Near-Optimal Signal Preprocessor for Positive Cardiac Arrhythmia Identification

Extremely high reliability of waveform detection is fundamental for computer-assisted identification of cardiac arrhythmias. The problem is formulated in terms of one of the basic aspects in digital signal processing, namely, detection of an unknown deterministic signal in noise. The signal is considered band limited and the noise as Gaussian with zero mean. The theory presented leads to an energy based detector. No a priori assumption is made about the morphology of the waveforms to be detected; only energy thresholding is necessary for event detection. The addition of a linear filter as a front end to the energy detector has resulted in an entire detection process described as suboptimal matched nonlinear filtering. The SNR gain at the input of the energy detector has given performances almost identical to the matched filter. The method has been implemented with an original fast algorithm to allow reasonable execution time for the processing of continuous long-duration signals (1 h). The design of such a fillter is described along with its application to the processing of two simultaneously recorded cardiac signals. The two signals are an auricular electrogram (OEG from an esophageal pill electrode) and an ECG (lead II or III). Systematic evaluation and detection performance results show that the proposed method could be seriously considered as a near-optimal approach to waveform detection, since it is based on proven signal processing theory and is far superior to heuristic methods.