ECG Compression - ECG Acquisition and Automated Remote Processing - page 96

t

P

h i 2

N

x ð n Þ _ ð n Þ

n ¼ 1

rms ¼ ð 5 : 5 Þ

N

2 3

P

N

2 ½ x ð n Þ x

4 5

n ¼ 1

SNR ¼ 10 log ð 5 : 6 Þ

h i 2

P

N

x ð n Þ _ ð n Þ

n ¼ 1

where

N total number of samples in the dataset;

x[n] actual value of sample n;

corresponding reconstructed value.

_ [n]

'Max' operator extracts maximum element of sample to sample difference
array.
A combined performance index indicating compression and PRD together,
called quality score (QS) is used, defined as:

QS ¼ CR
PRD

ð 5 : 7 Þ

The direct compression schemes utilize amount of redundancies among a group
of neighboring ECG samples and operate on inter-sample correlation. Recon-
struction of data is carried out by a linear prediction or interpolation technique. A
prediction algorithm uses a priori knowledge of some previous samples, whereas
interpolation technique requires the same for previous and succeeding samples.
Theoretical background of such compression techniques is provided in [ 2 - 5 ]. In
tolerance comparison techniques, a preset or threshold is used around a sample
point. Whenever the first difference between the reference sample and the following
exceeds this value, a line is generated whose parameters (slope and length) are used
to represent intermediate points. Otherwise, the following sample is ignored. The
amplitude zonal time epoch coding (AZTEC), fan/scan along polynomial approx-
imation (SAPA), turning point (TP), coordinate reduction time encoding system
(CORTES) algorithms for ECG compression rely on some principle of tolerance
comparison method. AZTEC, developed for real-time ECG compression, is suit-
able for QRS detection. The ECG data are converted into slope and plateaus,
considering three consecutive samples at a time. A CR of 10:1 is reported in [ 6 ].
However, the reconstructed data show steplike quantization, which is not suitable
for visual analysis by a cardiologist. Few modified AZTEC encoding principle is
described in [ 7 , 8 ]. The TP algorithm utilizes an adaptive down-sampling, where
half of the samples are discarded depending on relative slope magnitudes between
three consecutive samples of original data. Thus, a CR of 2:1 is straightway
achieved without any further coding [ 9 ]. CORTES scheme is a real-time