ECG Distortion Measures and their Effectiveness

Abstract

Measurement of quality is of fundamental importance to electrocardiogram (ECG) signal processing applications. A number of distortion measures are used for ECG signal quality assessment. A simple and widely used distortion measure is the percentage root mean square difference (PRD). It is an attractive measure due to its simplicity and mathematical convenience. But PRD is not a good measure of the true compression error and results in poor diagnostic relevance. In this paper, we discuss the advantages and drawbacks of the objective distortion measures using different compressed signals. With extensive analysis it is shown that although some distortion measures correlate well with the subjective evaluation for distortions resulting from a given compression method, they may not be reliable for evaluation of some other compression distortions. It is also concluded that a distortion measure should be subjectively meaningful in order to correlate a large or small quantitative distortion measure with bad and good quality. This work attempts to evaluate the closeness of the objective quality measures with subjective measure and investigation may help to suggest a better quality criterion for optimizing rate-distortion algorithms. Experimental results show that wavelet energy based diagnostic distortion (WEDD) measure is significantly better than other measures. This measure is sensitive to ECG feature changes and insensitive to smoothing of low-level background noise.