A fast ECG diagnosis by using spectral artificial neural network (SANN) approach

Abstract

Electrocardiogram (ECG) diagnosis is a widely-used clinical approach because it has been proven as an efficient way to monitor and diagnose cardiac diseases. However, because of a large amount of the raw ECG signal data, it is time-consuming to analyze the ECG signal. Moreover, ECG signal analysis is a nonlinear problem, which worsens the difficulty to diagnose the ECG signal. Therefore, many Neural Network (NN)-based ECG analysis approaches were proposed to perform ECG diagnosis in time domain in recent years, which can improve the ability of ECG analysis. By decomposing the ECG signal, the P, Q, R, S, and T waves can be acquired for the further analysis based on the information of the features of these waves such as the amplitude and waves interval. However, because of the complex pre-process for the signal purification and feature extraction, this kind of time-domain ECG signal process still suffers from the long computation time. To solve the problem, we propose a Spectral Artificial Neural Network (SANN) approach for the fast ECG diagnosis in this paper. Compared with the conventional time-domain-based approaches, the SANN analyzes the ECG signal in frequency domain. Because most of the noises in the raw ECG signal are high-frequency signals, the proposed SANN focuses to analyze the low-frequency signals in ECG spectrum. By this way, the proposed SANN not only reduces the pre-processing time but the diagnosis time. To obtain the proper window size for the precise ECG diagnosis, we further propose a heuristic window size adjustment in this paper, which helps to extract the suitable features. The experimental results show that the proposed SANN approach can reduce the ECG diagnosis time by 80% compared with the conventional ECG analysis with only 5% average diagnosis accuracy loss of the cardiac diseases.