Noninvasive imaging of cardiac electrophysiology has taken on increasing clinical significance. Classical efforts all depend on body surface potential maps (BSPMs) as the single data source, and the solutions are confined to equivalent sources or transmembrane potentials (TMPs) on the heart surface. A novel and integrated paradigm for noninvasive 3D TMP imaging from patient clinical data is presented. Based on the Bayesian perspective and system approach, the rationale is to depict 3D TMP dynamics by incorporating a priori model-based physiological knowledge, and integrating structural (tomographic image sequence) and functional (BSPMs) data in the presence of associated uncertainties....[ Read more ]

Noninvasive imaging of cardiac electrophysiology has taken on increasing clinical significance. Classical efforts all depend on body surface potential maps (BSPMs) as the single data source, and the solutions are confined to equivalent sources or transmembrane potentials (TMPs) on the heart surface. A novel and integrated paradigm for noninvasive 3D TMP imaging from patient clinical data is presented. Based on the Bayesian perspective and system approach, the rationale is to depict 3D TMP dynamics by incorporating a priori model-based physiological knowledge, and integrating structural (tomographic image sequence) and functional (BSPMs) data in the presence of associated uncertainties.

Firstly, we propose that a Bayesian paradigm, based on system modeling and statistic inversion, is desirable to combine a priori physiological knowledge and patient specific data within their respective uncertainties. The cardiac electrophysiological system is fully developed and customized to patient heart and torso structure derived from tomographic medical images. After properly reformulating it into a stochastic state space representation, sequential data assimilation is utilized to estimate 3D TMPs by combining the information in the prior system and patient BSPMs within associated uncertainties. The sensitivity of this framework to practical modeling and data errors is analyzed, and its application to cardiac arrhythmia imaging is presented.

Based on this model-constrained Bayesian framework, we further propose to integrate structural (tomographic image sequence) and functional (BSPMs) data to extend the spatiotemporal coverage of cardiac information and obtain more patient specific estimates. An unknown systematic error is modeled and introduced into the cardiac electrophysiological system, which provides a ready platform for the data integration. Algorithm of adaptive data assimilation is developed where tomographic image sequence dynamically guides the TMP imaging from BSPMs through cardiac electromechanical coupling. Experiments on cardiac arrhythmia imaging demonstrate the benefits of data integration over using single data sources.