A template matched-filter based scheme for detection and estimation of t-wave alternans

Microvolt T-wave alternans (TWA) is among the important risk stratifiers for sudden cardiac death, having significant prognostic ability. A new detection theoretic technique, called template matched-filter detector (TMFD), is proposed for TWA detection and estimation. Unlike classical schemes, which estimate the TWA amplitude directly, the proposed strategy targets alternant energy within the repolarization complexes. The scheme is evaluated using realistically synthesized ECG signals and alternant waveforms with varying temporal distributions, synthetic noises and real artifacts. The implementation of TMFD is discussed in the context of two different templates, i.e., median (TMFD-1) and mean (TMFD-2). Detection performance and estimation accuracy is validated against three classical schemes, i.e., spectral method (SM), modified moving average method (MMAM) and correlation method (CM). The proposed scheme outperforms CM over the entire range of SNR with both the templates and all noise types. Detection performance of TMFD approaches SM when the signal is corrupted with real artifacts. In Gaussian noise, performance of TMFD-1 is comparable with SM when detection probability is maximized, and is 2 dB degraded for Laplacian noise. In terms of estimation accuracy, SM has the least value of bias for all noise types over the low range of SNR. In better signal conditions (SNR = 25 dB), the bias of TMFD approaches SM for alternant magnitude > 40 μV in the Gaussian case and for magnitudes > 20 μV with real noises. The bias of TMFD for real noises is comparable to MMAM for all TWA amplitudes and there is a gap of approximately 20 μV under Gaussian noise for SNR = 10 dB. The proposed scheme is also promising for dynamic TWA tracking as it successfully indicates temporal origins and termini of transient alternan episodes through beat to beat estimation statistics.