Multiple-model Bayesian approach to volumetric imaging of cardiac current sources

Abstract

Noninvasive cardiac electrophysiological imaging aims to mathematically reconstruct the spatio-temporal dynamics of cardiac current sources from body-surface electrocardiography data. This ill-posed problem is often regularized by imposing a certain constraining model on the solution. However, it enforces the source distribution to follow a pre-assumed spatial structure that does not always match the spatio-temporal changes of current sources. We propose a Bayesian approach for 3D current source estimation that consists of a continuous combination of multiple models, each reflecting a specific spatial property for current sources. Multiple models are incorporated into our Bayesian approach as an Lp-norm prior for current sources, where p is an unknown hyperparameter with prior probabilistic distribution defined over the range between 1 and 2. The current source estimation is then obtained as an optimally weighted combination of solutions across all models, the weight being determined from posterior distribution of p inferred from electrocardiography data. The performance of our proposed approach is assessed in a set of synthetic and real-data experiments on human heart-torso models. While the use of fixed models such as L1- and L2-norm only properly recovers sources with specific spatial structures, our method delivers consistent performance in reconstructing sources with different extents and structures.