On projection matrices and dictionaries in ECG compressive sensing - A comparative study

In this communication we propose and discuss comparatively several techniques for ECG signal compression inspired from the fundamentals of compressed sensing (CS) theory, focusing on acquisition techniques, projection matrices and reconstruction dictionaries and on the effects of the preprocessing involved. Essentially, we investigate and discuss two approaches. The first approach for ECG signal compression relies on the direct CS acquisition of the signal with no preprocessing of the waveforms before taking the projections, neither for the construction of the dictionaries. This “genuine” CS we will call patient specific classical compressed sensing (PSCCS) since the dictionary is built from patient initial recordings. The second approach implements a specific preprocessing stage designed to enhance sparsity and improve recoverability, based on segmenting the signal into single heart beats (also known as cardiac patterns) - denoted further as cardiac patterns compressed sensing - (CPCS) since in this case the acquired signals and the dictionary atoms are preprocessed segmented cardiac beats without or with centering of the R wave.