A signal decomposition approach to morphological modeling of P-wave

Morphological modelling of electrocardiographical P-waves could simplify the detection of signals' morphological features employed in risk stratification. We compared four different approaches, based on signal decomposition, for morphological modeling of signal-averaged P waves. The functional models included: trigonometric, Bézier, B-spline, and Gaussian basis functions. The comparison between models was performed at a common fixed number of parameters (ranging between C=3 to 21). The performances of the approximations were evaluated using compression efficiency measures, like the percentage of root-mean-square differences (PRD). Nonlinear iterative parameter identification was employed for Gaussian models, while the parameters of the other basis functions were calculated through closed formulas. We tested the effectiveness of the several methods on the PhysioNet PTB diagnostic ECG database (561 subjects, 10 s each, 12 leads). Trigonometric and B-spline models proved to be the most effective in following the details of the morphology (PRD: 0.51% ± 0.62% and 0.99% ± 0.96%, respectively, on lead VI at C=21), possibly as they form an orthogonal basis for the specific signal. This property is not shared by Bezier curves and Gaussian basis functions (PRD: 2.47% ± 2.17% and 3.57% ± 6.83%).