Deep learning for heart disease anomaly detection: performance factors and algorithms

Abstract

Heart disease is a prevalent concern for individuals in every age group, as it significantly impacts their health and remains a leading cause of mortality today. An effective heart disease detection method is essential for people to assess their heart conditions accurately. Over the past decades, heart disease detection techniques, whether based on machine learning or deep learning, have evolved considerably—from relying on handcrafted features to automatically learned features, from using single models to hybrid models and from applications confined to clinical settings to deployment in daily life through wearable and remote sensing technologies. This paper provides a comprehensive review of various types of heart-related data and popular deep learning algorithms employed for heart disease detection. In addition, we conduct a comparative experiment that showcases the existing gaps between clinical and wearable signals and highlights the potential for further research on wearable devices. This paper also identifies the key challenges, emphasises the primary factors affecting detection performance and discusses potential solutions.