Physics-Constrained Deep Learning for Robust Inverse ECG Modeling

The rapid development in advanced sensing and imaging brings about a data-rich environment, facilitating the effective modeling, monitoring, and control of complex systems. For example, the body-sensor network captures multi-channel information pertinent to the electrical activity of the heart (i.e., electrocardiograms (ECG)), which enables medical scientists to monitor and detect abnormal cardiac conditions. However, the high-dimensional sensing data are generally complexly structured. Realizing the full data potential depends to a great extent on advanced analytical and predictive methods. This paper presents a physics-constrained deep learning (P-DL) framework for robust inverse ECG modeling. This method integrates the physics law of the cardiac electrical wave propagation with the advanced deep learning infrastructure to solve the inverse ECG problem and predict the spatiotemporal electrodynamics in the heart from the electric potentials measured by the body-surface sensor network. Experimental results show that the proposed P-DL method significantly outperforms existing methods that are commonly used in current practice. Note to Practitioners—This article is motivated by the remarkably increasing applications of advanced medical sensing and imaging technique for data-driven disease diagnosis and treatment planning. For instance, body surface potential mapping (BSPM) can be non-invasively acquired to delineate the spatiotemporal potential distribution on the body surface, enabling medical scientists to infer the electrical behavior of the heart (i.e., heart surface potential (HSP)). However, the reconstruction of HSP from BSPM is highly sensitive to measurement noise and model uncertainty. This paper presents a novel physics-constrained deep learning (P-DL) framework by encoding the physics-based principles into deep-learning for robust HSP prediction. Experimental results demonstrate the extraordinary performance of the proposed P-DL model in handling measurement noise and other uncertainty factors in inverse ECG modeling.