Higher Order Time-Frequency Domain Tensor-Based Method for the Detection of Atrial Abnormalities Using 12-Lead ECG Signals

Atrial fibrillation (AF) and atrial flutter (AFL) are significant atrial abnormalities with many similarities, making them more challenging to diagnose. It is essential to precisely diagnose AF and AFL to prevent the occurrence of other life-threatening cardiac abnormalities. This article proposes a higher order time-frequency domain tensor-based method for detecting AF and AFL-based atrial abnormalities using 12-lead ECG signals. Initially, 12-lead ECG signals are segmented beatwise to extract multilead beats. Then these multilead beats are further decomposed leadwise in the time-frequency domain. A fourth-order (4-D) tensor containing the lead, beat, time, and frequency information is formed corresponding to each patient. Each 4-D tensor is decomposed by the tensor decomposition method to obtain the tensor domain features. Various classifiers classify these tensor domain features to differentiate atrial abnormalities from healthy control (HC). The 12-lead ECG recordings from two public databases are used to evaluate the proposed method. The results show that the fourth-order tensor-based approach has obtained the classification accuracy values of 90.2%, 91.7%, and 73.9%, respectively, for HC versus AF, HC versus AF plus AF AFL, and AF versus AFL, respectively.