Hide/Show Apps

A Learning Based Statistical Approach for Combining Multiple Measurements in Electrocardiographic Imaging

Kalman filter has been applied in literature to in-verse electrocardiography problem as a spatio-temporalmethod. However, there is still an open question of howthe essential parameters in the state-space representationare found without claiming strong assumptions. In thisstudy, we proposed a maximum likelihood (ML) estimationbased method which incorporates multiple body surfacemeasurements to estimate the parameters that are essen-tial to use Kalman filter.Our proposed approach, Maximum Likelihood Infer-ence & Filtering (MLIF), was compared with zero orderTikhonov regularization and Bayesian maximum a poste-riori estimation (BMAP) by using three different trainingsets under two different measurement noise levels. Theresults showed that mean correlation coefficient (CC) forTikhonov regularization is 0.60, and mean CC ranges 0.64to 0.82, and 0.66 to 0.99 for Bayesian MAP and MLIF un-der 30 dB SNR measurement noise, respectively. Under10 dB SNR, mean CC is 0.37 for Tikhonov regularization,and mean CC ranges 0.53 to 0.78, and 0.53 to 0.98 forBayesian MAP and MLIF, respectively.