Convolutional Neural Network Based ECG Quality Assessment Using Derivative Signal

Abstract

Automatic electrocardiogram (ECG) signal quality assessment is most essential for reducing false alarm rates and energy consumption of health monitoring devices. In this paper, we propose an automatic ECG signal quality assessment method by using derivative ECG (dECG) signal and convolutional neural network (CNN). The proposed dECG-CNN method is tested by using the noise-free ECG and noisy-ECG signals collected from standard databases. On the benchmark performance metrics, the proposed method had an accuracy of 100%. Evaluation results further showed that the proposed method can achieve false alarm reduction rate of 100% for a total number of 30850 noisy segments. It is observed that the proposed method outperforms the recent ECG quality assessment methods such as pretrained CNN models, deep belief network. The method has great potential in improving accuracy and reliability of cardiovascular diseases (CVD) diagnostic systems and improving energy efficiency of battery powered wearable and portable medical devices. The optimal CNN model for ECG signal quality assessment is implemented on Raspberry pi-4 as real time computing platform. © 2022 IEEE.