An Effective Method for CHF Diagnosis via Attention-based RNN Using ECG Signals

Congestive heart failure is a kind of cardiovascular disease that poses a severe threat to human health and the early, convenient and automated detection of it is of great clinical significance. Therefore, we proposed a method based on fixed number of consecutive ECG heartbeats via bidirectional RNN combined with attention mechanism to detect it. In this study, we established a CHF dataset for experiments on Physionet. After signal noise removal, windows containing a fixed number of consecutive heartbeats were extracted as samples, which were divided into training set and testing set based on the "interpatient" scheme. Our attention-based algorithm could automatically assign different weights to different heartbeats in the detection task and the accuracy of this method was greatly improved by simulating the thought that doctors usually paid more attention to the useful information when making diagnosis decisions. The results show that our algorithm has great advantages with accuracy, sensitivity and specificity reaching 93.42%, 93.12%, and 93.72% respectively. The method studied in this paper has great practical value and can be applied in medical software applications.