Neural Network ECGI MRI Integration for Cardiac Assessment

The patent uses machine learning models as an intermediary to bridge the gap between non-invasive ECGI measurements and endocardial electrical activities. The ML model learns the mapping relationship from epicardial potentials (measured non-invasively) to endocardial electrical states, enabling accurate inference without direct endocardial measurement or invasive procedures.

The patent creates a virtual copy of the endocardial electrical activities through machine learning inference. Instead of directly measuring endocardial potentials (which would require invasive catheter insertion), the system generates a computational replica of endocardial electrical states based on non-invasive epicardial measurements and anatomical models.

The patent replaces complex physics-based computational models with machine learning models. Instead of solving forward and inverse electrocardiography problems using computationally intensive physics simulations, the system uses trained neural networks that rapidly infer electrical activities from measured data, dramatically reducing computational burden while maintaining accuracy.

The patent performs preliminary training of machine learning models using comprehensive physics-based simulations and labeled data. This preliminary action pre-computes the complex relationships between electrical activities and body surface potentials, so that during actual clinical use, only lightweight inference is needed rather than full physics-based reconstruction.