The Study On Differential EvolutionOptimization Algorithm And Extreme Learning Method For Solving The Inverse ECG Problem

The inverse ECG problem is noninvasive reconstruct the cardiac transmembrane potentialdistribution from the remote body surface potential distribution. As we know, the transmembranepotential distribution can provide more detailed information than the surface ofelectrophysiological potentials and therefore improve the accuracy of heart disease diagnosis.The inverse ECG problem can also be treated as a regression problem with multi-input (bodysurface potential distribution) and multi-output (cardiac transmembrane potential distribution). Inthis paper, the Extreme Learning Machine (ELM) method is used to solve the regressionproblem.In the first experimental setting, the body surface and cardiac transmembrane potentialcan be simulated by using the heart surface source model. In the second experimental setting, theheart surface and body surface potential information of the heart WPW syndrome is obtainedform the ECGSim software. In order to make the Extreme learning machine with goodgeneralization ability and fitting accuracy, the model parameters must be selected effectively. Inthis article, the differential evolution algorithm (DE) is used to optimize the parameter selectionof the Extreme learning machine method, so that the regression model can be built effectively.Based on the regression model, the cardiac transmembrane potential can be accurately predictedfrom the distribution of the surface potential. In this paper, the support vector regression (SVR),ELM, ELM-kernel are also used to solved the inverse ECG problem, and some comparisons arealso made to analysize the performances of each method.The experimental results show that the above mentioned four algorithms can solve theinverse ECG problem. However, compared with SVR methods, ELM and ELM-kernel methodhas better performances in terms of reconstruction accuracy and reconstruction speed. Moreover,compared with the ELM method, the kernelized ELM method features good approximation andgeneralization ability when reconstructing the TMPs.