CNN-LSTM Fusion: An Intelligent Framework for Classifying Heart Failure Severity

Heart failure (HF) is a chronic condition where the heart is either too weak to pump sufficient blood through arteries or lacks the elasticity to fill arteries adequately. Accurate assessment of the degree of HF severity is critical in identifying the most appropriate treatment. Based on physical limitations during activity, the NYHA functional classification system categorizes HF into four stages. This paper describes a robust deep learning approach for HF stratification based on short-term RR interval signals. The method employs a hybrid deep learning model, offering timely warning information for early detection and personalized disease management. Notably, the proposed deep model necessitates minimal pre-processing of RR interval signals and does not require engineered features. The model, trained and tested on a balanced dataset from the BIDMC Congestive Heart Failure (CHF) and CHF RR intervals databases, achieves an accuracy of 88.46%, a sensitivity of 76.26%, and a specificity of 91.62%.