Validation of μ-volt T-wave alternans analysis using multiscale analysis-by-synthesis and higher-order SVD

Detection of microvolt T-wave alternans (TWA) and finding its clinical significance remains a challenging task. In this work, we propose a new TWA detection method based on multiscale analysis-by-synthesis followed by higher-order singular value decomposition (MAS-HOSVD). The multilead ECG (MECG) data, after R-peak detection, were represented as a third-order tensor. Then HOSVD was applied on the subband reconstructed T-wave tensor to determine a tensor containing T-wave information (TensorTinfo). The p-value, TWA ratio, and the probability of detection (PD) were computed, based on the information present in TensorTinfo, to validate the presence of TWA or not. The proposed method was evaluated with semi-synthetic and real signals by adding several types of noise at different SNR levels and was compared to a multilead and two single-lead schemes. Simulation results showed that the proposed method detected TWA accurately with the sensitivity of 95.2% for the alternans level of 10 μV. Also, the method detected TWA with an SNR at least lower by 10 dB than other methods. The MAS-HOSVD method was tested to be accurate in TWA analysis of MECG data. This method can be considered as an automatic tool for prior prognosticator of sudden cardiac death, consequently, can help reduce the mortality rate.