The Use of Deep Learning in the Diagnosis and Prediction of Heart Failure: A scoping review

This scoping review presents a comprehensive analysis of the current implementation of deep learning techniques in heart failure diagnosis and prediction. We investigated the use of various deep learning models, focusing on their application in analyzing medical images and electronic health records. A thorough search across four electronic databases yielded 503 prospective studies, with 17 meeting our inclusion criteria. These studies predominantly originated from the United States and China and were primarily journal articles. Our review identified two main categories of deep learning models: those processing medical images and those analyzing clinical parameters from electronic health records. The most commonly used models were recurrent neural networks (RNN) for prediction and convolutional neural networks (CNN) and natural language processing (NLP) for diagnosis. The studies demonstrated a wide range of imaging modalities, with electrocardiograms being the most prevalent. Additionally, the review highlighted a variety of clinical parameters used for prediction and diagnosis, emphasizing the significance of artificial intelligence in medical research. Despite the promise shown by these models, challenges such as inconsistent performance, lack of detailed methodology, and limited geographical diversity in study sources were identified. Our findings underscore the potential of deep learning in enhancing heart failure diagnosis and prediction, but also point towards the need for more rigorous and diversified research to fully realize this technology's capabilities in healthcare.