Karhunen-Loeve Coding of ECG signals

The utility of using the Karhunen-Loeve transform (KLT) to code electrocardiogram (ECG) waveforms is explored. The results on reproducibility and compression ratio are considered. The average values of the maximum absolute error and of the root-mean-square error have been computed using a database of 1220 cases. The reconstruction with 10, 15 and 20 basis vectors has been considered. With the best approximation (20 vectors) the compression ratio varies from 3 to 5.5 (mean value 4). Several wave measurements have also been automatically extracted from the original as well as from the reconstructed signals in order to evaluate the impact of reconstruction errors on the usual diagnostic parameters. The comparison shows that unacceptable variations are generated. This finding suggests that KLT compression cannot be used when classification methods based on conventional diagnostic parameters are applied to the reconstructed signals. However, the utility of KLT coefficients for QRST filtering and classification is not excluded