Computational reduction for noninvasive transmural electrophysiological imaging

Noninvasive transmural electrophysiological imaging (TEPI) combines body-surface electrocardiograms and image-derived anatomic data to compute subject-specific electrical activity and the relevant diseased substrates deep into the ventricular myocardium. Based on the Bayesian estimation where the priors come from probabilistic simulations of high dimensional EP models, TEPI engages intensive computation that hinders its clinical translation. We present a reduced-rank square-root (RRSR) algorithm for TEPI that reduces computational time by neglecting minor components of estimation uncertainty and improves numerical stability by the square-root structure. Phantom and real-data experiments demonstrate the ability of RRSR-TEPI to bring notable computational reduction without significant sacrifice of diagnostic efficacy, particularly in imaging and quantifying post-infarct substrates.