Research Of ECG Inverse Problem Based On Deep-learning Approaches

Cardiovascular disease is one of the most important diseases that threaten human health nowadays.Cardiac electrical activity,which is characterized by changes in the transmembrane potential of the heart surface,is an important characterization of cardiovascular disease.The discovery and localization of specific signals and abnormal activities play an important role in the discovery and treatment of heart disease.However,due to the existing cardiac surface potential detection methods either based on clinical experience or need to cause trauma to the patient’s body,therefore,it is of great and important significance to study the ECG inverse problem of non-invasive and quantitative reconstruction of cardiac surface electrical activity.The research objective of the inverse problem of ECG is to establish the relationship between the surface potential of the heart and the surface potential by analyzing the physical information model of the human heart-thoracic cavity based on the potential information detected on the surface of the human body,thereby quantitatively reconstructing the surface potential information of the heart.Due to the ill-posed characteristic of the construction of the human body model and the complexity of the cardiac node potential,the slight noise present in the input body surface potential value will have a great impact on the final reconstruction result.In this paper,the following work is done on the inverse problem of ECG:Firstly,the ADMM algorithm is used as the basic algorithm model for solving the inverse problem of ECG,which not only ensures the convergence of the results,but also can be passed in the case of determining the regularization parameters.Multiple iterations to get relatively accurate results.In order to avoid the L-curve which is complex and has a high error rate in the traditional regularization method,the regularization parameter method is selected.The paper combines the ADMM algorithm with the deep learning method,constructs its iterative process into a neural network architecture,and proposes a new ADMM-The net method is used to solve the inverse problem of ECG.Secondly,the commonly used iterative algorithm ISTA for solving the inverse problem is used as the basis of the algorithm model.It is combined with the neural network model to improve the generalization degree of the result and improve the accuracy and convergence of the result.Sexuality,proposed a new ISTA-net method;finally,considering that the ECG signal generated by the cycle has obvious timing characteristics,the paper uses the latest LSTM network model as a method to solve the ECG inverse problem in order to process the timing signal.There is a better effect on it.In the experiment,ECG model data and body surface data simulated by ECGsim,which is dedicated to the ECG inverse problem,were used as input and output of the model.The simulated cardiac surface potential results were used as a verification set to reconstruct the corresponding epicardium and visualize it.The results show that the proposed method can reconstruct the ECG characteristics better than the traditional regularization method.It can realize the automatic selection of parameters and maintain the physiological characteristics of the results.It is an effective method to solve the ECG inverse problem.