Metamodelos e algoritmos genéticos para a resolução de um problema inverso de eletrofisiologia cardíaca

Abstract

The modeling of the electrical activity of cardiac cells has a large medical and scientific interest. The main objective of this work is to avaliate a metodology based in Genetic Algoritms (GA) that aims to automaticaly adjust a celular electrophysiology model to experimetals data. Those inverse problems are, computationally, extremely expensive. Thus, were evaluated two different techniques to accelerate the GA execution: Parallel Computation and Metamodels. The GA were parallelized using a master-slave classic model. A metamodel was implemented whit a purpose is replace the cellular model evaluation by less expensive estimations. This metamodel is dynamically created using an historic of individuals previously evaluated by the cellular model and stored in a database, which was implemented of two forms: an contiguous list and a struct known as Kd-Tree. The GA execution in 20 processors was twice quicker than in 10 processors. Furthermore, the Kd-Tree structure accelerated the functions of search into database in up to 4000 times when compared with the implementation using contiguous list. However, was possible to estimate up to 40% of the individual’s fitness using the metamodel with no quality losses into parameters adjust, what reduced the total execution time of 3 hours (without metamodel) to 2 hours.