Automatic Heart Failure Stratification Using a Convolutional Neural Network

Heart failure (HF) occurs when the heart is too weak to pump enough blood through the arteries or lacks the elasticity to fill the arteries adequately. It is a chronic, progressive disease that, depending on its severity, requires prompt diagnosis and treatment. HF is classified into four levels by the New York Heart Association (NYHA) classification system based on physical limitations during physical activity. The electrocardiogram (ECG) is a noninvasive test used to diagnose HF by analyzing RR intervals, heart rate, or heart rate variability (HRV). This paper describes a ten-layer deep convolutional neural network (CNN) model that automatically determines the stage of HF. The proposed CNN model does not require feature extraction and only minimal preprocessing of HRV signals. The model is trained and tested on a balanced dataset extracted from the RR interval signal databases, and it achieves an accuracy of 93.55%, a sensitivity of 87.14%, and a specificity of 95.70%.