On the compression of ECG records employing triangular elements and analysis-by-synthesis modeling

In this work, non-traditional coding of ECG signals using greedy decompositions is analyzed. For this purpose one starts from a dictionary of triangular waveforms, and using the Matching Pursuit algorithm one searches the dictionary for the elements that will be employed for representing a given signal. In this way, compression is performed together with signal analysis. This analysis models the signal utilizing a subset of elements from the dictionary, which may be overlapping and non-orthogonal. The extracted triangles are coded in terms of their centers and their bases´ lengths and asymmetries. One of the benefits of the discussed approach is that it provides a flexible signal model rather than just a plain waveform coding or the coarse-detail approximation model of wavelet based compression systems. Therefore, it can be used for pattern characterization of different signal blocks, providing an efficient way to find pathological phenomena, i.e. to analyze the signal, without reconstructing it. The proposed compressor is evaluated in terms of PRD versus bit rate.