QRS Detection in ECG Signal Based on Residual Network

Abstract

This paper presents a powerful and accurate convolutional neural network (CNN) model of QRS detection in electrocardiogram (ECG) based on one dimensional residual network. The one dimensional CNN model can obtain the time-domain characteristics of QRS waveform and determine whether each sampling point in ECG signal belongs to the QRS wave. Because of denoising and normalization of ECG signal before being input into the model, the model has a great generalization ability. The main advantages of the model are reducing the complex preprocessed steps of ECG signal and achieving the detection of QRS end to end, which greatly improve the efficiency of detection. Compared with traditional methods, our model is more robust to noise and it's easier to implement. We use 30 records in mitdb to train the model and use 16 records in mitdb to test the model. The positive predictivity rate and sensitivity are 99.98% and 99.92% respectively in test set.