Biomedical Engineering Reference
In-Depth Information
5.2 Review of ECG Compression Techniques
In general, data compression techniques can be classified into two broad catego-
ries, viz. lossless and lossy. In lossy compression techniques, some information is
'thrown away.' In lossless compression, the signal can be perfectly reconstructed
from the compressed data. Most of the biomedical signals include adequate
redundant information within the signal, and hence, popular biosignal compression
methods employ lossy techniques. For biomedical signal compression, one of the
following techniques is used, viz. time domain or direct data compression meth-
ods, transformation techniques, and parameter extraction techniques [ 1 ]. In direct
compression techniques, processing of raw samples (sometimes after a prepro-
cessing) is applied. Such techniques are popular in real-time applications. In
general, transformation techniques involve preprocessing of the samples using a
linear transform and then encode the expansion coefficients. During reconstruc-
tion, an inverse transform is carried out to get back original data, with some error.
In parameter extraction methods, some signatures of the ECG dataset such as
extreme locations (R-peak, S-peak), zero crossing points, or average slope in
individual peaks are computed and stored. Reconstruction is carried out by using
appropriate interpolation schemes.
The performance indices used by researchers for describing compression per-
formances and error estimates in the reconstruction are available in the literature.
Most common parameters are compression ratio (CR), percentage root mean
squared difference (PRD), PRD normalized (PRDN), maximum absolute error
(MAE), root mean square error (RMS), signal to noise ratio (SNR), etc. They are
defined as follows:
Input (raw) data file size
output (compressed) file data size
CR ¼
ð 5 : 1 Þ
P
t
2
N
x ½ n _ ½ n
n ¼ 1
PRD ¼ 100
ð 5 : 2 Þ
P
N
x ½ n ð 2
n ¼ 1
t
P
2
N
x ½ n _ ½ n
n ¼ 1
PRDN ¼ 100
ð 5 : 3 Þ
P
N
Þ 2
ð
x ½ n x
n ¼ 1
E max ¼ max x ½ n _ ½ n
ð 5 : 4 Þ
Search WWH ::




Custom Search