Real-time adaptive microstimulation increases reliability of electrically evoked cortical potentials

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Cortical neuroprostheses using fixed electrical stimulation parameters face significant trial-by-trial variability in evoked potentials.
• This variability, stemming from ongoing cortical processing, hinders precise neural activity imprinting for prosthetic applications.

Purpose of the Study:

• To investigate real-time adaptation of stimulus intensity to stabilize electrically evoked potentials.
• To explore the use of local field potentials (LFPs) to drive neuroprosthetic stimulation toward a target potential.

Main Methods:

• Utilized local field potentials measured by a microelectrode near the stimulation site.
• Employed machine learning (support vector regression with a custom kernel) to model the relationship between cortical activity, evoked potential, and stimulus intensity.
• Varied stimulus intensities randomly across trials to train the model.

Main Results:

• Real-time adaptation of stimulus intensity successfully drove evoked potentials toward a desired target.
• Minimal deviation from the target potential was achieved using low stimulus intensities.
• The precision of evoked potentials was time-sensitive, diminishing with delays between data acquisition and stimulation.

Conclusions:

• Local field potentials contain adequate information about ongoing cortical processing to stabilize electrically evoked potentials.
• Adaptive low-intensity microstimulation shows promise for enhancing future cortical prosthetic devices aimed at restoring sensory functions.