Sarah Geneser, Ph.D.

Thesis

Heart Position











Quantification of Stochastic Behavior in Cardiac Electrophysiological

Models

Mathematical models of intrinsically stochastic systems have typically been reduced to deterministic forms to avoid the computational expense and cumbersome implementation commonly associated with solutions to stochastic systems. The generalized polynomial chaos theory provides an efficient means of representing stochastic processes within random systems. In combination with either the stochastic Galerkin or stochastic collocation numerical techniques, one wields an extremely powerful tool for efficiently computing the stochastic solutions of random systems. Stochastic modeling has proven useful for accomplishing a number of goals. Firstly, intrinsically stochastic physiological phenomena are often modeled with deterministic mathematical systems to reduce complexity. In focusing on mean behaviors only, important aspects of a system's behavior may go unnoticed. With the gPC representations coupled with either the stochastic Galerkin or stochastic collocation method, it is now possible to quantitatively examine ranges of behavior outputs to distributions of model parameters or inputs. Secondly, these techniques can be applied to quantitatively determine the impact uncertainty in directly measured model inputs and parameters has upon model outputs. A third utility of the method applies to systems where the model parameters cannot be measured directly. In this case, parameter values must instead be fit to recorded response (output) data. Typically, modelers utilize statistics for the response, yet seek only to fit the mean parameter values. The excellent computational efficiency of the generalized polynomial chaos-stochastic collocation gPC-SC methods allows one to seek to fit not only mean model parameter values, but higher statistical moments as well, thus quantifying the uncertainty inherent in such parameters.

In this thesis, we apply both the gPC-SG and gPC-SC techniques to cardiac electrophysiological models for: (1) characterization of simulated behavior response, (2) uncertainty quantification and sensitivity analysis, and (3) determination of higher stochastic moments in fitted model parameters for cardiac electrophysiological systems of interest within the bioengineering community. For breadth, we focus on both microscopic level Markovian ion channel models as well as a high level physiological system- that of the forward problem of electrocardiography. During the process, we explore and expand the mathematical and computational aspects of the stochastic Galerkin and collocation methods.