Synthetic ECG produced by 2D reaction-diffusion model to fit real time electrocardiographic signals

Shane Loeffler and Joseph Starobin
UNCG


Abstract

Early detection of cardiac diseases is essential to choose the best strategies to prevent cardiac related deaths. Currently, real time diagnosis using surface ECG signals have some limitations. To improve the early detection of cardiac abnormalities we propose to use reaction-diffusion equations in a simplified 2D geometry to produce synthetic ECGs for real time fitting of electrocardiographic signals.     The cardiac electrical system is represented by computationally efficient Bueno-Orovio-Fenton-Cherry (BOFC) reaction-diffusion model. We solve this model numerically in a simplified 2D horseshoe shape which is indicative of the coronal cross section of the left ventricle. We use a finite difference grid in which forward Euler methods are applied to solve the non-linear BOFC reaction diffusion equations. Synthetic ECG signals are computed using average dimensions of an adult human heart. We consider BOFC equations within inhomogeneous domains comprised of epicardial, endocardial and mid-myocardial layers which allow us to simulate synthetic ECG signals with adequate temporal and topological characteristics.     Performing multiple numerical simulations we identified the major model parameters which had the greatest effect on the shape and temporal dynamics of synthetic ECG. By fitting these parameters to surface ECG signals we were able to trace corresponding variations related to specific cardiac abnormalities. In particular, we monitored variations of QT interval duration, ST segment elevation/depression, T-wave and QRS-complex durations/amplitudes and categorized them in terms of our model’s parameter changes.     Altogether the ability of real time monitoring of model parameters which cause specific variations of synthetic ECG allows one to get more accurate assessment of specific cardiac abnormalities, thus improving the precision of potential responses in emergency care. Also, data collected using the simplified but still sufficiently robust BOFC model can be effectively applied within machine learning algorithms. Machine learning could provide better assessment of real time electrical changes of the heart.