Generating Conceptual Explanations for DL based ECG Classification Model

Resumen

Deep learning techniques are being used for heart rhythm classification from ECG waveforms. Large networks using end-to-end learning such as convolutional neural networks are not easily interpretable by end-users such as doctors. This is because most of the state of art explainability techniques focus on explanations for data-scientists who have technical knowledge. However, end-users of such systems are normally doctors who are not familiar with the technical details of machine learning. Therefore, to address this gap, we propose a framework that provides explanation of ECG classification model in the language of clinicians. We leverage state of art knowledge-based systems to map domain concepts with outcomes of machine learning model. Our results show that domain concepts based explanations are useful for clinicians and can greatly reduce their cognitive load. This shall lead to larger deployment of these models in real world.