Data compression techniques for electrocardiograms

Many different techniques have recently been proposed for efficient storage of ECG data with data compression as one of the main objectives. Although high compression ratios have been claimed for some of these techniques, the techniques did not always include the word length considerations with regard to the parameters representing the compressed signal. The authors feel that any technique can be meaningfully evaluated only if the resulting compression is expressed in bits/sample rather than the compression ratio that is often used in this field. This paper provides a critical evaluation of two classes of techniques, the direct data compression technique and the transformation technique. It is shown through exhaustive analysis that the direct data compression technique utilizing adaptive least-squares curve fitting yields a relatively fast and efficient representation of ECG signals at about 1.6 bits/sample, while maintaining visual fidelity and a normalized mean-squared error less than 1%.