ECG Signal Analysis - ECG Acquisition and Automated Remote Processing - page 18

In the following sections, a brief review of ECG analysis algorithms are pro-
vided, divided into three parts, viz., preprocessing methodologies, R-peak detec-
tion techniques and feature extraction techniques.

2.3 Review on Computerized ECG Processing Techniques

2.3.1 Denoising Techniques

ECG denoising is aimed to eliminate (or, at least minimize) the unwanted signals
from an ECG record, without hampering the clinical information contained within
the signal itself. A detailed review of the denoising methods is available in [ 4 ]. The
available literature on ECG denoising mostly discusses PLI and BW removal. In
many reported works, the researchers have validated the algorithm by introducing
a simulated noise with a 'clean' ECG signal to generate a composite noisy signal
and then denoising it using their proposed algorithm.
In general, the ECG denoising techniques can be classified into one of the
following categories:
(a) Digital filtering techniques—adaptive and non-adaptive;
(b) Source separation methods—principal component analysis (PCA) and inde-
pendent component analysis (ICA);
(c) Neural networks;
(d) Wavelet-based methods;
(e) Other non-adaptive methods like empirical mode decomposition (EMD), etc.

The early approaches for noise filtering, prior to 1980s, were based on digital
filters [ 5 ]. For computerized processing of ECG, the recommendation for filtering
bandwidths and other specifications is guided by AHA circulation [ 6 ]. Design of a
digital filter with integer coefficients for microprocessor implementation is
reported [ 7 ]. The paper analyzes the performance and errors due to quantization,
rounding-off operation of the filter coefficients, their representation, and overflow
while implemented in NSC800 and 680E52 microprocessors. Using sampling rate
of 360 Hz, a comparative analysis of non-adaptive and adaptive notch filters for
PLI reduction is carried out in [ 8 ] in terms of computational efficiency (number of
multiplications), distortion of the signal, and residual signal entropy. The adaptive
filter implemented by Tompkins et al. [ 9 ] with an internally generated reference
was found to be efficient than the non-adaptive counterpart. A common problem
with the notch filter is the transient response which affects its performance for real-
time operations. Reference [ 10 ] deals a technique for suppression of transient
responses at the expenses of some computational load in the initial stages. A new
adaptive technique for PLI reduction is described in [ 11 ] where the line interfer-
ence signal on the patient body is separately recorded using a hardware arrange-
ment. A common problem with linear phase filtering is large number of