Research On Automatic Diagnosis Of Heart Failure Based On ECG Signals

Heart failure is the final stage of the development of various heart diseases.It has the characteristics of high morbidity,high mortality,high cost and poor prognosis,and has become a major public health problem in the world.ECG is the most common non-invasive diagnostic tool,which can accurately and intuitively reflect potential activity of heart,and can provide important information for clinicians’ diagnosis.In this paper,Three automatic diagnostic models of heart failure are studied by combining signal processing and machine learning techniques,which are applied to PhysioBank public database.The purpose of this paper is to study an efficient and fast method for automatic diagnosis of heart failure,which is of great significance to the objective evaluation and diagnosis of heart failure.These three models are as follows:(1)Automatic diagnosis of heart failure based on flexible analytic wavelet transform.Firstly,the pre-processed ECG beats were decomposed into five layers by using flexible analytic wavelet transform,and the fuzzy entropy of decomposed sub-band signals are computed.Secondly,feature selection method is used to rank the entropy features of different sub-band signals,so the redundant features are removed and the optimal feature sets are obtained.Finally,different traditional machine learning methods are used to diagnose heart failure,ten-fold cross validation and confusion matrix are employed to evaluate theperformances of different classifiers.The results show that support vector machine and least squares support vector machine have great potential in automatic diagnosis of heart failure and the highest accuracy is 94.63%.(2)Automatic diagnosis of heart failure based on one-dimensional deep neural network.From the perspective of one-dimensional characteristics of ECG signals,a deep convolution neural network model suitable for processing one-dimensional time series is established.The ECG beats segmented by R point detection and non-R point detection are fed into it to compare.The results show that the model is insensitive to R-point detection of input beats,and the highest accuracy is96.71%;(3)Heart failure diagnostic model combining signal characteristics and deep learning.From the perspective of high efficiency of deep learning in picture data processing,the third-order cumulant spectrum of ECG signals are extracted,which can better display the non-linear characteristics of ECG signals.It is converted into a more suitable input form for deep convolution neural network,and AlexNet,ResNet and DenseNet models are used respectively.The results show that combined signal features and deep learning can better assist the diagnosis of heart failure,with the highest accuracy rate of 97.86%.