Generating Optimally Focal and Intense Current Patterns in tES via Metaheuristic L1-L1 Search: Interior-Point vs. Simplex Algorithms (2209.11474v1)

Abstract: This numerical simulation study investigates solving the L1-norm fitted and regularized (L1-L1) linear programming (LP) problem to find a well-localized volumetric current density in transcranial electrical stimulation (tES), where a current pattern is attached through contact electrodes attached to the skin to create a stimulus in a targeted brain region. We consider a metaheuristic optimization process where the problem parameters are selected so that the final solution found is optimal with respect to given metacriteria, e.g., the intensity and focality of the volumetric current density in the brain. We focus on the effect of the LP algorithm on the solution. We examine interior-point and simplex algorithms, which constitute two major alternative ways to solve an LP task; the interior-point algorithms are based on determining a feasible solution set to allow finding an optimizer via Newton's method, while the simplex methods take steps along the edges of a polytope, subdividing the set of candidate solutions into simplicial polygons. Interior-point methods stand out among them for their complexity and predictability in terms of convergence properties at the same time. To find the current pattern, we apply five alternative optimization toolboxes: Matlab, MOSEK, Gurobi, SDPT3, and SeDuMi. We suggest that the mutual differences between these vary based on the placement of the target as well as the resolution of the two-parameter lattice. In the numerical experiments, we investigate maximizing the focality and intensity of the L1-L1 optimized stimulation current and, in the latter regard, examine its relationship to the reciprocal current pattern, maximizing the focused current density.