Multi-Channel Temporal Interference Retinal Stimulation Based on Reinforcement Learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Ophthalmology

Background:

• Retinal degenerative diseases cause vision loss, with current electrical stimulation lacking precision.
• Temporal interference stimulation (TIS) offers a non-invasive approach, but requires optimized parameters for effective retinal targeting.

Purpose of the Study:

• To develop and optimize a reinforcement learning (RL) framework for precise multi-channel electrode parameter optimization in TIS for retinal stimulation.
• To evaluate the focal precision and computational efficiency of the RL-based TIS optimization.

Main Methods:

• A whole-head finite element model with detailed ocular structures was created.
• Reinforcement learning (RL) was employed for multi-channel electrode parameter optimization in TIS.
• The JAX framework was utilized to accelerate envelope calculations for enhanced computational efficiency.

Main Results:

• Focal precision of TIS improved with increased channel numbers across all models.
• RL significantly outperformed genetic algorithms (GA) and unsupervised neural networks (USNN) in focusing capability.
• Optimization time was reduced by nearly an order of magnitude (approx. 2 minutes per run), demonstrating practical feasibility.

Conclusions:

• The study presents a novel, computationally efficient RL methodology for precise non-invasive neuromodulation parameter optimization.
• This approach is highly applicable to retinal diseases and potentially other neurological conditions requiring targeted stimulation.
• Optimized TIS parameters achieved through RL offer a promising advancement for vision restoration therapies.