Abstract
To model variability of cardiac action potentials (APs), a population of models (PoM) consisting of different sets of a model's parameter values can be created and calibrated to match observed variability in properties such as AP duration (APD). However, producing appropriate parameter sets for the PoM can be difficult and time-consuming. We adapted a particle swarm optimization (PSO) optimization technique to generate a population of models efficiently. Our population PSO (PPSO) algorithm discourages convergence to a local minimum, and instead guides the search to explore low-error areas of parameter space, yielding many parameter sets that can reproduce the variability of biomarkers seen in real tissue data. Using canine ventricular microelectrode recordings and a synthetic dataset, we extracted sets of APD- and voltage-based biomarkers, allowing ±10% and ±30% variations of the base biomarker values to represent variability. We created 5000- and 2500-member PoMs fitting the parameters of the Fenton-Karma (FK) and ten Tusscher-Noble-Noble-Panfilov (TNNP) models to the biomarker ranges using PPSO. Compared to a random approach, our novel PPSO method produced PoMs matching biomarkers with similar coverage of parameter space for both the FK and TNNP cases, but with greater computational efficiency, accepting up to 10 times more candidate parameter sets.