Compressão de sinais de eletrocardiograma utilizando análise de componentes independentes

Abstract

The continuing demand for high performance and low cost electrocardiogram processing systems have required the elaboration of even more efficient and reliable ECG compression techniques. The objective of this work is to evaluate the performance of an electrocardiogram (ECG) compression algorithm based on independent components analysis (ICA). To each of the ECG signal we processed, using ICA, vectorial subspaces composed of its basis functions were obtained, for the signal can be expressed as a linear combination of them. The ECG signal was subdivided into m fixed length windows, and each of them was projected in the subspace, resulting in a vector w of coefficients for each window. A simple quantization process was performed over the m vectors w, according to defined levels of quantization, each one generating different levels of reconstruction error. It was observed that the storage of the coefficients implies the use of less space in memory in comparison to that one used by the corresponding windows of the electrocardiogram signal. The reconstruction error measure traditionally used, the percent root mean-square difference (PRD), was used into the evaluation of the algorithm. The results had been compared with those obtained using the Karhunen Lo´eve transform (KLT).