ECG characteristic points detection using general regression neural network-based particle filters

Characteristic points (CPs) detection is still an open problem for the automatic analysis of electrocardiogram (ECG). Past Kalman Filter-Based efforts to extract CPs rely on a locally linearized approximation of the nonlinear ECG dynamical model and fail to detect all CPs accurately for strong noisy ECG. In this study, an improved particle filters-based algorithm is developed to track the dynamical ECG morphology and localize its characteristic points in strong noisy environments. Experiments on real ECG records contaminated by different coloration noise clearly show the superior performance of the presented approach over the Kalman Filter method.