ECG Compression - ECG Acquisition and Automated Remote Processing - page 105

consecutive zero elements are found out from the second encoded byte (following
255), and an equal number of zeros are placed in array q[]. The algorithm steps for
this operation are provided at the End of Chapter Appendix III .
The array q[] consists of elements, some of which are encoded sign bytes, and
others encoded magnitude elements, either in combined or in uncombined form. In
this stage, the sign as well as magnitude decoding is performed to generate a new
array r[]. The first element is a sign byte and decoded to generate the signs of 8
decoded elements of first group. The algorithm step for magnitude and sign
decoding is given at the End of Chapter Appendix IV .
The array r[] can be considered to be consisting of groups, each consisting of
1002 elements. The first three elements of each such group are split values of
normalized original sample, followed by 999 normalized first difference elements.
At first, the normalization constant is obtained by combining the 2nd to 5th
elements of array w[]. For each group, the first element is (original sample)
generated by de-normalizing and combining the first three elements of array r[].
The following 999 elements are de-normalized to generate corresponding original
first difference. From the first original element, the consecutive original samples
are reconstructed by successive addition of de-normalized SSD elements from
index 2, using the algorithm given at the End of Chapter Appendix V .
If it is desired to generate a time-stamped ECG data, the sampling time array is
generated by an arithmetic progression series by de-normalizing the first element
of the array w[]. Finally, a time-stamped two-column data file is generated, where
the reconstructed ECG samples are arranged corresponding to their sampling
instants.
5.3.3 Test Results
The encoding-decoding process is tested with MIT-BIH arrhythmia data (mit-db),
MIT-BIH compression test data (c-db), and PTB diagnostic ECG database from
PhysioNet [ 32 ].
A total of 240 different normal and abnormal leads from ptb-db are used.
Table 5.2 shows some test results with 240 leads for lead I, lead III, lead aVF, and
lead v5 from different ptb-db files. The average PRD obtained in these leads are
0.723, 0.721, 0.701, and 3.62, respectively. Considering all 12 leads, an average
CR of 25.11 with a PRD 1.525 is obtained. The algorithm is also tested with 60 s
mit-db data files. The mit-db files contain 2-lead ECG samples at 360-Hz sam-
pling. At first, these data are up-sampled to 1 kHz using an interpolation tech-
nique. A total of 25 different leads are tested. Table 5.3 summarizes the test results
with mit-db data. An average CR of 21.05 and PRD of 3.822 are obtained.
Figure 5.6 represents a qualitative representation of one normal and one abnor-
mal ptd-db lead plotted before compression and after reconstruction. Figure 5.7
shows the lead plot of one mit-db before compression and after reconstruction.