Metaheuristics applied to the thermographic detection of multicentric breast tumor

In recent years, the clinical diagnosis of breast tumors employing thermal analysis has received increasing attention as a non-invasive and less expensive alternative to classic techniques. In this paper, we propose to use and compare different metaheuristics applied to the inverse problem for estimating a multicentric tumor geometric parameters through thermography. To this end, a multicentric circular tumor in a multilayered breast model comprised of five tissue layers is considered, and from simulated temperature measurements taken on the breast skin surface, the geometric parameters are then estimated by adopting metaheuristics. To assess the effects of real data in the inverse solution, a random noise with the same order of magnitude that occurs in modern infrared cameras is inserted into the simulated measurement data. A two-dimensional steady-state nonlinear bioheat model governed by the Pennes’ equation is considered for the forward problem, which is efficiently solved by the finite element method (FEM) via FEniCS through a Newton–Raphson scheme. The inverse problem is formulated as an optimization one and then solved here by three algorithms, namely, differential evolution (DE), genetic algorithm (GA), and a self-adaptive differential evolution (SaDE). Our numerical experiments with and without noise show that SaDE outperformed the remaining techniques, indicating that this approach is a suitable alternative in this type of inverse problem.