Time-Incremental Convolutional Neural Network for Arrhythmia Detection in Varied-Length Electrocardiogram

Abstract

Automatic arrhythmia detection plays an important role in early prevention and diagnosis of cardiovascular diseases. Convolutional neural network (CNN) introduced a simple, end-to-end solution to multi-class arrhythmia classification, but the restriction that it could only accept fixed-length input resulted in noises or key information losses in training. Meanwhile, CNN's high memory consumption and computation cost also limited its application. To address these issues, we proposed a time-incremental convolutional neural network (TI-CNN), which utilized recurrent cells to introduce flexibility in input length for CNN models, and featured halved parameter amount as well as more than 90% computation reduction in real-time processing. The experiment results showed that, TI-CNN reached an overall classification accuracy of 77.3%. In comparison with a classical 16-layer CNN named VGGNet, TI-CNN achieved accuracy increases of more than 6% in average and up to 22% in detecting paroxysmal arrhythmias. Combining all these excellent features, TI-CNN offered an exemplification for all kinds of varied-length signal processing problems.