ECG Compression - ECG Acquisition and Automated Remote Processing - page 97

combination of TP and AZTEC, which are applied in parallel to ECG samples. TP
reveals clinically significant QRS regions, and AZTEC compresses isoelectric
regions. The data reconstruction is produced by expanding the AZTEC plateaus,
smoothing them, and interpolating between each pair of TP data [ 10 ]. A CR of 4.8
and PRD of 7 are reported to be achieved in [ 11 ] using CORTES. The fan algorithm
[ 12 , 13 ] was developed for ECG signal transmission. This technique uses a mini-
mum slope principle between two samples to include all intermediate points in the
'coverage angle.' The intermediate points are considered as redundant within this
coverage angle. The reconstruction is achieved by expanding the line into discrete
points by interpolation. SAPA-2 algorithm includes an additional slope in addition
to two lines representing the coverage angle described in fan. DPCM is perhaps the
simplest compression technique where the first difference of the samples is enco-
ded. The compression based on this scheme is called 'delta coding' [ 14 ]. For
enhancing the CR, a modified delta coding is described in [ 15 ] where the differ-
ences between successive ECG samples are compared with a threshold and if
exceed this threshold, are kept, and otherwise discarded. At 1-kHz sampling of
ECG data, this method achieved a CR of 10. An exhaustive discussion and test
results with direct data compression techniques and their suitability for telemedi-
cine are available in [ 16 , 17 ].
Many of the transform domain techniques have been employed for multilead
ECG signals. Here, after preprocessing of the signal, a liner orthogonal transform is
used to convert the ECG samples from time domain to a different domain. The
coefficients generated are then appropriately encoded to compress the data.
Popular transform techniques employ either discrete cosine transform (DCT),
Karhunen-Loève transform (KLT), or wavelet transform (WT). A DCT-based
compression technique is described [ 18 ], where a quantization scheme is adopted to
approximate the coefficients for reconstruction of the ECG. Three levels of quan-
tization, viz. 1-, 3-, and 4-bit were used with a fixed and variable threshold level
separately. A multichannel ECG compression technique is described in [ 19 ]. In this
work, a preprocessor discards the redundant ECG channels (III, aVR, aVL, and aVF)
and the rest of the channels are fed to the liner transformer. The paper compares the
compression performance using KLT and DCT. In [ 20 ], the authors report orthog-
onal transforms, Haar and Cosine to achieve a 3:1 compression for ECG signals.
WT-based ECG compression has been extensively used by researchers in telecar-
diology applications for the last decade. In [ 21 ], a new approach using orthogonal
zonal wavelet packet compression (OZWC) to model a GSM-based mobile tele-
cardiology system is described. Using three different quantization levels, viz. 8, 12,
and 16 bits, the authors report CR values of 18.32, 8.16, and 5.45, respectively, and
PRD values of 0.5967, 0.5778, and 0.5759, respectively, using record 100 from MIT-
BIH arrhythmia data. For reconstruction performance, the authors used different
wavelets like Haar, Coiflets, Daubechies, Symlets. In another application for con-
tinuous ECG transmission [ 22 ], the authors describe a wavelet-based low delay ECG
compression algorithm (WLDECG) to reduce the delay associated with frame size in
wavelet transformation as much as possible, without causing deterioration in the
reconstructed ECG signal quality. The approach divides an ECG beat into two cycles