Reinforcement Learning-Based Focality Optimization for Multi-Electrode Temporal Interference Stimulation.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Modeling

Background:

• Noninvasive deep brain stimulation using Multi-electrode Temporal Interference Stimulation (TIS) shows promise.
• Clinical translation is limited by complex parameter optimization challenges.
• Existing optimization algorithms face computational cost and flexibility issues.

Purpose of the Study:

• To introduce and validate a novel framework for optimizing Temporal Interference Stimulation (TIS).
• To address the high-dimensional and hybrid parameter space limitations of current TIS methods.
• To enhance the precision and clinical robustness of noninvasive neuromodulation.

Main Methods:

• Developed a reinforcement learning (RL) framework for simultaneous optimization of electrode positions and intensities.
• Evaluated the RL framework using six finite element head models.
• Benchmarked the RL framework against genetic algorithms (GA) and unsupervised neural networks (USNN).

Main Results:

• The RL-based approach significantly improved stimulation focality compared to GA.
• RL performance was comparable to USNN, with added explicit control over active electrodes.
• Stimulation focality increased with up to 16 electrodes, showing diminishing returns thereafter.

Conclusions:

• The developed RL framework offers a powerful and flexible computational paradigm for TIS optimization.
• This framework provides guidelines for optimal system design, overcoming previous limitations.
• The approach accelerates the clinical translation of TIS for precise noninvasive neuromodulation.