Intelligent Optimization Algorithm And Support Vector Regression Application In The Inverse ECG Problem

The typical inverse ECG problem is to non-invasively reconstruct the heart’s electricalactivity from body surface potentials (BSPs), such as epicardial, endocardial and thetransmembrane potentials. The reconstructed potential values on the cardiac surface or within themyocardium can provide far more complicated and detailed electrophysiological informationthan that of BSPs. Therefore, the inverse ECG problem is of great clinical value in the study ofheart disease diagnosis. In this paper, the electrical function imaging of the heart which is basedon myocardial transmembrane potentials reconstruction can be treated as the inverse ECGproblem and it is a regression problem with multi-inputs and multi-outputs which can be solvedby the support vector regression (SVR) method, that is deduce the multi-outputs problem of themyocardial transmembrane potentials of heart surface from the multiple input of the bodysurface.Feature extraction is an important task for pre-processing the original input data. This paperproposes the application of principal component analysis (PCA) and kernel principal componentanalysis (KPCA) to the SVR method for feature extraction of the body surface potentialdistribution sample training set. The experiment shows that the SVR method with the two featureextractions (PCA-SVR and KPCA-SVR) can perform better than single SVR in terms of thereconstruction of the TMPs on cardiac surfaces. Compared to PCA-SVR method, KPCA-SVRfeatures good approximation and generalization ability in reconstructing the TMPs.In order to obtain an effective SVR model with optimal regression accuracy andgeneralization performance, the hyper-parameters of SVR must be set carefully. As for themethods to optimize parameters of SVR model, three different intelligent optimization methods,i.e. Genetic Algorithm (GA), Differential Evolution (DE) algorithm, and Particle SwarmOptimization (PSO) are proposed to determine the hyper-parameters of the SVR model and toachieve the purpose of making the SVR model optimized. Then, the myocardial transmembranepotentials of heart surface will be predicted based on the regression model established and thebody surface potentials. In this paper, we attempt to investigate which one is the most effectiveway in reconstructing the cardiac TMPs from BSPs, and a full comparison of their performancesis also provided. The experimental results show that these three optimization methods are well performed in finding the proper parameters of SVR, and can yield good generalizationperformance in solving the inverse ECG problem. Moreover, compared with DE and GA, PSOalgorithm is more efficient in parameters optimization, and performs better in solving the inverseECG problem, leading to a more accurate reconstruction of the TMPs.