ECG signals’ detection using a Dropout Deep Convolutional Neural Network

Abstract

In the common application of ECG, a technique of detecting and analyzing ECG signal waveform based on deep learning for large samples of ECG signal is proposed to replace artificial image recognition. For higher accuracy, it’s necessary to improve the existing CNN classifiers. First, it’s to pre-process the ECG signal to remove noise for QRS detection. Secondly, it’s to extract the QRS complex features, which contain the R wave— with the largest amplitude, can be used to capture and locate protruding morphological data, then send the datasets to the model to train. Eventually, the results are obtained. Using the proposed Dropout-DCNN model, the overall accuracy of the detection reaches 99.52% when we compare it with manual ones marked by experts in the MIT-BIH library. It can be considered that this module better than other ones and can be applied in clinical practice.