Explaining Deep Neural Networks for Knowledge Discovery in Electrocardiogram Analysis

ABSTRACT Deep learning-based tools may annotate and interpret medical tests more quickly, consistently, and accurately than medical doctors. However, as medical doctors remain ultimately responsible for clinical decision-making, any deep learning-based prediction must necessarily be accompanied by an explanation that can be interpreted by a human. In this study, we present an approach, called ECGradCAM, which uses attention maps to explain the reasoning behind AI decision-making and how interpreting these explanations can be used to discover new medical knowledge. Attention maps are visualizations of how a deep learning network makes, which may be used in the clinic to aid diagnosis, and in research to identify novel features and characteristics of diagnostic medical tests. Here, we showcase the use of ECGradCAM attention maps using a novel deep learning model capable of measuring both amplitudes and intervals in 12-lead electrocardiograms.