ECG Compression - ECG Acquisition and Automated Remote Processing - page 100

developed compression scheme, encoding is performed using the following two
rules:

Rule (A): In P and T wave regions, the first difference elements can be suitably
normalized (in a scale of 0-100) and can be represented by a nibble instead of a
byte. Hence, in principle, magnitudes of two such consecutive elements can be
represented by a single byte. Thus, a direct 2:1 compression is achieved.
Rule (B): In isoelectric regions (TP, ST, and PQ), many consecutive first difference
elements, normalized in a scale of 0-100, assume zero values. Each such zero
sequence can be represented by a single byte, representing the total number of
consecutive zero elements.

5.3.1 Stages of Encoding

Encoding of the single-channel ECG is initiated with a two-column text file
containing data samples against the respective sampling instants. Since the sam-
pling interval is uniform, a single byte is used to encode the sampling information
which is embedded with the final encoded data stream. Hence, compression stages
are applied on millivolt samples only. The stages of compression are described in
Fig. 5.2 . At first, removal of high-frequency noise from the data samples is done.
The first difference array is computed using the following formula:

y ð i Þ¼ x ð i þ 1 Þ x ð i Þ ð 5 : 8 Þ

To account for the sign of the first difference array elements, a separate
encoding scheme is used. For this, data grouping is done and then Rule (A)
followed by Rule (B) is applied in sequence. During the reconstruction (decom-
pression) process, the samples will be generated by cumulative addition of first
difference values with the first sample.
Finally, the encoded elements are stored as 8-bit ACSII characters. The fol-
lowing subsections represent the stages of compression.
At first, the raw sample dataset is smoothed depending on the span of the
dataset by 'spline' smoothing function to eliminate the high-frequency noise. The
smoothing factor is empirically selected as 0.001 fraction of the amplitude span of
the raw dataset. The first difference array elements, computed by Eq. 5.8 , are
normalized using a factor (k) determined in the following manner:

99
max ½ abs ð y max Þ

k ¼ ð 5 : 9 Þ

where y max is the maximum amplitude value of first difference array.
Now, since the elements near the isoelectric region are having low amplitude,
the first element h is selected near the mid-height level of QRS:

h abs ð y max Þ= 2 ð 5 : 10 Þ