Lattice layout and optimizer effect analysis for generating optimal transcranial electrical stimulation (tES)...

Area of Science:

• Computational neuroscience
• Biomedical engineering
• Optimization algorithms

Background:

• The L1-norm fitted and penalized (L1L1) optimization method is used for multichannel transcranial electrical stimulation (tES).
• The L1L1 framework defines tES montages through linear programming, optimizing objective functions based on hyperparameters.

Purpose of the Study:

• To assess the computational performance and reliability of various optimization packages, algorithms, and search methods with the L1L1 method.
• To identify the most efficient solvers for generating tES current patterns.

Main Methods:

• Compared optimization solvers from Matlab, MOSEK, Gurobi, CVX (SeDuMi, SDPT3).
• Utilized linear programming techniques: Interior-Point (IP), Primal-Simplex (PS), and Dual-Simplex (DS).
• Implemented exhaustive, recursive, and heuristic direct search algorithms for the L1L1 optimization.

Main Results:

• Gurobi's IP solver was the preferred Interior-Point method.
• MOSEK's PS and DS solvers were preferred Simplex methods.
• These preferred methods demonstrated significant computational time efficiency for the L1L1 task across different search strategies.

Conclusions:

• The L1L1 method, with optimized solvers like Gurobi's IP or MOSEK's Simplex, is effective for maximizing tES focality and intensity.
• Solver discrepancies may arise from parameter sensitivity or lattice resolution differences, impacting bipolar configuration generation.