Hybrid Ensemble Learning For Robust QRS Complex Detection
Abstract
QRS complex identification with a certain accuracy is an essential step in electrocardiogram (ECG) signal processing and any subsequent cardiovascular diagnosis. However, in realistic monitoring scenarios, the accuracy of traditional QRS identification can be largely diminished by baseline wandering effects, low-frequency interference, as well as physiological morphological variability in the recorded ECG signals. This paper proposes an ensemble-based signal processing system for ECG signals by utilizing synchronous cardio-oscillation ensemble generation with correlation refinement and signal centering. This strategy will facilitate synchronous ensemble generation for PQRST waves associated with identified R-peaks in the recorded signal, apply low-frequency filtering and statistical signal centering, as well as select the representative ensemble members based on correlation assessments among signal segments. This particular strategy will enhance overall waveform homogeneity while removing non-synchronous interference before QRS feature identification in recorded ECG signals. This proposed model will establish an efficient signal-level processing strategy usable within traditional detectors or those involving learning for feasible noisy-channel ECG signal processing.