Parameter Estimation for the Mitchell-Schaeffer Cardiac Model Using the Particle Filter
Author(s)
Fortner, John Allen
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Cardiac action potential models are an important tool for understanding cardiac dynamics. One important tool for adapting these models to experimental data is parameter estimation, a technique by which optimal parameter values in a given model are estimated to best fit a set of experimental data. One data assimilation technique which can be used for parameter estimation is the particle filter, which works by sending numerous particles – sets of estimates of each state and parameter value - through a probability distribution and optimizing the distribution in order to best estimate the distribution of state values over time. In this project, we aimed to demonstrate the efficacy of the particle filter for the purpose of parameter estimation in the Mitchell-Schaeffer cardiac model. By analyzing the impact of adjustments to several hyperparameters of the particle filter, we demonstrated the tradeoffs involved in their selection and the capability of the filter to be fine-tuned depending on the details and difficulty of the problem. While different combinations of changes to the set of hyperparameters can have surprisingly varied effects, we viewed the changes created by modifying each individually in order to shed some light on the impact each change can have. We were able to demonstrate the effectiveness of the filter in estimating a single parameter, as well as limited effectiveness in estimating multiple parameters at the same time. We were thus able to show that the particle filter is a viable method for parameter estimation in this context.
Sponsor
Date
Extent
Resource Type
Text
Resource Subtype
Undergraduate Research Option Thesis