Preprocessing Method for Improving ECG Signal Classification and Compression Validation
Liviu Goras, Catalina Monica Fira
A method for improving electrocardiogram (ECG) signal classification in time domain is presented. The main idea is to preprocess the segmented waveforms in order to obtain an “alignment” of the ECG with respect to the maximum value of the R beat while keeping the information on its initial position as a feature.
We propose two simple preprocessing methods basically consisting in the alignment of the R-waves either by translation or by a slight nonlinear time scaling. It is shown that the methods significantly improve the classification rates obtained with two independent methods, one using a MLP artificial neural network and the other one based on the k-nearest neighbor (k-NN).
These techniques are also used to improve the validation of ECG compression by comparing the classification results obtained with original patterns and with reconstructed ones, a higher classification rate in the former case reflecting better compression results.