Tensor-Domain Machine Learning Based Cardiac Diseases Detection Using 12-lead ECG

The analysis and processing of 12-lead electrocardiogram signals enable the rapid and accurate diagnosis of cardiovascular diseases (CVDs). Each morphological change detected in the 12-lead ECG recordings corresponds to a specific cardiovascular disease. The physician must manually observe these morphological changes, which complicates and lengthens the ECG interpretation. To detect CVDs, an automated software-based computer-assisted diagnosis system is necessary. The purpose of this article is to propose a tensor-domain based cardiac disease detection method from 12-lead ECG data using a multivariate technique. The 12-lead ECG signal is initially segmented into multi-lead beats using a QRS detection algorithm. A multivariate projection-based Fixed-boundary empirical wavelet transform (MPFBEWT) is used to decompose segmented multilead beats into modes, resulting in a three-dimensional tensor containing information about leads, samples, and modes. A tensor decomposition technique is used for tensor decomposition and feature extraction. The K-nearest neighbor (KNN) classifier is used to classify the obtained features in various cardiac conditions such as myocardial infarction (MI), bundle branch block (BBB), heart muscle disease (HMD), and healthy control (HC). The proposed method is evaluated using 12-lead ECG recordings from the Physikalisch Technische Bundesanstalt (PTB) diagnostic ECG database. Our method outperforms existing methods for detecting cardiac disease with a classification accuracy of 97.96%.