . i . . . . i ......... i . . . . . . . . . . . . . . i

9 0

--: . . . . :-. -i . . : . . - :. 9 . . . . : . . . . 9 . . . . : 9 } . . . . . . . . ! . . . . 9 .: . . . . :9 9 . . . . : . . . . : . . . . :
... : . . . . . . . . . . . : . . . . . . . : . . . . . . . . . . . . . . . . . : . . . . . . . . : . . . . . . . . . . . . . . . . . . . . . . . . . . : . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . : . . . . . . . . . . . . . . . . . . : . . . . . . . . . . . . . . . . . . : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

~ !i! 9149149149
i! 9149149149149
!i 9149
i!!:: 9149149149
. . . . . . . . ii 9149149149
(b)

. . . . . : . . . : . . . . . . . . . . . . . . . . . . . . . . . . . . . . : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . : . . . . . . .
........ : ........ . ................ : ................................................................................................ : ................... : ......... : .............

5 0 5 2 5 4 5 6 5 8 6 0

T i m e (s)

F i g u r e 7.2: (a) Electrocardiographic baseline wander due to sudden body move-

ments. The amplitude of the baseline wander is considerably larger than that of
the QRS complexes. (b) A close-up in time (10x) of the ECG signal framed in (a).

7.1.1 Linear, Time-Invariant Filtering

The design of a linear, time-invariant, highpass filter involves several consid-
erations, of which the m o s t crucial are the choice of filter cut-off frequency
and phase response characteristic. The cut-off frequency should obviously
be chosen so t h a t the clinical information in the E C G signal remains undis-
torted while as much as possible of the baseline wander is removed. Hence,
it is essential to find the lowest frequency component of the E C G spec-
trum. In general, the slowest heart rate is considered to define this par-
ticular frequency component; the P Q R S T waveform is a t t r i b u t e d to higher
frequencies. During bradycardia the heart rate may drop to approximately
40 b e a t s / m i n u t e , implying t h a t the lowest frequency contained in the E C G
is approximately 0.67 Hz [12]. Since the heart rate is not perfectly regular