Temporal basis function models for closed-loop neural stimulation

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Closed-loop neural stimulation shows promise for treating neurological disorders like Parkinson's disease (PD).
• Current AI approaches face challenges in sample efficiency, training time, and latency for real-time brain activity adaptation.
• Tailoring AI for personalized, responsive neural stimulation requires advanced modeling techniques.

Purpose of the Study:

• To introduce Temporal Basis Function Models (TBFMs) as a solution for AI-driven closed-loop neural stimulation.
• To evaluate TBFMs' ability to predict optogenetic stimulation effects on neural activity.
• To address limitations in sample efficiency, training time, and latency for AI in neural stimulation.

Main Methods:

• Developed and applied TBFMs for spatiotemporal forward prediction of optogenetic stimulation effects.
• Utilized TBFMs to analyze local field potentials (LFPs) in non-human primates.
• Assessed model performance against complex nonlinear dynamical systems and linear state-space models.

Main Results:

• TBFMs achieved high prediction accuracy (44% higher than nonlinear, 158% higher than linear models).
• Models demonstrated sample efficiency (<20 min training data) and rapid training (<5 min).
• Simulations showed successful closed-loop control of neural trajectories and optimized stimulation trade-offs (AUC=0.7).

Conclusions:

• TBFMs offer a computationally efficient and rapid AI approach for neural stimulation.
• This method bridges the gap between complex AI models and practical clinical applications.
• Optimized TBFMs pave the way for novel, personalized closed-loop stimulation therapies for neurological diseases.