Deep learning-based control of electrically evoked activity in human visual cortex

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Current visual cortical prostheses provide limited visual perception due to crude, variable responses from electrical microstimulation.
• The nonlinear and state-dependent nature of neural population activity in the human visual cortex complicates the relationship between stimulation and perception.
• Manual electrode calibration is time-consuming and does not scale effectively for complex visual prostheses.

Purpose of the Study:

• To develop and validate a deep learning framework for precise control of stimulation-evoked population activity in the human visual cortex.
• To improve the quality and consistency of percepts generated by visual cortical prostheses.
• To establish a population-level understanding linking microstimulation, neural activity, and perception.

Main Methods:

• A bidirectional cortical implant was used to record neural activity and deliver electrical stimulation.
• A deep learning framework was trained on trial-resolved neural recordings to shape population responses.
• Two control strategies were implemented: a learned inverse network for real-time stimulation synthesis and a gradient-based optimizer for targeting.

Main Results:

• The deep learning framework significantly outperformed conventional methods in controlling neural activity and eliciting perception.
• The system achieved target neural responses at lower stimulation currents, leading to more consistent percepts.
• Recorded population activity was a better predictor of reported percepts than stimulation parameters alone.

Conclusions:

• Deep learning offers a powerful approach to causally shape neural population activity for improved visual cortical prosthesis function.
• The developed framework provides a foundation for linking microstimulation, neural dynamics, and subjective perception in the human visual system.
• This work paves the way for more sophisticated and effective visual restoration technologies.