Feature Mining Of 12-lead ECG Signals Based On Tensor Space

Heart disease has a long course of treatment and is difficult to cure.In addition,because there are often no obvious signs before the onset of illness,once the onset is very easy to have life-threatening,so the research and prevention of heart disease has become a top priority.In the past,ECG signals were mainly collected through single leads and two leads.In order to analyze abnormal signals more fully,12-lead ECG signals were acquired using the RM62160 C physiological instrument.However,with the increase in the number of leads,the amount of data collected in a unit of time also increases,which would be inconvenient for doctors to quickly diagnose the patient's condition.Therefore,this dissertation focuses on the feature dimension reduction of 12-lead ECG signals,extracts the main features,and subsequent cluster analysis of the feature data.The main content of this article has the following four aspects:Firstly,a 12-lead ECG signal acquisition experiment was designed.Combined with the two existing RM6280 C multi-channel physiological instruments in the laboratory,referring to the hospital's standards for electrode attachment,12 people collected a total of 31 leads and 12 leads in a lying state.Due to the large amount of ECG data collected during the experiment,there is noise interference.However,this part of the noise data is basically redundant for the study,so it is necessary to perform denoising first,and then follow the single-cycle segmentation processing based on the R-wave feature extraction method.Secondly,quantify high-dimensional multivariate ECG signals.For the preprocessed ECG signal,the main component analysis method is used in the vector space to deal with the situation that easily loses the heartbeat and lead structure information.This paper adopts a tensor space-based processing method that is different from the previous ones—the ECG signal.Quantitative.Experiments show that this can avoid the loss of structural information in the ECG signal and thus more accurately analyze the abnormal condition of the patient's ECG signal.Again,dimensionality reduction is performed using multi-linear principal component analysis based on tensor space.Taking into account the large amount of 12-lead ECG signals after quantification,in order to quickly discover useful data features and reduce the time-series dimensions,multi-linear principal component analysis(MPCA)was used in this paper.Dimensionality reduction.This method was applied to the laboratory measurement of the 12-lead ECG signal dataset and the online Physionet dataset.The experimental results show that the MPCA can effectively reduce the ECG data,and the dimensionality reduction efficiency can reach more than 80%.Subsequent PCA was used to compare the reduced dimensions of ECG data after quantification,and it was found that MPCA based on tensor space has better dimensionality reduction.Finally,a spectral clustering method based on weighted Gaussian distance is proposed to cluster high-dimensional multivariate 12-lead ECG signals.Taking into account the characteristics of ECG signal nonlinearity,this paper chooses a spectral clustering method based on graph theory.Firstly,a spectral clustering method based on tensor distance is designed.In addition,spectral clustering based on weighted Gaussian distance is designed.The results show that the clustering method based on the weighted Gaussian distance is better for both the measured and Physionet data,and the clustering evaluation index RI and ARI both reach 0.80.