Research On Processing Method Of MECG Signal Based On Tensor

In recent years,people’s attention to health has increased.The World Health Organization’s statistics show that heart disease is one of the most deadly diseases,so it is very important to know the heart condition in time.With the development of telemedicine diagnostic systems,the use of wearable devices that monitor ECG signals is becoming more common,and the multi-lead ECG signals(MECG)monitored by this device can provide more accurate diagnosis for medical personnel.information.However,telemedicine diagnostic systems generate large-scale data for long-term monitoring of multi-lead ECG signals.To improve the transmission and storage efficiency of multi-lead ECG signals,these data need to be compressed.At the same time,because the remote diagnostic medical system may not be able to collect at all times or because of the instability of the transmission channel,only incomplete data can be obtained,and the data needs to be restored in order to accurately diagnose the heart disease.MECG signals have strong correlations in heartbeats,between heartbeats and leads.Traditional vectors and matrices are not easy to exhibit multiple correlation characteristics.In this paper,the correlation of MECG signal is used to construct a MECG signal tensor structure.The main research work in this paper includes the tensorization of the MECG signal and the compression and recovery of the MECG signal by the tensor decomposition method.The main research results of this paper are:(1)Aiming at the problem of different lengths of MECG heatbeats in the process of tensorization,a periodic normalization method for the end-value padding is proposed.After the R-peak identification of the MECG signal,the irregular heartbeat period is filled by the value at the end of the cycle,and the reconstruction accuracy is improved at a similar calculation time compared with the zero-padding method;compared to interpolation,it reduces the time overhead.(2)For the compression of MECG signals,a framework for decomposing the MECG signal tensor into three factor matrices using CP decomposition is proposed.According to the three-sided distribution is uneven of the MECG signal,the tensor is reduced by the matrix product of the tensor to obtain a core tensor,and then the core tensor is subjected to CP decomposition(CPTC).The compression process of the MECG signal tensor is divided into sub-problems that acquire the orthogonal factor matrix and the non-orthogonal factor matrix.The performance of the CPTC method was evaluated by simulation experiments.The average percent mean square error was 2.38%-5.63% and the CR was 17.53-19.07 in the PTB diagnostic database.And based on the comparison of the reconstructed signal and the R-peak recognition accuracy of the original signal,it is verified that the CPTC well preserves the characteristic information of the MECG signal.(3)For the recovery of incomplete MECG signals,a vertex least squares method of CP decomposition is proposed.The method iteratively updates the feature components represented by the vertices in the graph on the three-part graph representing the M ECG signal.Verifies the convergence of CP-VLS on the simulated data and the real multilead ECG signal data.The experimental results show that in the case of the missing rate between 20%-80%,the CP-VLS method obtains less than 1% reconstruction error at the sampling positions,unsampling positions and all positions.