Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 173

4.2.1.3 Processing of ECG

In ECG deterministic and stochastic features are manifested and the repertoire of
the methods applied for its analysis is very broad. It includes: morphological features
extraction, analysis in time and frequency domain, statistical methods, classification
procedures. From ECG signal another time series, namely HRV (heart rate variabil-
ity), is derived which in turn is a subject of statistical analysis.
In case of signal processing before applying a new method to the given type of
data or for the differentiation of a particular pathological state it is recommended
to compare the results with other methods and test it on standard data. In case
of ECG fortunately there exists a Physionet database [Goldberger et al., 2000],
which can be helpful in this respect. It is available at http://www.physionet.org/
physiobank/database/ . PhysioBank is a large and continuously updated archive
of well-characterized digital recordings of physiologic signals and related data that
may be downloaded at no cost. PhysioBank currently includes databases of multi-
parameter cardiopulmonary, neural, and other biomedical signals from healthy sub-
jects and patients with a variety of conditions with major public health implications,
including sudden cardiac death, congestive heart failure, arrhythmia (MIT-BIH ar-
rhythmia database). In PhysioNet the software for analysis of ECG and HRV may be
found as well.

4.2.1.3.1 Artifact removal The first step in ECG analysis is elimination of the
artifacts. ECG signals may be corrupted by the technical artifacts such as: power
line interferences, artifacts due to bad electrode contacts, quantization or aliasing er-
rors, noise generated by other medical equipment present in the patient care environ-
ment and biological artifacts: patient-electrode motion artifacts, muscular activity,
baseline drift—usually due to respiration. Technical artifacts may be avoided by de-
signing proper measurement procedures, however elimination of biological artifacts
is much more difficult and requires special signal analysis techniques. The filtering
techniques used for denoising ECG involve linear and non-linear methods, model
based or model free approaches.
For rejecting high frequency noise FIR filters or Butterworth 4-pole or 6-pole low-
pass digital filters appear to be suitable [Weaver, 1968]. The cut-off frequency is
usually set around 40 Hz. For baseline wander high-pass linear filters of cut-off fre-
quency up to 0.8 Hz may be used. Cut-off frequency above 0.8 Hz would distort the
ECG waveform. Among the linear, model based approaches to denoising, Wiener
filtering is an often used method. Unfortunately, one of the Wiener filter assumptions
is that both signal and noise are stochastic. However, ECG has a prominent deter-
ministic character, which diminishes the performance of the Wiener filter. One of its
drawbacks is the reduction of the signal amplitude.
Wavelet transform is a model free method of ECG denoising. The discrete wavelet
transform may be used to decompose ECG into time-frequency components (Sect.
2.4.2.2.4) and then the signal may be reconstructed only from presumably free of
noise approximations. It is essential to choose the mother wavelet properly; it has
to be maximally compatible with the ECG signal structures. In [Clifford, 2006]