Probabilistic reachability for multi-parameter bifurcation analysis of cardiac alternans

TitleProbabilistic reachability for multi-parameter bifurcation analysis of cardiac alternans
Publication TypeJournal Article
Year of Publication2018
AuthorsIslam, Md. Ariful, Cleaveland Rance, FENTON FLAVIO H., Grosu Radu, Jones Paul L., and Smolka Scott A.
JournalTheoretical Computer Science
Date Published02/2018
KeywordsBifurcation analysis, Cardiac alternans, Cardiac modeling, Formal methods, Hybrid systems, Probabilistic reachability

Using a probabilistic reachability-based approach, we present a multi-parameter bifurcation analysis of electrical alternans in the two-current Mitchell–Schaeffer (MS) cardiac-cell model. Electrical alternans is a phenomenon characterized by a variation in successive Action Potential Durations generated by a cardiac cell or tissue. Alternans are known to initiate re-entrant waves and are an important physiological indicator of an impending life-threatening arrhythmia such as ventricular fibrillation. The multi-parameter bifurcation analysis we perform identifies a bifurcation hypersurface in the MS model parameter space, such that a small perturbation to this region results in a transition from highly likely alternans to highly likely non-alternans behavior. Our approach to this problem rests on encoding alternans-like behavior in the MS model as a five-mode, multinomial hybrid automaton. To perform multi-parameter bifurcation analysis of cardiac alternans, we first treat the parameters in question as bounded random variables. We then apply a sophisticated guided-search-based probabilistic reachability analysis to compute a bounded bifurcation region (possibly very tight) that contains the bifurcation hypersurface (BH). Our probabilistic reachability analysis uses a technique that combines a δ-decision procedure with statistical tests. In the process of computing the bifurcation region, we further partition the parameter space into two more regions such that any valuation chosen from one of the regions will either produce alternans or non-alternans behavior with a probability greater than a user-defined threshold.