Computational models of cardiac electrophysiology have contributed significantly to the elucidation of arrhythmic mechanisms in humans during the last decades. Using multi-scale approaches, mode of action as well as limitations of therapeutic approaches have moreover been identified. However, a substantial share of patients still suffers from cardiac arrhythmias such as atrial fibrillation or ventricular tachycardia today.
Lately, more and more evidence for significant differences between different patient subgroups piled up calling for group-specific investigations. These differences include comorbidities, arrhythmia history, genotype, and medication. However, current state of the art computational models consider only few of these characteristics.
In this work, two approaches for parametrizing ion current formulations according to experimental data representing specific patient subgroups will be compared. In general, the parametrization is a high-dimensional, non-linear problem. A further challenge stems from the fact that parameter identifiability might not be known a-priori.
The first approach regards the parameter estimation task as an optimization problem. To overcome local minima, a hybrid strategy combines gradient-based and population-based heuristic optimization. The second approach uses multivariate forward and inverse metamodeling. The advantage of this statistic approach is the implicit sensitivity analysis yielding information on parameter identifiability. However, this recent strategy has not been applied to ion current formulations before.
In this, work the applicability of the statistic approach will therefore be assessed to identify an optimal parameter estimation and identifiability assessment strategy as a basis for future individualized computational modeling of cardiac electrophysiology.
Comparison of statistic and optimization-based approaches for parameter estimation of ion current formulations