ECG Signal Analysis - ECG Acquisition and Automated Remote Processing - page 23

through difference equations in microcomputer-based systems for Holter tape
analysis [ 41 , 42 ]. An arrhythmia detection algorithm based on filtering, differen-
tiation, energy collector, and a classifier is described [ 43 ]. This algorithm gener-
ated a delay of 1 s while implemented in real time with 250-Hz sampling. Pan
Tompkins algorithm [ 44 ] is considered as one of the early pioneering works of
microprocessor-based real-time QRS detection and provided benchmark for per-
formance analysis for other researchers during the following decades. The algo-
rithm is based on digital analysis of slope, amplitude, and width and QRS energy.
The algorithm was implemented on Zilog 80 microprocessor, with a very good
accuracy (total detection failure of 0.67%) with 487 records. The improved version
of this algorithm [ 45 ] showed an improved performance (99.68% sensitivity and
99.63% predictivity) based on an event detector from the preprocessing stage of
QRS detection. In [ 46 ], a comparison between the first-derivative-based methods,
especially

Pan Tompkins, Hamilton-Tompkins with Hilbert transform (HT)

methods, is drawn.
In [ 47 ], an adaptive algorithm based on maximum a posteriori (MAP) esti-
mation is described. A mathematical model of pulse-shaped waveform for QRS
detection is developed, and its estimation procedure is estimated. To adapt with
varying QRS morphology, the detector (estimator) adjusts the parameters with
incoming beats. In [ 48 ], the same model as in [ 47 ] is dealt with; however, a
simplification in the model is introduced.
QRS detection using mathematical morphology is described in [ 49 ]. In this
approach, a morphological operator followed by peak value extractor (PVE),
which suppresses the non-QRS regions and converts the QRS peaks into sharp
complexes. Finally, an adaptive threshold-based rule base determines the QRS
locations. A mathematical operator based on morphology of the QRS complex is
described in [ 50 ]. A new concept of dominant wavelet rescaled coefficients
(DWRC) is used, where the relation between QRS complex and their corre-
sponding wavelet coefficients is derived analytically. In the model, a typical ECG
beat is approximated as summation of two sinus functions as P and T waves and
three triangles as QRS complex. A morphology alignment method for using a
piecewise uniform re-sampling ECG waveform is prescribed in [ 51 ]. A heartbeat is
first delineated into stable and flexible segments. These segments are then
resampled at two different rates such that re-sampling frequency becomes uniform
for the same segments in all beats. The morphological features extracted from each
heartbeat are evaluated for alignment of an ideal morphology.
A number of non-linear transformations for QRS detection are available in
literature. Among them, [ 52 ] described a multiplication of backward difference
(MOBD) operator to multiply the successive differences to feed in a decision rule
based on adaptive threshold to detect the QRS regions. The HT [ 53 ] for QRS
detection is proposed in [ 54 ]. The first difference of the signal is computed and its
HT is used to detect the point of inflexions in waveform, which appears as peak in
HT dataset. Then, an adaptive threshold based on statistical parameters of the
Hilbert sequence is used for detecting R peaks.