Explainable Electrocardiogram Analysis with Wave Decomposition: Application to Myocardial Infarction Detection

Automatic analysis of electrocardiograms with adequate explain-ability is a challenging task. Many deep learning based methods have been proposed for automatic classification electrocardiograms. However, very few them provide detailed explainable evidence. In our study, we explore ECG through explicit decomposition single-beat (median-beat) signal. particular, every sample decomposed into five subwaves and each subwave parameterised by Frequency Modulated Moebius. Those parameters meanings interpretation. stead solving the optimisation problem iteratively which time-consuming, make use an Cascaded CNN network to estimate Our preliminary results show that appropriate position regularisation strategy, neural able P, Q, R, S, T events maintain good reconstruction accuracy (with R2 score 0.94 on test dataset PTB-XL) in unsupervised manner. Using estimated parameters, achieve generalisation performance myocardial infarction detection four different datasets. The features high importance are accordance clinical interpretations.