Optimizing Interferential Stimulation Focality with Multi-Electrode Based on Reinforcement Learning
Poster No:
18
Submission Type:
Abstract Submission
Authors:
Wennan Chan1, Sheng Hu1, Yingqiang Meng1, Runze Liu1, Muhammad Mohsin Pathan1, Xiaoxiao Wang1, Bensheng Qiu1, Yanming Wang1
Institutions:
1University of Science and Technology of China, Hefei, Anhui
First Author:
Co-Author(s):
Introduction:
Temporal Interference Stimulation (TIS) is an emerging technique for deep brain stimulation (Grossman et al., 2017), with numerous studies demonstrating its effectiveness in targeting deep brain regions (Violante et al., 2023 , Wessel et al., 2023). However, achieving precise stimulation with traditional two-pair TIS is challenging due to the complexity of brain anatomy. Multi-electrode TIS offers the potential to improve focality, but its nonlinear nature complicates the optimization process. Previous approaches for optimizing multi-electrode TIS have several limitations, including being time-consuming (Huang et al., 2020) and lacking control over the number of electrodes used (Bahn et al., 2022). Reinforcement learning (RL), which excels at solving complex, nonlinear problems, shows great promise for optimizing electrode configurations in multi-electrode TIS. This study proposes an improved RL paradigm to enhance the focality of multi-electrode TIS.
Methods:
The RL was applied to optimize electrode positions and current parameters. However, conventional RL typically employed either continuous or discrete output parameters, making it difficult to simultaneously optimize both electrode positions and current parameters. We proposed an improved RL to overcome this challenge. The nonlinear and non-convex envelope calculation was embedded into the environment and reward. The agent network generated both discrete and continuous parameters, with its core functionality centered on the connection between input data, electrode positions, and current intensity. To improve learning efficiency, the agent was updated using a policy that prioritized the first few sets of solutions in each iteration. These configurations enabled optimization by comparing the envelope field amplitude between target and non-target regions. The focality of the proposed RL was evaluated across varying numbers of electrodes in the right globus pallidus internus (rGPi) using six realistic finite element head models derived from the HCP dataset. Additionally, to evaluate the superiority of the proposed RL, the performance of 16-electrode RL was compared with that of traditional two-pair genetic algorithms (GA) and multi-electrode unsupervised neural networks (USNN).
Results:
As the number of electrodes increased, the peak-sum ratio (PSR) exhibited an upward trend (Fig. 1a). Notably, the focality improvement was substantial from 4 (mean PSR = 1.04) to 16 (mean PSR = 1.35) electrodes, but minimal from 16 to 32 (mean PSR = 1.37) electrodes. Significant differences were observed between 4 and 8, 8 and 12, and 12 and 16 electrodes (Fig. 1b). As the number of electrodes increased from 4 to 16, the envelope amplitude in the non-target region decreased. However, the approximate envelope distributions were observed between 16, 24 and 32 electrodes. The envelope field distributions for 4, 8, 12, 16, 24 and 32 electrodes were shown in Fig. 1c. In addition, RL outperformed GA, with the PSR value of the RL (mean PSR = 1.35) much higher than that of GA (mean PSR = 1.04) (Fig. 2a). Meanwhile, the performance of RL was virtually identical to that of USNN (mean PSR = 1.33). However, USNN optimized only the electrode currents while using all electrode positions. The optimization results for a representative subject were illustrated, with the optimal electrode montage of the proposed RL displayed in Fig. 2b. The envelope field distributions of RL, GA and USNN were illustrated in Fig. 2c-e. The envelope amplitude of RL (Fig. 2c) was lower than that of GA (Fig. 2d), but similar to USNN (Fig. 2e) in the non-target region.
·Figure1. Comparison of TIS performance with different number of electrodes using the proposed RL on rGPi.
·Figure 2. Comparison of TIS performance between the proposed RL，GA, and USNN on rGPi.
Conclusions:
These findings indicate that the 16-electrode TIS using the proposed RL achieves significantly better focality than the traditional two-pair TIS, while also being more efficient than USNN by requiring fewer electrodes to achieve similar performance. This paradigm not only proves more practical but also delivers faster optimization compared to other multi-electrode TIS approaches.
Brain Stimulation:
Non-invasive Electrical/tDCS/tACS/tRNS 1
Non-Invasive Stimulation Methods Other 2
Motor Behavior:
Brain Machine Interface
Keywords:
Computing
Machine Learning
Physical Therapy
Other - Neuromodulation, Temporal Interference Stimulation, Reinforcement Learning
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
GROSSMAN N, BONO D, DEDIC N, et al. 2017. Noninvasive Deep Brain Stimulation via Temporally Interfering Electric Fields. Cell [J], 169: 1029-1041.
HUANG Y, DATTA A, PARRA L C 2020. Optimization of interferential stimulation of the human brain with electrode arrays. Journal of Neural Engineering [J], 17: 036023.
VIOLANTE I R, ALANIA K, CASSARà A M, et al. 2023. Non-invasive temporal interference electrical stimulation of the human hippocampus. Nature Neuroscience [J], 26: 1994-2004.
WESSEL M J, BEANATO E, POPA T, et al. 2023. Noninvasive theta-burst stimulation of the human striatum enhances striatal activity and motor skill learning. Nature Neuroscience [J].