Machine learning-based multichannel short-duration high-dynamic microvolt T-wave alternans detection

Microvolt T-Wave Alternans (MTWA) is a widely recognized predictor of sudden cardiac death (SCD). However, its detection is challenging due to the microvolt amplitude of alternans, the presence of short-term high dynamics, and low signal-to-noise ratios in clinical electrocardiogram (ECG) signals. This paper proposes a multi-channel TWA detection method that integrates tensor decomposition with a Convolutional Neural Network (CNN). Tensor decomposition is used to capture multi-dimensional correlations and suppress noise, while CNN automatically extracts discriminative features from the decomposed factors for classification. Simulation and real database experiments demonstrate that the proposed method works much better than traditional methods. Under noisy conditions, the maximum accuracy improvement reaches 71.2%, and the F1 score increases by 69.5%. These results highlight the robustness and potential clinical value of the proposed method.