Calibration of single-cell model parameters based on membrane resistance improves the accuracy of cardiac tissue simulations

• Published in: Journal of computational science Vol. 53; p. 101375
• Main Authors: Pouranbarani, Elnaz, Berg, Lucas Arantes, Oliveira, Rafael Sachetto, Weber dos Santos, Rodrigo, Nygren, Anders
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.07.2021
• Subjects: Cardiac modeling; Membrane resistance; Optimization; Parameters; Tissue simulation; Tissue simulation; Cardiac modeling; Membrane resistance; Parameters; Optimization
• ISSN: 1877-7503; 1877-7511
• DOI: 10.1016/j.jocs.2021.101375

•The hypothesis that including Rm in cardiac cell model’s calibration enhances the precision of tissue-level simulation is tested.•Various tissue configurations are designed to verify the hypothesis.•Well-defined statistical metrics verify the hypothesis.•The insights of this study are beneficial for researchers to develop more precise cardiac cellular and tissue models. In the calibration process, replicating some cellular properties is the main focus, while the importance of membrane resistance (Rm) that is primal in the tissue-level modeling is often underestimated. Previously, we presented a framework in which Rm in addition to action potential (AP) waveform was considered in the cellular model fitting. In this paper, we test the hypothesis that this approach for tuning cellular model parameters improves the accuracy of simulations at the tissue level. In doing so, two different sets of single-cell models are generated via independent realizations of our multi-objective optimization approach. In the first set of calibration (Model I), root-mean-square error (RMSE) of AP, and absolute error (AE) of maximum upstroke velocity are included as optimization functions; however, in the second set of calibration (Model II), RMSE of Rm curve in the repolarization phase is also added to the optimization functions. The calibrated cell models are then used in several tissue configurations of physiological relevance. We adopt well-defined evaluation metrics to compare tissue models tuned using Models I and II. In the source-sink mismatch configuration, the average absolute relative error (ARE) of the critical transition border, defined as the smallest required window width between source and sink for AP propagation, is less than 4.7 % in Model II, while this error is increased to more than 8.9 % in Model I. In addition, in Model I, the average ARE of total time for activation of tissue is 3.3–6.3 %; however, in Model II, this error is reduced to 0.7–1.6 %. In the Purkinje-myocardium configuration, the average of RMSE of activation time map is reduced approximately 75 % in Model II. Finally, in the transmural APD heterogeneity configuration, the average AREs of AP duration (APD) and APD dispersion (i.e., the difference between maximum and minimum of APD) are about 13.2 % and 17.4 % in Model I and 5.8 % and 6.8 % in Model II, respectively. Overall, our results demonstrate that consideration of Rm in the single-cell optimization procedure yields a substantial improvement in the accuracy of tissue models.