Advanced metaheuristic approaches and population doping for a novel modeling-based method of positron emission tomography data analysis

Abstract

This paper proposes a metaheuristic approach to solve a complex large scale optimization problem that originates from a recently introduced Positron Emission Tomography (PET) data analysis method that provides an estimate of tissue heterogeneity. More specifically three modern metaheuristics have been tested. These metaheustics are based on Differential Evolution, Particle Swarm Optimization, and Memetic Computing. On the basis of a preliminary analysis of the fitness landscape, an intelligent initialization technique has been proposed in this paper. More specifically, since the fitness landscape appears to have a strong basin of attraction containing a multimodal landscape, a local search method is applied to one solution at the beginning of the optimization process and inserted into a randomly generated population. The resulting "doped" population is then processed by the metaheuristics. Numerical results show that the application of the local search at the beginning of the optimization process leads to significant benefits in terms of algorithmic performance. Among the metaheuristics analyzed in this study, the DE based algorithm appears to display the best performance. © 2011 Springer-Verlag.