An ADMM-net Solution to Inverse Problem of Electrocardiology

Abstract

In this paper, we proposed a new ADMM-net method to solve the inverse problem of electrocardiograph(ECG) and choose the parameter of regularization automatically and efficiently. In order to solve the inner ill-posed problem and make the result more precise, firstly, we built an Alternating Direction Method of Multipliers(ADMM) algorithm model to solve the inverse problem of ECG, unlike the classical methods such as Tikhonov, truncated total least squares(TTLS), truncated singular value decomposition(TSVD), the iterative property of the algorithm ensures the convergence of the results, then we built a neural network based on the ADMM algorithm model which allows regularization parameters to be automatically selected and more accurate results. To get enough cardiac surface and surface potential data, we use a real human model on the ECGsim software to obtain the simulated potentials on the cardiac surface and body surface for training. The result of our experiment shows that the proposed method has a better performance in reconstructing epicardial surface potentials distribution than the common regularization method such as Tikhonov, TTLS, TSVD method.