Seeking Optimal Montage for Single-Pair Transcranial Direct Current Stimulation Using Bayesian Optimization and Hyperband-A Feasibility Study

Cheolki Im; Jongseung Lee; Donghyeon Kim; Sung Chan Jun; Hyeon Seo

doi:10.1016/j.neurom.2024.09.475

Seeking Optimal Montage for Single-Pair Transcranial Direct Current Stimulation Using Bayesian Optimization and Hyperband-A Feasibility Study

Neuromodulation. 2024 Nov 9:S1094-7159(24)01188-7. doi: 10.1016/j.neurom.2024.09.475. Online ahead of print.

Authors

Cheolki Im¹, Jongseung Lee², Donghyeon Kim², Sung Chan Jun³, Hyeon Seo⁴

Affiliations

¹ School of Electrical Engineering and Computer Science, Gwangju Institute of Science and Technology, Gwangju, South Korea.
² Research Institute, Neurophet Inc, Seoul, South Korea.
³ School of Electrical Engineering and Computer Science, Gwangju Institute of Science and Technology, Gwangju, South Korea; Artificial Intelligence Graduate School, Gwangju Institute of Science and Technology, Gwangju, South Korea.
⁴ Department of Computer Science and Engineering, Gyeongsang National University, Jinju, South Korea. Electronic address: hseo0612@gnu.ac.kr.

PMID: 39520458
DOI: 10.1016/j.neurom.2024.09.475

Abstract

Objectives: Transcranial direct current stimulation (tDCS) is an emerging neuromodulation technique. The effect of tDCS can vary significantly depending on electrode position and current intensity, making it crucial to find an optimized tDCS montage. However, because of the high computational load, most tDCS optimization approaches have been performed with a limited number of candidates for electrode positions, such as 10-10 or 10-20 international channel configurations. This study introduced the Bayesian optimization and hyperband (BOHB) method to seek optimal tDCS montage for the entire human scalp without conventional constraints.

Materials and methods: The BOHB method is a probabilistic approach that iteratively refines the selection of the optimal montage on the basis of previous results. To determine the suitability of this approach for tDCS simulation, we compared it with random search, which randomly selects montages, and greedy search, which, considers all candidates. Next, the conditions in the greedy search were used as the initial conditions for BOHB for fast learning. The objective function of tDCS optimization was set to maximize the average electric field norm (|E|) in the region of interest (ROI), which is the motor area (M1) and left dorsal lateral prefrontal cortex.

Results: The BOHB method performed better than the conventional random search for the same number of iterations in both ROIs. For M1, the iteration index yielding the maximum evaluation metric in each trial was statistically smaller in the BOHB method than in the random search (p < 0.0001). Regarding the normalized |E|, the BOHB method showed a higher normalized |E| than did the random search for the M1 region.

Conclusions: The BOHB method performed better than did the random search approach. Thus, the BOHB method is feasible for tDCS optimization and can be used as an optimal stimulation montage seeker by fine-tuning some control parameters.

Keywords: Bayesian optimization; hyperband; optimization; stimulation montage; transcranial direct current stimulation.