Research On Standard12-lead ECG Synthesis Methods

With the increasingly common cardiac diseases and the improvement of people’s demand for self-monitoring, the wearable ECG signal acquisition systems for home monitoring and telemedicine have been attracting more and more attention. However, this kind of wearable devices could directly collect only less than3ECG leads, which is not suitable for complete ECG signal analyses. So standard12-lead ECG lead synthesis from relatively fewer ECG leads.is a very effective way for the applications.In this paper, firstly a method is proposed to reconstruct the standard12-lead ECG from a3-lead subset (Ⅰ, Ⅱ and V2) by optimizing the back propagation neural network with genetic algorithm (GA-BP). The non-linear method BP network is more suitable for ECG signal processing and GA is utilized to optimize the initial settings of the weights and biases in the BP network. Based on the results experimented on the study population of39subjects randomly selected from the PTB diagnostic ECG database, the proposed GA-BP method is proved to achieve accurate synthesis of the standard12-lead ECGs, showing significant improvements over the common BP network and linear transformation method in terms of correlation coefficient values and root-mean-square errors.This paper further presents a novel adaptive lead synthesis method based on ECG feature extraction and region partition. The ECG waveform is partitioned into four regions according to the ECG features and fluctuation characteristics, and these four regions’networks are trained separately. When used for synthesis, ECG waveform is reconstructed from a specific trained network adaptively. The experiments reveal the significant superiority of the proposed adaptive partition method not only in the synthesis of the overall ECG waveform, but also in reconstructing the ECG features.