Multilevel parallelism scheme in a genetic algorithm applied to cardiac models with mass-spring systems

In this work we propose an adaptive parallel genetic algorithm, which is able to automatically choose which loops should run in parallel in order to balance the workload sharing and the extra cost to manage threads' forks and joins. Our genetic algorithm finds parameters to our cardiac simulator based on mass-spring systems and cellular automata, in order to reproduce a cycle of contraction and relaxation of the human left ventricle. The ventricle geometry was obtained by extracting information from MRI data. Our multilevel parallelism scheme decreased the execution time up to 12 % and the GA adjusts the cardiac model so that it reproduced a cardiac cycle. The advantages and drawbacks of our method are discussed as well as its limitations.